1

INTERNATIONAL RESEARCH JOURNAL OF FINANCE AND ECONOMICS ISSN: 1450-2887
Issue 11 September, 2007
INTERNATIONAL RESEARCH JOURNAL OF FINANCE AND ECONOMICS

http://www.eurojournals.com/finance.htm Editor-In-Chief Mete Feridun, Cyprus International University Co-Editors Chiaku Chukwuogor-Ndu, Eastern Connecticut State University Yu Hsing, Southeastern Louisiana University Emmanuel Anoruo, Coppin State University Editorial Advisory Board Zeljko Bogetic, The World Bank Jwyang Jiawen Yang, The George Washington University Christos Giannikos, Columbia University Zhihong Shi, State University of New York Jan Dutta, Rutgers University Hector Lozada, Seton Hall University Jatin Pancholi, Middlesex University Leo V. Ryan, DePaul University Richard Omoteye, Virginia State University Dimitrios Tsagdis, The University of Hull Necla Kucukcolak, Istanbul Stock Exchange Constantinos Vorlow, University of Durham Jerry Kolo, Florida Atlantic University Felix Ayadi, Texas Southern University Robert Van Ness, University of Mississippi Athanasios Koulakiotis,University of the Aegean Richard J. Hunter, Seton Hall University Bansi Sawhney, University of Baltimore Jean-Luc Grosso, Eastern Connecticut State University Nicholas Papasyriopoulos, University of Macedonia John Mylonakis, Tutor - Hellenic Open University Anton Miglo, University of Guelph Dov Mishor, Ben-Gurion University George Skoulas, University of Macedonia Muhammad Abdus Salam, State Bank of Pakistan David Wang, Hsuan Chuang University Cornelis A. Los, Kazakh-British Technical University Wen-jen Hsieh, National Cheng Kung University Stylianos Karagiannis, Center of Planning and Economic Research M. Carmen Guisan, University of Santiago de Compostela H. Young Baek, Nova Southeastern University Gregorio Gimnez Esteban, Universidad de Zaragoza Mukhopadhyay Bappaditya, Management Development Institute Ebere Oriaku, Elizabeth City State University Neven Borak, Securities Market Agency Maria Elena Garcia-Ruiz, University of Cantabria
Ghadir Mahdavi, Kyoto University Emanuele Bajo, University of Bologna Wassim Shahin, Lebanese American University Neil Reid, University of Toledo Zulkarnain Muhamad Sori, University Putra Malaysia Christopher Balding, University of California, Irvine Lucian Cernat, UNCTAD Carlos Machado-Santos, UTAD University Portugal Aims and Scope International Research Journal of Finance and Economics is a bi-monthly, peer-reviewed international research journal. The journal aims to publish articles of high quality dealing with issues in international finance and economics which impact on national and global economies. While maintaining the high standards of a fully-refereed academic journal with technical, empirical and theoretical material, the journal is also accessible to non-specialists, policy-makers and practitioners. International Research Journal of Finance and Economics will concern itself with issues such as exchange rates, balance of payments, financial institutions, corporate finance, risk analysis, international banking and portfolio management, financial market regulation, Third World debt, economic development, European Monetary Union, the financial aspects of transition economies, financial instruments and international financial policy co-ordination. The journal is aimed at publishing high quality articles and short research notes particularly in: International economics International finance Financial economics International political economy Financial analysis Corporate finance Exchange rate modelling Forecasting financial markets Economic policy and financial markets Monetary and fiscal policy decision-making Portfolio and asset management Pricing and risk of financial instruments Advances in financial econometrics and statistics Public finance decision-making
Editorial Policies 1) The journal realizes the meaning of fast publication to researchers, particularly to those working in competitive & dynamic fields. Hence, it offers an exceptionally fast publication schedule including prompt peer-review by the experts in the field and immediate publication upon acceptance. It is the major editorial policy to review the submitted articles as fast as possible and promptly include them in the forthcoming issues should they pass the evaluation process. 2) All research and reviews published in the journal have been fully peer-reviewed by two, and in some cases, three internal or external reviewers. Unless they are out of scope for the journal, or are of an unacceptably low standard of presentation, submitted articles will be sent to peer reviewers. They will generally be reviewed by two experts with the aim of reaching a first decision within a two-month
period. Suggested reviewers will be considered alongside potential reviewers identified by their publication record or recommended by Editorial Board members. Reviewers are asked whether the manuscript is scientifically sound and coherent, how interesting it is and whether the quality of the writing is acceptable. Where possible, the final decision is made on the basis that the peer reviewers are in accordance with one another, or that at least there is no strong dissenting view. 3) In cases where there is strong disagreement either among peer reviewers or between the authors and peer reviewers, advice is sought from a member of the journal's Editorial Board. The journal allows a maximum of three revisions of any manuscripts. The ultimate responsibility for any decision lies with the Editor-in-Chief. Reviewers are also asked to indicate which articles they consider to be especially interesting or significant. These articles may be given greater prominence and greater external publicity. 4) Any manuscript submitted to the journals must not already have been published in another journal or be under consideration by any other journal. Manuscripts that are derived from papers presented at conferences can be submitted even if they have been published as part of the conference proceedings in a peer reviewed journal. Authors are required to ensure that no material submitted as part of a manuscript infringes existing copyrights, or the rights of a third party. Contributing authors retain copyright to their work. 5) The journal makes all published original research immediately accessible through www.EuroJournals.com without subscription charges or registration barriers. Through its open access policy, the journal is committed permanently to maintaining this policy. This process is streamlined thanks to a user-friendly, web-based system for submission and for referees to view manuscripts and return their reviews. The journal does not have page charges, color figure charges or submission fees. However, there is an article-processing and publication fee payable only if the article is accepted for publication. Submissions The International Research Journal of Finance and Economics is an international journal that aims at dissemination and advancement of research in all areas of economics and finance. All papers are subjected to a blind peer-review process. Manuscripts are invited from academicians, practitioners and businesspersons for publication consideration. The journal welcomes submissions in all areas related to economics and finance. Each manuscript must include a 200 word abstract. Authors should list their contact information on a separate paper and suggest JEL classification code for their articles. Electronic submissions are acceptable. Although the journal intends to have an empirical bias, it also welcomes theoretical and descriptive studies. Articles for consideration are to be directed to the editor through finance@eurojournals.com. In the subject line of your e-mail please writeIRJFE submission. All correspondence with the journal should be directed to the Editor-In-Chief at: Mete Feridun Faculty of Economics and Administrative Sciences Cyprus International University Nicosia, Cyprus Tel: +90 392 671 1111 (Ext. 2202) / Fax: +90 392 671 1122 The International Research Journal of Finance and Economics is published in the United States of America at Lulu Press, Inc (Morrisville, North Carolina) by EuroJournals, Inc.
International Research Journal of Finance and Economics

Issue 11 September, 2007
Contents
On the Effect of Index Futures Trading on Stock Market Volatility Panayiotis Alexakis A Birds Eye View of the Dividend Policy of the Banking Industry in Greece Nikolaos Eriotis, Dimitrios Vasiliou and Vasileios Zisis Finance Growth Nexus: Evidence from Turkey Ali Acaravci, Ilhan Ozturk and Songul Kakilli Acaravci Volatility Clustering in the Greek Futures Market: Curse or Blessing? Nikolaos L. Hourvouliades The Volatility of the Stock Market and News Rohitha Goonatilake and Susantha Herath Searching for a Scholarly Visibility: The Case of Ukraine Jean Mirucki The Effect of Model-Selection Uncertainty on Autoregressive Models Estimates Islam Azzam Foreign Aid to Developing Countries: Does it Crowd-out the Recipient Countries Domestic Savings? Munther Nushiwat Foreign Ownership and Firm Performance: Evidence from Turkey Nurhan Aydin, Mustafa Sayim and Abdullah Yalama Feldstein-Horioka Puzzle in Latin American and Caribbean Countries: Evidence from Likelihood-Based Cointegration Tests in Heterogeneous Panels N. R. Vasudeva Murthy Does Inflation Have an Impact on Conditional Stock Market Volatility? : Evidence from Turkey and Canada Fatma Sonmez Saryal Impacts of Higher Crude Oil Prices and Changing Macroeconomic Conditions on Output Growth in Germany Yu Hsing The Regulation of the Credit Card Market in Turkey Ahmet Faruk Aysan and Lerzan Yldz 7-20 21-29 30-40 41-52 53-65 66-78 80-92
94-101 103-110
112-121
123-132
134-139 141-153
Non-Linear Dynamics and Chaos: "The Case of the Price Indicator at the Athens Stock Exchange" Mike P. Hanias, Panayiotis G. Curtis and John E. Thalassinos Does Free Cash Flow Anomaly Exist in an Emerging Market? Evidence from the Istanbul Stock Exchange zgr Arslan and Mehmet Baha Karan Export Diversification in the Export Led Growth Process of Turkey Filiz Ozag and Suleyman Degirmen RASM: A Risk-Based Projects Auditing Selection Methodology for Large Scale Programs Othonas Zacharias, John Mylonakis and Dimitrios Th. Askounis
155-163
163-170 172-178 180-193
International Research Journal of Finance and Economics ISSN 1450-2887 Issue 11 (2007) EuroJournals Publishing, Inc. 2007 http://www.eurojournals.com/finance.htm
On the Effect of Index Futures Trading on Stock Market Volatility

Panayiotis Alexakis Department of Economics, University of Athens 5, Stadiou str., Athens 105 62, Greece E-mail: paleks@econ.uoa.gr Tel.: +30 210 3689400, +30 210 6936698346 Abstract The purpose of this article is to investigate the effect of the introduction of stock index futures on the volatility of the spot equity market and contribute in this way to the contrasting arguments with respect to the stability and destabilising effects of such products. To test the impact of the introduction of stock futures contracts, a GARCH model is modified along the lines of the GJR-GARCH model, especially to take into account the link between information and volatility. The statistical results indicate that the index of futures trading is fully consistent with efficient market operation as it exerts a stabilising effect in the spot market, reducing volatility asymmetries and improves the quality and speed of the flow of information. Empirical research can be further expanded by selecting and analysing high frequency intraday data and the inclusion of additional economic variables in the conditional variance equation. In terms of policy implications, the introduction of derivatives trading did not lead to detrimental effects on the underlying market, but instead an improvement in the transmission of news into prices. These results can assist investors and regulators, as index futures lead to new channels of information, and reduce the number of uninformed investors, and render more efficient the market. This paper provides evidence, analysing, for first time, a new derivatives market, that index futures introduction is consistent with efficient stock market operations, by recognising more adequately the link between information and volatility, in getting more appropriate conclusions. Keywords: Index futures, spot market, volatility, information, futures markets Jel Classification Codes: G14 G32
1. Introduction
The introduction of trading of stock index futures contracts has received considerable attention, while at the same time it has led to a controversy over the impact of index futures trading on the underlying stock market volatility (Brady, 1988). These become more profound during serious or mild stock market crises such as those of 1987, 1989 or even during prolonged adverse stock market conditions, such as those experienced in the 2001-2003 period. Furthermore, any increase in stock market volatility that has followed the onset of futures trading has generally been taken as justifying the
International Research Journal of Finance and Economics - Issue 11 (2007)
traditional view that the introduction of futures markets induces destabilizing speculation, leading to calls for greater regulation to minimise any detrimental effects. An alternative view is that futures markets provide an additional route by which information can be transmitted, and, therefore, increased spot market volatility may simply be a consequence of the more frequent arrival, and more rapid processing of information. Thus, futures trading may be fully consistent with efficiently functioning markets. In spite of the substantial amount of studies undertaken, applying models to stock indices, mainly, individual stocks and portfolios, in various capital markets, there continues to be little agreement on this matter, [Aggarawal (1988), Edwards (1988a, b), Harris (1989) Smith (1989), Conrad (1989), Schwert (1990), Schwert and Senguin (1990), Lamoureux and Lastrapes (1990), Brorsen (1991), Baldauf and Santoni (1991), Chan and Karolyi (1991), Lee and Ohk (1992), Bessembinder and Seguin (1992), Kim and Kon (1994), Bacha and Vila (1994), Harris, Sofianos and Shapiro (1994), Robinson (1994), Darrat and Rahman (1995), Brown-Hruska and Kuserk (1995), Foster (1995), Antoniou and Holmes (1995), Reyes (1996), Hogan, Kroner and Sultan (1997), Pericli and Koutmos (1997), Antoniou, Holmes and Priestley (1998), Chang, Cheng and Pinegar (1999), Kyriacou and Sarno (1999), Dennis and Sim (1999), Gulen and Mayhew (2000), Board, Sandmann and Sutcliffe (2001), Rahman (2001), McKenzie, Brailsford and Faff (2001), Bae, Taek and Park (2004) and Charupat (2006)]. The findings of these studies range from the one side suggesting that futures market trading activity has no significant effect on changes in spot market price volatility, or that it does not lead to an increase in spot price volatility or that it even leads to a decrease in underlying index or stock volatility, to the other side indicating that it has a significant positive effect on underlying index and spot market volatility. Despite the differences of opinion, the evidence presented seems to gradually accumulate on the side of reduction or no change in stock market volatility along with the introduction of index futures trading. At the same time, it should be mentioned that the effects of derivatives trading on volatility are practically an empirical question. The results vary depending on the indices as well as on the testing methodologies. Model specification and the construction of the activity variables are also important. All these affect the results of the empirical analysis, their interpretation and conclusions. Furthermore, this study maintains that previous studies seem to have failed to recognise adequately the link between information and volatility, resulting in inappropriate policy conclusions. It extends the relevant empirical literature in the following ways: First, most of the studies view the question about the impact of derivatives trading on spot price volatility from a stabilising or destabilising view-point by comparing spot price volatility during the pre- and postderivatives trading areas, neglecting the investigation of the link between information and volatility. Second, the conditional variance from the Glosten, Jagannathan, and Runkle (1993) GJR-GARCH model is found to be the appropriate process of volatility of the spot prices, enabling the investigation of the link between information and conditional volatility and of the market dynamics, as reflected by a change in the asymmetric volatility response. Third, if a stabilising/destabilising impact is found, we investigate whether the introduction of stock index futures trading is the only cause for a change in the spot market volatility or whether other factors may have affected market volatility as well. For this purpose, in the variance model, several other economic indicators are included as proxies for market factors. The rest of the paper is structured as follows: As this study is undertaken for the case of the recently (1999) created derivatives market of Athens Exchange (ATHEX), for first time, Section II refers to the presentation of the derivatives market of ATHEX, while Section III deals with the methodology for the analysis undertaken. Section IV covers the description of the data and preliminary statistics and Section V presents the empirical analysis and results. Finally, Section VI draws the conclusions and policy implications.
2. The Derivatives Market of the Athens Exchange

The operation of the organised derivatives market in Greece rests with the ATHEX and the Athens Derivatives Exchange Clearing House (ADECH). ATHEX is responsible for the organisation and support of the derivatives market and for the supervision of trading, as well as for its development.
ADECH is responsible for the recording and clearing of the trades, as it acts as the central counter party in all trades concluded in ATHEX. Transactions are conducted electronically (screen trading) via the Integrated Electronic Trading System (OASIS). The first product of ATHEX was the FTSE/ASE20 futures contract with the underlying asset being the blue chip FTSE/ASE-20 stock index. The FTSE/ASE-20 index was created in September 1997 by FTSE International and the Athens Exchange (ATHEX), and is based on 20 highly capitalised and liquid companies listed on the ATHEX. The contracts are cash settled in the sense that the difference between the traded price of the contract and the closing price of the relevant index, on the expiration day of the contract, is settled between the counter-parties in cash. The futures contracts are traded in index points, while the monetary value of the contracts is calculated by multiplying the futures price by the multiplier (5 EUR per point for the FTSE/ASE-20). The tick size of the FTSE/ASE-20 futures is 0.25 points, equivalent to 1.25 EUR. Following this product, other products were gradually introduced, such as options contracts on these two indices, options contracts on stocks and futures contracts on bonds and foreign exchange. However, the FTSE/ASE-20 futures remained the most liquid product of the market, capturing almost 50% of the total annual, approximately, 10 million contracts. This study examines to what extent the introduction of trading in stock index futures contracts in ATHEX, since August 1999, has impacted on the price volatility of the underlying spot market. If a large number of market agents are not hedging themselves against adverse stock price movements, but aims to speculate, using the stock index futures market, their actions may induce excess volatility and therefore destabilise the spot market. It is of interest for the authorities and the participants of the Greek capital market to examine if such a stabilizing/destabilizing effect exists in the newly created derivatives market of the ATHEX. If the stock index futures market causes a change in the level of volatility in the spot market and this, in turn, is associated with greater uncertainty and unduly higher required rates, then there may well be a case for the Hellenic Capital Market Commission to increase the regulation of this market. However, if this market leads to new channels of information being provided, more information due to more traders, then the stock index futures market provides a useful service and calls for its regulation are unwarranted. Alexakis et al. (2002) examine the price discovery function in daily returns and volatilities between price movements of stock index futures and the underlying cash index in ATHEX. They argue that futures lead the cash index returns, by responding more rapidly to economic events than stock prices. Thus, new market information may be disseminated faster in the futures market compared to the stock market, and that futures volatility spills some information over to the cash market volatility. This may imply that the futures markets can be used as price discovery vehicles. Moreover, Alexakis, Kavussanos and Visvikis (2003) examining the hedging effectiveness of the stock index futures market argue that the futures contracts serve their risk management function through hedging, as they provide considerably high variance reductions, which are in accordance with the evidence in the literature. Thus, the investors of the Greek cash market can benefit from this result by developing appropriate hedge ratios in each market, in order to reduce their price risk more efficiently, since the variability of their cash flows can be explained by the fluctuations of the futures prices.
3. Methodology and Theoretical Approach

From the view of market efficiency it is reasonable to argue that if derivatives trading does increase the amount of information available, then spot price volatility may increase. Cox (1976) implies that information becomes available earlier, and not necessarily that information becomes available which would not otherwise. Thus, the rate of information flow increases, as does the rate at which the information is impounded into prices. Hence, volatility of prices may increase. In the context of the impact of derivatives markets on spot market volatility, if derivatives trading do increase the rate of flow of information, then spot prices may exhibit increased volatility. Thus, it is important to use a model to take into account the link between information and volatility and of possible asymmetric responses to news, so as to be seen whether the introduction of stock index futures trading has
10
increased or decreased spot price volatility and to investigate the extent to which the introduction of stock index futures contracts also affected the nature of volatility in the underlying spot market. The ARCH model of Engle (1982) and the GARCH model of Bollerslev (1986) can capture time variation in return distributions. More extended studies consider the impact of the introduction of derivatives contracts on how the market responds to news. According to Antoniou et al. (1998), market dynamics related to the transmission of news may be responsible for asymmetries in the volatility response mechanism. Thus, to test the impact of the introduction of stock index futures contracts, a GARCH model is modified along the lines of the GJR-GARCH model of Glosten et al. (1993). Thus, the mean equation of the GJR-GARCH process can be defined as follows: St = 0 +
iSt-i + t; t ~ iid(0,ht)
i= 1
p 1
(1)
where, St is the natural logarithm of the daily spot price, is the first-difference operator and t are the residuals that follow a normal conditional distribution with mean zero and time-varying covariance, ht. The conditional variance of the process can be specified as follows: ht = a0 + a1ht-1 + a2D1ht-1 + 1 t21 + 2D1 t21 + 1 t21 D + 2D1 (2) where, D1 is a dummy variable that takes on a value of unity after the introduction of stock index futures contracts and D t1 is a dummy variable that takes on a value of unity if the error is negative and zero otherwise. The specification of the conditional variance in equation (2) allows the examination of the impact of stock index futures trading to the unconditional volatility of the spot market through the 2 coefficient. A significant positive (negative) 2 coefficient indicates increased (decreased) spot price unconditional volatility in the post- stock index futures period. The specification of equation (2) allows the investigation of whether stock index futures trading has changed the role of market dynamics in terms of the way in which volatility is transmitted, and therefore, inferences can be made on how information is incorporated into prices. When the coefficient on D t1 is equal to zero, the model of equation (2) is the symmetric GARCH model. A negative shock (D t1 = 1) can generate an asymmetric response. However, to address the issue of the relationship between information and volatility, and not simply investigate whether stock index futures trading has led to an increase or decrease in volatility in the spot market, the period under investigation is partitioned into two sub-periods relating to before and after stock index futures trading began. GJR-GARCH models of equation (2) are estimated for both sub-periods, without the D1 dummy variable for the existence of stock index futures trading: (3) ht = a0 + a1ht-1 + 1 t21 + 1 t21 D t1 where D t1 = 1 if t 1 < 0, D t1 = 0 otherwise. Comparisons can then be made on the estimated coefficients, in order to examine the impact of stock index futures trading on the nature of spot volatility and to assess if stock index futures trading has led to changes in the asymmetric response of volatility. It is possible that factors, other than the introduction of futures contracts, may affect the variables considered in each of the hypotheses tests. To this end a control procedure is implemented under which we augment the conditional variances of the spot stock indices by incorporating the conditional variances of other economic indicators [S&P 500 Index (S&P), FTSE100 Index (FTSE), XETRA DAX Index (DAX) and Dow Jones International Index (DJ)]. Thus, the augmented variance model is the following: ht = a0 + a1ht-1 + a2D1ht-1 + 1 t21 + 2D1 t21 + 1 t21 D t1 + 2D1 + 1Gt (4) where, Gt is the conditional variance, from a GARCH model, of an economic variable. A significant 1 coefficient indicates that the conditional variance of the economic variable affects the conditional variance of the spot stock index prices.
11
4. Description of Data and Preliminary Statistics

The impact of stock index futures trading, on the volatility of the underlying FTSE/ASE-20 stock index market in ATHEX, is investigated by estimating a model for a period which covers the time before and after the introduction of stock index futures contracts. In the ensuing analysis, August 1999 will be the threshold point that separates pre- and post- stock index futures trading in order for robust inferences to be made. The data set comprises daily closing observations of the spot stock index rates for aforementioned market. It covers the periods 23 September 1997 to 07 June 2004. The particular period and set of data for the empirical investigation are chosen in order to give emphasis on the examination and contribution of the effects of a newly created market on the completion and efficiency of the Athens Exchange market, in total. Such a result could also prove useful for investors in this market, in particular, as since mid-2001 the number of foreign institutional investors started exerting a dominating role in this market. FTSE/ASE-20 spot stock index prices are obtained from the ATHEX and daily prices for the S&P, FTSE and DAX and DJ indices are obtained from Datastream. All prices are transformed to natural logarithms. The descriptive statistics of logarithmic first-differences of the daily FTSE/ASE-20 spot stock index prices are reported in Appendix A, Table 1, which is divided into three periods. The first period (panel A) corresponds to the whole period of the analysis. The second (panel B) and third (panel C) correspond to the pre- and post-stock index futures periods, respectively. The results indicate excess skewness and kurtosis in the price series, departures from normality for the spot prices, while all variables are log first-difference stationary, all having a unit root on their log-levels representation. The figures obtained for the standard deviation estimates, provide an initial view of volatility, where, by comparing the two periods (pre- and post-derivatives), it seems that volatility has decreased over time. Wherever, this decrease is significant, it is a matter of the ensuing empirical research.
Table 1: Descriptive Statistics of Logarithmic First-Differences of Spot Stock Index Prices
SD 0.0187 SD 0.0341 SD 0.0245 Skew 0.107 Skew -0.104 Skew 0.234 Kurt 4.982 Kurt 3.799 Kurt 6.442 Panel A: Spot Prices; Whole Period J-B ADF (lags) Lev PP(8) Lev ADF (lags) 1st Diffs 342.45 -1.167 (1) -1.222 -34.112 (0) Panel B: Spot Prices; Pre-Futures Period J-B ADF (lags) Lev PP(8) Lev ADF (lags) 1st Diffs 15.324 -0.456 (1) -0.564 -17.684 (0) Panel C: Spot Prices; Post-Futures Period J-B ADF (lags) Lev PP(8) Lev ADF (lags) 1st Diffs 525.21 -1.232 (1) -1.021 -28.385 (0) PP(8) 1st Diffs -33.367 PP(8) 1st Diffs -17.921 PP(8) 1st Diffs -28.378
FTSE-20 FTSE-20 FTSE-20

Notes:
All series are measured in logarithmic first-differences. Figures in parentheses (.) indicate t-statistics.Skew and Kurt are the estimated centralised third and fourth moments of the data. J-B is the Jarque-Bera (1980) test for normality. ADF is the Augmented Dickey Fuller (1981) test. PP is the Phillips and Perron (1988) test. Lev and 1st Diffs correspond to price series in log-levels and log first-differences, respectively. The 5% critical value for the ADF (1981) and PP (1988) tests is 2.89.
Similarly to the authors of earlier studies in this area, we initially conduct equality of variance tests, shown in Table 2. The results, in panel A, reveal significant differences between the pre- and post-futures variances. Comparing the pre- and post-futures variances of the various economic indicators, in panel B, it is indicated that the level of the variances has changed between the two periods. This may indicate that besides the introduction of stock index futures contracts, there might be several other economic events that contributed to this change.

Table 2: Equality of Variance Tests for Pre- and Post- Stock Index Futures Trading
Panel A: Spot Stock Index Prices Bartlett Levene 8.421 [0.000] 26.345 [0.000] Panel B: Economic Indicators F-test Bartlett Levene 271.01 [0.000] 357.19 [0.000] 245.75 [0.000] 89.025 [0.000] 121.36 [0.000] 91.937 [0.000] 8.392 [0.000] 9.583 [0.000] 8.362[0.000] 24.203 [0.000] 23.164 [0.000] 21.472 [0.000] F-test 1.189 [0.000]
12
FTSE-20 S&P FTSE DAX DJ

Notes:
Brown-Forsythe 25.121 [0.000] Brown-Forsythe 312.38 [0.000] 84.472 [0.000] 8.237 [0.000] 22.456 [0.000]
The F-test has a F-distribution with nL 1 numerator degrees of freedom and nS - 1 denominator degrees of freedom. The Bartlett test compares the logarithm of the weighted average variance with the weighted sum of the logarithms of the variances. The Levene test is based on an Analysis of Variance (ANOVA) of the absolute difference from the mean. The Brown-Forsythe test is a modification of the Levene test in which the absolute mean difference is replaced with the absolute median difference. S&P is the S&P 500 Index; FTSE is the FTSE100 Index; DAX is the XETRA DAX index, and DJ is the Dow Jones International index.
Table 3 reports the Ljung-Box (1976) statistics, where the results indicate significant linear and non-linear temporal dependencies in the adjusted residual series and squared adjusted residual series, respectively. It is clear that these series cannot be treated as news as there is evidence of serialcorrelation. Therefore, in Table 4, autoregressive models are estimated in order to remove any linear temporal predictability. The results indicate that the autoregressive models capture the serialcorrelation in the adjusted residual series, suggesting that the adjustment procedure removes the predictable part of the return series.
Table 3: Ljung-Box Statistics for Linear and Non-Linear Temporal Dependences in Regression Models
Q(1) 42.128 [0.000] Q(1) 15.483 [0.000] Q(1) 19.487 [0.000] Panel A: Whole Period Q(12) Q2(1) 50.484 [0.000] 63.494 [0.000] Panel B: Pre-Futures Period Q(12) Q2(1) 24.202 [0.000] 21.590 [0.000] Panel C: Post-Futures Period Q(12) Q2(1) 30.494 [0.000] 19.484 [0.000] Q2(12) 254.33 [0.000] Q2(12) 50.303 [0.000] Q2(12) 164.29 [0.000]

Notes:
Figures in squared brackets [.] indicate exact significance levels. Q(L) and Q2(L) are the Ljung-Box (1978) Q statistics on the first L lags of the sample autocorrelation function of the series and of the squared series; these tests are distributed as 2(L).
Table 4:
Ljung-Box Statistics for Linear and Non-Linear Temporal Dependences in Autoregressive Models
Q(1) 1.023 [0.930] Q(1) 1.292 [0.236] Q(1) 2.202 [0.081] Panel A: Whole Period Q(12) Q2(1) 5.292 [0.943] 120.13 [0.000] Panel B: Pre-Futures Period Q(12) Q2(1) 5.494 [0.842] 22.292 [0.000] Panel C: Post-Futures Period Q(12) Q2(1) 12.393 [0.258] 130.13 [0.000] Q2(12) 254.93 [0.000] Q2(12) 45.202 [0.000] Q2(12) 160.20 [0.000]

Notes:
Figures in squared brackets [.] indicate exact significance levels. Q(L) and Q2(L) are the Ljung-Box (1978) Q statistics on the first L lags of the sample autocorrelation function of the series and of the squared series; these tests are distributed as 2(L). The order of the most parsimonious autoregressive models are in parentheses (.).
5. Empirical Results
To assess whether there has been a change in volatility after the inception of FTSE/ASE-20 futures trading, a GJR-GARCH(1,1) model of conditional volatility is estimated. A dummy variable that takes the value of 0 pre-futures and 1 post-futures is included. The estimates of the GJR-GARCH model of the spot stock index prices for the whole period of the analysis are presented in Table 5. The diagnostic
13
tests, on the standardised residuals and squared standardised residuals, indicate absence of linear and non-linear dependencies, respectively. Thus, the estimated model fits the data very well. The results in Table 5 indicate that stock index futures trading has had a negative impact, that is stabilising effect, on the level of price volatility of the underlying spot market (2 coefficient). The results of the Wald tests for the null hypothesis indicate that the relevant coefficients in the variance equation have significantly changed. The results of the 1 coefficient of the asymmetric effects suggest statistically significant asymmetric effects.
Table 5: GJR-GARCH Model Estimates of the Effect of FTSE/ASE-20 Futures Trading on Spot Market Volatility (Whole Period)
Panel A: Coefficient Estimates
FTSE/ASE-20 - AR(1) (23/09/97 07/06/04) Mean Equation 0 1 Variance Equation a0 a1 a2 1 2 1 2 Panel B: Residual Diagnostic LL Skewness Kurtosis J-B Q(24) Q2(24) ARCH(12) a2 = 2 = 0 A2 = 2 = 2 = 0 5.265.23 0.249 [0.005] 1.483 [0.000] 150.210 [0.000] 33.157 [0.110] 21.027 [0.532] 0.501 [0.849] 2.658 [0.348] 20.494 [0.000] 3.24E-05* (0.002) 0.167* (5.483) 1.42E-04* (2.678) 0.483* (5.002) 0.113 (0.926) 0.109** (1.737) 0.047 (0.440) 0.210* (4.829) -1.10E-04* (-2.120)
Notes:
Figures in parentheses (.) and in squared brackets [.] indicate t-statistics and exact significance levels, respectively. * and ** denote significance at the 5% and 10% levels, respectively. J-B is the Jarque-Bera (1980) normality test. Q(24) and Q2(24) are the Ljung-Box (1978) tests for 24th order serial correlation and Heteroskedasticity. ARCH(12) is the Engles (1982) F-test.
The estimates of the GJR-GARCH models of spot prices for the pre-futures period are presented in Table 6. The standard diagnostic tests of the residuals from the model confirm the absence of any further ARCH effects. The results of the coefficients of the lagged variance (a1) and lagged error-terms (1) indicate that the conditional volatility is time-varying, with no ARCH effects. The results of the asymmetry coefficient (1) suggest that there is a statistically significant and positive asymmetric effect, which implies that negative shocks elicit a larger response than positive shocks of an equal magnitude.

Table 6: GJR-GARCH Model Estimates for the Pre-Futures Period
Panel A: Coefficient Estimates FTSE/ASE-2e0 AR(1) (23/09/97 02/08/99) Mean Equation 0 1 Variance Equation a0 a1 1 1 Panel B: Residual Diagnostic LL Skewness Kurtosis J-B Q(24) Q2(24) ARCH(12)
Notes: See notes in Table 5.
14
0.0011 (0.493) 0.242* (5.203) 0.001* (2.732) 0.419* (3.785) 0.087 (0.832) 0.403* (3.152) 1,134.43 0.106 [0.000] 3.692 [0.000] 10.108 [0.000] 15.754 [0.923] 21.843 [0.505] 0.732 [0.721]
The issue of the impact of futures trading on spot market volatility is further investigated by estimating the GJR-GARCH model, Table 7, for spot returns for the post-futures period. The results of the impact of futures introduction on asymmetric market responses may be assessed via consideration of the asymmetry coefficient (1) that captures the nature of any bias in the post-futures period. The results indicate that the post-futures asymmetry coefficient is statistically insignificant. Thus, the introduction of FTSE/ASE-20 stock index futures contracts appears to have had an impact on the asymmetry of volatility in those routes, as a significant asymmetry coefficient in the pre-futures period results in an insignificant asymmetry coefficient in the post-futures period.
Table 7: GJR-GARCH Model Estimates for the Post-Futures Period
Panel A: Coefficient Estimates FTSE/ASE-20 - AR(1) (03/08/99 07/06/04) Mean Equation 0 1 Variance Equation a0 a1 1 1 LL Skewness Kurtosis J-B Q(24) Q2(24) ARCH(12)
See notes in Tables V and VI.
-0.0008** (-1.756) 0.139* (4.225) 3.10E-05* (3.826) 0.320* (13.306) 0.210* (2.437) 0.130 (1.536) Panel B: Residual Diagnostic 2.,875.12 0.209 [0.000] 4.787 [0.000] 147.31 [0.000] 30.158 [0.119] 28.925 [0.143] 0.983 [0.465]
The lagged error-term, 1, relates to changes in the spot price on the previous day, can be viewed as a new news coefficient. Hence, a higher value in the post-futures period implies that recent news have a greater impact on price changes. The results, from Tables 6 and 7, indicate that this holds, suggesting that information is being impounded in prices more quickly due to the introduction of futures trading. The coefficient of the lagged variance term, a1, can be thought of as reflecting the
15
impact of old news. A reduction in uncertainty regarding previous news can be regarded as an increase in the rate of information flow with the onset of futures trading (old news will have less impact on todays price changes). The results, from Tables 6 and 7, indicate that this holds, where the value of the a1 coefficient has been reduced in the post-futures period. The next step consists in examining whether the introduction of futures trading is not the only factor responsible for the reduction in the spot market volatility. To address this issue, the behaviour of the spot variances is adjusted for exposition to additional factors, which may affect spot market volatility. More specifically, the S&P, FTSE and DAX and DJ indices are used as economic indicators that can capture major world economic conditions, which may impact the spot market volatility. Thus, we test the hypothesis that futures trading is the only cause for the diminished volatility, testing the null hypothesis that the futures dummy coefficient (2) is zero. The estimates of the most parsimonious and well-specified (in terms of diagnostic tests) GJR-GARCH models are presented in Table 8. The estimates of the coefficients of the variance equation including: (i) the S&P variable are presented in panel A; (ii) the FTSE variable are presented in panel B; (iii) the DAX variable are presented in panel C; and (iv) the DJ variable are presented in panel D. The results in Table 8 indicate that the futures dummy variable (2 coefficient) has not been affected by only one economic variable (DAX) and only at the 10% significance level. Thus, this result supports not only the hypothesis of reduced spot volatility but also that the reduction in volatility may be a direct consequence of futures trading. However, this holds only for this economic variable. We notice that despite the negative sign and significance of the futures dummy variable (2 coefficient) three out of the four economic variables (DJ, FTSE, and SP) fail to contribute to the spot markets conditional volatility. Thus, the results do not present a clear answer as to whether the reduction in spot volatility is a direct consequence of futures trading.

Table 8:
16
GJR-GARCH Model Estimates of the Effect of Futures Trading and Other Economic Indicators on Spot Market Volatility (Whole Period)
Panel A: Variance Equation with DAX Variable a0 0.00012* (2.714) a1 0.610* (5.004) a2 0.099 (0.811) 1 0.078** (1.689) 2 0.017 (0.324) 1 0.152* (4.889) 2 -0.00016* (-2.042) 1 -0.0142** (-2.099) Panel B: Variance Equation with DJ Variable a0 0.00013* (2.785) a1 0.608* (5.020) a2 0.108 (0.797) 1 0.087** (1.695) 2 0.029 (0.421) 1 0.160* (4.899) 2 -0.00014* (-2.121) 1 -0.0130 (-0.593) Panel C: Variance Equation with FTSE Variable a0 0.00012* (2.752) a1 0.611* (5.009) a2 0.103 (0.825) 1 0.086** (1.699) 2 0.023 (0.475) 1 0.160* (4.896) 2 -0.00012* (-2.106) 1 -0.0040 (-0.122) Panel D: Variance Equation with SP Variable a0 0.00013* (2.795) a1 0.609* (4.992) a2 0.109 (0.858) 1 0.0850** (1.696) 2 0.0230 (0.431) 1 0.155* (4.970) 2 -0.00012* (-2.101) 1 -0.0179 (-0.759)
Notes:
However, given that most financial and commodity markets in developed economies impound information into prices rapidly, it is necessary to utilise high-frequency intraday data, as far as the impact of the onset of derivatives trading in terms of the speed of the price change, is concerned. In this study the most frequent, available, data are used namely daily data. This data set proves to be sufficiently frequent to identify the changes resulting from the onset of stock index futures trading. If information is continually flowing into the spot market then the fact that futures speeds up this flow may not be identified if the data set used is weekly or monthly, as the increase in the speed of information might by a matter of hours or even days.
6. Conclusions
This study examined the impact of futures trading and the activities of speculators on spot market price volatility in the FTSE/ASE-20 stock index market of the ATHEX, for the period 23 September 1997 to 04 June 2004. Most previous studies report mixed evidence. The results suggest that the onset of futures trading has had: (i) a stabilising impact on the spot price volatility; (ii) a desirable impact on the asymmetry of volatility (market dynamics); and (iii) a substantial improvement on the quality and speed of information flowing. After the inclusion of other explanatory variables in the conditional
AR(1)-GARCH(1,1)
AR(3)-GARCH(1,1)
AR(1)-GARCH(1,1)
AR(1)-GARCH(1,2)
See a note in Table V. DAX is the XETRA DAX index, DJ is the Dow Jones International index, FTSE is the FTSE100 Index, and S&P is the S&P 500 Index.
17
variance equation that may affect spot volatility, they do not seem to add additional information on the reduction of volatility as a result of stock index futures trading. Certainly, other economic variables could be added and checked, leading possibly to stronger results. But even if the reduction in volatility is also linked to these variables too, one could not exclude the case of being influenced by the entry of futures contracts. These findings have implications for the way in which the FTSE/ASE-20 futures market is viewed. Contrary to the traditional view of derivatives trading, the results indicate that the introduction of futures contracts has not had a detrimental effect on the underlying spot market. On the contrary, it appears that there has been an improvement in the way that news is transmitted into prices following the onset of futures trading. We can conjecture that by attracting more, and possibly better informed, participants into the market, futures trading has assisted on the incorporation of information into spot prices more quickly. Thus, even those market agents who do not directly use the FTSE/ASE-20 futures market have benefited from the introduction of futures trading. These results can provide regulators and practitioners with important insights into the stock index futures trading - spot market price volatility relationship, since the stock index futures market leads to new channels of information being provided, more information due to more traders, and a reduction in uniformed investors.
18
References
[1] [2] Aggarwal, R. (1988) Stock Index Futures and Cash Market Volatility, Review of Futures Markets, Vol 7 No 2, pp. 290-299. Alexakis, P., Kavussanos, M. G. and Visvikis I. D. (2002) The Lead-Lag Relationship in Returns and Volatility between Cash and Stock Index Futures: The Case of Greece, Conference Proceedings, European Financial Management Association Annual Meeting, 26th 29th June 2002, London, UK. Alexakis, P., Kavussanos, M. G. and Visvikis I. D. (2003) The Hedging Performance of Stock Index Futures: The case of the Athens Exchange, presented at Staff Research Seminars, Athens University of Economics and Business, 14 January 2004. Antoniou, A. and Holmes, P. (1995) Futures Trading, Information and Spot Price Volatility: Evidence for the FTSE-100 Stock Index Futures Contract Using GARCH, Journal of Banking and Finance, Vol 19, pp. 117-129. Antoniou, A., Holmes, P. and Priestley, R. (1998) The Effects of Stock Index Futures Trading on Stock Index Volatility An Analysis of the Asymmetric Response of Volatility to News, Journal of Futures Markets, Vol 18 No 2, pp. 151-166. Bacha, O. and Vila, A. F. (1994), Futures Markets, Regulation and Volatility: The Case of the Nikkei Stock Index Futures Markets, Pacific-Basin Finance Journal, Vol 2, pp. 201-225. Bae, S.C., Kwon, T.H. and Park, J.W. (2004) Futures Trading, Spot Market Volatility, and Market Efficiency: The Case of the Korean Index Futures Markets, Journal of Futures Markets, Vol 24 No 12, pp. 1195-1228. Baldauf, B. and Santoni, G. J. (1991) Stock Price Volatility: Some Evidence from an ARCH Model, Journal of Futures Markets, Vol 11, pp. 191-200. Bera, A. K. and Jarque, C. M. (1980) Efficient Tests for Normality, Heteroskedasticity, and Serial Independence of Regression Residuals, Economic Letters, Vol 6, pp. 255-259. Bessembinder, H. and Seguin, P. J. (1992) Futures Trading Activity and Stock Price Volatility, Journal of Finance, Vol 47, pp. 2015-2034. Board, J., Sandmann, G. and Sutcliffe, C. (2001) The Effect of Futures Market Volume on Spot Market Volatility, Journal of Business, Finance and Accounting, Vol 28, pp. 799-819. Board, J. and Sutcliffe, C. (1990) Information, Volatility, Volume, and Maturity: An Investigation of Stock Index Futures, Review of Futures Markets, Vol 9 No 3, pp. 533-547. Bollerslev, T. (1986) Generalised Autoregressive Conditional Heteroskedasticity, Journal of Econometrics, Vol 31, pp. 307-327. Bollerslev, T. (1987) A Conditional Heteroskedastic Time Series Model for Speculative Prices and Rates of Return, Review of Economics and Statistics, Vol 69, pp. 542-547. Bollerslev, T. and Wooldridge, J. M. (1992) Quasi-Maximum Likelihood Estimation of Dynamic Models with Time-Varying Covariances, Econometric Reviews, Vol 11, pp. 143172. Bollerslev, T., Chou, R. Y. and Kroner, K. F. (1992) ARCH Modelling in Finance: A Review of Theory and Empirical Evidence, Journal of Econometrics, Vol 52, pp. 5-59. Brady Commission (1988), Report of the Presidential Task Force on Market Mechanisms. Brorsen, B. W. (1991) Futures Trading, Transaction Costs, and Stock Market Volatility, Journal of Futures Markets, Vol 11, pp. 153-163. Brown-Hruska, S. and Kuserk, G. (1995) Volatility, Volume and the Notion of Balance in the S&P 500 Cash and Futures Markets, Journal of Futures Markets, Vol 15, pp. 677-689. Chan, K. C. and Karolyi, G. A. (1991), The Volatitlity of the Japanese Stock Market: Evidence from 1977 to 1990, in Ziemba, W.T., Bailey, W. and Hamao, Y. (Ed.), Japanese Financial Market Research, Contributions to Economic Analysis, No 205, North Holland, Amsterdam.
[3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20]
19 [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43]
International Research Journal of Finance and Economics - Issue 11 (2007) Chang, E. C., Cheng, J. W. and Pinegar, J. M. (1999) Does Futures Trading Increase Stock Market Volatility? The Case of the Nikkei Stock Index Futures Markets, Journal of Banking and Finance, Vol 23, pp. 727-753. Charupat, N. (2006) The Effect of Derivative Trading on the Underlying Markets: Evidence from Canadian Instalment Receipts Trading, International Review of Economics and Finance, Vol 15 No 3, pp. 276-293. Chatrath, A., Ramchander, S. and Song, F. (1996) The Role of Futures Trading Activity in Exchange Rate Volatility, Journal of Futures Markets, Vol 16 No 5, pp. 561-584. Conrad, J. (1989) The Price Effect of Option Introduction, Journal of Finance, Vol 44, pp. 487-498. Cox, J. C. (1976) Futures Trading and Market Information, Journal of Political Economy, Vol 84, pp. 1215-1237. Darrat, A. F. and Rahman, S. (1995) Has Futures Trading Activity Caused Stock Price Volatility? Journal of Futures Markets, Vol 15 No 5, pp. 537-557. Dennis, S. A. and Sim, A. B. (1999) Share Price Volatility with the Introduction of Individual Share Futures on the Sydney Futures Exchange, International Review of Financial Analysis, Vol 8 No 3, pp. 153-163. Edwards, F.R. (1988a) Does the Futures Trading Increase Stock Market Volatility? Financial Analysts Journal, Vol 44, pp. 63-69. Edwards, F.R. (1988b) Futures and Cash Market Volatility Stock Index and Interest Rate Futures, Journal of Futures Markets, Vol 8, pp. 421-439. Engle, R. F. (1982) Autoregressive Conditional Heteroskedasticity with Estimates of the Variance of United Kingdom Inflation, Econometrica, Vol 50 No 4, pp. 987-1008. Engle, R. F. and Ng, V. K. (1993) Measuring and Testing the Impact of News on Volatility, Journal of Finance, Vol 48 No 5, pp. 1749-1778. Engle, R. F. and Rothschild, M. (1992) Statistical Models for Financial Volatility, Journal of Econometrics, Special Issue on ARCH Models in Finance, Vol 52, pp. 1-4. Figlewski, S. (1981) Futures Trading and Volatility in the GNMA Market, Journal of Finance, Vol 36, pp. 445-456. Foster, A. J. (1995) Volume-Volatility Relationships for Crude Oil Futures Markets, Journal of Futures Markets, Vol 15 No 8, pp. 929-952. Glosten, L. R., Jagannathan, R. and Runkle, D. (1993) On the Relation Between the Expected Value and the Volatility of the Nominal Excess Return on Stocks, Journal of Finance, Vol 48, pp. 1779-1801. Gulen, H. and Mayhew, S. (2000) Stock Index Futures Trading and Volatility in International Markets, Journal of Futures Markets, Vol 20, pp. 661-685. Harris, L. (1989) S&P 500 Cash Stock Price Volatilities, Journal of Finance, Vol 46, pp. 1155-1175. Harris, L. (1989) The October 1987 S&P 500 Stock-Futures Basis, Journal of Finance, Vol 44, pp. 77-99. Harris, L., Sofianos, G. and Shapiro, J. (1994) Program Trading and Intraday Volatility, Review of Financial Studies, Vol 7, pp. 653-685. Hogan, K. C. Jr., Kroner, K. F. and Sultan, J. (1997) Program Trading, Nonprogram Trading and Market Volatility, Journal of Futures Markets, Vol 17, pp. 733-756. Kim, D. and Kon, S. (1994) Alternative Models for the Conditional Heteroskedasticity of Stock Returns, Journal of Business, Vol 67, pp. 563-598. Kyriakou, K. and Sarno, L. (1999) The Temporal Relationship between Derivatives Trading and Spot Market Volatility in the UK: An Empirical Investigation, Journal of Futures Markets, Vol 19, pp. 245-270. Lamoureux, C. G. and Lastrapes, W. D. (1990) Heteroskedasticity in Stock Return Data: Volume versus GARCH Effects, Journal of Finance, Vol 45, No 1, pp. 221-229.
International Research Journal of Finance and Economics - Issue 11 (2007) [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59]
20
Lee, S. B. and Ohk, K. Y. (1992) Stock Index Futures Listing and Structural Changes in TimeVarying Volatility, Journal of Futures Markets, Vol 19, pp. 493-509. Ljung, M. and Box, G. (1976) On a Measure of Lack of Fit in Time Series Models, Biometrica, Vol 65, pp. 297-303. McKenzie, M. D., Brailsford, T. J. and Faff, R. W. (2001) New Insights Into the Impact of the Introduction of Futures Trading on Stock Price Volatility, Journal of Futures Markets, Vol 21 No 3, pp. 237-255. Pagan, A. R. and Schwert, G. W. (1990) Alternative Models for Conditional Stock Volatility, Journal of Econometrics, Vol 45, pp. 267-290. Pericli, A. and Koutmos, G. (1997) Index Futures and Options and Stock Market Volatility, Journal of Futures Markets, Vol 17, pp. 957-974. Rahman, S. (2001) The Introduction of Derivatives on the Dow Jones Industrial Average and their Impact on the Volatility of Component Stocks, Journal of Futures Markets, Vol 21, pp. 633-653. Reyes, M. G. (1996) Index Futures Trading and Stock Price Volatility: Evidence from Denmark and France, Journal of Economics and Finance, Vol 20, pp. 81-88. Robinson, G. (1994) The Effect of Futures Trading on Cash Market Volatility: Evidence from the London Stock Exchange, Review of Futures Markets, Vol 14 No 2, pp. 429-459. Ross, S. A. (1989) Information and Volatility: The No-Arbitrage Martingale Approach to Timing and Resolution Irrelevancy, Journal of Finance, Vol 44, pp. 1-17. Santana, E. and Wadhwani, S. (1992) Feedback Traders and Stock Return Autocorrelations: Evidence from a Century of Daily Data, Economic Journal, Vol 102, pp. 415-425. Schwarz, T. V. and Laatsch, (1991) Price Discovery and Risk Transfer in Stock Index Cash and Futures Markets, Journal of Futures Markets, Vol 11, pp. 669-683. Schwert, G. W. (1989) Stock Market Volatility, Financial Analysts Journal, Vol 46, pp. 2334. Schwert, G. and Seguin, P. (1990) Heteroskedasticity in Stock Returns, Journal of Finance, Vol 45, pp. 1129-1155. Smith, C. W. (1989) Market Volatility: Causes and Consequences, Cornell Law Review, Vol 74, pp. 953-962 Stein, J. L. (1961) The Simultaneous Determination of Spot and Futures Prices, The American Economic Review, Vol 51, pp. 1012-1025 Stoll, H. R. and Whaley, R. E. (1988) Volatility and Futures: Message Versus Messenger, Journal of Portfolio Management, Vol 14, pp. 20-22.
A Birds Eye View of the Dividend Policy of the Banking Industry in Greece
Nikolaos Eriotis Associate Professor, National and Kapodistrian University of Athens 5 Stadiou Street, GR 105 62, Athens, Greece E-mail: neriot@econ.uoa.gr Tel: 0030 210 8259886-9 Dimitrios Vasiliou Professor of Bank Management, Hellenic Open University 16 Sahtouri Street & Ag. Andreou Street, GR 262 22, Patras, Greece E-mail: vasiliou@eap.gr Tel: 0030 210 8259886-9 Vasileios Zisis Lecturer (PD 407), Athens University of Economics and Business 76 Patision Street, GR 104 34, Athens, Greece E-mail: vzisis@aueb.gr Tel: 0030 210 8259886-9 Abstract Studies on dividend policy provided evidence of the relation between current period dividends and prior year dividends after controlling for earnings. This result implies the existence of a long term dividend policy that is partly unaffected by the level of current period earnings. The purpose of this study is to explore whether the dividend policy of financial institutions and especially the dividend policy of banks is different from the dividend policy of other industrial firms. We focus on banks because volatility of earning series is assumed to be higher for banks than other industrial firms due to accounting practices related reasons. We also explore the effect of differential earnings persistence due to the asymmetric timeliness in the reporting of good and bad news on the dividend policy of banks. We use data from the Athens Stock Exchange covering the period 1997 to 2001 as in recent studies that examine the dividend policy of Greek firms. Finally, we attempt to provide evidence related to bank managers dividend smoothing practices through components of cash dividends. Our results suggest that in financial institutions and, especially, in banks last year dividends are unrelated to current period dividends after controlling for earnings. This finding is consistent with the argument that due to high earnings volatility bank managers do not adhere to a long term dividend policy unaffected by the level of firms performance. Keywords: Accounting Practices, Asymmetric Timeliness, Dividends, Greek Banks Jel Classification Codes: G3, G32, G35, M41
22
1. Introduction
Research on corporate dividend policy suggests that firms tend to follow a dividend policy driven by the existence of a long-term target payout ratio. If current period dividends were closely aligned to current period earnings then dividends would be as volatile as earnings. Therefore, firstly, managers tend to smooth dividends and, secondly, to the extent that long-term dividend payout ratio is based on non-anticipated (by shareholders) future earnings, managers signal information to shareholders through changes in dividends.
In mid fifties, Lintner conducted a number of interviews with corporate managers. His findings highlighted the importance given by corporate managers to the stability of dividends and the role of current period earnings in the determination of current period dividends. Lintner (1956) model, in which change in dividends is regressed on current period earnings and last period dividends, explained 85% of the variability of current period change in dividends. Fama and Babiak (1968) found that prior year earnings have incremental explanatory power as an incremental independent variable in Lintner (1956) model 1. Related to the corporate dividend policy of Greek firms, research is limited. Patsouratis (1989) and Joannos and Philippas (1997) are examples of early work on dividend policy. Patsouratis (1989), using Brittain (1964) model, provided empirical results that showed that earnings are a more influential factor than prior year dividends in determining current period changes in dividends. Joannos and Filippas (1997) provided results that are in, generally consistent, with the argument that Lintner (1956) model best described, for their sample, the dividend policy of Greek firms. In a recent study, Vasiliou and Eriotis (2002) concluded that the Lintner (1956) model can be improved in two ways; either by altering all variables of interest in a level or in a change format. Vasiliou and Eriotis (2002), and Eriotis and Vasiliou (2004 and 2006) provided evidence in favor of the existence of a long-term payout ratio. They found that current period dividends are positively associated with prior period dividends for firms that trade in the Athens Stock Exchange. If dividends were perfectly correlated with stock returns, then current period dividend would have not been correlated with previous period dividends. Moreover, Eriotis and Vasiliou (2004) concluded that dividends are adjusted to the performance and effectivness of management. Eriotis and Vasiliou (2006) explored the effect of leverage on the dividend policy of Greek firms. Cash disbursement to shareholders might delay the implementation of investment plans especially for highly leveraged firms. However, Eriotis and Vasiliou (2006) empirical results suggest that the higher (lower) the leverage the higher (lower) the dividend. They argue that their finding is consistent with the expected return pattern across different levels of the economy stability. Hrysanthopoulou (2004) provided evidence consistent with the argument that size, risk, leverage ratio and ownership concentration influence Greek firms dividend policy 2. In this study we re-examine the dividend corporate policy of Greek firms and extend prior work on the basis of Vasiliou and Eriotis (2002) level model in a number of ways. We explore whether the dividend policy of financial institutions, and especially the dividend policy of banks,3 is different from industrial firms. Earnings of financial institutions such as banks should in theory be more volatile. Their assets are mainly financial assets and that is why it is more likely to be reported in fair values. Earnings, especially in a clean surplus accounting, measure changes in the book value of equity. Therefore, changes in the fair value of assets of financial institutions and, consequently, earnings might be more volatile than earnings of other industrial firms. As aforementioned, current period earnings is a significant factor in the determination of current period dividends and adherence to a long-term dividend payout ratio may drive correlation between current period earnings and dividends away from one if earnings are volatile. The smoothing, therefore, of dividends in cases in which earnings volatility might exist is more likely to increase the coefficient attached to last year dividends in Vasiliou and Eriotis (2002) model. If smoothing of
1 2
See also Brittain (1964), Lee et al. (1987) and Nakamura and Nakamura (1985) for alternative models. See also Holder, Langrehr and Hexter (1998), Short, Zhang and Keasey (2002), Frankfurter and Wood (2002), Barclay, Smith and Watts (1995), Soter, Brigham and Evanson (1996), Woolridge and Ghosh (1985) for studies of influential factors on dividends (using non-Greek data). Karathanassis and Philippas (1998) provided evidence consistent with the argument that stock prices of Greek banks are related to the level of dividends. For other studies (that use Greek data) on the effect of dividends on stock prices see Papaioannou, Travlos and Tsangarakis (2000).
23
dividends does not exist in banks, we would expect that current period dividends would be uncorrelated to last year dividends. Moreover, we examine volatility in the earnings series driven by the asymmetry in the reporting of good and bad news. Basu (1997) and Pope and Walker (1999) provided empirical results that are consistent with asymmetry in the reporting of good and bad news. Good news tends to be underrecognised while bad news tends to be over-recognised by accountants. This asymmetry in the reporting of good and bad news tends to create earnings series with lower persistence when bad news are recorded. Related to the Greek reporting framework, Eriotis et al. (2004) suggested that bad news tend to be reported in a more timely way than good news but not in a conservative way. We examine, therefore, whether implied asymmetry in the reporting of good and bad news influences dividend policy of banks. Furthermore, the Greek corporate law (2190/1920) imposes firms to distribute a minimum (obligatory) dividend. This minimum dividend (called first dividend) is defined as the higher amount between the 6% of the equity capital or the 35% of earnings after tax that are available to shareholders after the deduction of the statutory reserve (measured as 5% of earnings after tax that are available to shareholders). Therefore, we expect that if managers attempt to smooth dividends rather earnings when setting long-run dividend targets this smoothing should evolve through the additional (over first) dividend. In order to have comparable to Eriotis and Vasiliou results, we use data that cover the same sample period they have used. However, we focus on banks and alter the research design in order to accommodate the two aforementioned extensions; the effect of accounting related practices that impose volatility in earnings and institutional factors. The remaining of our study is organised as follows; in section 2 we develop the research design, in section 3, we present descriptive statistics, while in section 4 we analyse our empirical results. Finally, we conclude in section 5.
2. Research Design
We use Vasiliou and Eriotis (2002) modeling in order to incorporate institutional factors and volatility in earnings due to accounting related practices 4. The model they have used is as follows; (1) where DIV denotes dividends, E denotes earnings, a0 is the constant, a1 and a2 are coefficient estimates of independent variables, e1 is the error term of the regression and t denotes time subscripts (firm subscripts are omitted). Model (1) is examined for the whole sample of firm-year observations and for firms in the banking sector. In order to test our first prediction about differences in the coefficient estimates between industrial firms and financial institutions (including banks) we incorporate a dummy variable (denoted as DFI) which takes the value of one if a firm belongs to the financial sector 5, otherwise it takes the value of zero. Model (1) is altered to model (2) as follows; (2) where b0 is coefficient estimate of the dummy variable on the constant, while b1 and b2 are coefficient estimates that capture differences on Et and DIVt-1 between industrial and financial firms. In order to test whether the dividend policy of banks is altered for good and bad news period, we incorporate, in model (1), a dummy variable (denoted as DRET) which takes the value of one if current period news is negative (bad news), otherwise it takes the value of zero. Model (1) is altered to model (3) as follows; (3)
4 5
We disregard the leverage ratio because we do not have any strong prediction for differential effect across industrial firms and banks. Financial sector includes banks, investment firms, insurance firms and firms in the leasing sector.
24
where c denotes coefficient estimates related to the incorporation of the dummy variable DRET. Model (2) and Model (3) are more restricted models than Model (1) and they are used in order to capture managers dividend smoothing practices that may exist in cases in which earnings are assumed to be more volatile due to accounting practices. As aforementioned, we expect that dividend smoothing might not come from the first dividend, which is obligatory, but through the additional dividend. We examine this hypothesis for banks through model (4); (4) where FDIV denotes first dividend and ADIV denotes additional dividend. The sum of FDIV and ADIV equals DIV.
3. Data Selection and Descriptive Statistics

Our sample covers the period 1997 - 2001 that has been used by Eriotis and Vasiliou (2006) and data have been collected from the Athens Stock Exchange database. Our initial sample has been reduced after deleting 1. the top and bottom percentile of current period earnings per share (E); defined as in Eriotis and Vasiliou (2006) as earnings before tax, current period leverage ratio; defined as in Eriotis and Vasiliou (2006) as total debt on total 2. the highest 2% values of current period dividend per share (DIV) previous period dividend per share (DIVt-1). We further deleted firm-year observations with earnings per share higher than four. The final sample includes 598 firm-year observations (126 firms). The initial sample includes 7 banks and 23 firms in the financial sector. The sub-sample of 598 firm-year observations includes 92 firm-year observations from the financial sector. 20 out of 92 firm-year observations correspond to the banking sector. In regressions in which we only use bank data we consider all observations from the bank sector (before deletion of outliers) apart from the Bank of Greece. We exclude the Bank of Greece, firstly, because of the level of earnings per share (the lowest value of earnings per share for Bank of Greece is around 55) and, secondly, because of the governmental role of this bank.
Table 1: Descriptive statistics
Mean 0.187 0.482 0.223 0.202 0.021 Mean 0.934 2.835 0.940 0.695 0.245 Std. 0.288 0.683 0.314 0.309 0.066 Std. 0.913 2.910 0.985 0.733 0.641 Panel A: Total sample (n = 598)a b min Q1 0.000 0.021 -0.937 0.093 0.000 0.043 0.000 0.029 0.000 0.000 Panel B: Banks (n = 30) a b min Q1 0.000 0.147 0.101 0.507 0.000 0.132 0.000 0.132 0.000 0.000 Median 0.097 0.266 0.123 0.103 0.000 Median 0.892 1.987 0.775 0.312 0.000 Q3 0.207 0.612 0.264 0.230 0.000 Q3 1.291 4.291 1.291 1.027 0.202 Max 2.142 3.999 2.142 2.142 0.587 Max 3.815 12.115 3.815 2.715 2.794
DIVt Et DIVt-1 FDIVt-1 ADIVt-1 DIVt Et DIVt-1 FDIVt-1 ADIVt-1
a.
b.
The sample presented in Panel A includes 598 firm-year observations and we arrive at this sample after deleting observations classified as outliers. The sample in Panel B includes six banks before the outlier deletion technique DIV denotes dividends per share, E denotes earnings after tax per share FDIV denotes first dividend, ADIV represents additional dividend and t is a time subscript.
25
Summary descriptive statistics of the total sample of 598 firm year observations and of 6 banks are presented, respectively, in Panel A and in Panel B of Table 1. Average earnings per share are positive in both Panel A and Panel B (0.482 and 2.835, respectively). Despite our deletion outlier technique both dividends and earnings present positive skewness. In more than 75% of firm year observations firms distribute dividends to shareholders. Additional dividends 6 have been distributed to shareholders in around 25% of firm-year observations in the banking sector, but less than 25% of firmyear observations that are included in the whole sample. For firms included in the whole sample and distribute ADIV, the average ratio of last year FDIV to last year ADIV is around 13, while the respective ratio is around 1.6 for the banking sector. This result suggests that in the banking sector ADIV are more material in size relative to FDIV (when reported in financial statements) than in the industrial sector.
In Table 2, we present empirical results using model (1) and model (2) for the whole sample of 598 firm-year observations 7. Results from model (1) suggest that both current period earnings and last year dividends are positively associated with current period dividends. The coefficient estimates of a1 and a2 are positive (0.269 and 0.249) and significantly different from zero. The model explains 0.798 of the variability of current period dividends. Coefficients estimates, in our study are higher than those reported by Eriotis and Vasiliou (2006) potentially due to our outlier deletion technique. The positive coefficient attached to DIVt-1 suggests that only 0.249 of the shock that appeared in last year dividend will persist in current period dividend after controlling for current period earnings. The reader should note that the small magnitude of this coefficient seems natural if one considers the contemporaneous association earnings and dividends and the persistence property of earnings. An interesting finding is presented in Table 2 when model (2) is used. Firstly, coefficient estimate of b1 is positive (0.185) and significantly different from zero and, secondly, coefficient estimate of b2 is negative (- 0.169) and statistically different from zero. This implies that dividend policy of financial institutions, all other factors equal, is different from those of industrial firms. Firms in the financial sector give more weight to current period earnings than industrial firms and less weight to last period dividends than industrial firms. Our result suggests that dividend policy of firms in the financial sector that trade in the Athens Stock Exchange is less adhered to a long-term target dividend. The adjusted R2 of model (2) is higher (0.890) to the respective (0.790) of model (1) suggesting that model (2) explains better than model (1) the variability of current period dividends. Dividend shocks of firms in the financial sector persist less than respective shocks in industrial firms, after controlling for information in earnings. Therefore, dividends of firms in the financial sector revert more quickly to their long run mean than dividends of industrial firms. In Table 3, we present empirical results when using model (1), model (3) and model (4) for the sub-sample of 30 observations from the banking sector. Results from model (1), in Table 3, are consistent with the results presented earlier from model (2) in Table 2. Coefficient estimate of a1 is positive (0.311) and statistically different from zero, higher (no formal statistical test has been done) from the respective coefficient estimate of a1 for industrial firms and close to the sum of a1 and a2 (from model (2) of Table 2). Last period dividends are not associated with current period dividends, after controlling for current period earnings, because coefficient attached to a2 (0.058) is positive but not statistically
6
Firms are not required to report separately first and additional dividend. When we run regressions based on model (4) for banks we use two sets of FDIV and ADIV. The first set of FDIV and ADIV is based on data as reported in financial statements. The second set of FDIV and ADIV is extended, because we reconstruct FDIV and ADIV according to the Greek corporate law. Inferences are similar and only results from the first set of data are presented. Descriptive statistics for FDIV and ADIV, in Table 1, refer, as well, to the first way of collecting FDIV and ADIV. We present statistics based on White (1980) consistent estimators. Coefficient estimators are based on generalised least squares regressions. Inferences from ordinary least square regressions are similar and are available by authors upon request.
26
different from zero. This result is consistent with the argument that, in the banking industry, managers do not follow a long term dividend policy. Current period dividends are driven by the level of current period earnings. Considering the assumed high volatility of earnings in banks this result suggests that it is difficult for managers to adhere to a long term dividend policy unaffected by the level of current period earnings.
Table 2:
n = 598 Modelab (1) (2)
a.
The association between current period dividends with current period earnings and last period dividends for the total sample and after controlling for financial institutions
A0 Coef pv st. Err Coef pv st. err -0.005 0.000 0.001 0.002 0.000 0.000 b0 a1 0.269 0.000 0.004 0.188 0.000 0.003 B1 A2 0.249 0.000 0.011 0.344 0.000 0.008 b2 R2 0.798 -0.169 0.000 0.037 0.890
-0.012 0.281 0.011
0.185 0.000 0.016
Tests in Table 2 are based on the following models (1) and (2) where DFI denotes a dummy variable that takes the value of 1 if a firm is classified in the financial institution sector, otherwise it takes the value of zero. For definitions of other variables see notes in Table 1. n denotes the number of observations, R2 denotes adjusted R-squared, coef. denotes coefficient estimate, pv. denotes p.value against the hypothesis that the coefficient estimate is different from zero and st. err. denotes standard error. Coefficient estimates are based on generalised least squares regressions, while p. values and standard errors are based on White (1980) corrections.
b.
27
Table 3:
N = 30 Modela (1) (3) (4)
a.

The association between current period dividends with current period earnings and last period dividends for firms in the banking industry
A0 coef. pv. st. err. coef. pv. st. err. coef. pv. st. err. 0.024 0.545 0.039 - 0.012 0.904 0.094 0.023 0.568 0.039 c0 A1 0.311 0.004 0.100 0.355 0.001 0.093 0.316 0.007 0.108 C1 A2 0.058 0.730 0.165 0.045 0.811 0.187 0.067 0.675 0.164 C2 A3 R2 0.765 - 0.215 0.041 0.010 - 0.067 0.809 0.285 0.794 0.769
0.003 0.934 0.041
0.279 0.085 0.155
Tests in Table 3 are based on the following models (1) (3) (4) where DRET denotes a dummy variable that takes the value of 1 if a firms current period return is negative (bad news), otherwise it takes the value of zero (good news). Return is measured as the annual stock return for a period ending six months after the fiscal year end (as financial statements might delay to be published). For definitions of other variables see notes in Table 1 and Table 2.
The irrelevance argument of last year dividends appears to be valid even if we divide last year dividends to first dividend and additional dividend. Coefficients attached to a2 (0.067) and a3 (- 0.067) in model (4) are not statistically different from zero 8. The statistical insignificance of FDIV, which, in the banking sector is more material in size than FDIV in industrial firms, enforces the argument of the lack of a long-term dividend target in the banking sector. Results that consider asymmetric timeliness in the reporting of current period good and bad news show that the coefficient attached to c1 (0.279) is positive and statistically different from zero 9. Following asymmetric timeliness argument, if bad news (losses) are over-recognised managers will increase the percentage of earnings they give as dividends because they know that these losses will revert next year 10. Coefficient attached to a2 is positive (0.045) but not significantly different from zero that is consistent with prior results for banks. In other words, if good news appears in current period, last year dividend shocks do not persist. Even in good news period bank managers cannot adhere to last year dividend shocks. In contrast, the coefficient estimate attached to c2 is negative (0.215) and statistically different from zero. This negative coefficient suggest, as expected, that due to bad news last year dividends are reverting; the higher the last year dividend, the lower the dividend that firms distribute in the current bad news period. In general our empirical results imply that industrial firms tend to adhere to a corporate policy that follows an implied long term dividend target. In contrast we provide evidence consistent with the argument that firms in the financial sector and, especially, banks do not follow a long term dividend policy. For firms in the banking industry, last year dividends are not related to current period dividends after controlling for current period earnings.
9 10
See also footnote 3 for the way we construct ADIV and FDIV. We perform analysis based on model (4) for the sample of 598 firm-year observations using FDIV and ADIV as reported (when reported) in financial statements. According to those results (not tabulated) coefficients attached both to FDIV and ADIV are positive and significantly different from zero. When we introduce a dummy variable for financial institutions the interactive with the dummy variable coefficient attached to ADIV and FDIV is negative and significantly different from zero. C1 is not statistically different from zero when ordinary least squares regressions are used. An alternative explanation is that higher percentage of earnings is given as dividends when stock returns are negative so as to compensate shareholders for capital losses they incur.
28
5. Conclusions
In this study we explore whether dividend policy of financial institutions and, especially, banks is different from the dividend policy of other firms. We focus on banks because earnings series of financial institutions are subject to higher volatility of other firms due to the way accounting profits are recorded. Our main empirical finding suggests that bank managers do not adhere to a long term dividend policy, in contrast to managers of other industrial firms. In the banking industry, current period dividends are largely driven by the level of current period earnings and are unrelated to last year dividends, after controlling for current period earnings. Moreover, we explore whether asymmetric timeliness in the reporting of good and bad news influences the dividend policy of banks and whether smoothing in dividends comes from the additional cash dividend is distributed to shareholders. Examination of these two factors provides results that enforce the irrelevance argument of last year dividends after controlling for earnings. Especially, in bad news periods bank managers seem to distribute a higher percentage of current period earnings as dividends consistent with the argument of the over-recognition of bad news in current period earnings, while in good news period bank managers do not adhere to last year dividend shocks.
29
International Research Journal of Finance and Economics - Issue 11 (2007) Barclay M. J., Smith C.W. and Watts R.L., 1995. The Determinants of Corporate Leverage Policy and Dividend Policy, Journal of Applied Corporate Finance, Bank of America, Vol. 7 (4), pp. 4-19. Basu S., 1997. The Conservatism Principle and the Asymmetric Timeliness of Earnings, Journal of Accounting and Economics, Vol. 24 (1), pp. 3-37. Brittain J., 1964. The Tax Structure and Corporate Dividend Policy, American Economic Review, pp. 272-287. Eriotis N. and Vasiliou D., 2004. Dividend Policy: An Empirical Analysis of the Greek Market, International Business and Economic Research Journal, Vol. 3 (3), pp. 49-57. Eriotis N. and Vasiliou D., 2006. The Link between Dividend Policy and Corporate Leverage: An Empirical Analysis of the Greek Market, Spoudai, vol. 56, no. 1, January-March, pp. 64-75. Eriotis N., Vasiliou D. and Zisis V., 2004. The Asymmetric Timeliness in the Reporting of Good and Bad News of Firms that trade in the Athens Stock Exchange, European Research Studies, Vol. 7 (3-4), pp. 103 - 127. Fama E. F. and Babiak H., 1968. Dividend Policy: An Empirical Analysis, Journal of the American Statistical Association, Vol. 62 (, pp. 1132-1161. Frankfurter G. M. and Wood B. G. Jr., 2002. Dividend Policy Theories and their Empirical Tests, International Review of Financial Analysis, Vol. 11 (2), pp. 111-138. Hrysanthopoulou E. H., 2005. Dividend Policy of Greek firms; A Theoretical and Empirical Examination, Ph. D. Thesis, Athens University of Economics and Business. Holder M. E., Langrehr F. W. and Hexter J.L., 1998. Dividend Policy Determinants: An Investigation of the Influences of Stakeholder Theory, Financial Management, Vol. 27 (3), pp. 73-82. Joannos J. and Philippas N., 1997. Dividend Policy of Firms listed in the Athens Stock Exchange: An Empirical Analysis, Spoudai, Vol. 47 (3-4), pp. 249-285 Karathanassis G. and Philippas N., 1988. Estimation of Bank Stock Price Parameters and the Variance Components Model, Applied Economics, Vol. 20, pp. 497-507. Lee C. F., Wu C. and Djarraya M., 1987. A Further Empirical Investigation of the Dividend Adjustment Process, Journal of Econometrics, Vol. 35, pp. 267-285. Lintner J., 1956. Distribution of Incomes of Corporations among Dividends, Retained Earnings and Taxes, American Economic Review, Vol. 46 (2), pp. 97-113. Nakamura A. and Nakamura M., 1985. Rational Expectations and the Firms Dividend Behavior, Review of Economics and Statistics, Vol. 67 (4), pp. 606-615. Papaioannou G.J., Travlos N.G. and Tsangarakis N.V., 2000. Valuation Effects of Greek Stock Dividend Distributions, European Financial Management, Vol. 6 (4), pp. 515-531. Patsouratis V., 1989. Corporate Taxation and Dividend Behavior: An Empirical Analysis, Greek Economic Review, pp. 323-338. Pope P. F. and Walker M., 1999. International Differences in the Timeliness, Conservatism, and Classification of Earnings, Journal of Accounting Research, Supplement, Vol. 37, pp. 5387. Short H., Zhang H. and Keasey K., 2002. The Link between Dividend Policy and Institutional Ownership, Journal of Corporate Finance, Vol. 8 (2), pp. 105-122. Soter D., Brigham E. and Evanson P., 1996. The Dividend Cut Heard round the World: The Case of FPL, Journal of Applied Corporate Finance, Vol. 9 (1), pp. 4-15. Vasiliou D. and Eriotis N., 2002. The Determinants of Dividend Policy: Some Greek Data, The European Journal of Management and Public Policy, Vol. 1 (2), pp. 48 - 58. Woolridge J.R. and Ghosh C., 1985. Dividend Cuts: Do they Always Signal Bad News?, Midland Corporate Finance Journal, Vol. 3 (2), pp. 20-32.
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22]
Finance Growth Nexus: Evidence from Turkey

Ali Acaravci Department of Economics, Faculty of Economics and Administrative Sciences Mustafa Kemal University, Antakya-Hatay, Turkey E-mail: acaravci@mku.edu.tr Tel: +90 326 2455813/1233; Fax: +90 326 2455854 Ilhan Ozturk Faculty of Economics and Administrative Sciences Cag University, 33800, Mersin, Turkey E-mail: ilhanozturk@cag.edu.tr Tel: +90 324 6514800; Fax: +90 324 6514811 Songul Kakilli Acaravci Department of Finance and Accounting, Faculty of Economics and Administrative Sciences Mustafa Kemal University, Antakya-Hatay, Turkey E-mail: sacaravci@mku.edu.tr Tel: +90 326 2455813/1233; Fax: +90 326 2455854 Abstract This paper examines the causal relationship between financial development and economic growth in Turkey for the period 1986:1-2006:4 using dynamic time series models. The results of the cointegration analysis provide evidence of no long-run relationship between financial development and economic growth. Therefore, the empirical investigation is carried out in a vector autoregression (VAR) framework to analyze the short run effect of the financial intermediary development on economic growth. The results show a one-way causal relationship running from the financial development to the economic growth in Turkey. Keywords: Financial development, Growth, Turkey, Causality Jel Classification Codes: O11, O16, O52
1. Introduction
Academic research on the finance-growth nexus dates back at least to Schumpeter (1911) who emphasized the positive role of financial development on economic growth. The relationship between financial development and economic growth has been a subject of great interest and debate among economists for many years. The debate has traditionally revolved around two issues. The first relates to whether development in the financial system results in a faster economic growth, and the second relates to how financial development affects economic growth. A large body of literature has emerged, both at the theoretical and empirical level, attempting to answer the above questions. Although many empirical studies have investigated the relationship between financial depth, defined as the level of
31
development of financial markets and economic growth, the results are ambiguous (see, Pagano, 1993; and Levine, 1997, 2003 for a survey of the literature). On the one hand, cross country and panel data studies find positive effects of financial development on output growth even after accounting for other determinants of growth as well as for potential biases induced by simultaneity, omitted variables and unobserved country-specific effect on the finance-growth nexus1. On the other hand, time series studies give contradictory results. Demetriades and Hussein (1996) find little systematic evidence in favor of the view that finance is a leading factor in the process of economic growth. In addition they found that for the majority of the countries they examine, causality is bi-directional, while in some cases financial development follows economic growth. Luintel and Khan (1999) used a sample of ten less developed countries to conclude that the causality between financial development and output growth is bi-directional for all countries. All these results show that a consensus on the role of financial development in the process of economic growth does not so far exist. As a matter of fact, the role of financial sector has been well recognized in the development literature. The seminal work of Patrick (1966) has resulted in widespread investigations into the role of the financial sector as an engine for economic growth. Patrick points out two possible relationships between financial development and economic growth. First, as the economy grows, it generates demand for financial services, which he calls a demand-following phenomenon. According to this view, the lack of financial institutions in developing countries is an indication of lack of demand for their services. Second, the establishment and the widespread expansion of financial institutions in an economy may actively promote development, which Patrick called supply-leading phenomenon. This latter view, which has been dubbed the financial-led growth hypothesis, has been popular among governments in several developing countries as a means to promoting development (Habibullah and Eng, 2006). Several theoretical and empirical studies have suggested that the role of financial development in the economy may vary across countries because of differences in institutional and economic structures (see La Porta et al., 1997; and Bell and Rousseau, 2001, among others). There are those who argue that, in a given economy, it is the sector with high economies of scale that benefit more from financial development (Kletzer and Pardhan, 1987; Beck, 2002), implying that financial development is much more effective in promoting economic growth in more industrialized economies than in less industrialized or agricultural economies. On the other hand, there are those who contend that countries at their early stage of development benefit more from financial development (see McKinnon, 1973; Fry, 1995). Moreover, it is argued that the effectiveness of financial intermediaries and markets in promoting economic growth depends on the institutions set up to implement financial transactions The growing body of empirical research, using different statistical procedures and data sets, produces remarkably consistent results. First, countries with better-developed financial systems tend to grow fasterspecifically, those with (i) large, privately owned banks that funnel credit to private enterprises and (ii) liquid stock exchanges. The levels of banking development and stock market liquidity each exert a positive influence on economic growth. Second, simultaneity bias does not seem to be the cause of this result. Third, better-functioning financial systems ease the external financing constraints that impede firm and industrial expansion. Thus, access to external capital is one channel through which financial development matters for growth because it allows financially constrained firms to expand (Levine, 2003). In 1980, the Turkish government initiated a series of reforms to accomplish a major policy shift from import substitution to an export-led growth (ELG) strategy, mainly by liberalizing foreign trade. The lifting of repressive controls on financial markets, referred to as financial liberalization, was realized gradually over 1980s as a part of this policy change. Thus, the beginning of 1980s constituted a turning point in the economic life of Turkey. Turkey liberalized its capital accounts in 1989, taking an important step towards integrating its economy with the global economic system. The major policy
1
See for example King and Levine (1993a,b), Khan and Senhadji (2000) and Levine et al. (2000).
32
shift from import substitution to ELG substantially increased Turkeys export figures and, together with a dynamic private manufacturing sector, Turkey became a major manufacturing base for a number of multinational corporations. Since 2001, Turkey has carried out some of the most impressive and long-awaited structural reforms, which were recognized by the international community: the EU has agreed to open full membership negotiations with Turkey and the IMF declared Turkey a success story. The economy is growing by around 7.5% annually, now for the fifth consecutive year. As of 2006, the Turkish GDP was USD 390 billions, while the per capita income was over USD 5,000, which corresponds to USD 9,000 in PPP terms. The improved outlook of the economy promises a new potential growth rate around 6% annually for the next decade. After a very strong devaluation in 2001, the New Turkish Lira (YTL) has been left to float freely with minor interventions and has become overvalued against major currencies by some 45 per cent, according to the exchange rate index at the end of 2005. The share of foreign investors in the banking sector is the 26.16 percent for the 2006 and it is expected to be 35 percent in 2007 by selling the Oyak Bank and Halkbank to the foreign investors. In addition to these developments in the banking sector, the share of the foreign investors in the Istanbul Stock Exchange is also increasing everyday. The present paper addresses the empirical relationship between financial development and economic growth for the Turkey over the period 1986:1-2006:4. Turkeys financial system and its banking sector are virtually synonymous as a consequence of the countrys economic and historical development. Many of the transactions and activities in both the money and capital markets are carried out by banks, as the banking sector constitutes the major part of the Turkish financial system. In Section 2, we present a literature review about financial development and economic growth. The methodology and data are presented in Section 3. The empirical results are discussed in Section 4. The paper concludes with a summary and policy implications.
2. Financial Development and Growth

An extensive amount of empirical investigations have been conducted, aimed at testing the conflicting theoretical developments using different techniques. The results of these studies provide evidence of strong connection between the exogenous component of financial development and long-run economic growth. This is more or less consistent with the classical view on the relation between growth and financial development. On the empirical side, King and Levine (1993a) use IMF data and various financial indicators to conclude that there is a positive relationship between financial indicators and growth, and that financial development is robustly correlated with subsequent rates of growth, capital accumulation, and economic efficiency. They correctly emphasize that policies that alter the efficiency of financial intermediation exert a first-order influence on growth. Atje and Jovanovic (1993) examine the role of stock markets on development, and conclude that there is positive effect on the level as well as on the growth. They could not, however, establish a significant relationship between bank liabilities and growth. Levine and Zervos (1996) use various measures of stock market development, and conclude that there is a significant relationship. Arestis and Demetriades (1997) use time series analysis and Johansen cointegration analysis for the US and Germany. For Germany, they find an effect of banking development on growth. In the US, there is insufficient evidence to claim a growth effect of financial development, and the data point to the direction that real GDP contributes to both banking system and stock market development. Levine et al. (2000), using a sample of 74 developed and less developed countries over the period 19601995. They found that the strong positive relationship between financial development and output growth can be partly explained by the impact of the exogenous components like finance development on economic growth. They interpreted these results as supportive of the growthenhancing hypothesis of financial development. Demirguc-Kunt and Maksimovic (1998) estimate a financial planning model to find that financial development facilitates the firms growth. In this context
33
an active stock market and a well-developed legal system are crucial for the further development of the firms. The theoretical argument by Bencivenga and Smith (1991), Levine (1991), and Saint-Paul (1992) support the proponents of the supply-leading hypothesis proposed by Schumpeter (1934) and Patrick (1966). In addition, most of the models argue that the process of growth has a feedback effect on financial markets by creating incentives for further financial development, which means that the two variables are endogenously determined. On the other hand, Harrison et al. (1999), and Blackburn and Hung (1998) argue that financial intermediation encourages economic growth because it reduces the cost of project appraisal. As the number of projects increases in a growing economy, more banks enter the markets as banks activity and profit increases. This entry reduces the average distance between banks and borrowers, promotes regional specialization and reduces the cost of intermediation. Goldsmith (1969), McKinnon (1973), Shaw (1973), Fry (1988), Jung (1986), Gupta (1984) and King and Levine (1993a, 1993b) are among those who have provided evidence that financial development is a prerequisite for economic growth. Nevertheless, other researchers are skeptical with respect to the financial-led growth hypothesis. Using annual data from 1965 to 1992, Demetriades and Hussein (1996) found that among the Asian countries covered under the study; only in the case of Sri Lanka did the evidence support the financial-led growth hypothesis. For Pakistan, their result indicates that economic growth causes financial development. Further, Demetriades and Husseins study suggests that bidirectional causal relationships are evident for India, South Korea and Thailand. In another related study, Arestis and Demetriades (1996) further support the evidence that the relationships between financial development and economic growth for India and South Korea are bidirectional. Murinde and Eng (1994) test the financial-led hypothesis on Singapore using quarterly data for the period 1979 to 1990. Using an array of financial indicators, they found that the results strongly support the financial-led hypothesis for Singapore. Habibullahs (1999) study on seven Asian developing countries suggests that only the Philippines support the financial-led growth hypothesis. The cases of demand-following growth hypothesis are supported by Malaysia, Myanmar, and Nepal. On the other hand, a bi-directional causality between growth and finance are evident for Indonesia, Sri Lanka and Thailand. Using annual data from 1970 to 2001 for Turkey, Unalmis (2002) found that, in the long-run, the test results in the context of VECM for the coefficients of all cointegrated series show a two way causality between financial deepening and economic growth. Guryay et al. (2007) examined the relationship between financial development and economic growth for Northern Cyprus for the period of 1986-2004. The result showed that there is a negligible positive effect of financial development on economic growth of Northern Cyprus. Although Granger causality test showed that financial development does not cause economic growth, on the other hand there is evidence of causality from economic growth to the development of financial intermediaries. Al-Zubi et al. (2006) applied a model developed by Levine in 1997 using panel data for eleven Arab countries during the period 1980-2001. The results show that all financial indicators are insignificant and do not affect economic growth. The modified model shows that only public credit to domestic credit indicator has a significant and positive effect on economic growth, indicating the dominance of the public sector in economic activities and the financial sectors are still underdevelopment and need more efforts to be able to exert its functions effectively in the Arab countries. Apergis et al. (2007) investigate the causal linkages between financial development and economic growth in a large sample of 65 countries (15 OECD and 50 non-OECD countries), both developed and developing, over the period 19752000. Overall, the results support a positive and statistically significant equilibrium relation between financial development and economic growth for all different financial indicators that we test for and in all groups of countries. The supply-leading hypothesis is also supported by more recent studies by Calderon and Liu (2003) on 109 developing and developed countries, and Christopoulos and Tsionas (2004) on 10 developing countries. Both studies conclude that the supply-leading hypothesis is the dominant force behind the relationship between
34
finance and the sources of growth; in particular, financial depth contributes more to the causal relationship in developing countries.
3. Model Specification, Data and Methodology

We employ dynamic time series methods for relationship between financial development and economic growth. Following the literature, such a model may specified as: GDPt = a + FDt + t (1) where GDPt is the natural logarithm of real GDP (fixed at 1987 prices); FDt , is the measure of finacial development that is represented by the domestic credit provided by banking sector (percent of GDP), and t is error term. The quarterly time series data are taken for period 1986:1-2006:4 from the Central Bank of the Turkish Republic electronic data delivery system. All series seasonally adjusted to remove the seasonal effects by using Census X-12 quarterly seasonal adjustment method. The empirical investigation for the relationship between financial development and economic growth will be performed in two steps. First, we define the order of integration in series and explore the long run relationships between the variables by using unit root tests and cointegration test, respectively. Second, we test the long-run or short-run causal relationship between financial development and economic growth is carried out in a VECM or VAR framework. 3.1. The order of Integration Because the order of integration of a time series is of great importance for the analysis, we use the Augmented Dickey-Fuller (ADF) unit root test to examine for the stationarity of variables. The regression forms of the ADF unit root test below:
yt = yt 1 + i yt i + t
i =1 k
(2) (3) (4)
yt = a0 + yt 1 + i yt i + t
i =1
yt = a0 + yt 1 + a2t + i yt i + t
i =1
where a0 is intecept, t is linear time trend, k is the number of lagged first differences, and t is error term. The null hypotesis is unit root and the alternative hypothesis is level stationarity. The , and
T statistics are all used to test the hypotheses = 0 . Dickey and Fuller (1981) provide three additional F-statistics (called 1 , 2 and 3 ) to test joint hypotheses ( = a0 = 0 , a0 = = a2 = 0 and = a2 = 0 , respectively) on the coefficients (Enders, 2004: 183). If the coefficient of the lag of yt 1 ( ) is
significantly different from zero, then the null hypotesis is rejected. 3.2. Cointegration Test and Long-Run Causal Relationship Estimation of cointegration vectors and testing for long-run causal relationships in the context of errorcorrection representation of cointegrated variables is conducted using the Johansen (1988) and Johansen and Juselius (JJ) (1992) procedure. Following the maximum likelihood approach of Johansen (1988), a vector error correction (VECM) representation of a VAR(p) model can be written as:
Z t = Z t 1 + 1 Z t i + + Et
i =1
p +1
(5)
where Zt is a nx1 vector composed of non stationary variables, and are nxn matrices of coefficients, is a set of deterministic variables such as constant, trend, and dummy variables, and Et
35
is a vector of normally and independently distributed error terms. The rank of the matrix gives the dimension of the cointegrating vector. If its rank, r is (0<r<n), then can be decomposed into =' where , are nxr matrices containing the adjustment coefficients and the cointegrating vectors respectively. Hence, model (5) reduces to:
Z t = ' Z t 1 + 1 Z t i + + Et
i =1
p +1
(6)
This means 'Zt-1 contains all the long-run information on the process of Zt. Specifically, the rows of ' are interpreted as the distinct cointegrating vectors and the rows of are loading factors which indicate the speed of adjustment of the dependent variables towards the long-run equilibrium state. A test of zero restrictions on the (=0) is, therefore, a test of weak exogeneity (see Johansen, 1992) and as shown by Hall and Milne (1994) weak exogeneity in a cointegrated system is a notion of long-run causality. Hence, a bivariate causality, say between, economic activity (GDP) and financial development (FD), can be seen by rewriting model (6) in the following equivalent form. GDPt p 1t GDPt p 1 11 12 GDPt i 1 (7) FD = + FD + + [ 1 2 ] FD + t p i =1 21 22 t t i 2 t2 There are three possible cases of causality testing: If the 1 =0 is not rejected then FD does not cause GDP in the long run; similarly if the null 2 =0 is not rejected GDP does not cause FD in the long run. Likewise rejection of the null 1 =0 and 2 =0 means there is a bidirectional causal relationship between the two variables. 3.3. VAR Framework and Short-Run Causal Relationship Absence of a cointegration between two variables or working two stationary variables, two variable VAR is specified as: GDPt a10 p 1 a11 a12 GDPt i 1t (8) FD = a + a + t 20 i =1 21 a22 FDt i t 2 Here the lagged value of FD plays role in the determination of GDP. Thus, FD is said to Granger cause GDP.The hypothesis that FD does not Granger cause GDP could be tested simply by running the regression of GDP on lagged values of GDP and FD, and examining whether the coefficient of latter variable is significantly different from zero.
4.1. Unit Root Tests Results In this study, the Augmented DickeyFuller (ADF) unit root test is employed for the order of integration of variables in our models. It has been observed that the size and power properties of the ADF test are sensitive to the number of lagged terms (k) used. Several guidelines have been suggested for the choice of k (see Ng and Perron, 1995). The optimal lags for unit root tests are to include lags sufficient to remove any serial correlation in the residuals. k is determined according to the recursive tstatistics procedure proposed by Hall (1994). Following Enders, we start with three years of lags (k=12), with significance determined at 5% level of asymptotic normal distribution. After optimal lag selections, we use the approach of Dolado et al. (1990) for the augmented Dickey-Fuller (ADF). They suggest a procedure to test for a unit root when the form of data-generating process is completely unknown (Enders, 2004: 192). Table 1 presents the ADF unit root test results derived for the order of integration. Based on the results, the null hypothesis of unit-root cannot be rejected at levels but it is strongly rejected at the 1% significance level at their first difference. So we conclude that real economic activity (GDP) and financial development indicator (FD) series are I(1).
36
Table 1:

ADF Unit Root Test Results
Series Level GDP FD First Differences (GDP) (FD) Critical values at 1% Critical values at 5%
Notes: The
T
-2.95 -2.67
3
4.36 3.56
k 0 5
-0.13 -2.49 -5.91 -2.84 -3.51 -2.89
1
6.09 3.19 17.44 4.03 6.70 4.71
k 4 5 3 4
-1.29 -2.83 -2.60 -1.95
k 5 4
and T are the ADF test statistics for the coefficient of the lag of
-4.04 -3.45
8.73 6.49
yt 1 . F-test statistics, 1 and 3 , are to test joint hypotheses
= a0 = 0 and = a2 = 0 , respectively. k is number of lags.
4.2. Cointegration Test Results After defining of the order of integration, we can explore the long-run relationship between GDP and FD that the two variables are integrated of the same order. Table 2 reports the result of cointegration test using the maximum likelihood approach of Johansen (1988) and JJ (1992). The maximal eigenvalue and the trace statistic tests for each of the variables are reported in the table 2. The null hypothesis is that there is no cointegrating vector, and the alternative is there is one cointegrating vector. Since the validity of Johansens cointegration estimation technique is based on an assumption of white noise errors, the selected lag lengths represent the minimum lag length for which there is no significant autocorrelation in the estimated VAR residuals. There are several criterions to select lag lengths: sequential modified LR test statistic (LR), final prediction error (FPE), Akaike information criterion (AIC), Schwarz information criterion (SC), Hannan-Quinn information criterion (HQ). Besides this, the specification of the VAR models is made based on a number of diagnostic tests. The Lagrange Multiplier (LM) technique is used to determine whether the residual of the model approximates white noise. The results in Table 2 reveal that both the trace and maximum eigenvalue tests derived several cointegration tests according to different lag lengths. We cannot reject the null hypothesis of zero cointegrating vectors in favor of one cointegrating vector under investigation at the conventional 5 per cent significance level or better. These results provide no evidence of a long-run equilibrating relationship between the two variables.
Table 2: Johansen Cointegration Test Results
Trace statistics 3.8652 0.0442 3.2580 0.1105 11.0565 0.3042 10.7381 1.3528 Critical value at 5% 15.4947 3.8415 15.4947 3.8415 15.4947 3.8415 15.4947 3.8415 Maximum eigenvalues 3.8210 0.0442 3.1375 0.1105 10.7523 0.3042 9.3853 1.3528 Critical value at 5% 14.2646 3.8415 14.2646 3.8415 14.2646 3.8415 14.2646 3.8415 k 1 SC 2 HQ 6 LR 7 FPE, AIC LM(1) pvalues 0.0114 0.6326 0.1201 0.3387 LM(4) pvalues 0.0130 0.0146 0.9127 0.8833
Cointegrating Vectors r=0 r1 r=0 r1 r=0 r1 r=0 r1

Notes:
k is number of lags and its criterion is at lower line. LM(1) and LM(4) are serial correlation tests for first and fourth order autocorrelation of the VAR residuals and p-values are under the null of no serial correlation.
4.3. VAR Framework and Short-Run Causal Relationship Because of no cointegration relationship between real GDP and financial development, the next task is to examine the direction of causality using the VAR models. The short run effect of the financial intermediary development on economic growth is analyzed using several Granger causality tests. Results from the Granger causality test also indicate the validity of four hypotheses that have been
37
analyzed extensively in the literature: The supplyleading hypothesis, the demand-following hypothesis, the bidirectional causality hypothesis, and the independent hypothesis. Table 3 reports the Granger causality test results where H01 and H02 are the null hypothesis for the supplyleading hypothesis and the demand-following hypothesis, respectively. We explore the causality between growth of real GDP (DGDP) and change in financial development indicators (DBC and DPC) up to three years of lags (k=12). For the causality between financial development indicators and economic growth, there is a one-way causal relationship from the change of domestic credit provided by banking sector (DFD) to the growth of real GDP capita (DGDP) up to three years of lags. But there is no a one-way causal relationships from growth of real GDP (DGDP) to change in financial development indicators (DBC and DPC) at all lagged models.
Table 3: Granger Causality Test Results
k 1 2 3 4 5 6 7 8 9 10 11 12 H01 F-Statistics [P-Values] 12.42 [0.0007] 6.25 [0.0031] 4.25 [0.0080] 2.86 [0.0297] 2.25 [0.0590] 2.45 [0.0337] 2.23 [0.0434] 1.79 [0.0981] 1.55 [0.1537] 1.36 [0.2200] 1.79 [0.0981] 1.19 [0.3147] H02 F-Statistics [P-Values] 0.06 [0.8061] 0.73 [0.4834] 2.35 [0.0790] 0.92 [0.4556] 0.93 [0.4690] 0.76 [0.6008] 0.86 [0.5415] 1.01 [0.4354] 0.93 [0.5045] 0.88 [0.5584] 0.93 [0.9285] 0.73 [0.7120]
Notes:
H01: DFD does not Granger Cause DGDP; H02: DGDP does not Granger Cause DFD. k is number of lags.
5. Conclusion
In this paper, finance-growth nexus literature survey and the direction of causality between the financial development and economic growth is investigated for Turkey for the period 1986:1-2006:4. The main findings of the paper can be summarized as follows: First, the results show one-way causality from financial development to the economic growth. Second, the results do not provide evidence of a long-run causal relationship between financial development and economic growth in Turkey. In other words, Granger causality test results show that financial development leads to economic growth and support the supply leading hypothesis for Turkey. Thus, the domestic credit provided by banking sector and healthy banking sector has been assumed to contribute to the growth of the Turkish economy.
38
International Research Journal of Finance and Economics - Issue 11 (2007) Al-Zubi, K., Al-Rjoub, S. and Abu-Mhareb, E. (2006). Financial development and economic growth: a new empirical evidence from the Mena countries, 1989-2001, Applied Econometrics and International Development, 6, 3-11 Apergis, N., Filippidis, I. and Economidou, C. (2007). Financial deepening and economic growth linkages: a panel data analysis, Review of World Economics, 143, 179-98. Arestis, P. and Demetriades, P. (1996). Finance and growth: institutional considerations and causality, Department of Economics Working Paper No. 5, University of East London. Arestis, P. and Demetriades, P. (1997). Financial development and economic growth: assessing the evidence, Economic Journal, 107, 783799. Atje, R. and Jovanovic, B. (1993). Stock market and development, European Economic Review, 37, 623640. Beck, T. (2002). Financial development and international trade: is there a link?, Journal of International Economics, 57, 107-131. Bell, C. and Roussseau, P.L. (2001). Post-independence India: a case of finance lend industrialization, Journal of Development Economics, 65, 153-175. Bencivenga, V. and B. Smith (1991). Financial intermediation and endogenous growth, The Review of Economic Studies, 58, 195-209. Blackburn, K. and Hung, V.T.Y. (1998). A theory of growth, financial development and trade, Economica, 65, 107124. Calderon, C. and Liu, L. (2003). The direction of causality between financial development and economic growth, Journal of Development Economics, 72, 321334. Christopoulos, D.K. and Tsionas, E.G. (2004). Financial development and economic growth: evidence from panel unit root and cointegration tests, Journal of Development Economics, 73, 5574. Demetriades, P.O. and Hussein, K.A. (1996). Does financial development cause economic growth? time series evidence from 16 countries, Journal of Development Economics, 51, 387411. Demirguc-Kunt, A. and Maksimovic, V. (1998). Law, finance, and firm growth, The Journal of Finance, 52, 21072137. Dickey, D. and Fuller, W.A. (1981). Likelihood ratio statistics for autoregressive time series with a unit root, Econometrica, 49, 1057-1072. Dolado, J., Jenkinson, T., and Sosvilla-Rivero, S. (1990). Cointegration and unit roots, Journal of Economic Survey, 4, 249-273. Enders, W. (2004). Applied Econometric Time Series, 2nd Edition. John Wiley & Sons: USA. Fry, M.J. (1988). Money, Interest and Banking in Economic Development. The Johns Hopkins University Press, Baltimore, MD. Fry, M.J. (1995). Money, Interest and Banking in Economic Development, 2nd Edition. The Johns Hopkins University Press, Baltimore, MD. Goldsmith, R.W. (1969). Financial Structure and Development, Yale University Press, New Haven. Gupta, K. L. (1984). Finance and Economic Growth in Developing Countries. Croom Helm: London. Guryay, E., Safakli, O.V. and Tuzel, B. (2007). Financial development and economic growth: evidence from Northern Cyprus, International Research Journal of Finance and Economics, Issue 8, 57-62. Habibullah, M.S. (1999). Financial development and economic growth in Asian countries: testing the financial-led growth hypothesis, Savings and Development, 23, 279290. Habibullah, M.S. and Eng, Y. (2006). Does financial development cause economic growth? a panel data dynamic analysis for the Asian developing countries, Journal of the Asia Pacific Economy, 11, 377393.
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23]
International Research Journal of Finance and Economics - Issue 11 (2007) [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48]
39
Hall, A. (1994). Testing for a unit root in time series with pretest data-bsed model selection, Journal of Business and Economic Statistics, 12, 461-470. Hall, S. G. and Milne A. (1994). The relevance of p-star analysis to UK monetary policy, Economic Journal, 104, 597-604. Harrison, P., Sussman, O. and Zeira, J. (1999). Finance and growth: theory and new evidence, Federal Reserve Board Finance and Economics Discussion Paper No. 1999-35. Johansen, S. (1988). Statistical analysis of cointegration vectors, Journal of Economic Dynamics and Control, 12, 231-254. Johansen, S. and Juselius, K. (1992). Some structural hypothesis in a multivariate cointegration analysis of purchasing power parity and uncovered interest parity for the U.K., Journal of Econometrics, 53, 211-244. Jung, W.S. (1986). Financial development and economic growth: international evidence, Economic Development and Cultural Change, 34, 336-346. Khan, S.M. and Senhadji, A.S. (2000). Financial development and economic growth: an overview, IMF Working Paper, No.00/209. King, R.G. and Levine, R. (1993a). Finance and growth: Schumpeter might be right, Quarterly Journal of Monetary Economics, 108, 717-738. King, R.G. and Levine, R. (1993b). Finance, entrepreneurship, and growth, Journal of Monetary Economics, 32, 513-542. Kletzer, K. and Pardhan, P. (1987). Credit markets and patterns of international trade, Journal of Development Economics, 27, 27-70. La Porta, R., Lopez-de-Silanes, F., Sheleifer, A. and Vishny, R. (1997). Legal determinants of external finance, Journal of Finance, 52, 1131-1150. Levine, R. (1991). Stock market growth and tax policy, Journal of Finance, 46, 1445-1465. Levine, R. and Zervos, S. (1996). Stock market development and long-run growth, World Bank Economic Review, 10, 323339. Levine, R. (1997). Financial development and economic growth: views and agenda, Journal of Economic Literature, 35, 688-726. Levine, R. and Zervos, S. (1998). Stock markets, banks, and economic growth, American Economic Review, 88, 537-558. Levine, R. (2003). More on finance and growth: more finance more growth, Reserve Bank of St. Louis Review, 85, 31-46. Levine, R., Loayza, N. and Beck, T. (2000). Financial intermediation and growth: causality and causes, Journal of Monetary Economics, 46, 31-77. Luintel, K.B. and Khan, M. (1999). A quantitative reassessment of the finance-growth nexus: evidence from multivariate VAR, Journal of Development Economics, 60, 381-405. McKinnon, R. (1973). Money and Capital in Economic Development, Brookings Institution, Washington, DC, USA. Murinde, V. and Eng, F.S.H. (1994). Financial development and economic growth in Singapore: demand following or supply-leading?, Applied Financial Economics, 4,391404. Ng, S. and P. Perron (1995). Unit tests in ARMA models with data dependent methods for the selection of the truncation lag, Journal of the American Statistical Association, 90, 268-281. Pagano, M. (1993). Financial markets and growth: an overview, European Economic Review, 37, 613-622. Patrick, H.T. (1966). Financial development and economic growth in underdeveloped countries, Economic Development and Cultural Change, 14, 174189. Saint-Paul, G. (1992). Technological choice, financial markets and economic development, European Economic Review, 36, 763-781. Schumpeter, J.A. (1911). The Theory of Economic Development. Cambridge, Mass.: Harvard University Press.
40 [49] [50] [51]
International Research Journal of Finance and Economics - Issue 11 (2007) Schumpeter, J.A. (1934). The Theory of Economic Development, Oxford: Oxford University Press. Shaw, E.S. (1973). Financial Deepening in Economic Development, Oxford University Press, New York. Unalmis, D. (2002). The causality between financial development and economic growth: the case of Turkey, The Central Bank of the Turkish Republic, Research Department Working Paper, No:3, June.
Volatility Clustering in the Greek Futures Market: Curse or Blessing?

Nikolaos L. Hourvouliades Department of Economics, Aristotle University of Thessaloniki And Anatolia School of Business, American College of Thessaloniki, 55510 Pylaia, Greece E-mail: hourvoul@act.edu Tel: +30 6944368333; Fax: +30 2310530932 Abstract This paper examines the existence and nature of volatility clustering phenomena in the Athens FTSE20 index futures contract. An issue that permanently bewilders researchers, volatility clustering or persistence has been found present in most financial markets across the world. The purpose of this analysis is to offer an in-depth analysis of the characteristics of clustering in the domestic derivatives market. Our analysis shows that the time series of the Greek futures market exhibit volatility clustering, with negative shocks being more persistent than positive ones. The paper also applies GARCH-type, regression and exponential smoothing models in order to compare their forecasting power on volatility. The results show that GARCH models, albeit their heteroskedasticity-healing power do not prove their superiority, especially to the ES models that manage to give superior forecasts. In all, despite the fact that volatility clustering and many of its attributes are known and explored to a satisfactory level, there seems to be limited benefits in everyday trading, making its presence more of a curse than a blessing in forecasting the market. Keywords: Volatility clustering, GARCH models, exponential smoothing, forecasting Jel Classification Codes: C10; C53; G10; G13
1. Introduction
The issue of volatility clustering has been in the scope of researchers for a considerable time, but there seems to be still a long way to the cure. Financial time series of various assets have been consistently followed, including currencies, stocks, fixed income and commodities, both in the cash and derivatives markets. Nevertheless, research on the volatility characteristics of the Athens Derivatives Exchange (ADEX) has been limited, since there are few empirical applications dealing with volatility and its forecasting, with literature concentrating mainly on the spillover effects between futures and spot prices. The purpose of this paper is to examine, first, the nature of domestic volatility and, second, to compare the efficiency of different forecasting techniques. In particular, this study will examine the existence of volatility clustering, also known as persistence or GARCH-effect, and whether the negative and positive shocks influence in the same fashion the market. The time horizon rationale is the medium to long-term range, that mainly includes the actions of portfolio makers and hedging strategists, thus making daily observations most appropriate. On the contrary, high-frequency intraday observations seem to support mainly short-term speculative type investors, whereas weekly or monthly observations support extra long-term policy makers that follow the waves of the markets. Another
42
significant feature of our analysis is that of the cost-benefit involved in the methodology process: it is vital that the data are easily collected, the calculations easily verified and the results applicable in practice by market professionals. Research in the Greek market has mainly concentrated in the cash equities market and, later, in their relationship with the futures market. A first breakthrough paper was that of Koutmos, Negakis and Theodossiou (1993) dealing with the stochastic behavior of stock prices in the Greek market. That paper takes weekly data for a period of ten years and by applying the, back then, brand new EGARCHM model concludes that volatility depends asymmetrically on past shocks. Surprisingly enough and contrary to later research, they suggest that positive shocks have a greater impact on volatility than negative ones. Apergis and Eleptheriou (2001) also study the spot market of the Athens Stock Exchange, over a span of almost a decade, and find asymmetry in stock returns and strong persistence of shocks into volatility that is proven through a quadratic GARCH model. Phylaktis et al. (1999) in their study of the domestic market, suggest that volatility is not affected by the daily price limits; using ARCH modeling they show that price limits mainly act as a slow-down process and do not make investors reassess their behavior and reduce volatility. In their paper on the Athens Derivatives Exchange, Kavussanos and Visvikis (2004b) find the series of the spot and futures market cointegrated CI(1,1), and apply various models to forecast futures behavior. Hourvouliades and Kousenidis (2006) also examine the cash stock indices and their respective futures contracts by taking daily data, giving evidence that the series of the cash and futures market are cointegrated, with one-way causality existing from the futures to the cash market. This makes the study of futures volatility even more meaningful in a way that since this market paves the way, any additional knowledge of volatility behavior can enhance portfolio selections. This price discovery function of the Greek futures market is in agreement with the findings of Kenourgios (2004), Kavussanos and Visvikis (2004b) and Alexakis et al. (2001). Finally, in order to minimize the standard deviation of the constructed portfolio, Floros and Vougas (2004) get the best estimates of hedge ratios by applying M-GARCH models. A comprehensive study on forecasting volatility is that of Poon and Granger (2001) who offer an all inclusive survey of ninety three different articles dealing with volatility in various markets. These papers include different approaches in almost all aspects: various measures (proxies) for volatility, different frequencies and time span of data and various forecasting models. The study concentrates on forecasting volatility and divides research into two major sub-categories: those who use historical prices for forecasting and those who prefer implied volatility in option prices. Our study will concentrate on historical standard deviation of daily returns as a proxy for market volatility. Locke and Sarkar (2001) examine futures from CBOT and CME and find that the difference between the opening price and the closing price of the previous day accounts for most of the persistence in ARCH. Granger and Machina (2006) on the other hand offer a structural attribution to volatility clustering. Furthermore, Ghysels et al. (2006) perform a complicated mixed data sampling (MIDAS) regression to compare forecasting models, which vary in volatility measures, frequencies and lag lengths. Their study contradicts most of the accepted literature: they suggest that absolute returns and daily ranges are very good predictors instead of the widely accepted squared returns. Thomakos and Wang (2003) perform an in-depth analysis of realized volatility by taking intraday, 5-minute observations on four futures contracts for a time period of five years. They find that the logarithmic standard deviations exhibit long memory and approximate Gaussianity, in contrast with returns that exhibit no serial correlation. In addition, they find that volatility has a leptokurtic distribution and is highly skewed to the right. Bhar (2001) examines the Australian SPI futures contract and takes daily settlement prices for a period of nine years. The study applies a bivariate EGARCH model which suggests that the conditional mean and the conditional variance are better explained by its explanatory residual. Lux and Marchesi (1998) carry out a simulation in which they take for granted the existence of volatility clustering. In their effort to explain it they conclude that by applying chartist and fundamentalists strategies, one can explain an outbreak of volatility after a certain threshold value and then its return to tranquility.
43
From a forecasting point of view, Hansen and Lunde (2005) conduct a comparison of more than 330 ARCH-type models, on DM-$ and IBM stock, and find no evidence that a GARCH (1,1) is outperformed by more sophisticated models. This is also of major significance for the current study, since we are going to select only the GARCH and EGARCH models as representatives of the great ARCH family, and compare their forecasting ability with other less complicated models. Martens et al. (2004) select the sum of squared intraday returns as an unbiased measure of ex-post volatility for the S&P500 and ForEx index futures contracts, over a period of six years. They develop a nonlinear autoregressive fractionally integrated moving average model (ARFIMA) that successfully incorporates the leverage effect and yields efficient forecasts of volatility. Blair et al. (2001) take high-frequency returns of the UK index of S&P100 over a period of five years, and by using several ARCH-type models suggest a superiority of intraday observations in shortterm forecasts from one to twenty days. Martens (2002) seems to agree on the high-frequency significance of intraday observations compared to daily in forecasting. Ederington and Guan (2002) on the contrary use historical standard deviation on daily returns for a set of financial assets. Their findings suggest that GARCH puts too much emphasis on the recent observations and not enough on older ones, however the GARCH (1,1) model generally yields better forecasts than the rest of the models. Balaban et al. (2002) construct a set of weekly and monthly volatility measures based on daily observations, and perform a comparison among various forecasting models. Surprisingly enough, they find evidence that single exponential smoothing provides superior forecasts of volatility. Finally, there is a set of papers that focuses on the persistence of volatility in the markets. Chen et al. (2006) select various futures contracts and examine how different volatility proxies influence their findings. They suggest that when intraday high-frequency data are taken, they seem to yield results of higher persistency than daily ones. They also find that the type of GARCH model used for conditional volatility also influences results. Areal and Taylor (2002) use high-frequency five minute data for fifteen years on the FTSE100 index, and find persistent positive autocorrelation that displays long-memory effect. It becomes obvious that there is a wide range of different approaches to volatility modeling and forecasting. As stated before, this papers main goal is to provide useful insights in the Greek futures market, keeping in mind the cost-benefit factor. This papers intention is dual: to be as efficient as possible on the one hand, and produce robust results on the other. Under this rationale, daily closing prices are preferred, that are easy and quick to collect for anyone, for a length of five years, and calculations will be based on the historical volatility measure. Likewise, the models used in the end for forecasting volatility, are selected on a cost-reduction and dependability basis, i.e. regression, exponential smoothing and GARCH.
2. Data and Methodology

The futures contract used in this article is the Athens FTSE20, which is the composite index of the twenty major companies of the Athens Exchange in terms of capitalization and trading volume. This index contract is preferred to the mid-cap Athens FTSEMid40, mainly due to its significantly larger liquidity and trading volume, which represents the majority of all trading in the domestic futures market. The data set takes the daily closing price of the futures contract, for the period from January 2002 to November 2006. The total number of observations in the sample is 1223. Due to the monthly maturity effect, the study rolls over to the next closest contract one day prior to maturity. The paper is organized as follows: first, preliminary statistics for the return and variance of the data are conducted, second, a detailed analysis of the volatility follows, and third, there is a comparison of the forecasting efficiency of future volatility of five basic models: simple regression, single exponential smoothing, Holt-Winters multiplicative exponential smoothing, GARCH (1,1) and finally EGARCH (1,1).
44
The data taken in the study are the daily closing prices of the futures contract of the Athens FTSE20, big capitalization stock index. Then, the continuously compounded returns are calculated as follows: P Rt = ln t Pt 1 (1) where Rt is the daily return, Pt and Pt-1 are the closing prices of the index future contract on trading day t and t-1, respectively, and ln stands for the natural logarithm. Apart from this geometric return, there is the simple arithmetic return, used for simplicity by many researchers, i.e. Apergis and Eleptheriou (2001), and is given by the following ratio: P P rt = t t 1 Pt 1 (2)
Nevertheless, for small changes, the simple arithmetic and log returns approximate each other. Table 1 shows that the characteristics of both series are quite similar. As a result, this study will use formula (1) for the calculation of volatility. Having named the log return Rt, we define the dailyrealized volatility as the standard deviation of continuously compounded daily returns as follows:
n 1 where the mean daily return stands for R = (1 n ) R t

t =1 n
t =
( R R)
t =1 t
(3)
(4) This measure is judged as the most efficient when daily observations are taken. The study will also take into consideration an alternative volatility measure, the natural logarithm of the realized volatility ln(t) in an effort to smoothen the data and approximate normality, in a way similar to Andersen et al. (2003) and Martens et al. (2004). In general the volatility proxy measure varies, depending on the data frequency: some researchers such as Andersen and Bollerslev (1998) argue that, ideally, the daily volatility follows a continuous time diffusion process, whereas Thomakos and Wang (2003) use 5-minute intervals. Locke and Sarkar (2001) use a volume-weighted average of all expirations of a particular contract, while Bluhm and Yu (2000), Taylor (2004), Balaban et al. (2002), Vilasuso (2002), Yu (2002), Ederington and Guan (2005) prefer the daily squared returns as a proxy for volatility. More specifically, daily square returns are simply calculated as: t = ( Pt Pt 1 ) 2 (5) that is, the squared difference of the daily price change. Another proxy for volatility is the low-high range of values occurring within a daily session, for instance the Parkinson Range PARK-R daily volatility estimator, but it seems fit only for short spans of time, i.e. 3-6 months, instead of the fiveyear period of this study. Poon and Granger (2002) in their extensive review of ninety three papers dealing with volatility in financial markets, they show that the vast majority prefers the standard deviation of daily returns as an unbiased estimator of volatility. Additionally, it shows that a trade-off between observation frequency and time span exists, ranging from 5-minute to weekly intervals and one month to ten years duration, thus making the present paper robust through its daily observations over a five year period. In general, bearing in mind the cost-benefit ratio, the daily closing prices are preferred to the vastly costly, or even unavailable, 5 or 10 minute sampling, or high to low intraday range.
45
3. Preliminary Statistics
The preliminary statistics find that the logarithmic daily returns approximate normality to a great extend. The Ljung-Box Q-statistic reported for 12 lags, is found negative for autocorrelation, showing that returns are serially uncorrelated. The values are very similar for the simple arithmetic returns showing that it is safe to apply log returns in following calculations. On the other hand, the realized standard deviations do not exhibit a Gaussian behavior, even if the logarithmic values of standard deviation are taken. More precisely, the standard deviation series exhibit high kurtosis and are extremely skewed to the right, as in Thomakos and Wang (2003), whereas the returns show an almost ideal bell-shaped curve (see figure 1). Furthermore, the Ljung-Box statistic for 12 lags is found significant, showing patterns of autocorrelation in standard deviation. This is a positive sign that volatility has memory and previous observations affect future ones, that is, periods of large volatility are followed by large ones and those of small volatility by small ones.
Table 1: Summary statistics
Log returns 0.000404 0.000602 0.05866 -0.066465 0.013229 -0.062124 4.710392 149.8623 13.331 Simple returns 0.000492 0.000602 0.060421 -0.064304 0.013232 0.011362 4.705833 148.3082 13.404 Volatility 0.000283 0.000211 0.001912 1.83e-07 0.000251 1.724312 7.249988 1526.477 168.38* Log Volatility -8.630303 -8.464030 -6.259555 -15.51343 1.119457 -1.090349 5.152291 480.5852 36.710*
Mean Median Maximum Minimum Standard deviation Skewness Kurtosis JB Q(12)
*Statistically significant at the 5% level
Beyond the basic statistical measures, the graphical representations of the series are shown in Figure 1. The volatility clustering effect can be identified easily at the graphs: there are periods of high turbulence with many high peaks clustering together and there are periods of quiescence where volatility stays low. The Kernel density graphs show the Gaussian-like distribution of the returns in contrast with volatility that has a skewed distribution. As far as the volatility measures are concerned, it shows that taking the logarithmic value of the arithmetic volatility slightly changes the distribution; skewness and kurtosis still exist to marginally lower degree and normality is barely existing in the Kernel density graph (see Figure 1, graphs S and log S, respectively).

Figure 1: Graphical representations
46
4. Forecasting volatility
Forecasting measures The current study has used a set of forecasting indicators in order to compare the power of the various forecasting techniques applied. The first indicator is the Mean Absolute Error (MAE) and is equal to: 1 T MAE = Yt s Yt T t =1 Another useful measure is the Root Mean Absolute Error (RMSE), which is a relative indicator but still depends on the units of measurement of the variables:
RMSE =
1 T (Yt s Yt )2 T t =1
47
An important relative indicator is the Mean Absolute Percentage Error (MAPE) which has the advantage of not depending on the units of measurement of the variables and thus will be the main indicator for forecasting efficiency. It is calculated as follows: 1 T Y s Yt MAPE = t T t =1 Yt All afore mentioned indicators should be as small as possible, showing small differences between pairs of observed values and forecasts. Finally, Theils U is practically a compromise between absolute and relative measures, that usually takes values 0<U<1, where U=0 indicates a perfect fit. The most significant feature of Theils U is its three components that break up the forecast error and help give a better understanding of its nature: 1. The bias proportion, that shows the difference between the means of the two series, forecast and observed ones. It should be as small as possible. 2. The variance proportion, which shows the difference between the variance of the two series. It should be as small as possible. Both these measures represent the systematic error. 3. The covariance proportion that represents the unsystematic error. The higher this value the better the forecast. Regression Model
The simple linear regression model is the first to be evaluated, representing the simplest and most nave way of forecasting. This is a simple autoregression given by the following model: t2 = c + t21 + t 1 (6) which shows that todays volatility is a function of yesterdays observed volatility and a forecast error that follows : iid (0, ) .
Exponential Smoothing
The article will use two exponential smoothing techniques: single and Holt-Winters multiplicative. The single exponential smoothing, SES, gives greater weight to the recent observations but also keeps in mind all previous ones. SES calculates future volatility as a weighted function of the previously forecast value plus the continuously updated forecast error: Ft = Ft 1 + a ( t21 Ft 1 ) = a t21 + (1 a ) Ft 1 (7a) where a is the smoothing constant, and 0<a<1. The speed at which older values are smoothed depends on a: for high values of a (a 1) there is a greater weight on the previous observation, while low values of a put more emphasis on the forecast value and gives considerable smoothing. When a=1 then SES yields a forecast that is equal to the last Yt observation. The Holt-Winters multiplicative, HWM, allows for seasonality in the data. The rationale seems attractive: futures contracts have a one-month cycle, which may imply seasonality and trend in the data, thus giving a hope that an adaptive model like HWM may be able to yield superior evaluations compared to the simple exponential smoothing. The model is calculated as follows: Ft + k = ( t + bt k )ct + k (7b) where Ft+k is the forecast k periods ahead, is the intercept, is the trend and ct is the multiplicative seasonal factor. They are calculated as follows: yt a(t ) = a + (1 a)(a(t 1) + b(t 1)) ct (t s ) b(t ) = (a (t ) a(t 1)) + (1 )b(t 1) y ct (t ) = t + (1 )ct (t s ) a(t )
48
In the above system, , , are the smoothing factors and they are all smaller than 1 (,,<1), and s is the seasonal frequency (seasonal cycle). Garch (1,1) The standard symmetric GARCH (p,q) model of Bollerslev (1986) for conditional volatility takes into consideration lagged values of past volatility: Yt = a0 + a1 X t + t t I t 1 : N (0, t2 ) (7c)
(8) with (7) being the conditional mean equation and (8) the conditional variance equation. In other words, the conditional volatility today depends on yesterdays squared forecast error and yesterdays conditional volatility. In our study we are going to use the standard GARCH (1,1) model: t2 = a0 + a1 t21 + 1 t21 (9)
i =1 j =1
t2 = a0 + ai t2i + j t2 j
that goes one period back, which in our case is a lagged value of one day.
Egarch (1,1) One basic characteristic of standard GARCH models is their assumption that good and bad news have a symmetrical effect on volatility. This fact is usually violated in many financial time-series, the socalled Leverage Effect, Black (1976), and in the case of the Athens futures market this seems to be occuring. The basic advantage of EGARCH models is that they do not hold the assumption of symmetrical effect on volatility, allowing for different handling of good and bad news. Our previous analysis initially suggests that there is serial correlation, thus persistence, in the series, and we will further show if negative news are more persistent on volatility than positive ones. The exponential general autoregressive conditional heteroskedastic model of Nelson (1991) also caters for the effect of time, as old values gradually wear off and recent ones influence more the forecast value. Volatility in the GARCH model is an additive function of the lagged error terms, t, whereas in the EGARCH it is a multiplicative function of lagged innovations that can react asymmetrically to good and bad news. The model we apply is the following EGARCH (1,1): ln( t2 ) = t + ln( t21 ) + a t 1 2 + t 1 t 1 t 1 (10) The degree of persistence of the conditional volatility is different under the last two models: in the GARCH (1,1) it is given by 1+1, whereas in the EGARCH (1,1) it is equal to .
Table 2 summarizes the equation outputs for the five different methods. The outputs are very similar for the two exponential smoothing techniques, as well as for the two GARCH family ones. Starting from the exponential smoothing methods, it is shown that the estimated values of a are very low, close to 0.05, which means that there is more emphasis put on the forecast Ft-1 instead of the actual observed t-1. This means that there is a considerably high level of smoothing in the estimated series. On the other hand, the GARCH type models have a slightly bigger difference in their estimated values. In particular, the volatility persistence in the GARCH (1,1) is equal to a+=0.99 and in the EGARCH (1,1) is equal to = 0.965. This translates to marginally higher levels of persistence calculated by the first model, however their combination safely lead to the conclusion that there are indeed high level of volatility persistence in the Greek futures market. The efficiency of the forecasting methods is given in table 3, which gives a detailed overview of the estimated forecasting methods. The exponential smoothing models offer a superior forecast according to all measures; comparing in between the single exponential smoothing and the Holt-
49
Winters multiplicative, it seems that there is very small difference, with SES yielding better results. The regression method has little explaining power. Likewise, the GARCH type models fail to surpass the forecasting efficiency of the exponential ones, in agreement with Balaban et al. (2002) conclusions. Another significant outcome of this study comes from the analysis of the normality of the residuals, shown in figure 2, where the residuals of all estimated techniques are graphed, with the help of the theoretical Quantile Quantile method, which proves extremely helpful: the straight line in the middle of the box represents the exact normal distribution; any deviation from the line shows a persistent higher effect of the relative difference, positive or negative. A closer look into all methods leads to the conclusion that there is an asymmetry to the effect that positive or negative news have in the market: negative news, thus negative shocks in the market, seem to have a more persistent, higher deviation from normality.
Table 2: Estimated equations under each model
Output Ft = 0.000273 + 0.0339t-1 Ft = 0.948Ft-1 + 0.052t-1 Ft = 0.943 Ft-1 + 0.057t-1 0= 7.42e-10 1= 0.0632 1= 0.9271 t= -0.6321 = 0.0566 = 0.9650 = 0.0685
Regression Single Exponential HWM GARCH (1,1)
EGARCH (1,1)
Table 3:
MAE RMSE MAPE Theils U Bias Variance Covariance
Forecasting measures
Regression 0.000189 0.000251 438.2198 0.37965 0.0000 0.9979 0.0021 SES 0.0000242 0.0000308 8.414 0.05295 0.02400 0.34495 0.63105 HWM 0.0000349 0.0000410 12.045 0.07383 0.6390 0.0624 0.2986 GARCH 0.000183 0.000253 388.9106 0.40095 0.0146 0.9831 0.0023 EGARCH 0.000181 0.000252 394.2912 0.39823 0.0116 0.9860 0.0024
Further examination of the volatility residuals under each method, offers interesting insights in the Greek markets characteristics. The simple linear regression graph is probably the only one that seems to support that positive deviations are as important as negative ones. In fact, positive shocks after some point seem to deviate a lot from Gaussianity. On the contrary, the exponential smoothing techniques are showing a different situation: residuals are closer to the normality central line (due also to the better forecast efficiency) and deviations are mainly found in the negative side of the graph. Thus, negative shocks to the market seem to be more powerful and move the forecasting system out of balance to a greater degree. Finally, the two GARCH models yield an almost identical graph: there are deviations in both sides (positive and negative) but negative ones clearly show to be of greater size. The respective correlograms estimated under each forecasting method agrees to the above conclusions. The Ljung-Box Q statistic for serial correlation estimated for each situation is given in table 4:
Table 4:
Q(12)
the Ljung-Box Q-statistic for serial correlation

REG 147.82* SES 10.939 HWM 2763.4* GARCH 36.710* EGARCH 36.710*
* Statistically significant for lag=12
The Q-Q graphs of the residuals normality are as follows:

Figure 2: Q-Q graphs of normality of residuals
8 6 Normal Quantile
Normal Quantile 4 3 2
50
4 2 0 -2 -4 -.0005
1 0 -1 -2 -3 -4
.0000
.0005
.0010
.0015
.0020
-5 -2.0E-19
0.0E+00 RESID_SES
2.0E-19 3.0E-19
RESID_REGRESSION
4 3
4 2 Normal Quantile 0 -2 -4 -6 -8 -8 -6 -4 -2 0 2 4 RESIDGARCH
2 Normal Quantile 1 0 -1 -2 -3 -4 -.00008 -.00004 .00000 .00004 .00008 .00012 RESID_HWM

4 2 Normal Quantile 0 -2 -4 -6 -8 -8 -6 -4
-2
RESIDEGARCH
6. Conclusions
The papers contribution is multiple in the domestic futures literature. First volatility clustering is found, with returns following a normal distribution, and volatility not. Second, exponential smoothing seems to offer superior forecast efficiency, despite the more sophisticated GARCH models. Third, the single exponential smoothing method is offering a better explanation than the seasonal HWM method, showing that the market has no seasonal trends. GARCH type models have similar results, showing the persistence of volatility and that serial correlation decays slowly. Nevertheless, the EGARCH model was not able to show its superiority in forecasting non-symmetric effects. On the explanation of the markets behavior it is proved that negative shocks seem in general to be more powerful and persistent. Beyond the quantitative measures, the qualitative ones, especially the Theils U, show the breakdown of the systematic and unsystematic error. Further analysis should pose the question of the correct proxy to volatility, which is the standard deviation of returns and other measures, in order to better discover the markets characteristics and yield useful results for forecasting efficiency.
51
International Research Journal of Finance and Economics - Issue 11 (2007) Alexakis, P., Kavussanos, M. and Visvikis, I. (2002) An Investigation of the Lead-lag Relationship in Returns and Volatility Between Cash and Stock Index Futures: the Case of Greece, Conference Proceedings, European Financial Management Association, Annual Meeting. Apergis, N. and Eleptheriou, S. (2001) Stock Returns and Volatility: Evidence from the Athens Stock Market Index, Journal of Economics and Finance, 25(1): 50-61. Andersen, T. and Bollerslev, T., Diebold, F. and Labys, P. (2003) Modeling and Forecasting Realized Volatility, Econometrica, 71(2): 579-625. Andreou, E. and Ghysels, E. (2002) Detecting Multiple Breaks in Financial Market Volatility Dynamics, Journal of Applied Econometrics, 17: 579-600. Antoniou, A. and Holmes, P. (1995) Futures Trading, Information and Spot Price Volatility: Evidence from the FTSE-100 Stock Index Futures Contract Using GARCH, Journal of Banking and Finance, 19(1): 117-129. Areal, N. and Taylor, S. (2002) The Realized Volatility of FTSE-100 Futures Prices, The Journal of Futures Markets, 22(2): 627-648. Bhar, R. (2001) Return and Volatility Dynamics in the Spot and Futures Markets in Australia: An Intervention Analysis in a Bivariate EGARCH-X Framework, The Journal of Futures Markets, 21(9): 833-850. Balaban, E., Bayar, A. and Faff, R. (2002) Forecasting Stock Market Volatility: Evidence from Fourteen Countries, University of Edinburgh Center for Financial Markets Research, working paper. Blair, B., Poon, S-H. and Taylor, S. (2001) Forecasting S&P100 Volatility: The Incremental Information Content of Implied Volatilities and High-Frequency Index Returns, Journal of Econometrics, 105(1): 5-26. Bluhm, H. and Yu, J. (2001) Forecasting Volatility: Evidence from the German Stock Market, University of Auckland, working paper. Bollerslev, T. P. (1986) Generalized Autoregressive Conditional Heteroskedasticity, Journal of Econometrics, 31: 307-327. Chen, Z., Daigler, R. and Parhizgari, A. (2006) Persistence of Volatility in Futures Markets, The Journal of Futures Markets, 26(6): 571-594. Darrat, A., Rahman, S. and Zhong, M. (2002) On the Role of Futures Trading in Spot Market Fluctuations: Perpetrator of Volatility or Victim of Regret?, Journal of Financial Research, 25(3): 431-444. Ederington, L. and Guan, W. (2002) Is Implied Volatility an Informationally Efficient and Effective Predictor of Future Volatility?, Journal of Risk, 4(3). Floros, C. and Vougas, D. (2004) Hedge Ratios in Greek Stock Index Futures Market, Applied Financial Economics, 14: 1125-1136. Gallo, G. and Pacini, B. (2000) The Effects of Trading Activity on Market Volatility, The European Journal of Finance, 6: 163-175. Granger, C. and Machina, M. (2006) Structural Attribution of Observed Volatility Clustering, Journal of Econometrics, 135: 15-29. Ghysels, E., Santa-Clara, P. and Valkanov, E. (2006) Predicting Volatility: Getting the Most Out of Return Data Sampled at Different Frequencies, Journal of Econometrics, 131(1-2): 5995. Hansen, P. and Lunde, A. (2005) A Forecast Comparison of Volatility Models: Does Anything Beat a GARCH (1,1)?, Journal of Applied Econometrics, 20: 873-889. Hourvouliades, N. and Kousenidis, D. (2006) Stock Index Futures, Trading Volume and Open Interest: Efficiency and Forecasting in ADEX, 5th Annual Conference of the Hellenic Finance and Accounting Association.
References
[1]
[2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20]
International Research Journal of Finance and Economics - Issue 11 (2007) [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36]
52
Kavussanos, M. and Visvikis, I. (2004a) Market Interactions in Returns and Volatilities Between Spot and Forward Shipping Freight Markets, Journal of Banking and Finance, 28(8): 2015-2049. Kavussanos, M. and Visvikis, I. (2004b) The Predictability of Non-overlapping Forecasts: Evidence from the Derivatives Market in Greece, Hellenic Finance and Accounting Association Conference. Kenourgios, D. (2004) Price Discovery in the Athens Derivatives Exchange: Evidence for the FTSE/ASE20 Futures Market, Economic and Business Review, 6(3): 229-243. Koutmos, G., Negakis, C. and Theodossiou, P. (1993) Stochastic Behavior of the Athens Stock Exchange, Applied Financial Economics, 3: 119-126. Locke, P. and Sarkar, A. (2001) Liquidity Supply and Volatility: Futures Market Evidence, The Journal of Futures Markets, 21(1): 1-17. Lux, T. and Marchesi, M. (1998) Volatility Clustering in Financial Markets: A MicroSimulation of Interactive Agents, Quantitative Methods in Finance, Conference, Sydney. Martens, M. (2002) Measuring and Forecasting S&P500 Index-Futures Volatility Using HighFrequency Data, The Journal of Futures Markets, 22(6): 497-518. Martens, M., VanDijk, D. and DePooter, M. (2004) Modeling and Forecasting S&P500 Volatility: Long Memory, Structural Breaks and Nonlinearity, Tinbergen Institute Discussion Paper. Nelson, D.B. (1991) Conditional Heteroskedasticity in Asset Returns: A New Approach, Econometrica, 59: 347-370. Pericli, A. and Koutmos, G. (1997) Index Futures and Options and Stock Market Volatility, The Journal of Futures Markets, 17(8): 957-974. Phylaktis, K., Kavussanos, M. and Manalis, G. (1999) Price Limits and Stock Market Volatility in the Athens Stock Exchange, European Financial Management, 5(1): 69-84. Poon, S-H. and Granger, C. (2002) Forecasting Volatility in Financial Markets: A Review Taylor, J. (2004) Volatility Forecasting with Smooth Transition Exponential Smoothing, International Journal of Forecasting, 20(2): 273-286. Thomakos, D. and Wang, T. (2003) Realized Volatility in the Futures Markets, Journal of Empirical Finance, 10: 321-353. Vilasuso, J. (2002) Forecasting Exchange Rate Volatility, Economics Letters, 76(1): 59-64. Yu, J. (2002) Forecasting Volatility in the New Zealand Stock Market, Applied Financial Economics, 12: 193-202.
The Volatility of the Stock Market and News

Rohitha Goonatilake Department of Mathematical and Physical Sciences Texas A&M International University, 5201 University Boulevard Laredo, Texas 78041-1900 E-mail: harag@tamiu.edu Susantha Herath Department of Business Computer Information Systems St. Cloud State University, 720 Fourth Avenue South, St. Cloud Minnesota 56301-4498 E-mail: sherath@stcloudstate.edu Abstract The volatility of stock market indicators goes beyond anyones reasonable explanations. Industry performance, economic, and political changes are among the major factors that can affect the stock market. This paper focuses on the effect of news that surfaces throughout the day in the stock market. While there are many influences behind the constant changes in stock market performance, we can statistically analyze the influence of news on the DJIA, NASDAQ, and S&P 500. From the analysis of the data collected over a ten-week period of time, we were able to conclude that there is an association between news items and the market fluctuations, measured by an increase, a decrease, or unchanged. The association between the market fluctuations of the DJIA and crude oil stock prices is adequately explored to obtain a regression model in this paper. Keywords: News, market indicators, volatility, fluctuations, Black-Scholes option pricing formula, Markovian analysis Jel Classification Codes: A12, C02
1. Introduction
Market performances are not entirely data driven; with the exception of the simplest open-air markets, trade is rarely a face-to-face transaction between a producer and a consumer [1]. There are many factors that influence the market indicators on a given day. Rising investor confidence, fueled by soaring corporate profits and new market opportunities, are made possible by globalization and the Internet, generate nearly a decades worth of unprecedented growth in major US stock markets [2]. The stock market fluctuations exhibit several statistical peculiarities which are still waiting for a satisfactory interpretation [3]. News magazines such as Futurific interpret the importance of these for the average investor [4]. It is widely believed that news events dominate the markets on any given day. Due to the technological advancements of the times, the details of global and local events are presented at the forefront of market activities. One motivation for this study comes from the understanding that good news and bad news, when exposed to the market, determine whether to buy or sell the stocks. The preliminary study was conducted for a period of 10 weeks. The effect of the Lebanon-Israel war that started on July 23, 2006, was not considered in this study. However, we noted that the markets had
54
amazing flexibility after the initial impact of the war, and tried to discount the situation as quickly as possible. The fact that the market is able to adjust to some news events can be investigated as a separate component of this study. Following the observation period, we witnessed the market dramatically outperform to record levels, at the same time as oil prices, both crude oil and gasoline, were dramatically reduced. Risk assessment is considered a matter of great importance because it expands valuable resources when good events are observed [11]. One way to expand resources is to invest in stock markets. Investment in self-protection and self-insurance can affect the risk that people actually face within the market conditions, given that factors such as the influence of good news and bad news can transcend these limitations. Probabilistic risk analysis can be pursued along these lines, accommodating all the possible factors affecting the markets. A somewhat reverse implication has been given consideration in [5]. News stories about publicly traded companies are labeled positive or negative according to price changes in the company stock. The study seeks a statistical, probabilistic, or deterministic model that provides a justifiable interpretation of news for the market. Fluctuations occurring in the stock are believed to depend on the economic data as well as the measured nature of news items the market movers confront on a daily basis. The potential correlation between economic news and the stock market has been studied for the UK market [6]. In that study, the data set was collected for the month of November 2001. It consists of a corpus of 30 news items per day published on the Reuters news site between 9:00 am and 5:30 pm and the corresponding closing value of the FTSE 100 for the day. The corpus contains time stamped business news from Britain (UK), US, and Europe. We shall assume that these findings remain the same and carry forward across the pond. Good news, bad news, and the GARCH, Generalized Autoregressive Conditional Heteroskedasticity, model effects on stock return data and some other variations of them in the spirit of personal and organizational financial securities have been already studied [7, 8, 9, & 10]. Another study analyzed the emergence of news on the performance of the currency market [26]. Section 2 of this paper describes the methodology and approach. Section 3 presents an analysis of the stock market. Section 4 provides brief accounts of empirical studies. Section 5 provides a glimpse of current discussion taking place on the relevancies of this study to stock markets, personal wealth and tax cuts. Section 6 presents conclusions and future work.
2. Methodology and Approach

For this study, the opening and closing numbers of three prominent market indicators; the DJIA, NASDAQ and S&P 500, are monitored and recorded for a consecutive period of ten weeks beginning July 3, 2006. Media to Watch CNN Headline News Channel, Fox News Channel, www.yahoo.com, and The New York Times (web version-front page) are used as news outlets for identifying news items. The individual outlets are considered irrelevant in identifying these news items that are signaling good news (G), bad news (B), and that of neutral news (N) on a given day for this study. Types of Possible News Items Some of the main and sub news categories shown in Table 1 may be either interrelated, or not all mutually exclusive. However, this is not a substantial factor in the study because their overall impacts are ascertained in order to determine the market performances on a given day. Some categories have a greater impact on the market than the other. In a possible separate study, their contributions can be highly weighted compared to other categories in deciding the impact they have on market activities as suggested therein. Some items are measured objectively (viz., 1 & 3) whereas the rest is done
55
subjectively (viz., 2 & 3) considering their impact using a four point scale (e.g., 1-little, 2-somewhat, 3-sufficent, and 4-greater). The top ten most dominant or significantly important news items are identified as good, bad, or neutral news for the entire period of observation including during the daily operations of the market, holidays, and weekends. The website, www.cnn.com identifies the top ten news items based on those hit most in a single day. Most of the references cited in this article did not identify the news items as good, bad or otherwise individually. There was particularly no mention about them in their references too. We categorized the news items into groups on the basis of their impact on the general public. Overall, the good news items stay as good news items and the bad news items stay as bad except their measured impact on performing the market activities. Choosing news items is done by considering its news-worthiness and interest to the general public. There may also be a news item that, while negative may lead to good news later. Identification of these mixed news items is difficult, based on which side of the story is accepted and how they are categorized. These matters need to be sorted out before any long term studies are carried out.
Table 1: News Main Categories, Weighted factors, and Subcategories
Weighted Factors 4 Subcategories a. Terror incidents/reports/changers/captures b. Economic successes/strengths c. Governments new/overthrown Death/assassination of world leaders d. Natural/man-made disasters a. Daily energy prices b. Weekly job data/interest rates c. Lay offs/new hiring d. Corporate profits/losses a. Global political successes, announcements, proclamations, or reforms b. Corporation mergers/acquisitions c. Important international delegations/visits d. Major buy-out/financial reports a. Socio-economic status b. Major sport championships/victories c. Hollywood marriages/divorces d. Box-office successes/failures
Main Categories 1. Global stability
2. Economic data
3. World business
4. Social-cultural changes
News Categories In general, global and local events fall into several categories: security status, political stability (either local or global), social situations, economic conditions, and technological/scientific advancements. Table 1 shows the four main categories, their weighted factors, and four subcategories of news identified for the relevant studies. News is available 24/7. It never stops nor does it rest for weekends and holidays. The influences these market conditions may have are notable in the sections to follow. As a corrective measure, we consider that market indicators remain unchanged during these days, while the news items are included in the analyses. The analysis of weekly aggregate data rather than daily data safeguards this consideration to some extent. In the determination of the effect of news items, a preliminary study was conducted during the observation period, July 3 September 10, 2006.
3. Analysis of Stock Market

In determining whether there is an association between two variables, one of the most commonly used statistical procedures is the chi-square ( 2 ) independent test. This procedure requires that all expected frequencies are 1 or greater; almost 20% of the expected frequencies are less than 5; and finally, the data is collected using simple random sampling. Table 2 presents the two variables under consideration for this test, the number of news items in each category and the market indicator(s) fluctuations
56
measured in terms of increase (), decrease (), or unchanged (). As far as increases and decreases are concerned, the three market indicators; DJIA, NASDAQ and S&P 500, essentially behave in the same manner with some rare exceptions. Therefore, only the DJIA was analyzed using this test.
Table 2: Contingency Table for Number of News Items and Market Fluctuations
Number of Good News Items (G) 69 (70.97) 72 (70.97) 66 (65.06) Number of Bad News Items (B) 105 (100.46) 103 (100.46) 85 (92.09) Number of Neutral News Items (N) 66 (68.57) 65 (68.57) 69 (62.86)
DJIA vs. News Increase Indicator 24 Decrease Indicator 24 Unchanged Indicator 22
The hypotheses for the test are: H 0 : Variable in question are not related (independent) H a : Variable in question are associated (dependent) For this test, the 2 test statistics is 140.206, which is greater than 9.488 ( 2 0.05 with df = 4 ). From the data gathered for the 10-week period ending on September 10, 2006 we concluded that at the 5% significance level, the data provides sufficient evidence to determine that there is an association between the nature of the number of news items and the market fluctuations. News Item Analysis Weekly news items have been identified as good news, bad news, and neutral news items for tabulation during the period.
Figure 1: Weekly Good News, Bad News, and Neutral News Frequency Distribution
W e e k ly G o o d N e w s , B a d N e w s , a n d N e u t r a l N e w s D i s t r i b u t io n
40 35 30 F re qu e n 25 20 15 10 5 0 1 2 3 4 5 6 W eek 7 8 9 10
G o od N ew s Item s B a d N e w s Item s N e u t ra l N e w s I t e m s
Figure 1 depicts a summary of news items on the basis of good news, bad news, and neutral news and their respective frequencies during the period for analysis. Anova Single Factor Analysis The analysis of variance (ANOVA) provides methods for comparing several population means, that is, the means of a single variable for several populations. This type of ANOVA is called one-way analysis of variance because it compares the means of a variable for populations that result from a classification by one other variable. We need to compare three news categories. The hypothesis being tested is that
57
the populations of news items means are equal, i.e., the null hypothesis, H 0 : 1 = 2 = 3 , where 1, 2, and 3 are used to represent three news item categories considered in the study. To test this hypothesis, the estimate of each mean is calculated using the data collected on each news category. We examine any differences between the sample means to see whether the differences are evidence between the population means or merely due to any expected experimental error, thus, asking if the sample means are significantly different. If the sample means are significantly different, then we conclude that there was a difference between the population means. Hence, H 0 is H 0 : 1 = 2 = 3 and H 1 is H 1 : not all the i ' s are equal. The significance level of 0.05 is adopted elsewhere unless stated otherwise. The appropriate test of H 0 (as described) depends upon certain assumptions [12]. The ANOVA carried out is subject to these assumptions. Table 3 shows descriptive statistics for each category of weekly aggregated news items. With the assumption of homogeneity of variance it maintains close standard errors. A violation of this assumption increases the likelihood of making a Type 1 or Type 2 error [13]:
Table 3:
(News Items) Groups Good Bad Neutral Total
Descriptive Statistics
Weekly Count (N) 10 10 10 30 Std. Deviation 5.0343 3.3350 5.2493 6.1942 95% Confidence Interval for Mean Lower Bound 17.099 26.914 16.245 21.020 Upper Bound 24.301 31.686 23.755 25.646
Mean
Std. Error
Minimum
Maximum
20.700 29.300 20.000 23.333
1.5920 1.0546 1.6600 1.1309
15 25 9 9
31 35 27 35
An analysis was carried out on the data which represents counts of news items in each three news categories. Table 4 is the output obtained from the test of homogeneity of variances. This test normally reveals a significance difference between the variables. The test resulted in a probability (Sig.) is above 0.05. Therefore, a one-way analysis of variance concludes that the test maintains the assumptions of homogeneity. Thus, the test of homogeneity has been verified.
Table 4: Test of Homogeneity of Variances RESP
Df1 2 Df2 27 Sig. 0.509
Levene Statistic 0.693
This test concludes from Table 5 that the population means of three news item categories are not all the same. The multiple comparison test is helpful in addressing this question.
Table 5: ANOVA Results
Sum of Squares 536.467 576.200 1112.667 Df 2 27 29 Mean Square 268.233 21.341 F 12.569 Sig. 0.000
Source of Variation Between Groups Within Groups Total
Table 6 is the ANOVA single factor output obtained for the Least Significant Difference, LSD, test. This test further asserts that the ANOVA F test is also significant. Multiple comparisons analysis shows that good news items vs. bad news items and bad news items vs. neutral news items are statistically significant. Unfortunately, the LSD test was not strictly valid while the pairs of means being tested were assumed to be independent proven to be wrong. In practical terms this means that the 5% significance level is much higher i.e., a chance of only 5% of making an error with a higher
58
probability. To ensure that the significance level for the whole experiment remains at 5% we can use Duncan's multiple range test. SPSS [13] accommodates these applications. With this assumption supported, we now turn into the pooled (equal) variance test to evaluate the results of these contrasts. Of these assumptions, those pairing good news with bad news and bad news with neutral news items and vice versa were statistically significant.
Table 6:
(I) Type 1 2 3
LSD Multiple Comparison Results for Dependent Variable

Mean Difference (I-J) (J) Type 2 3 1 3 1 2 -8.600 0.700 8.600 9.300 -0.700 -9.300 Std. Error 2.0659 2.0659 2.0659 2.0659 2.0659 2.0659 Sig. 0.000 0.737 0.000 0.000 0.737 0.000 95% Confidence Interval Lower Bound Upper Bound -12.839 -4.361 -3.539 4.939 4.361 12.839 5.061 13.539 -4.939 3.539 -13.539 -5.061
* The mean difference is significant at the 0.05 level.
Table 7 shows the DJIA and oil price fluctuations. From the 2 test, the test statistics is 36.482, which is greater than 3.841 ( 2 0.05 with df = 1 ). This concludes that at the 5% significance level, from the data gathered during the ten weeks ended on September 10, 2006, the data provides sufficient evidence to conclude that there is an association between the market fluctuations of DJIA and crude oil stock prices as measured as increase and decrease.
Table 7: Contingency Table for DJIA and Oil Price Fluctuations
Increase 8 (11.02) 15 (11.98) 23 Decrease 15 (11.98) 10 (13.02) 25 Total 23 25 48
DJIA vs. Oil Price Increase Decrease Total
The time-series configuration of stock market indicators, DJIA, NASDAQ, and S&P 500 obeys similar patterns with regard to crude oil prices as shown in Figure 2. Scales such as 5, 10, and 160 have been chosen to elevate the indicators, NASDAQ, S&P 500, and crude oil prices to the level of DJIA, respectively for this configuration assessment.
Figure 2: Price Configuration of DJIA, NASDAQ, S&P 500, and Crude Oil
59
Polynomial Regression Models It has long been suspected that there is a strong correlation between DJIA and crude oil prices. In the following, various possible regression models are considered. The polynomial regression of degree six seems to obey the stock price pattern with a strong coefficient of determination ( R 2 ).

Table 8: Possible Regression Model Equations
Degree Coefficient of Determination ( R
60
Model Linear Exponential Power Logarithmic Polynomial Polynomial Polynomial Polynomial
Regression Model Equation
).
1 2 3 4 5
y = 0.0096 x + 180.48 y = 323.21e 0.0001x
R 2 = 0.4648 R 2 = 0.4620 R 2 = 0.4566 R 2 = 0.4594 R 2 = 0.6244 R 2 = 0.6581 R 2 = 0.7007 R 2 = 0.7529
y = 7 10 7 x 1.4696 y = 106.13 ln( x) + 1062.4 y = 3 10 5 x 2 + 0.5581x 2974.9 y = 7 10 8 x 3 + 0.0022 x 2 24.245 x + 88816 y = 4 10 10 x 4 2 10 5 x 3 + 0.2701x 2 2009.2 x 6 10 6
y = 2 10 12 x 5 1 10 7 x 4 + 0.0027 x 3 30.022 x 2 + 166233x 4 10 8 y = 5 10 15 x 6 3 10 10 x 5 + 1 10 5 x 4 0.14 x 3 + 1159.2 x 2 5 10 x + 9 10

6 9
Polynomial
R 2 = 0.7599
However, any regression model listed in Table 8 predicts DJIA from corresponding crude oil prices or vice versa, where the correlation coefficient varies between 0.682 and 0.872. One should not believe that the degree of polynomial model can be increased to come up with a perfect model. There is also an optimal degree needed to be identified considering the accumulation of round off errors (including underflow and overflow) [14]. There is a fundamental problem with the polynomial regression models even though several models are described by polynomial equations. This means that the best-fitting results can rarely be interpreted in terms of these models. Polynomial regression can be useful to create a standard curve for interpolation purposes, or to create a smooth curve fitting for graphing. But the polynomial regression is rarely useful for fitting a model to economic data, because a proper laymans interpretation has to be created for business communities and for the general public. For the validity of polynomial regression, see [15]. Figure 3 depicts the scatter plot and the polynomial regression model of the sixth degree for DJIA and crude oil prices:
61

Figure 3: Scatterplot Drawn Using DJIA and Crude Oil Stock Prices from July 30 September 10, 2006
4. Brief Accounts of Empirical Studies

Through re-examination in terms of a simple Markov chain, the question of dependency among the changes in values of selected stocks can be given consideration. The findings showed that the stock market does have a short-term memory in those prices are dependent on immediate preceding changes [16]. But, at the same time, this dependency has non-stationary transition probabilities, and thus the steady-state transition matrix cannot be constructed. This method used Markovian analysis of the changes in natural logarithms of a set of stock prices over the time period, both for the stock market as a whole, in terms of a vector Markov chain and for a single stock in terms of its particular Markov chain [16]. A set of values in which the process takes place must be first defined using three states; high and low and a third state, no change or rather small change for the process in terms of the mean absolute deviations of changes in the natural logarithm of prices similar to the comparison of stocks carried out between 1963 to 1968 in [16]. Our study considers the three-state processhigh , low , and no change . During the first few weeks of data collection of the New York Stock Exchange (NYSE) we discovered that it would be better to collect daily stock prices of EMC Corp., Lucent Technologies Inc., iShare Japan Index Fund, and General Electric Co. among most active stocks during the observation period [17] for similar analyses. Finally, an interesting phenomenon from mathematical economics, the Black-Scholes optionpricing formula is worth-mentioning [18]. Options are the types of financial instruments that give the investor the right to buy or sell security at a future time at a fixed price. This price is called the exercise price or striking price [19]. The exercise of the option that occurs on a fixed date at which time the option expires is called European options. Thus, this phenomenon deserves to be thoroughly understood by those interested in advancing the two types of options, calls and puts options, and strengthening its relevancy to the types of news items.
62
5. Stock Markets, Personal Wealth, and Tax Cuts

The findings of this study have important implications for both scholarly work and for business communities. Firstly, the results of the stock markets performance are naturally the reflection of someones personal wealth. However, a small portion of sums exchanged each day on the worlds stock markets through companies depends on their financial needs and prospects. Actually, preprogrammed computers carry out necessary recognition of their own stock prices and arrange investments almost instantaneously. The investing components of these transactions are mostly mathematics using hyper-speed electronic equipment [27]. It is known that Americans are attracted to the thought of amassing significant wealth. One of the most basic ways of determining the scope of financial inequality is to focus on the total annual income of individuals and families who generally invest in stock market [21]. This measure takes into account wages and salaries, rental income, investments in stocks and bonds and other direct sources of income. Some believe that a sense of social value rather than survival motivates much activity in relation to high income societies. In the context of social stratification gaps, survival concerns outweigh social concerns [21]. Yet the longterm survival of individuals and groups may not be functional for the long-term survival of society. Functionalism is how society performs various functions (socialization, economic institutions, etc.) and how an institution persists and survives by meeting the needs of the people. There are two macro-level approaches in sociology to describe them. Functionalist theory basically assumes that the economy functions for social survival and benefits society as a whole. Conflict theory, however, assumes that there are social class divisions in society and that what benefits the elite, who are increasing their share of the wealth, does not necessarily benefit the (shrinking) middle class or the working class and poor. A synthesis of these two approaches only explains why the US is peaking. The elite expansion of social control over wealth will soon lead to intense conflict. It is a global one because the accumulation of wealth rapidly integrates the nations of the north with the south to those disadvantaged. The wealth accumulation disparity perhaps is an important topic of study because it was very one-sided in the 1970s and it has remained as such with only certain Asian nations escaping from this anomaly. What we all wonder is, when capital flight moves to the US, correlated with a rise in the prices of oil, is it because we are the world's largest economy, that we are a "safe haven" from any long-term economic disaster? Or is this connected to any investment occurring in the displacement of oil to hydrogen or solar energy, etc.? This causes inflation to impact various nations disproportionately. Those who are not in favor of the rise in stock markets (such as [22, 23, & 24]) argue that the movement of the stock market and the tax cuts theory of economics is, over the long run, associated with a greater concentration of wealth in the hands of the elite and less in the hands of the middle class, the working class, and the poor. It is believed that the middle class is shrinking, which makes us wonder what group of individuals was most impacted by the dotcom boom. We tend to think there was a group of people trying to move into the elite, but according to the recent publication of the 20th annual ranking of the 400 richest Americans, Forbes Magazine, 2006 [25], we do not know if the elite "suffered" much monetary fluctuation in their family fortunes. Tax cuts and stock market changes appear to have occurred concurrently with the movement towards hiring people abroad, virtually, or by moving industrial assembly line work, therefore removing the skilled working class of industrial workers, by and large, from the auto industry, etc. The stock market is a vehicle of global economic integration. The 1990 dotcom boom, which ended with the collapse of the NASDAQ in 2000, resulted in a massive movement of wealth to the elite; from 1973-2000, ninety percent of the US population saw their real income decline by 7% while the real income of the elite increased by 143%, the top 1% increased by 343% and the second top 1% increased by 599%. Wealth has been created much more rapidly than ever, but society has become more unequal than any time in the history of US since the 1920s. By 1996, the US had become the world's most unequally developed nation, where the richest 1% had 40% of the nation's entire wealth. US wages declined below Europe and the rate of poverty is three times that of Northern Europe. Meanwhile, in the middle class, credit card debt was estimated to
63
be US $734 billion in 2004. This overall causes challenges and opportunities for sustained economic growth and development, as well as risks and uncertainties for the future of the world economy.
6. Conclusions and Future Work

Using the preliminary analysis conducted during the 10-week period as a case study, this paper has sought to expound the way the performances of stock market pervade the dissemination of news. A long-term study can be accommodated after remedying the circumstances of this study. The comments of financial analysts dominate actual news coverage about the market when we tuned in to any TV channels during market activities. They should be disregarded in selecting and identifying the news items. These news items can reflect a bias in selecting the news items caused by the authors of this article. The data concluded that at the 5% significance level, the data provides sufficient evidence to conclude that there is an association between the nature of the number of news items and the market fluctuations of DJIA and the crude oil prices. It is also noticed that on a given day, if the stocks have an upward tendency, then all three indicators will have the same trend, and vice versa to conclude that if the buyers want to buy, they go on a buying spree and DJIA, NASDAQ, S&P 500, and crude oil prices share the same time-series configuration. We were able to obtain a set of regression models that predicts crude oil prices from the DJIA and vice versa with a strong correlation coefficient. If a major news story continues even after it has first surfaced, it can still affect the market for a period and diminishes after some time. The selection of news items as good or bad must be streamlined for future studies. We could also investigate the variations of highs and lows using the plots for comparison of all indicators. Analysis using categories of news such as world, local, US, financial, and international may have some impact on the stock market. News analyses by reporters and their affiliated news channels could easily interfere with market activities [20]. Similar analyses can be done for the comparison of gasoline prices in the stock markets as they vary in different states due state, local, and environmental taxes. Graphs similar to [6] and line graphs of stock markets, various news items and their weekly frequencies can provide evidence for similar behavior.
Acknowledgement
Dr. Roberto R. Heredia (Bob) and Dr. Judith A. Warner of Texas A&M International University, Laredo, Texas 78041-1900 have commented on various aspect of this article from the time to time when I was gathering topical thoughts for this article. Their comments as well as Bobs statistical superiority helped tremendously to improve this article at various stages of its development.
64
References
[1] [2] [3] [4] [5] Scott Sernau, WORLDS Apart: Social Inequality in a Global Economy, Second Edition, 2006, Pine Forge Press, an Imprint of SAGE Publications, Inc., Thousand Oaks, CA Christopher J. Kollmeyer, Good News/Bad News: Elite Discourse on the Performance of Californias New Economy (Grant # SB 10018), 2001, Department of Sociology, University of California, Santa Barbara, CA J.-P. Bouchaud and R. Cont, A Langevin Approach to Stock Market Fluctuations and Crashes, The European Physical Journal B, Vol. 6, 1998, pp. 543-550 Futurific, Newsmagazines of the Future, Volume 30, #7 (July), #8 (Aug.), #10 (Oct.), and #11 (Nov.), 2006, Futurific, Inc., New York, NY Moshe Koppel and Itai Shtrimberg, Good News or Bad News? Let the Market Decide, AAAI Spring Symposium on Exploring Attitude and Affect in Text: Theories and Applications (AAAI-EAAT 2004), March 22-24, 2004, Stanford University http://www.cs.biu.ac.il/~koppel/papers/sentimentchapter-koppel.pdf Lee Gillam, Khurshid Ahmad, Saif Ahmad, Matthew Casey, David Cheng, Tugba Taskaya, Paulo C F de Oliveira and Pensiri Manomaisupat, Economic News and Stock Market Correlation: A Study of the UK Market In (Ed.), TKE 2002 Conference (Retrieved from http://www.computing.surrey.ac.uk/ai/TKE/economic_news_stock.doc) Craig A. Depken II, Good News, Bad News and Garch Effects in Stock Return Data, Journal of Applied Economics, Vol. IV, No. 2, Nov. 2001, pp. 313-327, http://www.cema.edu.ar/publicaciones/download/volume4/depken.pdf Go for the Bananas! A Chimpanzee's Guide to Financial Security http://www.scienceprojects.com/bananas.htm Donald Robertson and Stephen Wright, The Good News and the Bad News about Long-run Stock Market Returns, Sept. 24, 1998, http://www.econ.cam.ac.uk/faculty/wright/stocks.pdf S. Ghon Rhee, Short-Sale Constraints: Good or Bad News for the Stock Market? Fifth OECD Round Table on Capital Market Reform in Asia, Tokyo, Japan, Nov. 2003, pp. 19-20, http://www.oecd.org/dataoecd/4/63/18454115.pdf Kjell Hausken, Probabilistic Risk Analysis and Game Theory, Risk Analysis, An International Journal, Vol. 22, No. 1, 2002, pp. 17-27 Richard A. Johnson and Gouri K. Bhattacharyya, STATISTICS: Principles and Methods, Fifth Edition, 2006, John Wiley & Sons, Inc. D. M. Shannon and M. A. Davenport, Using SPSS to Solve Statistical Problems: A SelfInstruction Guide, 2001, Prentice-Hall, Inc, Upper Saddle River, NJ Richard L. Burden and J. Douglas Faires, Numerical Analysis, Sixth Edition, 1997, Brooks/Cole Publishing Company, Pacific Grove, CA Elliot M. Cramer, Mark I. Appelbaum, The Validity of Polynomial Regression in the Random Regression Model, Review of Educational Research, Vol. 48, No. 4, 1978, pp. 511-15 Bruce D. Fielitz and T. N. Bhargava, The Behavior of Stock-Price relativesA Markov Analysis, Operation Research, Vol. 21, No. 6, 1973, pp. 11831199 MOST-ACTIVE STOCKS, New York Stock Exchange, Poughkeepsie Journal.com http://customwire.ap.org/dynamic/files/static/moneywire-basic/dc_tables/actv.html David Williams, Probability with Martingales, 1991, Cambridge University Press, London, UK Stephen G. Kellison, The Theory of Interest, Second Edition, 1991, Irwin/McGraw-Hill, a division of the McGraw-Hill Companies Niklas Luhmann, The Reality of the Mass Media, Stanford University Press, 2000 Robert A. Rothman, Inequality and Stratification: Race, Class, and gender, Fifth Edition, 2002, Pearson Prentice Hall, Upper Saddle River, NJ Paul Krugman, The Death of Horation Alger, The Nation, Dec. 23, 2003 Chuck Collins and Felice Yeskel, Economic Apartheid in America Economic Apartheid in America: A Primer on Economic Inequality and Security, First Edition, 2000
[6]
[7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23]
65 [24] [25] [26] [27]
International Research Journal of Finance and Economics - Issue 11 (2007) Charles Derber, Corporation Nation, International Social Science Review, Spring-Summer, 2002 America's Richest, The Forbes 400, 20th Annual Ranking of the 400 Richest Americans, Forbes Magazine, 2006 http://www.forbes.com/2002/09/13/rich400land.html Desh Peramunetilleke, Raymond K. Wong, Currency exchange rate forecasting from news headlines, Australian Computer Science Communications, Proceedings of the 13th Australasian database conference Vol. 5 ADC '02, Jan. 2002, pp. 131-139 Alvin Toffler and Heidi Toffler, Revolutionary Wealth, Alfred A. Knopf, New York, 2006
Searching for a Scholarly Visibility: The Case of Ukraine

Jean Mirucki F.D.E.G. Sciences Economiques Universit de Valenciennes E-mail: jean.mirucki@univ-valenciennes.fr Abstract The purpose of this paper is to position the place of economics in Ukraine and its scholarly visibility at the international level. Using the EconLit database, the analysis covers the affiliation of contributing authors, the gender composition, the patterns of collaboration between Ukrainian and foreign economists and a review of the different thematic treated in the articles. Results show that Ukrainian authors have a general profile of co-authorship similar to western practices, a significant proportion of publications from governmental circles, a low female representation and a small number of Ukrainian authors having published in journals also included in the S.S.C.I. database. Keywords: N.I.S.; Ukraine; EconLit; S.S.C.I.; Scientific production Jel Classification Codes: A140; P210
Introduction
For the past few decades, the analysis of scientific production of economists has met an ever growing popularity in the profession (see Liebowitz and Palmer, 1984; Portes, 1987; Frey and Eichenberger, 1993; Laband and Piette, 1994; Goodwin and Sauer, 1995; Hodgson and Rothman, 1999; Mirucki, 2000; Ellison, 2002; Baser and Pema, 2003; Ramrattan and Szenberg, 2003; Lubrano and Protopopescu, 2004; Faria, 2005; among others). The usual arguments come repeatedly to justify the existence of new material on that topic. Firstly, it may provide useful information on journal ranking and provide a gain of information search to the reader. Secondly, efficient financial and operational management of library acquisitions largely depend on accessibility of up-to-date information on the most valuable research findings. Similarly, this type of crucial data is essential to authorities at time of decision on evaluations and promotions of academic faculties. Lastly, somehow, the lambda researcher may satisfy his general curiosity or even gratify his personal ego which, in fine, could stimulate competition among peers and contribute to the improvement of general research objectives. While these criteria will also apply in this endeavor, a more specific reason may be invoked here: positioning the place of Ukraine economics with the recent development of western practices and projecting the future evolution of its scientific visibility at the international level. In order to do so, the analysis is carried out using the EconLit database from the American Economic Association. This will require a sketchy introduction of the aims and the characteristics of this useful panoply of specialized instruments. Following a general presentation of the different types of publication, a more detailed analysis of journal articles will cover the affiliation of contributing authors, the gender composition, the patterns
67
of collaboration between Ukrainian and foreign economists and a review of the different thematics of the articles. A summary of the main results will then be presented in light of the potential for Ukrainian visibility on the international scene in the production of scientific knowledge in economics.
Presence of Ukraine in Academic Economic Literature 1 Since 1969
It is important to present first the general characteristics of the EconLit database. It is followed then by the method of analysis and the description of the investigation used in the framework of the paper. The survey of the general content of the database by type of document, and especially in the case of journal articles allows, finally, to get a general profile of the Ukrainian-related contributions present in EconLit. Introduction to the EconLit database Like a majority of publications applied on the organization of the scientific production in economics, the analysis is carried out using the AEA-EconLit database. This reference instrument of bibliographic information is under the jurisdiction of the American Economic Association, a private body serving the needs of about 24000 subscribers, including 4500 libraries, located throughout the world. EconLit includes, periodically, selected bibliographic records on journal articles, books, collective work articles, dissertations and working papers. Information is identified by different field indicators which can then be manipulated to provide useful results in specific analysis. They include variables such as TI (title), AU (author), AF (author affiliation), SO (source of publication), DT (document type), PY (publication year), GE (geographic descriptor) or DE (subject code), among others. Method of analysis The ongoing update of the EconLit database with continuous and irregular submissions, on one hand, and the periodical release of overlapping versions in the AEA-EconLit accessible to individual researchers, on the other, require the development of an initial strategy as to the organization of an operational database fit to the objectives of this study. Then, a very direct process of identification of Ukraine-related publications included in EconLit has been used to extract records containing all possible variants of the root-word "Ukraine" and to check them for relevance to the investigation. Given time limitations and knowing that a vast majority of publications has adopted that option, the choice was made to deepen the analysis mainly on journal articles. While material provides individual information on each author, it should be reminded that the analysis is carried out on a categorical rather a nominative basis. That is, the profile of the group of Ukrainian authors is compared with that of the group of foreign authors (non-residents), all of these contributors being interested in the treatment of Ukraine-related thematic which ended up with publications included in the EconLit database during the 1992-2002 period. As a rule, this material must be accepted by scholarly journals which had themselves been first selected during the screening process of the AEA Evaluation Committee. Publications by type of document Publications included in EconLit started with journals published since 1969 followed by other type of documents with different starting coverage dates: collective volumes (1984); working papers (1984); books (1987); dissertations (1987); book reviews (1993).
While the analysis is entirely based on the EconLit bibliometric database, the results do not carry any implication as to the quantitative production of economists who have published their material thru outlets not included in that media. However, articles accepted by refereed journals validated by EconLit are usually considered, in the profession, as research satisfying scholarly standards.
68
Recent versions of the EconLit CD ROM provide the following information on the document type (DT) ventilation for the 1992-2002 period.

Table 1: Distribution of publications by type of document: 1992-2002
Total (1) 199 738 19 425 109 395 10 502 28 654 367 714 % (2) 54.3 5.3 29.8 2.9 7.8 100 UA (3) 194 85 180 0 35 494 %UA (4) 39.3 17.2 36.4 0 7.1 100
69
Type of document Journal article Book Collective volume article Dissertation Working paper Total:
%UA / % (5) 0.7 3.3 1.2 0 0.9 1
In Table 1, the Total EconLit figures in column (1) are confronted with the Number of documents related to Ukraine in column (3) in order to obtain the relative ratio between the two groups indicated in column (5). This relative ratio is treated here as an index of comparison for each type of documents, where values superior to unity imply an overrepresentation of entries with Ukraine-related thematic against all EconLit records, and vice-versa. For example, out of 367 714 documents present in EconLit it shows that 19 425 of them are categorized as books, for a ratio of 5.3 %. Referring to Ukraine as the thematic term, the ratio becomes 17.2 % (85 out of 494), which corresponds to a much higher relative ratio of books treating of Ukraine than the overall number of books contained in EconLit and the index value is 3.3 since it corresponds to the following relative ratio (17.2 % by 5.3 %). The inclusion of Ukraine-related thematic documents depends exclusively of Ukrainian residents and foreign contributors submissions to scholarly publishers (academic journals, scientific books, selected universities, renowned research laboratories) accepted by the AEA-EconLit Evaluation Committee of the J.E.L. (Journal of Economic Literature). Graph 1 provides an overall view of all Ukraine-related publications (Total UA) and journal articles (Jnl) included in the EconLit database during the 1969-2002 period.
Graph 1: Publications related to Ukraine
If one should refer to the 1991 year of independence of Ukraine, the production until that date is clearly almost inexistent for the first 23 years 2. So, both for historical and pragmatic reasons, the
2
EconLit indicates only two journal articles by foreign authors for the 1969-90 period and three in 1991.
70
selected sample will begin with 1992 and end with 2002 the last complete year of data available now in EconLit3, giving about a decade of observations (11 years) to perform the proposed analysis.
Articles by journal group

The number of journals included in the EconLit database has varied from 182 at the beginning in 1969 to pass 750 in 2005. In our actual sample of articles related to the Ukraine thematic during the 1992-2002 period, there are 71 different journal titles included in EconLit and 26 of them are also included in the S.S.C.I. database (Social Science Citation Index), which computes journal citations. A total of 41 Ukrainian contributions, made up of 33 individuals 4, corresponds to the production of 39 journal articles (see Annex 1: data comes directly from EconLit, such as entered by authors at time of publication, with possibly some russification of names or cities). For these Ukrainian contributors, Table 2 indicates that there is a total of 13 different journal titles present in EconLit and that the first group of 5 are also journals included in the S.S.C.I. database 5 . Column 1 shows the number of articles related to the Ukrainian thematic authored by Ukrainian for each journal title, while total articles (Ukrainian and/or others) are indicated in column 2. Specific Ukrainian contributors are named at the level of the journal title (column 3), with the number of multiple interventions, whenever applicable. Noticeably, among the 8 Ukrainian authors with 9 contributions to journals in the S.S.C.I. group, there is a majority of government officials. It is showing that more theoretical articles are produced by practitioners, at the S.S.C.I. level, by persons from governmental circles, especially when some of them have a sound academic background. The remaining 32 interventions are listed in the second group of 25 different authors, knowing that Pynzenyk is computed in both groups and that Alexander Petrik corresponds to Oleksandr Petryk. The majority of the titles suggests, in each group, a vocation for publishing papers with EastEuropean contents 6 or an orientation towards socialists systems and transitional economies 7 (20 out of 29 entries in the S.S.C.I. group and 55 out of 63 in the standard EconLit group) with a total of 9 out of 14 different journal titles.
6 7
Using the 03/2005 version of the EconLit database, inclusion of all records is normally ensured for 2003 for american journals and 2002 for other countries. Still, updates may happen occasionally at later dates for journals and even much later for other type of publications. EconLit provides information as it gets it, and authors should be: a) very careful the way they provide their personal data at the time of publication and b) maintain their own standard representation their brand image all throughout their publishing career. Here there are two names affiliated with the National Bank of Ukraine: "Petrik Alexander" and "Petryk, Oleksandr Ivanovych". In fact, they should be viewed as a same person and the number of individual authors comes up to be 33 and not 34. Journal articles present in the S.S.C.I. database are used to reference citations to other articles and reflect, thereby, the importance of past publications to the progression of the research process. They include journals like: Eastern European Economics; Russian and East-European Finance and Trade; Ukrainian Economic Review. The other group refers to: Emerging Markets Review; Journal of Emerging Markets; Post-Soviet Affairs; Post-Communist Economies; Post-Soviet Geography and Economics; Problems of Economic Transition.

Table 2: Contributions of Ukrainian authors by journal group
71
Group 1. Journals included in EconLit and S.S.C.I. 3 8 Eastern European Economics: Teriokhin; Yekhanurov; Malysh 2 8 Economic Policy: A European Forum: Pynzenyk; Lanoviy 2 10 Post-Soviet Affairs: Khmelko (2) 1 1 Brookings Papers on Economic Activity: Korobov 1 2 Post-Communist Economies: Dodonov Group 2. Journals included in EconLit 10 14 Russian and East-European Finance and Trade: Petryk (3); Mortikov (2); Bolharyn; Oksymets; Pynzenyk; Shpek; Yushchenko 7 29 Ukrainian Economic Review: Popovkin; Pyrozhkov; Starostyna; Lazarenko; Shevchuk; Yastremsky; Ustenko 5 5 International Regional Science Review: Popov (2); Chumachenko; Doroguntsov; Onishchenko 5 9 Problems of Economic Transition: Lukinov; Mamutov; Burakovsky; Petrik; Boldakov 1 1 Atlantic Economic Journal: Akimova 1 1 Emerging Markets Review: Mertens 1 1 Journal of Emerging Markets: Voronova 1 1 Post-Soviet Geography and Economics: Pugachov 1 2 Public Budgeting and Finance: Slukhai
With 29 entries, Ukrainian Economic Review was naturally the most prolific journal for the general thematic, but it is now discontinued. Its publication was limited to five issues and covered only the 1995-98 period. The detailed list of the 71 journal titles is included in the Annex 2. It contains, in the 26 S.S.C.I.-type journals, two top-10 level journals: the European Economic Review and the Quarterly Journal of Economics.
Patterns of collaboration between authors

Over past decades, collaboration between authors of journal articles seems to increase regularly in social sciences, giving rise to debates on the truly beneficial effects of such practices, like the real added-value to production of scientific knowledge. Regardless of its possible controversial aspects, the comparative analysis of the Ukrainian and foreign attitudes towards cooperation in scientific endeavors seems worth investigating, particularly when it considers also the geographic dimension. Based on a total of 278 contributions in 194 journal articles, the general collaboration profile for all authors is established as follows: 18 articles with 4 or more authors (10 %) 16 articles with 3 authors (8 %) 53 articles with 2 authors (27 %) 107 articles with 1 author (55 %) Among the 41 Ukrainian contributions, only two articles are written each by two Ukrainian authors from the same institution, giving a total of 39 journal articles written by 33 individuals from Ukraine. Collaboration of 19 Ukrainian contributors with 19 non-resident (8 in a trio and 11 in a duo) are established with participations from USA (11), Russia (4), Canada (2), UK (1) and Germany (1). Out of an overall total of 60 collaborations, only 56 of them are strictly between Ukrainian and foreign researchers. That amounts to 30 individuals from Ukraine co-writing with 15 non-resident researchers from USA (8), Russia (3), Canada (2), UK (1) and Germany (1.) Overall, comes out a predominance of collaborations from North America (10), with Western European countries coming last (2) For Ukrainian authors, the collaborations by type of article are as follows: 1 article with 4 or more authors (3 %)
72
4 articles with 3 authors (10 %) 13 articles with 2 authors (33 %) 21 articles with 1 author (54 %) Overall, the differences between the two groups are not very serious, except for the last category of "4 or more authors", with a lower frequency in the group of Ukrainian authors 8. Already now, almost half of Ukrainian authors engage in cooperative work, and the trend should very well be sustained in the future, particularly if it involves an increasing number of contacts with foreign colleagues.
Authors' geographic location, gender and institutional affiliation

To a certain extent, geographic location may lead to natural hypothesis: concentration is associated with higher economic activity while centralization is compatible with administrative and political centers of power and influence, usually attracting some of the intellectual leading groups. Out of 194 journal articles, there was 41 participations by Ukrainian authors in the publication of 39 journal articles (20 %). The location of these contributions is very centralized in Kiev (34 for 28 individuals) while regional participations are present in the East (4 for 3 persons) and limited to a single entry each for Kherson and Lviv. This non-diversified pattern of scientific production in economics seems to be the heritage of a traditionally centralized political system of recent past. Power and influence can also be reflected by the gender composition of performing groups. There are only 4 women among the 33 authors and their global profile is in line with the rest of the Ukrainian group: publication in the S.S.C.I. group (1 out of 4); centralized geographical location (3 out of 4); academic affiliation (3 out of 4); a more pronounced tendency for collaboration (3 out of 4), but not in USA or Russia (2 in Canada and 1 in Germany). As pioneers in their genre, they certainly deserve to be named individually: Irina Akimova (Kharkiv State Polytechnical U.); Nadia Malysh (Office of the President of Ukraine); Alla Starostyna (Kiev Polytechnical Institute); Alla Voronova (International Management Institute, Kiev). Another similar analysis applied to type of affiliation of authors could suggest the general orientation of the publications. They may be associated either in the group of theoretical analysis for authors operating in academic environments or rather in the category of applied economics for individuals addressing specific problems, usually under the pressure of government priorities, mostly during political mandates. When regrouping the type of affiliation into those two groups of institutions, either mainly academic (Universities, Institutes, Academy of Sciences) or governmental/political bodies (Ukrainian Parliament, National Bank of Ukraine), the numbers are clearly in favor of the academic (27) rather than the governmental (16) group. It should be worth noting the presence, in the latter group, of personalities with economic experience who served later as past or actual ministers, or became officials 9 like Volodymyr Lanoviy (1994), Viktor Pynzenyk (1994 and 2000), Roman Shpek (2000), Serhyi Teriokhin (2000), the Prime Minister Yuri Yekhanurov (2000) and the President of Ukraine, Viktor Yushchenko (2000).
5. Thematic of publication in refereed journals

Originally, the subject descriptors were based on the J.E.L. classification system using a four-digit numerical code which was replaced, in 1991, by an alphanumeric code (letter plus 3-digit). Due to some compatibility problems between the two systems and the specific period of this analysis, the new JEL-1991 subject descriptor codes had to be retained.
8 9
This type of multiple authorship has become more frequent in western and american academic publications in social sciences. Reference to political situation of individuals in general is made as of the time of writing (2006) and not with respect to specific nomination during the 1992-2002 period of observation.
73
The general thematic pattern of documents included in EconLit (Table 3) can be compared with the codes associated with the Ukraine-related thematic applied to this sample. In order to do so, the percentage of the total number of codes (% in column 2), which corresponds to the average number for each code, has been extracted from a previous analysis covering a similar period (1991-2001) and is confronted with the existing results (%UA in column 4) to observe any significant irregularities (Index in column 5). This index of relative percentage of frequency (column 5) is calculated as the ratio of the percentage of Ukrainian contributions (UA % in column 3) over the percentage of the total number of codes (% in column 2), where unity corresponds to a case of similar behavior for both groups, like for code "A" (1.6 % each). Table 3 indicates that in 16 codes out of 19, the articles related to Ukraine are underrepresented (values ranging from 0 to 0.9) with particularly low values (ranging from 0 to 0.4) for 8 codes (B, C, D, G, I, K, M, N). The codes "E" for "macroeconomics and monetary economics" (index value of 1.2) and especially "P" for "Economic Systems" (index value of 10.7) are the leading topics for all Ukrainerelated articles.
Table 3:
Code A B C D E F G H I J K L M N O P Q R Z Total:
Distribution of J.E.L. codes

Number (1) 8 024 13 567 19 110 47 913 38 707 46 571 34 865 23 204 19 084 39 487 8 284 38 586 13 497 12 926 54 437 20 362 31 180 18 432 1 382 489 618 % (2) 1,6 2,8 3,9 9,8 7,9 9,5 7,1 4,7 3,9 8,0 1,7 7,9 2,8 2,6 11,1 4,2 6,4 3,8 0,3 100 UA (3) 8 0 1 18 46 30 11 19 8 23 3 24 4 2 34 220 30 12 1 494 % UA (4) 1,6 0 0,2 3,6 9,3 6,1 2,2 3,8 1,6 4,7 0,6 4,9 0,8 0,4 6,9 44,5 6,1 2,4 0,2 100 Index (5) 1,0 0 0,1 0,4 1,2 0,6 0,3 0,8 0,4 0,6 0,4 0,6 0,3 0,2 0,6 10,6 0,9 0,6 0,7 1 Subject Code Descriptor General Economics and Teaching Schools of Thought and Methodology Mathematical and Quantitative Methods Microeconomics Macroeconomics and Monetary Economics International Economics Financial Economics Public Economics Health, Education and Welfare Labor and Demographic Economics Law and Economics Industrial Organization Business Administration; Marketing; Accounting Economic History Economic Development; Technol. Change; Growth Economic systems Agricultural and Natural Resource Economics Urban, Rural and Regional Economics Cultural Economics and Other Topics
Overwhelmingly, the "P" general category 10 has been used by a large number of articles, recalling that several codes may be included in each of them. Details are provided in the Annex 3. Out of a total of 494 codes 220 of them (45 %) selected by the authors are concerned by the "P" code and the 6 first entries, dealing with "Socialist Systems and Transitional Economies", cover 33 % of all the themes. They refer to the general themes of "Socialist Systems and Transitional Economies": planning, coordination and reform; public firms; overall performance; growth; factor and product markets; foreign trade and investment. Other major topics are parallel to the predominant preoccupations: prices; finance; housing; transportation; public finance and political economy; health, education, welfare and poverty; privatizations; monetary policy; currency; exchange rates; banking; wage level; economic integration. In fact, they concern many of the problems transitional economies are desperately struggling to resolve.
10
It includes the following main sub-categories with corresponding entries: P210 (43); P310 (34); P270 (28); P240 (22); P230 (19) and P330 (19).
74
At the bottom of the list, it may be surprising to observe a lower interest, judging from the author's contributions, for other type of issues: unemployment models; land reform; renewable resources and conservation; general migration; public pensions; regional and rural development; energy government policy. Even worse 11, some topics are barely examined (only one subject code) or totally ignored: bureaucracy; models of political processes; policy designs and consistency; tax evasion; incidence of subsidies; economics of the elderly; food policy; demographic public policy; retirement policies; agricultural labor markets; collective bargaining; unemployment insurance; discrimination public policy; regulation and business law; antitrust policy; utilities and transportation systems; technological innovation; intellectual property rights; multisector growth models; aggregate productivity, among others.
Conclusion
This study presents a direct analysis of existing academic information on a given thematic, such as "Ukraine" and all its derivatives, and visible by thousands of economists and others directly from CD ROMs or on the web. It is only since independence that this thematic has seriously interested number of Ukrainian authors (residents) and an even larger group of foreign contributors, a minority of those being part of the Diaspora of Ukraine. It gives the profile of the group of Ukrainian authors compared with the group of foreign authors, all of them interested in the treatment of Ukraine-related thematic, and having published scholarly articles, during the 1992-2002 period, in journals which passed the selection standards of the American Economic Association in order to be included in the EconLit database. When examining only the journal articles, results show that the group of Ukrainian authors has a general profile of co-authorship similar to western practices, an institutionally centralized presence of international contributors, a significant proportion of publications from academic groups, a low female representation and a smaller number of Ukrainian authors, some of them from governmental circles, having published in journals included also in the S.S.C.I. database. Overall, the different thematics treated in journal articles are partly biased in favor of the subject category of "Economic Systems", and more particularly "Socialist Systems and Transitional Economies", reflecting thereby a legitimate concern of the various contributors (Ukrainian and nonresidents) to address issues on contemporary problems facing emerging markets. In fact, if interested researchers are looking for new areas to explore for potential journal article publication, the dilemmas facing transitional economies, such as Ukraine, are certainly promising fields of investigation for attentive observers. Now that there is this unique opportunity for any author or any community of economists anywhere in the world to give an "international visibility" to its publications on a given specific thematic, it would be expected that, periodically and continuously, more new material from Ukraine be included in the EconLit database, mainly thru english-language outlets which represent 97 % of EconLit. So far, the only exception for Ukrainian-language material concerns a book authored by Stepan Zlupko, entitled "Osnovy istoriji ekonomichnoji teoriji" (Lviv, Ukraine: Ivan Franko National University, 2001) and included in the J.E.L. and the EconLit database in December 2002.
11
With a relatively modest number of articles (194) generating a total of 494 codes, it is understandable that all topics could not have been treated. However, the potential of article publication for aspiring authors remains.

Annex 1: Authors from Ukraine with journal articles in EconLit: 1992/2002 (data from EconLit, such as entered by authors at time of publication, with possible russification)
Kharkov State Polytechnic U; Otto von Guericke U National Bank of Ukraine National Bank of Ukraine; National Bank of Ukraine and St Cloud U Kiev-Mohyla Academy National U Ukrainian Academy of Sciences Institute of Econ Research and Policy Consulting, Kyiv Ukrainian Academy of Sciences Ukrainian Academy of Sciences (missing) U TX; International Institute of Sociology, Kiev-Mohyla Academy; CA Institute of Technology U TX; International Institute of Sociology, Kiev-Mohyla U; CIT Duke U; World Bank; Kiev State Econ U Center for Market Reforms, Kiev Kharkiv State U; George Washington U Ukrainian National Academy of Sciences Office of the President of the Ukraine Inst of Econ Legal Research, Ukrainian Academy of Sciences National U Kyiv-Mohyla Academy and Kyiv Taras Shevchenko U; City U London E Ukrainian State U; Novgorod State U E Ukrainian State U; Novgorod State U National Bank of Ukraine National Bank of Ukraine National Bank of Ukraine National Bank of Ukraine
75
Akimova, Irina; Schwodiauer, Gerhard Boldakov, Viktor Bolharyn, Ihor Viktorovych; Banaian, King Burakovsky, Igor V Chumachenko, Nikolai G. Dodonov, Boris, et al. Doroguntsov, Sergey I.; Onishchenko, Vladimir F. Hinich, Melvin J; Khmelko, Valeri; Ordeshook, Peter C Hinich, Melvin J; Khmelko, Valeri; Ordeshook, Peter C Johnson, Simon; Kaufmann, Daniel; Ustenko, Oleg Lanoviy, Volodymyr Lazarenko, Volodymyr; Zvihlianych, Volodymyr Lukinov, I Malysh, Nadia Mamutov, V K Mertens, Alexander; Urga, Giovanni Mortikov, Vitalii Vladimirovich; Volonkin, Vladimir Mikhailovich Mortikov, Vitalii Vladimirovich; Volonkin, Vladimir Mikhailovich Oksymets, Valeriy Illich Petrik, (misspelled) Petryk, Ivanovych Petryk, Ivanovych Alexander Oleksandr Oleksandr
Atlantic Economic Journal, March 2000; 28 (1): 48-59 Problems of Economic Transition, March 1996; 38 (11): 6-15 Russian and East-European Finance and Trade, May June 1998; 34 (3): 45-55 Problems of Economic Transition, January 1998; 40 (9): 35-55 International Regional Science Review, 15 (3), 1993, pages 235-46. Post-Communist Economies, June 2002; 14 (2): 149-67 International Regional Science Review, 15 (3), 1993, pages 247-55. Post-Soviet Affairs, April June 1999; 15 (2): 149-85 Post-Soviet Affairs, July Sept. 2002; 18 (3): 250-69 Ukrainian Economic Review, 1996; 2 (3): 112-16 Economic Policy: A European Forum, Suppl. December 1994; 9 (19): 191-95 Ukrainian Economic Review, 1996; 2 (3): 42-52 Problems of Economic Transition, August 1996; 39 (4): 5-12 Eastern European Economics, March April 2000; 38 (2): 18-23 Problems of Economic Transition, January 1995; 37 (9): 57-70 Emerging Markets Review, September 2001; 2 (3): 292-308 Russian and East-European Finance and Trade, March April 1999; 35 (2): 31-41 Russian and East-European Finance and Trade, March-April 1999; 35 (2): 8392 Russian and East-European Finance and Trade, May-June 1998; 34 (3): 37-44 Problems of Economic Transition, March 1996; 38 (11): 16-23 Russian and East-European Finance and Trade, May-June 1998; 34 (3): 56-59 Russian and East-European Finance and Trade, May-June 1998; 34 (3): 20-36
76
Petryk, Oleksandr Ivanovych; Banaian, King Popov, Mikhail I. Popov, Mikhail I.; Trumbull, William N. Pugachov, Mykola; Van Atta, Don Pynzenyk, Victor Pynzenyk, Viktor Pyrozhkov, Serhii; Popovkin, Valerii Samonis, Val; Voronova, Alla Shevchuk, Victor Shiller, Robert J.; Boycko, Maxim; Korobov, Vladimir Shpek, Roman V Slukhai, Sergii Teriokhin, Serhyi Yastremsky, Oleksander Yekhanurov, Yuri I Yushchenko, Viktor

National Bank of Ukraine; National Bank of Ukraine and St Cloud U Ukrainian Academy of Sciences Ukrainian Academy of Sciences; WV U Institute of Policy Reform U IL Ukraine Foundation for Support of Reforms Ukrainian Parliament and Institute of Reforms National Inst. for Strategic Studies, Kiev; National Inst. for Strategic Studies, Kiev U Toronto; International Management Institute, IMI Kyiv, Ukraine Lviv Academy of Commerce Yale U; Russian Academy of Sciences; Kherson Pedagogical Institute National Agency of Ukraine Reconstruction and Development National Kyiv Taras Shevchenko U Ukrainian Parliament U 'Kiev-Mohyla Academy' Ukrainian Parliament National Bank of Ukraine Russian and East-European Finance and Trade, May-June 1998; 34 (3): 7-19 International Regional Science Review, 15 (3), 1993, pages 281-89. International Regional Science Review, 15 (3), 1993, pages 225-27. Post-Soviet Geography and Economics, OctNov. 2000; 41 (7): 527-40 Economic Policy: A European Forum, Suppl. December 1994; 9 (19): 197-204 Russian and East-European Finance and Trade, Jan.-Feb. 2000; 36 (1): 59-77 Ukrainian Economic Review, 1995; 1 (12): 3-17 Journal of Emerging Markets, Summer 1996; 1 (2): 45-60 Ukrainian Economic Review, 199798; 3 (45): 155-66 Brookings Papers on Economic Activity, 0 (1), 1992, pages 127-81. Russian and East-European Finance and Trade, Jan. Feb. 2000; 36 (1): 7-27 Public Budgeting and Finance, Fall 2002; 22 (3): 46-66 Eastern European Economics, Jan.-Feb. 2000; 38 (1): 34-50 Ukrainian Economic Review, 1996; 2 (3): 53-72 Eastern European Economics, Jan.-Feb. 2000; 38 (1): 71-93 Russian and East-European Finance and Trade, Jan. Feb. 2000; 36 (1): 78-96
for

Annex 2:
UA 1 3 2 Nbr 3 1 8 8 1 8 2 2 3 1 1 2 2
77
Ukraine-related journals included in EconLit and in the S.S.C.I. Group 1. Journals included in EconLit and in the S.S.C.I.
Journal title American Journal of Agricultural Economics Brookings Papers on Economic Activity Eastern European Economics Economic Policy: A European Forum Economics Letters Economics of Transition Environmental and Resource Economics Eurasian Geography and Economics European Economic Review Hitotsubashi Journal of Economics IMF Staff Papers Journal of Comparative Economics Journal of Development Studies UA Nbr 3 1 1 1 1 1 2 2 10 1 1 1 1 1 1 Journal title Journal of Economic Education Journal of Economic Psychology Journal of Policy Modeling Journal of Public Economics Journal of the Japanese and International Economies Post-Communist Economies Post-Soviet Affairs Quarterly Journal of Economics Review of Income and Wealth Revue d'Etudes Comparatives Est-Ouest Small Business Economics World Development World Economy
Group 2. Journals included in EconLit

UA Nbr 1 1 1 1 1 5 4 1 1 1 1 3 1 1 1 1 1 1 1 1 1 2 1 Journal title Annals of the American Academy of Polit. and Social Science Antitrust Bulletin Atlantic Economic Journal Central European Journal for Operations Research Challenge Communist Economies and Economic Transformation Comparative Economic Studies Development and Change East-West Journal of Economics and Business Economia Internazionale Economic Review (Keizai Kenkyu) Economic Systems Economies et Socits Emerging Markets Review Environment and Planning C: Government and Policy European Journal of Development Research European Journal of International Relations Federal Reserve Bank of Cleveland Economic Review Finance and Development Industrial and Labor Relations Review International Journal of Urban and Regional Research International Labour Review International Monetary Fund Staff Papers UA 5 Nbr 5 1 1 3 1 1 1 1 1 1 1 1 1 1 1 16 5 1 9 2 1 1 1 14 30 Journal title International Regional Science Review Journal of Common Market Studies Journal of East and West Studies Journal of East-West Business Journal of Emerging Markets Journal of International Business Studies Journal of Socio-Economics Journal of World Business Konjunkturpolitik Labour MOCT-MOST: Economic Policy in Transitional Economies Momento Economico Pacific and Asian Journal of Energy Population Studies Post-Soviet Geography and Economics Problems of Economic Transition Public Budgeting and Finance Regional Studies Resources Policy Risk Decision and Policy Russian and East European Finance and Trade Ukrainian Economic Review
10 8
78
Annex 3:
* 43 34 28 22 19 19 12 11 10 8 8 7 7 7 6 6 6 5 5 5 5 4 4 4 4 4 4 4 4 4 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 2 1

J.E.L. Descriptors codes ranked by frequency
Subject code description Transitional Economies; Socialist Systems and Transitional Economies Planning, Coordination, and Reform (P210) Privatization; Privatizing; Socialist Enterprises and Their Transitions (P310) Socialist Systems and Transitional Economies Performance and Prospects (includes Spatial) (P270) Socialist Systems and Transitional Economies National Income, Product, and Expenditure; Money; Inflation (P240) Inflation; Wage; Socialist Systems and Transitional Economies Factor and Product Markets (Labor; Capital) (P230) Socialist Institutions and Their Transitions International Linkages (P330) Price Level; Inflation; Deflation (E310) Boundaries of Public and Private Enterprise; Privatization; Contracting Out (L330) Socialist Institutions Finance (P340) Monetary Policy (Targets, Instruments, and Effects) (E520) Exchange Rates; Socialist Systems and Transitional Economies Prices (P220) Housing; Real Estate; Residential; Socialist Systems and Transitional Economies Urban, Rural, & Regional Econ; Housing; Transportation (P250) Political Economy of Socialism (P260) Socialist Institutions and Their Transitions Public Finance (P350) Country and Industry Studies of Trade (F140) Exchange Rates; Stabilization; Foreign Exchange (Exchange Rates; Intervention; Foreign Exchange Reserves) (F310) Formal and Informal Sectors; Shadow Economy; Institutional Arrangements (O170) Economic Models of Political Processes Rent seeking, Elections, Legislatures, and Voting Behavior (D720) Banks; Other Depository Institutions; Mortgages (G210) Budget; Expenditure; Households; Income; Socialist Institutions and Their Transitions Household Behavior (P360) Energy Demand and Supply (Q410) Demand for Money (E410) Currency; Dual Currency; Monetary; Monetary Standards and Regimes; Government and the Monetary System (E420) General Outlook and Conditions (E660) Multinational Firms; International Business (International Competitiveness) (F230) Economics of Gender (J160) Wage Level and Structure; Wage Differentials by Skill, Training, Occupation, etc (industry, schooling, experience, tenure, cohort, etc) (J310) Unemployment Models, Duration, Incidence, and Job Search (J640) Entrepreneurship (New Firms; Startups) (M130) Economic Development: Human Resources; Income Distribution; Migration (nutrition, health, education, fertility, household labor) (O150) Inflation; Transitional Economies; Macroeconomic Aspects of Fiscal Policy; Public Expenditures, Investment, and Finance; Taxation (E620) Comparative or Joint Analysis of Fiscal and Monetary or Stabilization Policy (E630) Economic Integration (F150) International Monetary Arrangements and Institutions (F330) International Lending and Debt Problems (F340) Mergers; Acquisitions; Restructuring; Voting; Proxy Contests; Corporate Governance (G340) Pension; Social Security and Public Pensions (H550) Demographic Trends and Forecasts; General Migration (J110) Macroeconomic Analyses of Economic Development (includes macro models and analyses of patterns and determinants of development) (O110) Economic Development Financial Markets; Saving and Capital Investment (Financial Intermediation) (O160) Economic Development Regional, Urban, and Rural Analyses (O180) Economywide Country Studies Europe (O520) Micro Analysis of Farm Firms, Farm Households, and Farm Input Markets (Q120) Land Ownership and Tenure; Land Reform; Land Use; Irrigation (Q150) Renewable Resources and Conservation; Environmental Management Water; Air (Q250) Energy and the Macroeconomy (Q430) Energy Government Policy (Q480) General Spatial Economics Analysis of Growth, Development, and Changes (R110) Urban, Rural, and Regional Economics Regional Migration; Regional Labor Markets; Population (R230) = 20 remaining descriptors with 2 occurrences = 69 remaining descriptors with 1 occurrence
79
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] AXARLOGLOU, K. and V. THEOHARAKIS, "Diversity in Economics: An Analysis of Journal Quality Perceptions", Journal of the European Economic Association 1 (2003), pp. 1402-1423. BASER, O. and E. PEMA, "The Return of Publications for Economics Faculty", Economics Bulletin 1 (2003), pp. 1-13. ELLISON, G., "The Slowdown of the Economics Publishing Process", Journal of Political Economy 110 (2002), pp. 947-993. FARIA, J. R., "The Game Academics Play: Editors versus Authors", Bulletin of Economic Research 57 (2005), pp. 1-12. FREY, B. S. and R. EICHENBERGER, "American and European Economics and Economists", Journal of Economic Perspectives 7 (1993), pp.185-193. GOODWIN, T. H. and R. D. SAUER, "Life Cycle Productivity in Academic Research: Evidence from Cumulative Publication Histories of Academic Economists", Southern Economic Journal 61 (1995), pp. 728-743. HICKS, D., "The Difficulty of Achieving Full Coverage of International Social Science Literature and the Bibliometric Consequences", Scientometrics 44 (1999), pp. 193-217. HODGSON, G. M. and H. ROTHMAN, "The Editors and Authors of Economics Journals: A Case of Institutional Oligopoly?", Economic Journal 109 (1999), pp. 165-186. KALAITZIDAKIS, P., T. P. MAMUNEAS & T. STENGOS, "Rankings of Academic Journals and Institutions in Economics", Journal of the European Economic Association 1 (2003), pp. 1346-1366. LABAND, D. N. and M. J. PIETTE, "The Relative Impacts of Economics Journals: 19701990", Journal of Economic Literature 30 (1994), pp. 640-666. LINER, G. H., "Core Journals in Economics", Economic Inquiry 40 (2002), pp. 138-145. LUBRANO, M. and C. PROTOPOPESCU, "Density Inference for Ranking European Research Systems in the Field of Economics", Journal of Econometrics 123 (2004), pp. 345-369. MIRUCKI, J., "A Visibility Analysis of the Scientific Production of Ukrainian Economists: 1969-94", Journal of Socio-Economics 28 (1999), pp. 221-232. NEDERHOF, A. J. and E. WIJK, "Profiling Institutes: Identifying High Research Performance and Social Relevance in the Social and Behavioral Sciences", Scientometrics 44 (1999), pp. 487-506. PORTES, R., "Economics in Europe", European Economic Review 31 (1987), pp. 1329-1340. RAMRATTAN, L. B. and M. SZENBERG, "Ranking of Economic Journals: A Statistical Survey and Analysis", American Economist 47 (2003), pp. 82-90. SUTTER, M. and M. KOCHER, "Patterns of Co-authorship among Economics Departments in the USA", Applied Economics 36 (2004), pp. 327-333.
The Effect of Model-Selection Uncertainty on Autoregressive Models Estimates

Islam Azzam Assistant professor of Finance, Department of Management American University in Cairo, 113 Kasr El Aini Street P.O. Box 2511, Cairo 11511, Egypt E-mail: iazzam@aucegypt.edu Abstract Model-selection uncertainty corresponds to the uncertainty about the true lag order of the autoregressive process that should be picked. This paper shows that all modelselection criteria perform poorly in small samples. Model-selection uncertainty adds to the bias and variability in the coefficient estimates for these examples, and appears to be most relevant for small sample sizes, which is the case for most annual macro data. Even if we try to correct this bias using the bootstrap bias reduction method, the bootstrap biasreduction method performs well in decreasing the bias of the coefficient estimates, but not without a cost. In particular, this method adds considerably to the variance and mean square error of these estimates. A comparison of several commonly used model-selection criteria shows that the SC outperforms the AIC and performs like the Shaos criterion with small subsample size. However, the Shaos criterion has to satisfy restrictive conditions. The SC is better than bootstrap model-selection criteria, such as the Shaos criterion, because the SC does not require subsampling when the optimal subsample is unknown. For larger subsample size, the AIC and SC outperform the Shaos criterion. Both the AIC and the Shaos criterion are inconsistent when I use the entire sample, but the Shaos criterion performs much worse than SC and AIC in that case. The problem is how to choose the optimal subsample size. Shaos criterion is very sensitive to the changes in the subsample size. Both the AIC and SC are computationally less demanding than Shaos criterion. Accordingly, the SC is preferable to the AIC and Shaos criterion. Keywords: Model selection criteria, AIC, SC, Shao criterion, and bootstrap. Jel Classification Codes: Co1, C13, C14, C15
1. Introduction
Here model-selection uncertainty corresponds to the uncertainty about the true lag order of the autoregressive process that should be picked. Among the many information criteria suggested in literature to choose the lag order, the most commonly used are Akaikes information criterion (AIC) and Schwarzs criterion (SC). The AIC aims to find the best prediction, while the SC aims to find the model with the highest posterior probability of being the true model. The SC is consistent estimator for the true lag order, while the AIC is not always consistent. In Ltkepohl (1993), an estimator of the VAR lag order 0 is called consistent if P lim = 0 . The
T
AIC is not consistent under any null hypothesis of order less than 0 . Paulsen and Tjstheim (1985)
81
argue that the AICs limiting probability for overestimating the order declines with increasing dimension K and is negligible for K 5. So, asymptotically the AIC picks the correct order almost with probability one if the multiple time series has high dimension. A consistent lag order selection criterion such as SC is better suited for the analysis of finite-lag order VAR models than the less parsimonious AIC. In contrast, the AIC is more appropriate for infinite-order autoregressions. Shao (1996) shows that AIC (Akaikes information criterion), Mallows Cp, delete-one cross validation, and Efrons bootstrap used in Freedman, Navidi, and Peters (1988) are asymptotically equivalent. Unless the optimal model is the largest model, those criteria do not provide consistent estimates of the difference in this prediction error between two competing models. The bootstrap selection criterion, however, is consistent if either smaller bootstrap samples (m) are taken or the n n residuals are inflated by a factor of where and m . m might be between log log n m m and n / log log n to minimize the length of some bootstrap confidence intervals. The modified bootstrap selection criterion substantially outperforms the Cp criterion. White (2000) suggests a general model-selection criterion which compares moments of a set of candidate models with a benchmark. This can be based on mean square prediction error or other criteria, such as predictive log likelihood. Under the restrictive assumption that the normalized vector of moments converges to a multivariate normal distribution, selecting the model with the best performance relative to the benchmark will yield asymptotically correct model selection. White (2000) shows that the Politis and Romanos (1994) stationary bootstrap can be used to estimate a consistent sampling distribution of the normalized moment vector. McFadden (1997) uses the R 2 as a selection criterion. The problem is to jointly estimate the coefficients by nonlinear least squares. He shows that the resulting estimator is consistent and asymptotically normal and the asymptotic covariance of this estimator can be estimated by a GaussNewton auxiliary regression under additional smoothing conditions. None of the model-selection criteria performs well with small sample sizes, giving rise to the problem of model-selection uncertainty. This uncertainty increases the bias and variability of the estimated coefficients. Ltkepohl (1993) indicates that selecting a higher lag order than the true lag order causes an increase in the mean square forecast errors of the VAR, while selecting a lower lag order than the true lag order generates autocorrelated errors. Braun and Mittnik (1993) show that estimates from a VAR whose lag order differs from the true lag order are inconsistent. Kilian (1998b) shows that ignoring model selection causes under-coverage for traditional confidence intervals for impulse response estimates of the VAR. Braun and Mittnik (1993) examine the variability in the impulse response estimates due to changes in the lag order of the autoregressive process. Using the traditional impulse responses that are sensitive to the order of the variables, they show that the impulse response estimates are sensitive to changes in the lag structures of their autoregressive model. Few analyses in the literature incorporate the lag-order uncertainty in calculating the impulse response estimates and their coverage accuracy in autoregressive models. Faraway (1992), in discussing the links between model selection and inference, suggests using pairwise resampling of the original data to avoid conditioning on the selected regression model in inference. He notes that the resulting bootstrap distribution will tend to be wider, reflecting the additional uncertainty arising from model selection. Other possible approaches in the literature include sample splitting and jack-knife methods. Hjorth (1994) suggests combining cross-validation techniques with the model-free bootstrap methods for inference in regression models. However, model-free resampling methods for time series are known to require very large samples for reasonable accuracy. Kilian (1998b) introduces a modified bias-correction bootstrap algorithm that reflects the true extent of sampling uncertainty in the regression estimates (Kilian, 1998b).
82
This endogenous lag order bootstrap algorithm recognizes that lag order selection is an integral part of the sampling procedure by re-estimating the lag order in each bootstrap iteration. Kilian (1998b) suggests that the endogenous lag order bootstrap algorithm should replace the standard bootstrap algorithm in applications. Kilian (1998b) shows that endogenizing the lag order choice improves coverage accuracy in small samples. As the uncertainty about the true lag order declines in large samples, the performance of the standard interval converges to that of the endogenous lag order interval. There are some problems with Kilians (1998b) algorithm. First, he uses a standard nonparametric resampling technique which is not consistent. Under resampling, the bootstrap fails due to violations of continuity conditions. The consistency of a bootstrap approximation requires the bootstrap distribution to be a consistent estimator of the underlying population distribution. It also depends on the continuity of the mapping between the population distribution and the distribution of the statistic we are examining. This continuity condition requires that small changes in the data process not lead to discontinuous changes in the parameter estimates. Incorporating the lag order selection as does Kilians (1998b) algorithm violates this continuity condition. Small changes in the modelselection criterion across different lag orders may cause large discontinuous changes in the parameter estimates. For example, in going from VAR (5) to VAR (4), the value of the AIC may slightly change, while the estimated parameter on the fifth lag order changes from a finite value to zero. Politis and Romano (1994) show that reducing the bootstrap sample size (subsampling) can remedy this bootstrap failure. A key requirement for subsampling to work is that the ratio of the bootstrap subsample size (m) to the original sample size (n) goes to zero as the original sample size goes to infinity. Second, the validity of Kilians (1998b) procedure depends on the choice of a suitable lag order selection criterion. Both exogenous and endogenous lag order algorithms are asymptotically valid when a consistent lag order selection criterion is used. However, Kilian (1998b) uses the Akaike information criterion (AIC) which is only consistent if the true model is the largest model. This means both exogenous and endogenous lag order algorithms are not asymptotically valid. The previous literatures use bootstrap resampling to account for the effect of model-selection uncertainty on the model estimates; however, bootstrap resampling is inconsistent. In this paper, I use Monte Carlo simulation based on an empirical example to decompose the effects of the model-selection uncertainty to the bias of the coefficient estimates, the bias of their standard errors, and the variability of those estimated coefficients in the autoregressive models. Higher variability and bias in the coefficient estimates means the model is less reliable and less robust as predictions vary across different lag orders. When the coefficient estimates vary across models, the model-selection uncertainty becomes a larger issue with small sample sizes. The performance of the bootstrap bias-reduction method in reducing the bias in the coefficient estimates is evaluated at small sample sizes. I also make a comparison between the most popular model-selection criterion, the Akaike information criterion (AIC) with the Schwarz criterion (SC) and Shaos criterion which minimizes the bootstrap estimates of the prediction error. Shao (1998) derives a criterion for model selection that modifies the bootstrap selection procedure to make it consistent through subsampling which takes bootstrap sample size smaller than the original sample size. As the optimal size of the bootstrap subsample is unknown, I examine the sensitivity of Shaos criterion to the changes in the bootstrap subsample size.
2. Model-Selection Criteria
The AIC and SC are based on minus 2 times the average log likelihood function, adjusted by a penalty term as follows:
83
AIC = 2l + 2k n n SC = 2l + k log n n n where (1) nq n l = (1 + log 2 ) log 2 2 ' = det( ) n where k is the number of estimated parameters, n is the number of observation, l is the value of the realized log likelihood function, and q is the number of equations. Shao (1996) considers the case an autoregressive series of unknown order that must be estimated using the data. A series {yt , t = 0, 1, 2, } is called an autoregressive time series of order max if yt = 1 yt 1 + 2 yt 2 + + max yt max + t , (2) t = 0, 1, 2, where i , i = 1, , max , are unknown parameters, and the t are iid random variables with mean zero and variance 2 . Estimating the order can be formulated as a model selection problem in which we 1 select a model from A = { , , max } and each corresponds to autoregressive model of order : yt = 1 yt 1 + 2 yt 2 + + yt + t
zt = ( yt 1 , , yt )
(3)
t = 0, 1, 2, Under model , = (1 ) is estimated by least square estimation (LSE). The modified bootstrap model selection starts by letting t , t = 0, 1, 2, , be iid from the distribution putting ) mass n 1 on ri r , i = 1, , n , where ri = yi z i' max max is the i residual under the largest model = . The bootstrap analog * is estimated by replacing n by m and with y replaced by
* max
,m
y = z t + t*
* t
(4)
The model selected by the modified bootstrap, denoted n ,m , is the minimizer of the expected in sample prediction error n ( y z * )2 t ,m n ,m ( ) = E* t , = 1, , M (5) n t =1 n ( y z * )2 t , m I calculate t at each bootstrap iteration, and then take the average over all n t =1 bootstrap iterations to estimate ( ) .
n ,m
3. Model-Selection Uncertainty
Model-selection uncertainty potentially arises in most empirical time-series models involving empirical applications of economics, finance and many other fields, such as physics. Yet there is no wellaccepted way of dealing with this problem in the classical approach. Using Shaos (1996) definition of model-selection uncertainty, in regression analysis there is a vector x of explanatory variables to be used to fit a model between x and a response variable y. x = { yt 1 , yt 2 , , yt } . Because some of the components of x may not be related to y, using all
84
components of x need not produce a better model than using part of components of x. And, since the related performance of each model (corresponding to a set of components of x) is usually unknown, we have to select a subset of explanatory variables (components of x) based on a data set with a total number of observations n {( xi , yi ), i = 1, , n} where yi is the response at x = xi . This variable selection problem is equivalent to a model selection problem in which each model corresponds to a particular set of the components of x where true 0 is unknown. I collect quarterly data from International Financial Statistics about Austrias GDP and private consumption for the period from 1964-Q1 to 2006-Q1. These series are seasonally adjusted. I consider eight autoregressive representations, AR () models where =1,.,8 as follows: yt = 1 + 2 ct + 3 y t 1 + + + 2 y t + u t (6) and yt = log(GDP) t ct = log( private consumption) t I do not include lags for consumption ( ct ) to simplify the analysis. I start all the AR () process from the same point of time to have equal sample size for each process, so my sample after adjustment is from 1966:Q2 to 2006:Q1. The coefficient estimates of these eight-candidate autoregressive processes along with their standard deviations are reported in Table 1. There is no serial autocorrelation in any of these candidate models. For purpose of generating the data in the Monte Carlo simulation, AR (4) specification is taken as the true model. In Table 1, the constant 1 is significant across all AR () models with the exception of = 8, while and are significant
2
across all models. Table 1 shows that 2 is stable until the third lag then it significantly changes over lag orders indicating that model-selection uncertainty might be a critical issue to be examined. In order to mimic the true behavior of the data and the relation between yt and ct , I use a VAR
(4) specification to only generate ct which will be held constant and outside my Monte Carlo simulation loop. My estimated VAR (4) where the standard deviations are in parentheses and estimated coefficients with asterisk are significant at 95% confidence level is as follows: yt = 0.009* + 0.089 ct 1 0.091ct 2 + 0.082 ct 3 + 0.089 ct 4 0.168 yt 1
( 0.003) ( 0.078 ) ( 0.087 ) ( 0.086 ) ( 0.079 ) ( 0.100 )
0.016 yt 2 0.005 yt 3 + 0.461* yt 4

( 0.109 ) ( 0.111) * ( 0.098)
ct = 0.010 0.517 ct 1 0.518* ct 2 0.333* ct 3 + 0.320* ct 4 + 0.507 * yt 1

* ( 0.004 ) ( 0.103) ( 0.114 ) ( 0.113) ( 0.104 ) ( 0.132 )
(7)
+ 0.385 yt 2 0.223 yt 3 + 0.290 yt 4

* * ( 0.145 ) ( 0.147 ) ( 0.130 )
yt
ct
Residual correlation matrix = yt ct
1 0.5 0.5 1
85
Table 1:

The estimated coefficients of the AR () using actual data
AR (1) 0.0146* (0.0019) 0.4099* (0.0435) -0.2277* (0.0689) _ _ _ _ _ _ _ AR (2) 0.0157* (0.0026) 0.3969* (0.0479) -0.2305* (0.0692) -0.0495 (0.0754) _ _ _ _ _ _ AR (3) 0.0168* (0.0033) 0.3942* (0.0483) -0.2424* (0.0729) -0.0577 (0.0772) -0.0381 (0.0719) _ _ _ _ _ AR (4) 0.0085* (0.0033) 0.2574* (0.0488) -0.1750* (0.0658) -0.0046 (0.0692) 0.0123 (0.0644) 0.4190* (0.0712) _ _ _ _ AR (5) 0.0068* (0.0033) 0.2251* (0.0509) -0.2705* (0.0811) -0.0182 (0.0687) 0.0383 (0.0650) 0.4632* (0.0739) 0.1606* (0.0816) _ _ _ AR (6) 0.0072* (0.0035) 0.2184* (0.0535) -0.2607* (0.0846) 0.0028 (0.0852) 0.0399 (0.0654) 0.4624* (0.0742) 0.1527 (0.0839) -0.0357 (0.0847) _ _ AR (7) 0.0078* (0.0036) 0.2208* (0.0537) -0.2703* (0.0857) 0.0185 (0.0879) 0.0798 (0.0844) 0.4631* (0.0743) 0.1476 (0.0844) -0.0516 (0.0875) -0.0614 (0.0821) _ AR (8) 0.0054 (0.0036) 0.1994* (0.0528) -0.2585* (0.0835) 0.0529 (0.0864) 0.0258 (0.0845) 0.3330* (0.0862) 0.1391 (0.0822) -0.0477 (0.0852) 0.0039 (0.0833) 0.2237* (0.0806)
1 2
3 4 5 6 7 8
9
10
Estimated coefficients with asterisk are significant at 95% confidence level.
The correlation coefficient between yt and ct is 0.60. The first eight actual values of yt and ct are used as initial values. I recursively generate a new series of yt from the AR (4) specification using the ct , which is generated outside the loop from the VAR (4), for each Monte Carlo iteration. The new simulated series are used to estimate the coefficients on each of the AR () for =1,., 8. I tested for the stationarity of the VAR (4) in equation.7, by which I generate ct . This estimated VAR (4) is stationary. The estimated AR (4) for generating yt is also stationary. The analysis uses the AIC and SC for lag order selection because they are the most popular ones, easy to calculate and they are similar to many other criteria in inference. For each iteration, I calculate the AIC for each model and record the order of the process that minimizes this AIC. Then, I record 2 and its standard deviation for the process that was selected and for AR (4) which is considered to be the true model. Using the variance of instead of its standard
2
deviation does not make any difference in results. Table 2 compares 2 from one iteration of simulated data and from the true data with N = 140. The standard deviations of are between parentheses.
2 2
Table 3 shows the AIC probability distribution of selecting different lag orders at different sample sizes for 10,000 Monte Carlo iterations. From Table 3, we can see that the AIC is an inconsistent estimator. The probability for the AIC to pick the true model increases as I increase the sample size until N=140, then it is non-increasing. The AIC is not accurate with small sample sizes. The probability that the AIC selects the true model with N=40 is 31%, implying that there is a big chance not to pick the true model.

Table 2:
86
Comparing 2 values using true and one iteration of simulated data (N=140) 2
from true data 0.4099 (0.0435) 0.3969 (0.0479) 0.3942 (0.0483) 0.2574 (0.0488) 0.2251 (0.0509) 0.2184 (0.0535) 0.2208 (0.0537) 0.1994 (0.0528)
2 from simulated data

0.3763 (0.0430) 0.3603 (0.0446) 0.3509 (0.0472) 0.2979 (0.0482) 0.2903 (0.0496) 0.2860 (0.0492) 0.2900 (0.0496) 0.2902 (0.0498)
AR (1) AR (2) AR (3) AR (4) AR (5) AR (6) AR (7) AR (8)
The AIC tends to pick higher lag orders as I increase the sample size. It is worth mentioning that, in macroeconomic studies based on annual data, the researcher typically has at most 40 observations. For many developing countries, the sample size is even shorter.
Table 3: The AIC probability distribution of selecting AR (), = 1,.,8, using 10,000 Monte Carlo Repetitions
N=40 0.218 0.103 0.083 0.312 0.098 0.061 0.054 0.067 N=80 0.019 0.008 0.005 0.674 0.124 0.076 0.048 0.044 N=140 0.0005 0.0001 0 0.718 0.125 0.070 0.046 0.038 N=300 0 0 0 0.728 0.123 0.069 0.043 0.034 N=1000 0 0 0 0.732 0.120 0.065 0.044 0.037 N=10,000 0 0 0 0.742 0.118 0.066 0.041 0.031
P=1 P=2 P=3 P=4 P=5 P=6 P=7 P=8
What is the effect of this model-selection uncertainty on the variability and bias of 2 ? To estimate the variability due to model-selection uncertainty, I compare the conditional standard deviation of 2 , which is conditional on AR (4), with the unconditional standard deviation of 2 using the AIC as model-selection criterion. For each iteration, I save 2 and its standard deviation for the lag order that is selected by the AIC and for the AR (4) which is my true model in this example. The unconditional standard deviation of 2 is the standard deviation of all the 10,000 2 s saved over Monte Carlo repetitions for the model selected by the AIC. The unconditional standard deviation is the standard deviation of the histogram a in Figure 1. The conditional standard deviation of 2 is the standard deviation of all the 10,000 2 s estimated for AR (4) over repetitions. This is the standard deviation of the histogram b in Figure 1. The percentage difference between the unconditional standard deviation and the conditional standard deviation of 2 is the percentage variability in 2 due to model-selection uncertainty. To estimate the bias in due to model-selection uncertainty, I estimate the unconditional mean of
2 2
which is the mean of 2 for the lag orders that were selected by the AIC over Monte Carlo repetitions.
87
This is the mean of histogram a in Figure 1. I also estimate the conditional mean of 2 which is the mean of for AR (4) over repetitions. This is the mean of histogram b in Figure 1. The bias in
2 2
due to model-selection uncertainty is the difference between the unconditional and conditional mean of 2 . Finally, I estimate the bias in the conditional standard deviation of . The difference between
2
the conditional standard deviation of 2 , which is the standard deviation of the histogram b in Figure 1, and the mean of standard deviations of 2 for AR (4), which is the mean of the histogram d in Figure 1, is the percentage bias in the conditional standard deviation of .
2
Summarizing, we have 1. Unconditional standard deviation of 2 = the standard deviation of the 10,000 2 s estimated from the model selected by AIC over Monte Carlo iterations. 2. Conditional standard deviation of 2 = the standard deviation of the 10,000 2 s estimated from the AR (4), which is considered to be the true model, over Monte Carlo iterations. 2 due to model-selection uncertainty 3. The percentage variability in Unconditional Std of 2 Conditional Std of 2 . = Conditional Std of 2 4. The percentage bias in due to model-selection uncertainty =
2
Mean of 2 s for selected mod el Mean of 2 s for true mod el Mean of s for true mod el
2
5. The
bias in the conditional standard deviation of Conditional Std of 2 Mean of S tan dard deviations of 2 for true mod el Mean of S tan dard deviations of for true mod el
2
percentage
Table 4 reports the unconditional and conditional mean of 2 over 10,000 Monte Carlo repetitions. The last row in Table 4 shows the percentage bias in due to model-selection
2
uncertainty. It shows that the bias in 2 dies away as we increase the sample size. For all sample sizes, the bias of is less than one standard deviation of for the true model, which can be considered
2 2
insignificant. This bias of 2 in a small sample size of N = 40 is 10%. This bias with N = 80 and larger sample sizes is negligible. For a sample size of 10,000, the bias in is zero. In addition, the
2
unconditional and conditional means of 2 , using simulated data, are equal to the estimated 2 using the true data. Accordingly, model-selection uncertainty does not seem to affect the bias of .
2
Table 5 shows that the model-selection uncertainty is a larger issue in a small sample size of N = 40. With N = 40, the model-selection uncertainty account for 26 % of the variability in 2 . This model-selection uncertainty becomes insignificant at larger sample sizes. These results are consistent with the results in Table 3 which shows that the probability of the AIC selecting AR (4) for N = 40 is 30% while this probability jumps to 71% for N = 140. Then this probability becomes stable and nonincreasing.
88
Figure 1: A- histogram of 2 for model selected over 10,000 monte carlo (mc) repetitions; B- histogram of 2 for AR (4) over mc repetitions; C- the standard deviation of 2 for model selected over mc repetitions; D- the standard deviation of 2 for ar (4) over MC repetitions; with N = 40.
beta of model selected for n= 40-AIC 5000 4000 a 3000 2000 1000 -0.2 5000 4000 b 3000 2000 1000 -0.2 5000 4000 c 3000 2000 1000 -0.2 5000 4000 d 3000 2000 1000 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 Std of beta for AR(4) for n= 40 0.7 0.8 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
beta of AR(4) for n= 40
Std of beta for model selected for n= 40
Table 4:
The bias in 2 over 10,000 monte carlo repetitions

N=40 0.334 0.303 10.00% N=80 0.277 0.274 1.09% N=140 0.263 0.263 0.03% N=300 0.260 0.260 0.03% N=1000 0.258 0.258 0.00% N=10,000 0.257 0.257 0.00%
Uncond. mean of 2 Cond. mean of for AR (4)

2
Percent. bias in
89
In Table 5, the model-selection uncertainty for N = 140 and larger is less than one percent. At N = 10,000, variability and bias in 2 due to model-selection uncertainty are zeros. Accordingly, we should be watchful of the model-selection uncertainty in small sample sizes as it account for one fourth of the variability in the estimated coefficient in my example. This is likely to be very serious problem for the macroeconomist who conducts studies using annual data. Table 5 also shows that the bias in the conditional standard deviation of 2 is very small even for N = 40 where this bias accounts for 4%. The variability of due to model-selection uncertainty is
2
dying away as the sample size increases. I also used the SC as a model-selection criterion. With N = 40, the variability of 2 due to model-selection uncertainty is 23% using the SC as model-selection criterion. This variability due to model-selection uncertainty dies away as the sample size increases.
Table 5:
The effect of model-selection uncertainty on the bias and variability of 2 over 10,000 monte carlo repetitions, using the aic as model-selection criterion (STD = STANDARD DEVIATION)
N=40 0.1165 0.0924 0.0831 0.0967 4.43% 26.05% N=80 0.0688 0.0639 0.0625 0.0629 1.58% 7.66% N=140 0.0534 0.0529 0.0521 0.0522 1.34% 0.94% N=300 0.0313 0.0311 0.0308 0.0308 0.97% 0.64% N=1000 0.0168 0.0167 0.0166 0.0166 0.60% 0.59% N=10,000 0.0052 0.0052 0.0052 0.0052 0.00% 0.00%
Unconditional Std. of 2 Conditional Std. of 2 Mean of Stds of 2 for model selected Mean of Stds of 2 for AR(4) Percentage bias in the conditional Std of 2 Percentage variability in 2 due to the model-selection uncertainty
In order to examine the robustness of my results, I use another example. Using the same methodology and data range with the AIC as model-selection criterion, I examine the model-selection uncertainty using US GDP and consumption data for the same period. The analysis found that 2 changes little among different lag orders. This indicates that the effect of model-selection uncertainty is expected to be small. With N = 40, the percentage bias in 2 due to model-selection uncertainty is 0.8% while the percentage bias in the conditional standard deviation of is 2%. The AIC probability
2
to pick the true model is 55%. The variability in 2 due to model-selection uncertainty is 12% in this US example. This indicates that when we have little variability in among different lag orders,
2
model-selection uncertainty accounts for a small portion of the variability in 2 . But, returning to the results for the Austrian data, Table 4 shows that the bias in 2 for N = 40 is 10%, which might be significant for macroeconomists. Can we correct for this bias using the bootstrap bias reduction method? In the next section, I examine the performance of the bootstrap bias reduction method in correcting for bias in 2 .
4. Bootstrap Bias Reduction

How can one use the bootstrap to reduce the finite-sample bias of an estimator? Horowitz (1999) suggests the following procedure for bootstrap bias reduction: a. Use the data to estimate n .
International Research Journal of Finance and Economics - Issue 11 (2007) b-
90
cde-
Generate a bootstrap sample of size n by sampling the estimation data randomly with * replacement, and then compute n = g ( X * ) . Compute E * * by averaging the results of many repetitions of step b, and then set
n
* = E n n .
* n *
* Form the new reduced-bias estimator n n . * Estimate E ( n 0 ) and E ( n n 0 ) by averaging the results of many repetitions, and then estimate the mean square errors of and * by averaging the realizations
n n n
* of ( n 0 ) 2 and ( n n 0 ) 2 . Following the previous procedure with N=40 and 0 = 0.2574 which is the estimated 2 from the true data, I use paired bootstrap with 5,000 Monte Carlo and 500 bootstrap repetitions to reduce the bias of 2 . In paired bootstrap, we randomly sample pairs ( yi , xi ) with replacement and re-compute the least squares estimator * for each sample. Table 6 shows the mean and standard deviation of
n n
* and n n over Monte Carlo repetitions.

Table 6:
The bootstrap bias reduction for 2 using N=40 and 5,000 monte carlo repetitions with 500 bootstrap repetitions
N=40
Mean of n
Standard deviation of Mean of
0.3449
n * n n n
0.1182 0.2142 0.2194 0.0874 0.0216 0.0433
* n n
Standard deviation of Bias of
n * n n * n n
Mean squared error of Bias of
Mean square error of
0.0500
Table 6 shows that the bootstrap bias reduction method reduces the magnitude of the bias of 2 by 50%, while the mean square error increased by 131%. Hence, the bias of fell by one-half at the
2
expense of increasing the MSE by a factor of 1.3. The bootstrap bias reduction method succeeded in reducing the bias in 2 very well, but we cannot depend on this method because the MSE increased substantially as reflected in the increase in the standard deviation of * . It is not recommended to
2
use this bootstrap bias reduction method, as its disadvantages are much more than its advantages.
5. A Comparison Between The AIC, SC and SHAOS Criterion

The consistency of a bootstrap approximation depends on the continuity of the mapping between the population distribution and the distribution of the statistic we are examining. The Shaos criterion is consistent only when m and m 0 . I compare the probability distribution of the AIC, SC and n
91
Shaos criterion in picking the true model using the Austria example. As the optimal size of the subsample m is unknown, I examine the sensitivity of the Shaos criterion to the choice of m. Table 7 reports the probability distribution of the AIC and SC in picking AR () for = 1,,8, using 10,000 Monte Carlo repetitions. It also reports the probability distribution for the Shaos criterion to select different lag orders over 5000 Monte Carlo repetitions and 100 bootstrap repetitions using different subsample sizes (m). m might be between log log n and n/log log n to minimize the length of some bootstrap confidence intervals. So, for N = 80, N = 140 and N = 1000, m might be in these ranges 1 - 54, 1 - 87, and 1 - 517, respectively. In Table 7, m/n with asterisk is between log log n and n/log log n. Table 7 shows that the SC is consistent estimator as it has a probability of 0.99 for picking the true model at N = 1000 while the AIC is not a consistent estimator as the true model is not the largest model. Using a sample size of N = 80 or larger, the SC outperforms the AIC in picking the true model. Both the AIC and SC outperform the Shaos criterion for large m. For small m, the SC performs like the Shaos criterion, which has to satisfy restricted conditions. The SC is not required to satisfy certain conditions like the bootstrap model-selection criteria. The SC is much easier and needs less computing time than the Shaos criterion. Accordingly, the SC appears to be preferable to the AIC and Shaos criterion for sample size equal or bigger than 80. Using a small sample size of N = 40, neither the AIC, the Shaos criterion nor the SC perform well. With N = 40, the probabilities of the AIC, SC and Shaos criterion, with resampling, to pick the true model are 0.31, 0.20 and 0.33, respectively. It is hard to subsample from a small sample size, such as N = 40. None of the model-selection criteria considered here is reliable with small sample sizes. Table 7 shows that the SC tends to pick lower lag order for small sample sizes than the AIC and Shaos criterion. For N = 40, the probabilities that the SC and AIC pick order lower than AR (4) are 71% and 39%, respectively. Also, for a sample size of N = 80, these probabilities for SC and AIC are 22% and 2%, respectively. As the sample size increases, the SC tends to pick the true model. With N = 140, both the Shaos criterion with m/n = 30% and SC have a probability of about 94% to pick the true model, while the AIC has a probability of 71%. How sensitive is the Shaos criterion to changes in m? Table 7 shows that the probability of the Shaos criterion to pick AR (4) is very sensitive to the changes in m. For some choices of m, the AIC performs better than Shao. For N = 1000, Shaos criterion, without subsampling, has a 13% probability of selecting the AR (4) while the AIC has a probability of 73%. At the same time, for N = 1000 with m/n = 10%, the Shaos criterion has 99% probability of picking the AR (4)! For large m, the AIC outperforms the Shaos criterion. Accordingly, the Shaos criterion and any other bootstrap modelselection criteria that depend on subsampling are very sensitive to the choice of the subsample size and have to satisfy the condition m and m 0 to be consistent. Thus, bootstrap model-selection n criteria are not robust as the optimal size of m is unknown.

Table 7:
92
The probability distribution for the AIC, sc and shaos criterion to pick AR () using 10,000 monte carlo repetitions for the aic and sc, and 5000 monte carlo with 100 bootstrapping repetitions for the shaos criterion
AIC Shao m/n=100% SC AIC Shao* m/n =45% Shao* m/n =60% Shao m/n =100% SC AIC Shao* m/n =30% Shao* m/n =40% Shao* m/n =50% Shao m/n =70% Shao m/n =100% SC AIC Shao* m/n =10% Shao* m/n =30% Shao m/n =100% SC AR (1) 0.21 0.02 0.51 0.01 0.01 0 0 0.17 0 0 0 0 0 0 0.02 0 0 0 0 0 AR (2) 0.10 0.03 0.12 0 0.08 0 0 0.04 0 0 0 0 0 0 0 0 0 0 0 0 AR (3) 0.08 0 0.08 0 0 0 0 0.01 0 0 0 0 0 0 0 0 0 0 0 0 AR (4) 0.31 0.33 0.20 0.67 0.80 0.56 0.20 0.72 0.71 0.94 0.79 0.61 0.33 0.14 0.93 0.73 0.99 0.86 0.13 0.99 AR (5) 0.09 0.09 0.03 0.12 0.05 0.06 0.05 0.03 0.12 0.02 0.06 0.05 0.05 0.02 0.02 0.12 0 0.04 0.01 0 AR (6) 0.06 0.32 0.01 0.07 0.10 0.29 0.42 0 0.07 0.02 0.12 0.26 0.42 0.56 0 0.06 0 0.07 0.56 0 AR (7) 0.05 0.09 0 0.04 0 0.04 0.11 0 0.04 0 0.01 0.03 0.09 0.10 0 0.04 0 0 0.12 0 AR (8) 0.06 0.07 0 0.04 0 0.02 0.19 0 0.03 0 0 0.02 0.08 0.15 0 0.03 0 0 0.16 0
Conclusions
The examples considered in this paper show that all model-selection criteria perform poorly in small samples. Model-selection uncertainty adds to the variability in the coefficient estimates for these examples, and appears to be most relevant for small sample sizes, which is the case for most annual macro data. The bootstrap bias-reduction method performs well in decreasing the bias of the coefficient estimates, but not without a cost. In particular, this method adds considerably to the variance and mean square error of these estimates. As such, we must exercise caution when applying this bootstrap method to reduce the bias of our estimators. A comparison of several commonly used model-selection criteria shows that the SC outperforms the AIC and performs like the Shaos criterion with small subsample size. However, the Shaos criterion has to satisfy restrictive conditions. The SC is better than bootstrap model-selection criteria, such as the Shaos criterion, because the SC does not require subsampling when the optimal subsample is unknown. For larger subsample size, the AIC and SC outperform the Shaos criterion. Both the AIC and the Shaos criterion are inconsistent when I use the entire sample, but the Shaos criterion performs much worse than SC and AIC in that case. The problem is how to choose the optimal subsample size. Shaos criterion is very sensitive to the changes in the subsample size. Both the AIC and SC are computationally less demanding than Shaos criterion. Accordingly, the SC is preferable to the AIC and Shaos criterion.
N = 1000
N = 140
N = 80
N=40
Shao m/n with asterisk is between log log n and n/log log n. The numbers less than 1% are set to zero
93
International Research Journal of Finance and Economics - Issue 11 (2007) Andrews, D. W. K. (1997). A simple counterexample to the bootstrap. Cowles foundation Discussion Chapter No. 1157, Yale University, August, 1997. Braun, P. and Mittnik, S. (1993). Misspecifications in vector autoregressions and their effects on impulse responses and variance decompositions. Journal of Econometrics, 59, 319-341. Burnham, K. and Anderson, D. (2002). Model-selection and Multimodel Inference: A Practical Information-Theoretical Approach. Springer-Verlag, New York. Faraway, J. F. (1992). On the cost of data analysis. J. Comput. Graph. Stat. 1 (3), 213-29. Freedman, D. A., Navidi, W., and Peters, S. C. (1988). On the impact of variable selection in fitting regression equations. In T. K. Dijkstra, ed., On Model Uncertainty and its Statistical Implications, Springer-Verlag, New York. Hall, P. (1992). The Bootstrap and Edgeworth Expansion. Springer Verlag, New York. Hastings, W. K. (1970). Monte Carlo sampling using Markov chains and their applications. Biometrika 57, 97-109. Hjorth, J. S. U. (1994). Computer Intensive Statistical Methods: Valuation, Model-selection, and Bootstrap. London: Chapman and Hall. Horowitz, J. (2001). "The Bootstrap." Handbook of Econometrics, Vol. 5, J.J. Heckman and E.E. Leamer, eds., Elsevier Science B.V., Ch. 52, pp. 3159-3228. International Financial Statistics. Data CD, IMF. Kilian, L. (1998). Accounting for lag order uncertainty in autoregressions: The endogenous lag order bootstrap algorithm. Journal of Time Series Analysis. 19 (5), 531-547. Ltkepohl, H. (1993). Introduction to Multiple Time Series Analysis. Berlin, Heidelberg: Springer-Verlag. McFadden, D. (1997). A note on selecting variables to maximize their significance. Department of Economics, University of California, Berkeley, February 1997. Paulsen, J. and Tjstheim, D. (1985). On the estimation of residual variance and order in autoregressive time series. Journal of the Royal Statistical Society. B 47, 216-228. Politis, D. and Joseph P. R. (1994). The stationary bootstrap. Journal of the American Statistical Association 89(428): 1303 1313. Shao, J. (1996). Bootstrap model-selection. Journal of the American Statistical Association, June 1996, 91 (434), 655-665. White, H. (2000). A reality check for data snooping. Econometrica, September 2000, 68 (5), 1097-1126.
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17]
Foreign Aid to Developing Countries: Does it Crowd-out the Recipient Countries Domestic Savings?
Munther Nushiwat Division of Business, Marymount Manhattan College, New York, NY Abstract Earlier papers written on the subject of the effect of foreign aid on domestic savings disagreed on whether this effect is positive or negative. However they were empirically consistent in finding this correlation to be negative. In reassessing this correlation empirically in this paper, hypotheses are tested on whether lags in the effects of aid, or the source of aid were of importance in determining the sign of this correlation. The conclusion of this cross-country empirical investigation; is that, while the lags factor is not substantiated; estimating aid as separate variables according to its sources, bilateral aid or aid from multilateral institutions, suggests that the correlation between bilateral aid and savings is generally positive. Because the purpose of this paper is to investigate the cause of the negative correlation in the other papers despite the strong theoretical arguments for a positive effect, equations similar to those used in most of the papers on this subject are used here. Jel Classificaiton Codes: F35, O19
I. Introduction
Three stages can be distinguished in the debate, since the late sixties, on the subject of the effect of foreign aid to developing countries on the recipient countries domestic savings1. The first, considered foreign aid as additive to domestic savings and, therefore, would cause an increase in economic growth and domestic savings (Chenery and Strout, 1966)2. The second, stressed the negative effects of foreign aid on economic growth and domestic savings of the recipient countries. Arguing that causality runs from foreign aid to domestic savings and based on the negative correlation found in their econometric estimates, they concluded that the effect of aid on savings is negative (Griffin and Enos, 1970; Griffin, 1970; and Weisskopf, 1972)3. The third, suggested plausible savings functions that may imply positive or negative effects of aid on savings, and concluded that the effect of aid on savings is not necessarily negative (Papanek, 1972, 1973, and Over, 1975)4 . The major part of the papers written, after the early three stages of the debate, tried to estimate this relation using different models and making adjustments to the original models used. However, the results were consistent in indicating that the correlation between foreign aid and domestic savings is negative. For example, negative correlation was found in the estimates of Chenery and Eckstein
1 2
The original three stages of the debate are represented by the articles of the authors mentioned in this paragraph. In a Harrod-Domar model of growth in Chenery and Strouts (1966) article, foreign aid, added to domestic savings, relieves the two constraints on economic growth. These are the savings constraint and the foreign exchange constraint. The explanation given for the negative effect is: First, only a part of foreign aid is used for investment, the rest is used for consumption. Second, aid may lower the capital-output ratio causing growth to be lower and, therefore, lowering domestic savings. Some of these plausible functions do not apply to foreign aid, they apply to the other types of inflows.
95
(1970), Griffin and Enos (1970), Weisskopf (1972), Papanek (1973), Gupta (1975), Over (1975), Mosely (1980), Singh (1985), Bowles (1987), Snyder (1990), and Reinhart and Talvi (1998)5. The basis, in economic literature, of a beneficial effect of foreign aid is the original two gaps model, where foreign aid may ease the two constraints on economic growth of developing countries, namely the savings constraint and the foreign exchange constraint. The following two arguments can be advanced to support the plausibility of a positive effect of foreign aid on domestic savings: 1. The equation S = I F, where S, I, and F are domestic savings, gross investment, and net total foreign inflows respectively, is a behavioral equation of flows of these variables over time. It is not an accounting identity. An increase in the foreign inflows does not necessarily cause a reduction in domestic savings even if some of the inflows are used for consumption. This is because an increase in consumption demand can be met by an increase in the domestically produced consumption goods. This increase requires an increase in investment and output. Savings may increase depending on the savings function of the economy. However, domestic supply constraints may limit the increase in domestic output causing an increase in prices of non-tradable goods relative to those of tradable goods6. The domestic supply constraints may be eased when the recipient countries pursue active economic policies that target the barriers to economic growth as was demonstrated by the newly industrialized countries. The following three studies, of Eshag (1971), Burnside and Dollar (1997), and Levy (1988), support this argument. Eshag (1971) pointed out that even if foreign aid is used for consumption, the increase in consumption must be accompanied by an increase in output although some of it is met by an increase in imports. In Burnside and Dollars (1997) study, where foreign aid happened to coincide with good economic policies, it had a strong positive effect on economic growth. Levy (1988) found, in his study, that aid is positively related to the growth rate of both traded and non-traded goods in Sub-Saharan Africa. The increase in consumption demand is not likely to be all met by an increase in the price level or an increase in imports, especially with the prevailing interventionist economic policies in the developing countries. As a result of growth, savings are likely to increase since per capita income is the main determinant of savings. 2. Foreign aid, in the studies, meant in general the total foreign inflows to a recipient country. Usually it included foreign aid, foreign borrowing, and foreign investment. Foreign aids effect on economic growth and domestic savings can be distinguished from that of foreign borrowing and foreign investment in that, unlike the later two types, foreign aid does not cause an outflow of funds to pay back debt or repatriate profits and capital. A positive effect of foreign aid is more plausible than that of the general effect of all the foreign inflows. This paper is an attempt to reassess the correlation between foreign aid and domestic savings. This reassessment is important because the empirical negative correlation is the main support for the second stage of the debate on the effect of aid on savings. We take into consideration the following limitations which may be some of the causes of the empirical negative correlation in the earlier studies: 1. Using the Current Account Balance The very early estimates used the current account balance as a proxy variable for foreign aid of the recipient countries included in the studies7. It is plausible that the recipient developing countries
5
In Overs (1975) estimate, the correlation was negative when foreign aid was treated as the exogenous variable, but was positive when domestic investment was treated as the exogenous variable while foreign aid was assumed endogenously determined. The purpose of Overs article was to demonstrate that a minor difference in assumptions reverses the sign of the estimated parameter. In Reinhart and Talvis (1998) estimate, there was no difference in the negative correlation between the two regions, in their study, Latin America and Asia. The inflow of aid may cause an appreciation of the real exchange rate of the recipient country, negatively affecting its exports competitiveness. Reduction in exports and output may have a negative effect on domestic savings. According to Tsikatas (1998) survey of empirical studies, not much empirical work has been done on this subject, called the Dutch decease, but the few studies undertaken highlight the importance of an appropriate macroeconomic mix. For example, Chenery and Strout (1966), Griffin and Enos (1970), and Weisskopf (1972). Papanek (1972, 1973) pointed out the incorrect use of the current account balance as a proxy for foreign aid. He used, in stead, the disaggregated inflows (foreign aid, foreign debt, and direct foreign investment)) and estimated their correlation with growth and savings. Over (1975) used only foreign aid in his estimate. Gupta (1975) defined foreign
96
current accounts balances were narrowing as a result of their growing exportables and import substitutes (consider the case of South Korea as an example). At the same time, as a result of their economic growth, their domestic savings were increasing causing the correlation between foreign aid, measured as mentioned above, and domestic savings to be negative. The following two studies, of Mosley (1980) and Stewart (1971), support this argument. In comparing the experience of the seventies with that of the sixties, Mosley (1980) indicated that a number of countries (South Korea, Brazil, Taiwan, Colombia, and Iran) grew during the sixties while receiving aid, continued to grow during the seventies while their intake of aid declined suggesting a negative correlation between aid and savings. Stewart (1971) pointed out that high per capita income countries tend to have higher savings rates and smaller current accounts deficits than low per capita income countries. This negative correlation between savings and the current account deficit constitutes the main support for the literature that advocated a negative effect of aid on savings. 2. Lagged Effects of Aid The effect of foreign aid on economic growth and therefore on domestic savings may occur with two lags. In the first lag, foreign aid, at least initially, could have been used to build the basic infrastructure and the basic industries. The increase in economic growth could have followed with a lag of at least one year. The second lag is between economic growth and domestic savings. At the early stages of growth, savings are not expected to increase significantly until income had increased sufficiently. 3. The Main Determinant of Savings When the equations used to estimate the effect of aid on savings were modified, the correlation between them, although negative, was much smaller8. The equations used by Gupta (1975), Mosley (1980), and Snyder (1990) included per capita income which had a high positive, and statistically significant, effect on domestic savings suggesting that the negative effect of foreign aid in other estimates was overestimated. Snyder (1990) concludes, after testing, that per capita income is an important omitted variable from the savings equations of the earlier studies. The negative correlation, between aid and savings in Snyders estimate, became small and insignificant in the presence of the per capita income variable. 4. The Aid Variable The differences between bilateral and multilateral foreign aid to developing countries may include differences in their correlations with the recipient countries domestic savings, in addition to the differences in their sources and motivations. Bilateral aid is mainly determined by political considerations and in some cases granted specifically for development projects. Aid from multilateral institutions comes in most cases during periods of poor economic and political conditions, natural disasters, wars, and civil wars; therefore, it is not expected to be accompanied by significant economic growth and increased savings9. Regressing aid as two explanatory variables, bilateral and multilateral, does not imply that there are significant differences in their effects on savings. It is a difference in correlation of each of the two types with savings, although bilateral aid may have a stronger effect when assigned to specific development projects. Hellers (1975) study found that shifting from on type of aid to the other does not change the growth rate of a country.
aid as net transfers received by the recipient government plus official long-term borrowing. In general, the studies published after 1972 stopped using the current account balance as a proxy for foreign aid. Papanek (1973) included the other foreign inflows as determinants of domestic savings. Singh (1985) included a variable for the degree of state intervention in the economy. Papanek (1973) included the other foreign inflows as determinants of domestic savings. Singh (1985) included a variable for the degree of state intervention in the economy.
97
II. The Hypotheses

Based on the above four limitations of the previous studies, the correlation between foreign aid and domestic savings, when estimated, may be found positive if the estimate of this correlation: 1. Included an accurately defined foreign aid variable. 2. Allowed for a sufficient lag periods for the foreign aid variable. 3. The equation being estimated included the most likely determinant of savings, per capita income. 4. The foreign aid variable was represented by the two variables of its two main components, bilateral aid and multilateral aid, since it is plausible that each type shows a different correlation with domestic savings. Using multiple regression single equations in the estimates allows the comparison with the earlier studies, and is based on the following two assumptions: Assumption (1) Foreign aid, bilateral and multilateral, is exogenously determined. The findings by McKinlay (1978), and McKinlay and Little (1977, 1978, 1979) substantially support the hypothesis that the donors interest, not the recipients interest, is the major determinant in the allocation of bilateral aid in the cases of three major bilateral donors, The United States, The United Kingdom, and Germany10. Also, the testing results of Maizel and Nissanke (1984) confirm that bilateral aid allocations are made, largely or solely, to support the donors interest, while multilateral aid is allocated essentially based on the recipients need. Papaneks (1972, 1973) argument that foreign aid and savings are determined by exogenous factors following a crisis or natural disaster can be applied to multilateral aid only. Assumption (2) Per capita income is the major determinant of domestic savings. In reviewing the studies relating savings to income in developing countries, Mikesell and Zinser (1973) concluded that most studies show strong evidence of the positive effect of per capita income on savings. Specific other determinants can be important in individual cases. The sources of data are the International Financial Statistics, the Statistical Yearbook, and the National Accounts Statistics, Main Aggregates. Gross savings data include public and private savings. The countries included are from South America, South East Asia, South Asia, the Middle East, and Africa. They are included because they are recipients of significant amounts of foreign aid with available data.11 Three hypotheses are tested in this paper to support the papers argument that the correlation between foreign aid and domestic savings is generally positive: Hypothesis (1): The foreign aid variable is included as a determinant of domestic savings with sufficient lags. Its negative coefficient may become smaller or may become positive when the lag is extended to a longer period of time. Hypothesis (2): The foreign aid variable is represented by its two components, bilateral development assistance and assistance from multilateral institutions, as determinants of domestic savings with sufficient lags. At least, the coefficient of the bilateral aids variable, if negative, may become smaller or may become positive when the lag is extended to a longer period of time. Hypothesis (3): The estimate includes only the countries that have a higher ratio of bilateral to multilateral aid. The coefficient of the bilateral foreign aid variable, if negative, may become smaller or become positive if the lag was extended to a longer period of time. Each of the above three hypotheses is tested separately, and all the equations include per capita GDP as an explanatory variable.
10
11
The recipients need may play a role in this determination. A sudden need, like a natural disaster, may generate a request for aid from the needing country (Mosley, 1981). Data are available upon request from the author.
98
Based on the above discussion, stating the above hypotheses leads to the expectation that the correlation between aid and savings will be positive: 1. If the aid variable was sufficiently lagged. 2. If the aid variable is represented by its two components, at least bilateral aid will have a positive correlation with savings.
III. Estimation
The First Estimate The first hypothesis is tested using equation (1): (1) s = a0 + b0f + c0y Where, s = domestic savings divided by GDP. f = foreign aid divided by GDP. y = per capita GDP. The estimates of the regression equations below use 2003s data for domestic savings and per capita income in all equations. The foreign aid variable (total) is that of 2003s data in the first equation, then that of a one year lag, two years lag, and three years lag respectively. The estimates, in table (1), indicate that foreign aid has a negative, and statistically significant, correlation with domestic savings at the 10% level of significance. Per capita income has a positive, and statistically significant, effect on domestic savings at the 10% level of significance. In all the equations of table (1) the per capita income coefficient is equal to one plus a very small fraction. However, using a lagged variable of foreign aid does not show a positive correlation, nor does it show a reduction in the negative correlation. In the last equation the coefficient becomes higher but statistically less significant12. The findings here are consistent with the previous studies in finding a negative correlation between aid and savings.
Table 1:
(1) (2) (3) (4)
Regression Equations of the First Hypothesis

st st st st = = = = 0.169 (10.07) 0.172 (10.03) 0.167 (10.24) 0.154 (9.8) Foreign Aid is Aggregated in One Variable: 0.661ft + 1.0yt R2 = 0.304 (-3.27) (2.4) 0.757ft-1 + 1.0yt R2 = 0.319 (-3.51) (2.37) 0.706ft-2 + 1.0yt R2 = 0.305 (-3.28) (2.42) 0.979ft-3 + 1.0yt R2 = 0.257 (-2.46) (2.8) F = 13.54 F = 14.54 F = 13.59 F = 10.71 N = 64 N = 64 N = 64 N = 64
The values in parentheses are the coefficients t statistics. s = domestic savings divided by GDP. f = foreign aid divided by GDP. y = per capita GDP.
The Second Estimate The second hypothesis is tested using equation (2): s = a1 + b1f1 + c1f2 + d1y (2) Where, s = domestic savings divided by GDP. f1 = bilateral aid divided by GDP. f2 = multilateral aid divided by GDP. y = per capita GDP. The estimates of the regression equations below use 2003s data for domestic savings, multilateral aid, and per capita income in all equations. The bilateral foreign aid variable is that of 2003s data in the first equation, then that of a one year lag, two years lag, and three years lag respectively. The estimates, in table (2), indicate that the multilateral aid variable has a strong negative, and statistically significant, correlation with domestic savings, at the 10% level of significance. The
12
Using current and lagged aid variables shows that the most recent aid variable has a higher and statistically significant coefficient. For example: st = 0.169 0.657ft 0.009ft-3 + 1.0yt R2 = 0.304 F = 08.88 N = 64 (9.95) (-2.03) (-0.14) (2.38)
99
bilateral aid variable has a far lower and statistically insignificant coefficient, indicating a weak correlation with domestic savings relative to that of multilateral aid. Using lagged variables of bilateral aid does not show a positive correlation nor does it reduce the negative correlation13.
Table 2:
(1) st (2) st (3) st (4) st = = = =
Regression Equations of the Second Hypothesis

Foreign Aid is in Two Variables, Bilateral Aid and Multilateral Aid: 0.171 0.184f1t 1.310f2t + 1.0yt R2 = 0.325 (10.22) (-0.46) (-2.54) (2.39) 0.172 0.253f1t-1 1.266f2t + 1.0yt R2 = 0.326 (10.24) (-0.62) (-2.48) (2.33) 0.171 0.233f1t-2 1.297f2t + 1.0yt R2 = 0.326 (10.25) (-0.55) (-2.60) (2.36) 0.173 0.373f1t-3 1.242f2t + 1.0yt R2 = 0.330 (10.33) (-0.83) (-2.58) (2.33) F = 09.77 F = 09.86 F = 09.82 F = 10.01 N = 64 N = 64 N = 64 N = 64
The values in parentheses are the coefficients t statistics. s = domestic savings divided by GDP. f1 = bilateral foreign aid divided by GDP. f2 = multilateral foreign aid divided by GDP. y = per capita GDP.
The Third Estimate The third hypothesis is tested using equation (2), the same equation used in testing the second hypothesis. However, the data will include only those countries that have a higher ratio of bilateral aid to multilateral aid14 . The estimates, equations 1, 2, and 3 in table (3), indicate that, when the data include only the countries with high ratios of bilateral to multilateral aid, the coefficient of the bilateral aid variable becomes positive, although statistically insignificant. The coefficient of the multilateral aid variable is higher and still negative although statistically less significant compared with the estimates in table (2).
Table 3: Regression Equations of the Third Hypothesis Foreign Aid is in Two Variables, Bilateral Aid and Multilateral Aid.
= = = = 0.154 (4.63) 0.158 (4.70) 0.159 (4.82) 0.165 (4.79) Includes Only Countries with High Bilateral to Multilateral Aid. + 0.351f1t 2.003f2t + 1.0yt R2 = 0.226 (0.52) (-1.62) (1.14) + 0.084f1t-1 1.675f2t-1 + 1.0yt R2 = 0.218 (0.111) (-1.28) (1.07) + 0.044f1t-2 1.619f2t-2 + 1.0yt R2 = 0.218 (0.063) (-1.35) (1.07) 0.208f1t-3 1.420f2t-3 + 1.0yt R2 = 0.221 (-0.314) (-1.33) (0.97) F = 02.62 F = 02.52 F = 02.51 F = 02.46 N = 30 N = 30 N = 30 N = 30
(1) st (2) st (3) st (4) st
The values in parentheses are the coefficients t statistics. s = domestic savings divided by GDP. f1 = bilateral foreign aid divided by GDP. f2 = multilateral foreign aid divided by GDP. y = per capita GDP.
IV. Conclusions
While their theoretical arguments varied in advocating positive and negative effects, the earlier studies on the subject of the effect of foreign aid on the recipient countries domestic savings were consistent in finding an empirical negative correlation between the two variables. As mentioned earlier, this negative correlation was the main support for the argument of a negative effect of aid on savings. Some of the limitations of earlier studies were taken into consideration in this paper. However, the estimates, in this paper like earlier estimates, suffer from the limitations of the single equation international crosscountry analysis. Changing assumptions, countries, and periods can change the results significantly.
13
14
Using lagged multilateral variable does not change the results of table (2). For example: st = 0.169 0.220f1t 1.51f2t-3 + 1.0yt R2 = 0.311 F = 09.17 (10.03) (-0.53) (-2.26) (2.44) Except for one country that has a close to equal bilateral to multilateral ratio. This brings the sample size up to 30 countries.
N = 64
100
The results of the estimates are only suggestive. Also, as stressed by Mikesell and Zinser (1973), caution must be exercised in cross-country analysis of data because of the unreliability of data and the varying methodology employed in the derivation of data between countries. The following conclusions can be drawn from the results of the estimates in this paper: 1. There are no lags in the effect of foreign aid on domestic savings. The results of the estimates show consistently that domestic savings are more likely to be correlated with current rather than lagged foreign aid. The argument that there is a lag in the effect of foreign aid is not substantiated. There remains the possibility that during the early stages of economic development, foreign aid affected savings with a significant lag because the recipient countries economies had not grown sufficiently to allow for savings to increase (the 1950s and 1960s)15. 2. The results of the second and third estimates, tables II and III, although only suggestive, are consistent in indicating that the source of foreign aid is important in determining its coefficient. The statistically significant negative correlation is between multilateral aid and domestic savings. When foreign aid is aggregated, like in all earlier studies, its strong negative correlation is caused by the inclusion of multilateral aid in the aggregated variable. This does not mean that foreign aid, bilateral or multilateral, crowds-out the recipient countrys domestic savings, although it is not ruled out in some cases. 3. The effect of bilateral aid on the recipient countries economic growth and domestic savings is likely to be positive as suggested by the above estimates. 4. Despite multilateral aids high and statistically significant negative correlation with savings, its effect on savings is not necessarily negative. Without multilateral aid during difficult times, the recipient countrys domestic savings are likely to be even lower. High multilateral aid and low savings are caused by exogenous conditions like deteriorating economic and political situations, disasters, draughts, wars, and civil wars. As papanek (1972, 1973) pointed out, the strong negative correlation between domestic savings and foreign aid, in many cases, is more likely to be the result of exogenous factors affecting both of them rather than a causal relationship between them. Papaneks argument, as suggested by the estimates results in this paper, appears to apply only to multilateral aid. Poor countries and countries passing through a crisis have low savings and generally have high inflows of multilateral aid. 5. The equations used, in the above estimates, explain less than one third of the variation in domestic savings, in statistical terms. And the per capita income coefficient (the marginal propensity to save) is consistently close to one. This suggests that the equations used, although they allow for comparison with earlier studies, can be further improved to get more accurate results.
15
Mosleys estimate found that foreign aid is positively correlated with the rate of economic growth in the group of the poorest countries for the period 1970-1977, and significantly so if the lag is five years.
101
International Research Journal of Finance and Economics - Issue 11 (2007) Bowles, Paul. (1987). Foreign Aid and Domestic Savings in Less Developed Countries: Some Tests for Causality, World Development, 15, pp. 789-796. Burnside, Craig and Dollar, David. (1997). Aid, Policies, and Growth, World Bank Policy Research Paper #1777. Chenery, H. and Eckstein, P. (1970). Development Alternatives for Latin America, Journal of Political Economy, 78, pp. 966-1006. Chenery, H. and Strout, M. (1966). Foreign Assistance and Economic Development, The American Economic Review, 66, pp. 679-733. Eshag, Eprime. (1971) Comment on Griffin, Oxford Bulletin of Economics and Statistics, 33, pp. 149-156. Griffin, K and Enos, J. (1970). Foreign Assistance: Objectives and Consequences, Economic Development and Cultural Change, 18, pp. 313-327. Griffin, Keith .(1970). Foreign Capital, Domestic Savings, and Economic Development, Oxford Bulletin of Economics and Statistics, 32, pp. 99-112. Gujarati, Damodar. (1978). Basic Econometrics, Mcgraw-Hill Book Company, New York. Gupta, Kanhaya. (1975). Foreign Capital Inflows, Dependency Burden, and Savings Rates in Developing Countries: A Simultaneous Equations Model, Kyklos, 28, pp. 358-374. Heller, Peter. (1975). A Model of Public Fiscal Behavior in Developing Countries: Aid, Investment, and Taxation, American Economic Review, 65, pp. 429-445. International Monetary Fund, International Financial Statistics, different issues. Levy, Victor. (1988). Aid and Growth in Sub-Saharan Africa: The Recent Experience, European Economic Review, 32, pp. 1777-1795. Maizels, Alfred and Nissanke, Machiko. (1984). Motivation for Aid to Developing Countries, World Development, 12, pp. 879-900. McKinlay, R. (1978). The German Aid Relationship: A Test of the Recipient Need and the Donor Interest Models of the Distribution of German Bilateral Aid 1961-1970, European Journal of Political Research, 6, pp. 235-257. McKinlay, R. and Little, R. (1977). A Foreign Policy Model of U.S. Bilateral Aid Allocation, World Politics, XXX, pp 58-86. McKinlay, R. and Little, R. (1978). A Foreign Policy Model of the Distribution of British Bilateral Aid, 1960-70, British Journal of Political Science, 8, pp. 313-332. McKinlay, R. and Little, R. (1979). The U.S. Aid Relationship: A Test of the Recipient Need and the Donor Interest Models, Political Studies, XXVII, pp. 236-250. Mikesell, Raymond, and Zinser, James. The Nature of the Savings Function in Developing ountries: A Survey of the Theoretical and Empirical Literature, Journal of Economic Literature, 11, pp. 1-26. Mosley, Paul. (1980). Aid, Savings, and Growth Revisited, Oxford Bulletin of Economics and Statistics, pp. 42: 79-95. Mosley, Paul. (1981). Models of the Aid Allocation Process: A Comment on Mckinlay and Little, Political Studies, XXIX, pp. 245-253. Over, Mead. (1975). An Example of the Simultaneous Problem: A Note on Foreign Assistance: Objectives and Consequences, Economic Development and Cultural Change, 23, pp. 751-756. Papanek, Gustav. (1972). The Effect of Aid and Other Resource Transfers on Savings and Growth in Less Developed Countries, The Economic Journal, 82, pp. 934-950. Papanek, Gustav (1973). Aid, foreign Private Investment, Savings, and Growth in Less Developed Countries, The Journal of Political Economy, 81, pp. 121-130. Singh, Ram. (1985). State Intervention, Foreign Economic Aid, Savings, and Growth in LDCs: Some Recent Evidence, Kyklos, 38, pp. 216-232. Reinhart, C. and Talvi, E. (1998). Capital Flows and Savings in Latin America and Asia: A Reinterpretation, Journal of Development Economics, 57, pp. 45-66.
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25]
International Research Journal of Finance and Economics - Issue 11 (2007) [26] [27] [28] [29] [30] [31]
102
Snyder, Donald. (1990). Foreign Aid and Domestic Savings: A Spurious Correlation?, Economic Development and Cultural Change, 39, pp. 175-181. Stewart, Francis. (1971). Comment on Griffin, Oxford Bulletin of Economics and Statistics, 33, pp. 138-149. Tsikata, Tsidi. (1998). Aid Effectiveness: A Survey of the Recent Empirical Literature, IMF Paper on Policy and Analysis. PPAA/98/1. The United Nations, National Accounts Statistics, Main Aggregates, different issues. The United Nations, Statistical Yearbook, different issues. Weisskopf, Thomas. (1972). The Impact of Foreign Capital Inflows on Domestic Savings in Underdeveloped Countries, Journal of International Economics, 2, pp. 25-38.
Foreign Ownership and Firm Performance: Evidence from Turkey

Nurhan Aydin Professor (Finance), The Social Sciences Institute of Anadolu University Anadolu Universitesi sosyal Bilimler Enstitusu 3 kat Yunusemre Kampusu 26470 Eskisehir-Trkiye E-mail: naydin@anadolu.edu.tr Mustafa Sayim Research Associate (Finance) Eskiehir Osmangazi University School of Business Administration E-mail: msayim@ogu.edu.tr Abdullah Yalama Research Associate (Finance) Eskiehir Osmangazi University School of Business Administration E-mail: abdullay@ogu.edu.tr Abstract The study investigates whether foreign owned firms perform significantly better than domestically owned Turkish corporations quoted on Istanbul Stock Exchange (ISE), in Turkey. In the study, t-test statistics is applied to examine if there are significant differences on Operating Profit Margin (OPM) of firms, Return on Assets (ROA) and Return on Equity (ROE) between foreign owned participation firms and domestic firms listed in ISE. The results reveal that the firms with foreign ownership operating in Turkey perform better than the domestic owned ones in respect to ROAs. The evidence supports the hypothesis that foreign ownership participation increases performance of firms. The research is applied to all quoted firms on the Istanbul Stock Exchange (ISE) for the period 2003-2004. The findings may guide the foreign investors interested to make investment in Turkey. The study can be considered as one of the comprehensive research testing the effects of foreign ownership participation level on profitability performance of all firms listed on the ISE. Keywords: Foreign Direct Investment, ISE, Turkey, Foreign Ownership Share, Firm Performance Jel Classification Codes: F21, F37, F23, N20, C40
Introduction
The effect of Foreign Direct Investment (FDI) on corporate governance has long been issue of interest by academics and policy-makers. The main challenging question in the international business strategy study is the outcomes gained form FDI. Therefore, it is mainly accepted that FDI plays a crucial role
104
for economic growth and development, particularly in developing and transition economies (Grg and Greenaway 2004). Buckley and Casson (1976) tried to explore the reasons for existence of Multinational Enterprises (MNEs). Therefore, MNEs have been a major and important influences shaping of world economy (Qian 1996). Especially last two decades levels of FDI all over the world have been increasing. In the last decade, developing and transition economies of the world became increasingly crucial targets for FDI. The state of Turkey is one such developing economy. Due to Turkeys strategic location at eastern border of European Union (EU) and close to the oil rich Middle East, one may think of that it would make it region for inward FDI. However, in reality, Turkey has been underperforming in respect to its counterparts (Foster and Alkan 2003) until year 2004 and 2005. Especially, in year 2005, the total amount of FDI inflows to Turkey is around $ 9 billion USD due to establishing economic, political stability and recovering from economic crises occurred in 1999 and 2001. A numerous studies conducted to explore effects of different types of ownership structures; for instance, family owned, state owned and foreign owned etc. on firm performance in both developed and developing or transitions economies. Driffield et al. (2005) tested how family and managerial shared ownership structure can affect performance of firms in South East countries, Indonesia and Korea. Another study made by Bozec et al. (2002) examined state ownership influences on the 20 largest Canadian firms performances. They concluded that the profitability of firms can be significantly increased under state ownership. Although, the role of ownership structure on firm performance is investigated widely, foreign ownership effects on corporations listed in ISE remains unexplored in Turkey. Therefore we focus on investigating whether foreign capital participation has significant different effects on performance of firms between foreign-owned and domestically owned Turkish corporations quoted on Istanbul Stock Exchange (ISE), in Turkey. The paper is organized as follows. The next section describes the results of earlier researches, tries to fit them with respect to available theories, which is going to be tested in this study. This section followed by the data, research methodology and discussions of the empirical findings. The study closes up with some conclusions.
Literature Review
Earlier and recent empirical studies conclude that the MNEs have performed better than the domestically owned firms. Therefore, the foreign ownership has positive influences on the firms performance. This might be true for developed countries; however, in developing and transition economy, some findings are in contrast with earlier empirical findings. In this section, the empirical results of foreign ownership effects are reviewed in developed and developing countries. Researches on firms with foreign ownership operating in developed countries, Goethals and Ooghe (1997) conducted a study to investigate the performance between 25 Belgian firms and 50 foreign companies, which are Belgian taken over by foreigners. They calculated twenty-eight financial ratios for both foreign and domestic firms and concluded that foreign takeovers have positive impacts on the performance of firms by using regression analysis. Moreover, the firms with foreign ownership performed better than their domestically owned counterparts. Besides Alan and Steve (2005) also looked at the short and long term performance of UK corporations acquired by foreigners. The findings on 333 overseas acquisitions by UK limited companies for the period 1984-1995 reveal significant positive returns on the firm performance. Grant (1987) and Qian (1998) assessed the relationship between the return performance and multiple explanatory factors per se multinationality. Grants study to investigate performance on overseas production using based on three methods; static, dynamic regression and disaggregated analysis for the period 1972-1984 revealed that profitability for the 304 largest UK manufacturing firms drawn from The Times 500 list of British largest companies was positively correlated to their
105
level of multinationality. If the overseas production increases, it means an increase in sales and profitability. Qian included in his analysis the 164 largest US industrial corporations on the Fortune 500 listings for the period 1981-1992 and applied regression method to examine data. He found that the return performance is different due to differences in the level of foreign involvement. Most importantly, multinationality has a significant impact on the performance of MNEs. This might be due to market imperfections providing a powerful motivation to MNEs to explore the multinational ownership advantages; for instance, know-how, managerial skills and marketing ability. Liu et al. (2000) looked at the issue from different angle and examined intra-industry productivity spillovers from FDI on manufacturing sector in UK. Their empirical study is consisted of 48 different UK manufacturing industries over the 1991-95 periods. Caves-type single equation model was applied to test for effect of technology spillovers. The findings indicate that FDI existence has a positive spillover on the productivity of UK owned firms. Piscitello and Rabbiosi (2005) extended the study to Italy to investigate the influence of inward FDI coming into existence through acquisitions. The empirical results consist of foreign acquisitions that occurred in Italy for the period 1997-1997. Their sample was based on 113 foreign acquisitions, 74 of them undertaken by European MNEs, 31 by US MNEs and 8 from other countries. The evidence from OLS regression analysis shows that acquisitions improve the target firms performance in the medium term. This is due to the transfer of ownership benefits and also being part of international network of the acquiring MNE. Although, it is commonly agreed on that foreign owned firms and MNEs have been performing better than their domestically owned counterparts, some conflicted results in respect to the conclusions mentioned above were come cross in the literature. One of them, Kim and Lyn (1990) made a research to evaluate MNEs performance operating in U.S. using t-test and regression analysis. Their empirical sample for the study is based on the 54 largest foreign corporations operating in US in the period of 1980-1984. The corporations are grouped into the different industries. Nine of them are in mining, twenty-nine in manufacturing and sixteen in other industries. The results indicate that foreign owned firm operating in USA is less profitable than randomly selected domestically owned U.S corporations. Reasons for that might be US firms are less R&D incentives and more advertising oriented than foreign owned firms. More studies of the firms operating in developing and transition economies, Konings (2001) extended a research to test the effects of FDI on productivity performance of firms on three developing economies: Bulgaria, Romania and Poland. The dataset is consisted of 2,321 firms in Bulgaria for the period of 1993-1997, 3.844 firms in Romania between 1994 and 1997 and 262 firms in Poland over the period of 1993-1997. The evidence shows that foreign corporations do not perform better than domestic ones, except in Poland. This might be due to taking time for foreign ownership impacts on performance. Moreover, he found no evidence of positive spillovers of foreign investment to domestic firms. Khawar (2003) also finds no evidence of spillovers in Mexican manufacturing industry based on econometric approach. She derived the data set from the Mexico Industrial Survey for the year 1990. However, her study reveals that foreign firms are more productive than domestic firms, because of indication the presence of a strong foreign ownership on the productivity of individual firms. Aitken, and Harrison (1999) find no evidence of spillovers from foreign firms to domestic owned firms in Venezuela. They employed the data set of 43,010 observations covered form 1976 trough 1989 was gathered directly from Venezuelas National Statistical Bureau. They estimated log-linear production functions to investigate if foreign ownership is related to an increase in the productivity of plant and whether foreign equity participation has positive or negative spillovers to domestic firms. The findings show no evidence supporting the presence of technology spillovers to domestic firms from foreign firms. Another research conducted by Dauma et al. (2003) in developing economy tested foreign ownership effects on performance of 1005 Indians firms in 1999 and 2000 by applying regression method. They observed that foreign ownership positively affects firm performance.
106
Akimova and Schwdiauer (2004) examine the impact of ownership structure on corporate governance and performance of privatized corporations in Ukrainian transition economy. Their analysis was based on data from a survey conducted in 2001 on 202 medium and large industrial corporations in Ukraine between the periods of 1998-2000. OLS regressions were used for estimations. Empirical evidence shows that there are significant ownership effects on the performance, but which is non-linear relation. In other words, its effect is positive within lower range but negative from the point that closes to majority ownership. The most recent study conducted by Barbosa and Louri (2005) investigate if MNEs operating in Portugal and Greece perform differently than their domestic counterparts. The final sample is consisted of 523 manufacturing firms data produced by Portuguese Ministry of Labour in 1992 and based on standard survey that must be answered by firms with wage earners every year. In the Greek situation, 2,651 firms are used and data has been obtained from ICAP directory in 1997. After using the robust method of quintile regression, the results suggest that ownership ties do not make a significant difference with corresponding to the firm performance in Portugal and Greece. We can say that conclusions drawn from earlier and recent studies, foreign ownership on the performance of firm in developed countries has a positive impact on the productivity. In other words, findings present a reconciliation of these prior studies. However, when we look at the effect of ownership structure on the performance of firm in developing and transition economies, some findings are in contrast with earlier studies. For example, the results of ownership structure in Portugal and Greece, the theory of FDI, which foreign ownership has a positive impact on the performance; therefore MNEs have superior performance over their domestic counterparts, are not compatible with findings obtained from the studies conducted in developed countries. Moreover, Gonenc (?) examined the effect of concentrated ownership on performance of 185 industrial firms using OLS and 2SLS estimations for the period 1992-1998 in Turkey. He found that foreign ownership does not yield better performance. Another study about foreign owned firms performance in Turkey was conducted by Gunduz and Tatoglu (2003). They employed the one-way analysis of variance (ANOVA) to investigate the effect of foreign ownership on performance of 202 non-financial firms listed on ISE in 1999. The findings reveal that foreign owned firms have significantly better performance than domestic firms regarding with ROA, but not in other financial performance ratios. Isik et al. (2004) assessed the relationship among liberalization, ownership and performance of public, private and foreign banks for the period 1981-1990 in Turkey by applying DEA-type Malmquist productivity index approach. The findings indicate that the foreign owned banks were the most productive followed by private and public banks respectively. The reason for foreign banks being more productive is due to technological changes in their operation process. Therefore, in the paper, we examine the effect of foreign ownership on performance of firms with foreign ownership listed on ISE in Turkey as a developing country and evaluate findings whether they are compatible with earlier studies. Moreover, we compare results whether the firms with foreign ownership perform better than domestic firms in Turkey.
Data
Dataset used in the study for the period 2003-2004 was gathered from firms ends of year balance sheets and income statements values published by FINNET. We obtained list of firms with foreign ownership from web site of Foreign Investors Association of Turkey (FIA). Than we verified which of them quoted in ISE to be included to our analysis. Therefore, we included the total number of 42 firms with foreign ownership and 259 domestic corporations listed on ISE in Turkey in our analysis.
107
Methodolgy
In this study t-test statistics is performed to test whether there are any significant differences on Return on Asset (ROA), Return on Equity (ROE) and Operating Profit Margin (OPM) ratios between the firms with foreign ownership participation ranging from %0-%100, %25-%100 to %50-%100 in capital structure and domestic firms. We define ROA as the operating profits before interest, depreciation and taxes to the book value of total assets. OPM is operating profits before interest, depreciation and taxes to the book of sales. And also ROE is defined as profit before taxes to the book value of total equity.
The Emprical Results

The empirical findings obtained from t-test are summarized as follow:
Table 1:
Group The firms with FO (%0-%100) The firms with FO (%25-%100) The firms with FO (%50-%100) The firms without FO 2003 2004 2003 2004 2003 2004 2003 2004
Model1 T-test for comparison level of ROAs on the firms with FO and domestic firms
N 41 41 31 31 15 15 259 259 Average return 21,38 23,70 22,28 19,18 23,46 19,83 14,59 15,92 Standard Deviation 17,17 13,47 15,57 10,86 20,59 10,20 15,61 20,83 T -2,550 -2,314 -2,589 -0,857 -2,097 -0,721 df 298 298 288 288 272 272 p 0,011 0,021 0,010 0,392 0,037 0,471
The hypotheses to test effect of foreign ownership presence on ROA are; H0: There is no significant difference between firms with FO and domestic firms in corresponding to ROA. H1: There is significant difference between firms with FO and domestic firms in corresponding to ROA. As it can be seen in Table I, for year 2003 (0%-100% FO), H0 is rejected. (p=0,011<0,05).When ROAs are compared, it is found that the firms ROAs with FO (Foreign Ownership) and domestic firms are different significantly after t-test was performed (t0,05: 298 = -2,550). According to the finding, ROA of firms with FO, which is 21.38 %, is higher than ROAs of domestic firms, which is 14.59 %. Moreover, for year 2004 (0%-100% FO), H0 is also rejected. (p=0,021<0,05). It also is reached to same conclusion in 2004 that there is significant difference on the firms ROAs with FO and domestic firms (t0,05: 298 = -2,314). Based on this outcome, ROA of firms with FO, which is 23.70 %, is higher than ROAs of domestic firms, which is 15.92 %. For year 2003 (25%-100% FO), H0 is rejected. (p=0,010<0,05). When ROAs are compared, it is found that the firms ROAs with FO and domestic firms are different significantly after t-test was performed (t0,05: 288 = -2,589). According to the finding, ROA of firms with FO, which is 22.28 %, is higher than ROAs of domestic firms, which is 14.59 %. For year 2003 (50%-100% FO ownership), H0 is rejected. (p=0,037<0,05). When ROAs are compared, it is found that the firms ROAs with FO and domestic firms are different significantly after t-test was performed (t0,05: 272 = -2,097). According to the finding, ROA of firms with FO, which is 23.46 %, is higher than ROAs of domestic firms, which is 14.59 %. However for year 2004 between 25%-100% and 50%-100% foreign ownership participation level, there are no significant differences between firms with FO and domestic firms in corresponding to ROA.
108
The results on ROEs can be shown in Table II.

Table 2:
Group The firms with FO (%0-%100) The firms with FO (%25-%100) The firms with FO (%50-%100) The firms without FO 2003 2004 2003 2004 2003 2004 2003 2004
Model2 T-test for comparison level of ROEs on the firms with FO and domestic firms.
N 41 41 31 31 15 15 259 259 Average return 26.67 38 32,40 31,68 27,84 45,52 14,77 -5,00 Standard Deviation 78.87 55.95 83,08 46,89 120,38 27,97 264,95 832,36 T -0.285 -0.330 -0,368 -0,245 -0,190 -0,968 df 298 298 288 288 272 272 p 0.776 0.741 0,713 0,807 0,850 0,815
The hypotheses to test effect of foreign ownership presence on ROE are; H0: There is no significant difference between firms with FO and domestic firms in corresponding to ROE. H1: There is significant difference between firms with FO and domestic firms in corresponding to ROE. As can be seen Table II for year 2003-2004 among foreign ownership participation level groups, there are no significant differences between firms with FO and domestic firms in corresponding to ROE. The results on OPMs can be shown in Table III.
Table 3: Model3 T-test for comparison level of OPMs on the firms with FO and domestic firms.
2003 2004 2003 2004 2003 2004 2003 2004 N 41 41 31 31 15 15 259 259 Average return 21.32 22.08 22,67 19,17 25,09 22,38 6,44 15,97 Standard Deviation 19.08 19.58 15,65 21,90 19,81 20,42 106,67 40,84 T -0.890 -0.941 -0,845 -0,429 -0,676 -0,603 df 298 298 288 288 272 272 p 0.374 0.348 0,399 0,668 0,500 0,547
Group The firms with FO (%0-%100) The firms with FO (%25-%100) The firms with FO (%50-%100) The firms without FO (%50%100)
The hypotheses to test effect of foreign ownership presence on OPM are; H0: There is no significant difference between firms with FO and without FO on corresponding to OPM. H1: There is significant difference between firms with FO and without FO on corresponding to OPM. The findings indicate that for year 2003-2004 among foreign ownership participation level groups, there are no significant differences between firms with FO and domestic firms in corresponding to OPM.
Conclusion
In the study, we undertook an analysis of performance of firms with foreign ownership to test its effects on the firm performance and examine whether there is any significant performance difference between the firms with foreign ownership and domestic Turkish corporations listed on ISE in Turkey. In the paper, t-test statistics confirms the hypothesis that firms with foreign ownership in respect to ROAs have performed better than domestic firms. Therefore, our results show that firms with foreign ownership perform better than domestic firms in Turkey for the period 2003-2004. Our
109
findings matches the prior studies conclusions of performance of foreign owned firms and domestic ones in developed, developing and transition economies (Goethals, J. and Ooghe H. 1997, Khawar 2003, Douma et al. 2003, Gunduz 2003, Isik 2004 and Akimova et al. 2004). We provided some evidence of foreign ownership benefits positively on performance of firms listed in ISE. There may be couple reasons why foreign ownership increases firm performance. One reason might be ability to monitor or control or give incentives for managers leading manage a firm more seriously and avoiding initiatives reducing the corporate values. Another one would be transferring new technology by foreign firms generating savings on operating expenses. Regarding the future study agenda, it would be interesting to investigate in order to explore the effect of inward FDI on the domestic firms in Turkey. In other words, to research whether there is a positive spillover impacts on the local economy would make valuable addition into the existence literature regarding Turkey as a developing country.
110
International Research Journal of Finance and Economics - Issue 11 (2007) Alkan, . And Foster, J. (2003), Understanding Patterns of FDI: the Case of Turkey and its Auto Industry, European Business Journal, 15 (2), pp.61-69. Aitken, B. and Harrison, A. (1999), Do Domestic Firms Benefit from Direct Foreign Investment? Evidence from Venezuela, The American Economic Review, 89 (3), pp. 605-617. Akimova, I. and Schwdiauer, G. (2004), Ownership Structure, Corporate Governance and Enterprise Performance: Empirical Results from Ukraine, International Advance in Economic Research, 10 (1), pp. 28-42. Alan, G. and Steve, M. (2005), Foreign Acquisitions by UK Limited Companies Short and Long-run Performance, Journal of Empirical Finance, 12 (1), pp. 99-125. Barbosa, N. and Louri, H. (2005), Corporate Performance: Does Ownership Matter? A Comparison of Foreign and Domestic owned Firms in Greece and Portugal, Review of Industrial Organization, 27, pp. 73-102. Bozec, R., Breton, G. and Cote, L. (2002), The Performance of State-Owned Enterprises Revisited, Financial Accountability & Management, 18 (4), pp. 383-407. Buckley, P. and Casson, M. (1976), The Future of the Multinational Enterprise, Holmes & Meier: New York. Chung, W., Mitchell, W. and Yeung, B. (2003), Foreign Direct Investment and Host Country Productivity: the American Automotive Component Industry in the 1980s, Journal of International Business Studies, 34, pp. 199-218. Douma, S., George, R., and Kabir, R. (2003), Foreign and Domestic Ownership, Business Groups and Firm Performance: Evidence From a Large Emerging Market, Discussion Paper, provided by Tilburg University, Center for Economic Research in its series with number 104. Avaliable: greywww.kub.nl:2080/greyfiles/center/2002/doc/104.pdf Driffield, N., Vidya, M. and Sarmistha, P., (2005), "How Ownership Structure Affects Capital Structure and Firm Performance? Recent Evidence from East Asia," Finance 0505010, Economics Working Paper Archive EconWPA. Goethals, J. and Ooghe H. (1997), The Performance of Foreign and National Take-Overs in Belgium, European Business Review, 97 (1), pp. 24-37. Gonenc, H, Ownership Concentration and Corporate Performance: A Simultaneous Equation Framework for Turkish Companies, Forthcoming Chapter in The Turkish Economy, accepted to be published by Routledge. (http://ssrn.com/abstract=498263). Grg, H. and Greenaway D. (2004), Much A do About Nothing? Do Domestic Firms Really Benefit From Foreign Direct Investment, The World Bank Research Observer, 19, pp. 171-197 Gunduz, L., and Tatoglo, E., (2003), A comparison of the Financial Characteristics of Group Affiliated and Independent Firms in Turkey, European Business Review, 15 (1), pp. 48-54. Grant, R. (1987), Multinationality and Performance Among British Manufactoring Companies, Journal of International Business Studies, 13 (3), pp.79-89. Isk, I., Gunduz, L., and Klc, O., (2004), Assessing the Relationship Between Liberalization, Ownership, and Performance: The Case of Turkish Banks, International Business & Economics Research Journal, 3 (1), pp. 39-52. Khawar, M. (2003), Productivity and Foreign Direct Investment- Evidence From Mexico, Journal of Economic Studies, 30 (1), pp. 66-76. Kim, W. and Lyn, E.(1990), FDI Theories and the Performance of Foreign Multinationals Operating in the US, Journal of International Business Studies, 21 (1), pp. 41-53. Konings, J. (2001), The Effects of Foreign Direct Investment on Domestic Firm, Economics of Transition, 9 (3), pp. 619-633. Lui, X., Siler, P.,Wang, C. and Wei, Y. (2000), Productivity Spillovers from Foreign Direct Investment: Evidence from UK Industry Level Panel Data, Journal of International Business Studies, 31 (3), pp. 407-425.
References
[1] [2] [3] [4] [5] [6] [7] [8] [9]
[10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20]
International Research Journal of Finance and Economics - Issue 11 (2007) [21] [22] [23]
111
Piscitello, L. and Rabbiosi (2005), The Impact of Inward FDI on Local Companies Labour Productivity: Evidence from The Italian Case, International Journal of the Economics of Business, 12 (1), pp. 35-51. Qian, G. (1996), The Effect of Multinationality Measures Open the Risk Return Performance of US Firms, International Business Review, 5 (3), pp. 113-132. Qian, G. (1998), Determinants of Profit Performance for the Largest US Firms 1981-92, Multinational Business Review, 6 (2), pp. 44-51.
Feldstein-Horioka Puzzle in Latin American and Caribbean Countries: Evidence from Likelihood-Based Cointegration Tests in Heterogeneous Panels
N. R. Vasudeva Murthy Professor of Economics, College of Business Administration Creighton University 2500 California Plaza Omaha, NE 68178 E-mail: vmurthy@creighton.edu Tel: (402) 280-2112 Abstract This research paper, using data from a panel consisting of fourteen Latin American and four Caribbean countries over the period 1960-2002, applies the recently developed Larsson, Lyhagen and Lothgren [2001] maximum likelihood panel cointegration procedure to examine the validity of the famous Feldstein-Horioka puzzle (F-H Puzzle) in this panel. Despite many criticisms, the F-H Saving-Investment correlation hypothesis is still used in the literature to infer the degree of capital mobility among countries. The finding of the presence of cointegration of the savings and investment ratios and the observed magnitude of the estimated average savings-retention coefficient for the panel reveal that for this panel of Latin American and Caribbean countries, the F-H puzzle is not valid and the long-run international solvency condition is maintained in most of these countries. The observed low savings-retention coefficients for these countries imply a moderate degree of capital mobility and the absence of the Feldstein-Horioka puzzle. This finding of the prevalence of a moderate degree of capital mobility is consistent with the macroeconomic experience of these countries during the period under investigation.1 Keywords: Saving-Investment Correlation, Panel Unit Roots, Panel Cointegration and Capital mobility Jel Classification Codes: C32, F21, F32
1. Introduction
Economists have always been interested in finding out whether different countries and groups of countries exhibit capital mobility or immobility, especially in the long-run. Knowledge of the prevalence of a high degree of capital mobility in a country is beneficial for many economic and political reasons. Capital mobility in an economy, besides making savings available for economic development, helps in assessing the efficiency in capital allocation, rendering the possibility of consumption smoothing and evaluating the effects of a countrys fiscal and monetary policies, in identifying the burden of investment taxes and discerning an economys costs of adjustment to negative exogenous shocks such as a natural disaster. One of the ways of inferring prevalence of the degree of capital mobility in an economy is to examine whether the celebrated Feldstein-Horioka
1
A revised version of a paper presented at the Academy of Economics and Finance Annual Meeting held in Jacksonville, FL during February 2007
113
[1980, 1983] hypothesis, which refers to the correlation between the savings ratio [hereafter, (S/Y)] and investment ratio [hereafter, (I/Y)] in a country. It is widely understood in the literature that in the presence of a sufficient degree of capital mobility in a country, domestic investors are not constrained by the insufficiency of domestic savings as they can borrow from abroad and thus, the countrys (I/Y) and (S/Y) need not be correlated. Feldstein and Horioka [1980], using the average cross-sectional data for16 OECD [Organization of Economic Development Countries] countries during the period 19601974, estimated the regression model, (I Y ) = + (S Y ) + . Their estimated coefficient is 0.887 with a standard error of 0.07 and a coefficient of determination, R 2 = 0.91. They attributed the high value of 0.887,which is statistically significant, to the absence of capital mobility, because in the presence of capital mobility the value of would be 0, as domestic savings would follow a high rate of return or interest rate else where. Furthermore, Feldstein and Horioka (1983) tested the hypothesis by extending the sample to 1980 and found the results again supporting the FH-puzzle. As in most of these OECD countries, other macroeconomic indicators such as the increased integration of world financial markets, the recent development of deregulation of financial markets in many countries, purchasing power parity, common shocks affecting both (I/Y) and (S/Y), and the existence interest rate differential among countries, reveal that in reality some degree of capital mobility prevails and therefore Feldstein-Horioka evidence has been called the Feldstein-Horioka puzzle (hereafter, FHpuzzle). Some empirical studies inspired by Feldstein-Horioka studies do support the FH-puzzle [see, Baxter and Crucini (1993), Capiro and Howard (1984) and Dooley et al. (1987)]. However, some other economists contend that saving-investment correlation may not be after all informative about the prevailing degree of capital mobility and in fact, there have been many theoretical and empirical attempts to show that (I/Y) and (S/Y) may be correlated or cointegrated even if capital were mobile because of the presence of certain factors, such as common causes, the big country effect, endogenous shocks [see, Herwartz and Xu (2006), Coakley et al. (2003), Coakley and Kulasi (1997), Coakley, Kulasi and Smith (1998), Bayoumi (1990), Obstfeld and Rogoff (2000), Murphy (1984)]. It has been contended that unless a country violates its inter-temporal budget constraint, the saving-investment correlation or cointegration is an empirical regularity because it has to maintain inter-temporal budget balance in the long-run in order to borrow in the future and repay its external budget. A country will be unable to borrow for ever and therefore, it has to maintain its inter-temporal budget constraint. Therefore in the long-run, savings and investment ratios will be correlated or cointegrated irrespective of the prevalence of capital mobility. Despite cointegration of the savings and investment ratios, it has been argued in the literature that the existence of relatively small magnitude of the savings-retention coefficient implies a moderate degree of capital mobility. While there have been many theoretical and empirical attempts that deal with the F-H puzzle for the developed countries in general and the OECD countries in specific, there is a paucity of empirical verification of this puzzle for developing countries, especially for Latin American and Caribbean countries [see, Payne and Kumazawa (2005), De Wet and Van Eyden (2005), Kim et al. (2005), Levy (2000), Murthy (2005), Hoffman (2004), Sinha (2004), Ho (2003), Banerjee and Zanghieri (2003), De Wet and Koekemer (2003), Isaksson (2001), Rocha (2000), Hussein and Mello (1999), Jansen (2000), Hauqe and Montiel (1990), Montiel (1994), Dooley, Frankel and Mathieson (1987) and Murphy (1984)]. In light of the level and volatility of capital inflow experienced by many Latin American and Caribbean countries during the period 1960-2000, an investigation of the validity of the FH-puzzle is highly warranted. Therefore, this paper using the data for the period 1960-2002 countries empirically tests whether the FH-puzzle is valid for a panel of fourteen Latin American and four Caribbean countries by employing a likelihood-based panel cointegration procedure to determine the cointegration rank and estimate the savings-retention coefficient in a heterogeneous panel, recently developed by Larsson, Lyhagen and Lothgren (2001) [also, see Pedroni (2004)]. The countries included in the panel, studied in this paper, are Barbados, Chile, Colombia, Costa Rica, Dominion Republic, Ecuador, El Salvador, Guatemala, Guyana, Honduras, Jamaica, Mexico, Nicaragua, Paraguay, Peru, Suriname, Trinidad and Tobago, and Venezuela. Panel data, besides providing more information, gives greater power and less size distortions than the standard time-series unit root and
114
cointegration tests [see, Levin et al., (2002]. Furthermore in the literature to the authors knowledge, no panel cointegration analyses for developing countries having such a panel dimension are found.
2. Model Specification
To test whether the savings and investment ratios are cointegrated in a heterogeneous panel of 14 Latin American and 4 Caribbean countries, the recently developed maximum likelihood-based panel procedure to determine the cointegration rank in a heterogeneous panel developed by Larsson, Lyhagen and Lothgren (2001) [hereafter, the LLL method] is employed in this paper. The LLL method is based on the famous Engle and Granger (1987) and Johansen (1988, 1995) error-correction representation theorem. For each country, consider a data vector X t as X = [(S Y )t , (I Y )t ] Furthermore, a reparameterized vector error-correction (VECM) form of X t is specified as follows: (1) Where (I/Y) and (S/Y) are respectively, the savings and investment ratios. Dt is a vector of deterministic variables. The disturbance term, t ~ IN (0, ) .The matrix i = i i and is of reduced rank with i and i as two N * r matrices representing the speed adjustment and cointegrating vectors, respectively. The data on these variables for the period 1960-2002 are obtained from World Development Indicators CD-ROM 2005 [World Bank (2005)]. The availability of the data on the relevant variables for a reasonable length of time-series has dictated the number of countries chosen for analysis and estimation. A visual inspection of the data for most of the countries reveals that the series exhibit a linear trend. This has influenced the specification of the deterministic terms in the cointegration modeling. [Also, see Gerdtham and Lothgren (2000)]. For each country in the panel, the Johansen cointegration test (1988, 1995) is conducted as a sequential procedure by way of testing the null hypothesis that rank, i = ri < r for all the 18 countries against the alternative that the rank of i = p For each country in the panel the null hypothesis, that r = 0 is tested using the observed trace statistic and, if this hypothesis is rejected, then the hypothesis that r = 1 is tested. This sequential testing procedure is ended when the null hypothesis, that r = ri is no longer rejected and at this stage the rank estimate of r is determined. For determining the LLL panel trace test, the statistic YLR is
i =1
X t = X t 1 + i X t 1 + Dt + t
k 1
Where E (Z k ) and Var (Z k ) are the mean and variance of the asymptotic trace statistics. These moments can be obtained from either a stochastic simulation following the procedure described by Johansen (1995) or by the simulated moments, for the case where no deterministic terms are included in the underlying VAR model, reported in a table by Larsson, Lyhagen and Lothgren (2001). Recently, Breitung (2005) has extended the LLL procedure showing alternative representations of the YLR statistic by consider various deterministic components in the underlying VAR model. The computed YLR statistic, which will be used in the sequential procedure to determine the cointegration rank of a heterogeneous panel, has been demonstrated to be normally distributed asymptotically. In the panel test, the null hypothesis states that all of the N countries in the chosen panel have a maximum common rank of r cointegrating relationships among the p variables against the full rank alternative for all the countries, although each country can have its own ri number of cointegrating vectors. Before conducting the LLL panel cointegration procedure, it is required that panel unit root tests are conducted to find out whether a unit root is present in the series, (I/Y) and (S/Y). If all the
obtained by standardizing the average of the N countries individual observed trace statistics, LR NT as follows: N LR NT E (Z k ) YLR = N (0,1) (2) Var (Z k )
115
series are found to be stationary in levels, then the Ordinary Least Squares technique can be used to estimate the model in question. But, if the series are suspect to be non-stationary in levels, then unit root testing will have to be conducted prior to cointegration analysis. Specifically, the order of integration of the series must be determined in order to know whether the series are stationary in levels, I(0), or in their first-differences, I(1). Therefore, to test for the presence of a unit root in the savings and investment ratio series, in addition to the individual country univariate unit root tests, the recently developed panel unit root tests of Im, Pesaran and Shin [1997 and 2003] (hereafter, IPS) Maddala and Wu [1999] (hereafter, MW), Breitung [2000] and Hadri [2000] are applied. As these tests have become standard tests in panel cointegration analysis, the details these tests are not given in this paper. A survey and detailed discussion of these first-generation panel tests are presented in Baltagi [2005], and Breitung and Pesaran [2005]. In Hadris test [2000], the null hypothesis is stationarity or presence of no unit root in any of the cross-sectional units as against the alternative of the presence of a unit root in all cross-sectional units. In IPS, Breitung and MW panel unit root tests, the null hypothesis is that all the N individual countrys time series have a unit root against the alternative of no unit root. The Breitung and Hadri panel unit root tests assume that the autoregressive coefficients across cross sections in the AR(n) process for panel data remain the same, where as in the IPS and MW unit root tests, they may vary across cross-sections. The IPS is a group-mean panel unit root based on individual countries augmented Dickey-Fuller (ADF) statistics with the heterogeneous alternative. IPS test statistic is expressed as below: N t i E (t i i = 0) (3) t IPS = VAR[t i i = 0]
Where N is the number of cross-sectional units (individual countries) in the panel and t is the average of the computed ADF statistics for all cross-sectional units included in the panel under study. Pi is the autoregressive unit. The moments, E [t i i = 0] and Var [t i i = 0]. Show respectively, the average and variance obtained from Monte Carlo simulation as reported by IPS [1997, 2003]. It is shown that t IPS approaches in probability a standard normal distribution as N and T reach infinity. Thus, the IPS test statistic is the cross-section average of the t-statistics corrected by the simulated moments. The MW test statistic has a chi-square distribution with 2N degrees of freedom under the null hypothesis is expressed as:
= 2 InPi
i =1
(4)
Where, Pi are the probability values from the individual ADF unit root tests for all the countries included in the sample and N is the number of cross-sectional units. The Breitungs unit root test [2000] corrects for a substantial loss of power of the IPS test when the individual ADF regression, used for computing the IPS statistics, includes a trend in its specification.
3. Empirical Findings
Before conducting a panel cointegration analysis, it is required that the stochastic properties of the variable series included be investigated. If the series are non-stationary in levels and stationary in their first-differences, then one can ascertain whether they are cointegrated. Table 1 presents the results of various panel unit root tests.

Table 1:
Series (I/Y) (S/Y)
116
Panel Unit Root Tests
t IPS
-1.173 (0.120) -0.360 (0.359) -8.337* (0.000) -7.664* (0.000)
MW 40.843 (0.266) 32.514 (0.635) 134.604* (0.000) 120.976* (0.000)
Breitung Test -2.499 (0.006) -2.623 (0.004) -8.186* (0.000) -8.237* (0.000)
Hadri Test 5.872* (0.000) 4.981* (0.000) 4.498* (0.000) 3.294 (0.0005)
(I Y ) (S Y )
P-values, obtained from Eviews 5.0, in parentheses. * shows rejection of the null-hypothesis. All the tests are conducted with individual effects and a linear trend. Lags = 3 for both levels and first-differences. For the MW test, the degrees of freedom are 2N (=36).
The IPS, MW and Breitung tests results show that the level series, (I/Y) and (S/Y) are nonstationary at the 5% level of statistical significance and their firstdifferenced series are stationary at the 1% level. Hadri tests results, while confirm that for (I/Y), (S/Y) are non-stationary and (I/Y) is stationary, for (S/Y) series, it rejects the null hypothesis of stationary in support of a common unit root in the present panel. According to a recent simulation study, the Hadri panel unit test has been found to perform poorly [see, Hlouskova and Wagner (2005)]. But, the overall panel unit root test results indicate that (I/Y) and (S/Y) are integrated of the order of one, I ~ (1) , where as, (I/Y) and (S/Y) are integrated of the order zero, I ~ (0 ) . It has been demonstrated by a recent simulation study by Karlsson and Lothgren [2000] that while using panel data enhances the power of unit root tests, these tests may actually lack power when a large fraction of the panel is stationary. Therefore, they recommend a careful joint analysis of both the individual country unit root test for each cross-sectional unit and the panel unit root tests when the stationary properties of the panel series need to be evaluated. Therefore, in Tables 2 and 3, the results of the individual country ADF unit root tests for both the level and first-differenced series are reported. With he exceptions of Colombia, Guyana and Suriname, the level series of (I/Y) and (S/Y) for individual countries, are non-stationary. The first-differenced (I/Y) series, barring the exception El Salvador, Guatemala and Jamaica, are stationary. The first-differenced (S/Y) series, with the lone exception of El Salvador, for all other individual countries included in the panel, are stationary at the 5% level of significance. Thus, the results of both the panel and individual country unit root test results overwhelmingly confirm that the (I/Y) and (S/Y) series are integrated of the order one and their first-differenced series are integrated of the order zero.
117
Table 2:

Unit Root Test Results of Individual Country (Levels)
ADF Test -2.404 -2.605 -4.042 -1.341 -2.048 -2.272 -2.673 -2.777 -1.476 -2.130 -1.437 -2.567 -2.904 -1.068 -2.784 -2.771 -2.653 -2.389 -2.222 -2.114 0.742 -1.173 (0.120) P-value 0.372 0.280 0.015 0.860 0.558 0.439 0.253 0.214 0.821 0.514 0.834 -0.297 0.172 0.922 0.211 0.216 0.261 0.380 (S/Y) ADF Test -1.392 -2.564 -1.559 -1.474 -2.261 -2.222 -2.580 -2.367 -3.454 -2.671 -1.806 -1.396 -2.317 -1.639 -2.511 -3.442 -3.442 -1.815 -2.188 -2.114 0.742 -0.360 (0.360) P-value 0.848 0.298 0.791 0.822 0.444 0.465 0.291 0.390 0.059* 0.254 0.683 0.846 0.415 0.759 0.321 0.060* 0.060* 0.678
(I/Y) Country Barbados Chile Colombia Costa Rica Dominican Republic Ecuador El Salvador Guatemala Guyana Honduras Jamaica Mexico Nicaragua Paraguay Peru Suriname Trinidad & Tobago Venezuela Average
E (t ) E (Var )
IPS
* Significance at the 6% level. Lags = 3. Deterministic terms in the ADF regressions are individual effects and individual linear trends.
Individual country and panel cointegration tests are reported in Table 4. For individual country Johansen cointegration test, the number of lags used in the VAR model, is equal to four and furthermore, it is assumed that the level data and the cointegration equation have linear trends (Option 4 in Eviews 5.0). This assumption regarding the deterministic terms are consistent with linear trends in both (I/Y) and (S/Y) series. Furthermore, the results are broadly similar when alternative deterministic terms and various lags are attempted. Following the sequential procedure suggested by Johansen [1988, 1995], we find that for a majority of individual countries, the observed trace statistics determine the rank of r = 1 The existence of cointegration for 10 countries (Chile, Dominion Republic, Ecuador, Guatemala, Guyana, El Salvador, Nicaragua, Suriname, Trinidad and Tobago and Venezuela) is consistent with the long-run solvency of these countries inter-temporal budget constraint. However, for the remaining 8 countries (Barbados, Colombia, Costa Rica, Honduras, Jamaica, Mexico, Peru and Paraguay) the rank of r = 0 shows that for these countries, the inter-temporal budget is violated and not binding at least in the short-run, not withstanding the fact that the Johansen procedure is highly sensitive to the implied lag structure and the presence of the deterministic term. Also for individual countries, small size of the sample may be a problem. This

Table 3: Unit Root Test Results of Individual Country (First-Differences)
118
(I Y )
Country Barbados Chile Columbia Costa Rica Dominican Republic Ecuador El Salvador Guatemala Guyana Honduras Jamaica Mexico Nicaragua Paraguay Peru Suriname Trinidad & Tobago Venezuela Average ADF -3.602 -3.662 -4.119 -3.734 -4.113 -3.548 -2.833 -3.007 -5.960 -3.168 -2.872 -3.740 -3.754 -3.773 -4.939 -4.545 -3.535 -3.733 -3.813 -2.112 0.750 -8.337 0.000 P-value 0.043 0.038 0.013 0.032 0.013 0.048 0.195* 0.147* 0.000 0.106* 0.183* 0.032 0.031 0.029 0.002 0.004 0.049 0.032 ADF -3.830 -3.516 -3.364 -3.134 -3.462 -4.199 -3.146 -3.452 -4.811 -3.287 -3.438 -3.522 -3.559 -3.325 -3.702 -4.160 -3.819 -4.442 -3.801 -2.112 0.750 -7.66 (0.000)
(S Y )
P-value 0.026 0.052 0.072 0.113* 0.058 0.011 0.111* 0.060 0.002 0.084 0.061 0.051 0.047 0.078 0.034 0.011 0.026 0.006
E (t ) E (Var )
IPS
*Indicates significance at the 11% level and greater. The p-values are from Eviews 5.0. Lags = 3. Deterministic terms: Individual effects and Individual linear trends.
result of lack of cointegration between (I/Y) and (S/Y) can also be viewed as an indication of short-run capital mobility in these countries. It has been demonstrated by Coakley et al., (2003) that in the presence of persistent shocks to a countrys current account, saving and investment rates may not cointegrate. Since, the focus of the paper is on panel cointegration and the panel data used in estimation by providing more information, increases the power of the cointegration test, the results of the panel rank test is appropriate in this context [see, Larsson et al., (2001) and Gerdtham and Lothgren (2000). For the panel cointegration rank test, the observed Z LR statistic of 4.52 is greater than the critical Z statistic of 1.64, we reject the null hypothesis of a largest rank = 0. But, we cannot reject the null hypothesis of a largest rank of r = 1 because the Z LR value of 0.29 is less than 1.64. Therefore, the LLL panel test for cointegration determines the common rank of r = 1 and therefore, (I/Y) and (S/Y) are cointegrated. This finding indicates that for the panel as a whole, the inter-temporal solvency condition that (I/Y) and (S/Y) is not violated as their gap cannot continue permanently. The magnitude of the observed savings- retention coefficient for 10 countries is less than one. For Chile, Ecuador and Nicaragua, the (I/Y) and (S/Y) are cointegrated, and the savings-
119
Table 4:

LikelihoodBased Cointegration Tests: Panel and Individual Country Test Results
LRiT (H (r ) H (2))
Country Barbados Chile Colombia Costa Rica Dominican Republic Ecuador Guatemala Guyana El Salvador Honduras Jamaica Mexico Nicaragua Peru Paraguay Suriname Trinidad & Tobago Venezuela Panel Tests
r=0
12.20 29.71 11.34 23.02 33.93 21.92 25.13 29.98 19.74 16.72 18.94 11.87 31.11 16.28 16.48 25.33 21.13 25.15 16.28 25.50 4.52
r =1
2.59 10.11 4.20 5.93 8.69 7.01 4.67 11.71 3.01 3.72 5.94 4.48 10.76 5.56 6.01 9.37 8.27 4.78 6.27 10.45 0.29
0.29 0.50* 0.58 0.62 1.11* 0.61* 1.28* 1.07* 0.78 1.02 1.29 0.20 0.46* 0.64 1.00 1.13* 0.98 1.33*
Rank (ri ) 0 1 0 1 1 1 1 1 0 0 0 0 1 0 0 1 1 1
E (Z k ) Var (Z k )
Z LR
r=0
r =1
* Significant at the 5% level. Lags = 4. The critical value for the panel test is 1.645. Deterministic terms are the level data and the cointegrating equations have linear trends (Option # 4 in Eviews 5.1). Asymptotic moments for standardizing the test statistic are from Breitung (2005). For individual country tests, the 5 % critical values for testing r = 0 and r = 1, respectively are 25.87 and 12.52.
retention-coefficient is significantly less than one, indicating that in these countries the inter-temporal budget constraint is maintained in the longrun and a high degree of capital mobility prevails. In Dominion Republic, Guatemala, Suriname and Venezuela, not only the savings and investment ratios are cointegrated, but also the savings-retention ratio is significantly different from unity. The finding that for the panel as a whole the savings and investment ratios are cointegrated and for majority of countries, the savingsretention is relatively small and (I/Y) and (S/Y) are cointegrated indicates that for the Latin American and Caribbean countries included in the sample, the Feldstein-Horioka puzzle does not hold valid. Furthermore, for these countries the evidence is consistent with the presence of a moderate degree of capital mobility, applying the cut-off value of 0.60 suggested by Murphy (1984).
4. Conclusions and Policy Implications

This research paper by employing the recently developed maximum likelihood-based panel cointegration procedures, empirically asserts that the Feldstein-Horioka puzzle is not valid for the panel, considered in this paper that consists of 14 Latin American and 4 Caribbean countries during the period 1960-2002. The individual and panel unit root tests conducted in the paper demonstrate that the savings and investment ratios are non-stationary in levels and stationary in first-differences. The Johansen cointegration tests for individual countries show that for majority of countries, a single cointegrating vector is present, and thus, these countries maintain their inter-temporal budget solvency in the long-run. For these countries, the savings-retention coefficient is either significantly low or not statistically significant indicating a moderate degree of capital mobility. For the remaining eight countries, no cointegrating relationship between savings and investment ratios is found, revealing perhaps the lack of inter-temporal imbalance in the short-run or due to the lack of robustness and
120
sufficient power of the cointegration tests in small samples. The absence of cointegration in these countries could also be due to the persistence shocks that they may have experienced relative to their current account, productivity and population. The evidence from the panel cointegration tests which indicate the presence of a single cointegrating vector and a moderate magnitude of the average savingsretention coefficient, confirms the prevalence of a moderate degree of capital mobility. The empirical evidence of a reasonable degree of capital mobility is consistent with the macroeconomic evidence of a large capital flows and volatility of capital flows, deregulation of capital markets, integration of world financial markets, and finally, the information technology revolution experienced by these countries during the period 1970-2000, the recent volatility of capital flows. The prevalence of capital mobility under both flexible and fixed exchange systems, have tremendous implications for the effectiveness of fiscal and monetary policies, exchange rate stabilization, co-operation among various countries monetary authorities and independence of the central banks. Increased capital mobility, despite its potential to induce volatility and macroeconomic instability, has an additional benefit of augmenting economic growth by providing an additional source of much warranted savings. A further implication of increased capital mobility is the shifting of the tax burden and an externality generated on domestic factors, such as labor which are not readily mobile across countries. Capital mobility also imparts fiscal discipline in these countries.
121
International Research Journal of Finance and Economics - Issue 11 (2007) Baltagi BH (2005) Econometric Analysis of Panel Data, John Wiley & Sons, New York. Banerjee A, Zanghieri P (2003) A New Look at the Feldstein-Horioka Puzzle Using an Integrated Panel. CEPII Working Paper. Baxter M, Crucini M (1993) Explaining SavingInvestment Correlations. American Economic Review 83: 416-436. Bayoumi T (1990) Saving-Investment Correlations: Immobile Capital, Government Policy or Endogenous Behavior? IMF Staff Papers 37: 360-387. Breitung J (2000) The Local Power of Some Unit Root Tests for Panel Data in BH Baltagi (ed) Nonstationary Panels, Panel Cointegration, and Dynamic Panels. Elsevier Science, Amsterdam, 161-178. Breitung J (2005) A Parametric Approach to the Estimation of Cointegration Vectors in Panel Data. Econometric Reviews 24: 151-173. Breitung J, Pesaran MH (2005) Unit Roots and Cointegration in Panels. Cambridge Working Papers in Economics, University of Cambridge, UK. Coakley J, Fuertes A, Spagnolo F (2003) The Feldstein-Horioka Puzzle is not as Bad as You Think. Money, Macro and Finance Research Group Conference Paper 17: 1-21. Coakley J, Kulasi F (1997) Cointegration of Long Span Saving and Investment. Economics Letters 54: 1-6. Coakley J, Kulasi F, Smith R (1998) The Feldstein-Horioka Puzzle and Capital Mobility: A Review. International Journal of Finance and Economics 3: 169-188. DeWet AH, Van Hyden R (2005) Capital Mobility in Sub-Saharan Africa: A Panel Data Approach. South African Journal of Economics 73: 1-22. Dooley M, Frankel J, Mathieson D (1987) International Capital Mobility: What do SavingInvestment Correlations Tell Us? IMF Staff Papers 34: 503-530. Engle RF, Granger CWJ (1987) Cointegration and Error Correction: Representation, Estimation and Testing. Econometrica 50: 251-276. EVIEWS 5.0. Quantitative Microsoft, Irvine, CA. Feldstein M, Horioka C (1980) Domestic Saving and International Capital Flows. Economic Journal 90: 314-329. Feldstein M, Horioka C (1983) Domestic Saving and International Capital Movements in the Long Run and Short Run. European Economic Review 21: 129-151. Gerdtham UG, Lothgren (2000) On Stationarity and Cointegration of International Health Expenditure and GDP. Journal of Health Economics 19: 461-475. Hadri K (2000) Testing for Stationarity in Heterogeneous Panel Data. Econometrics Journal 3: 148-161. Haque N, Montiel P (1990) How Mobile Capital is Developing Countries? Economics Letters 33: 359-362. Herwartz H, Xu F (2006) University of Kiel, Department of Economics Working Paper 200606: 1-20. Hlouskova J, Wagner M (2005) The Performance of Panel Unit Root and Stationary Tests:From a Large Scale Simulation Study. European University Institute Working Paper. EUI ECO 2005/05. Ho TW (2002) The Feldstein-Horioka Puzzle Revisited. Journal of International Money and Finance 21: 555-564. Hoffman M (2004) International Capital Mobility in the Long-run and the Short-run: Can We Still Learn from Saving-Investment Data? Journal of International Money and Finance 23: 113-131. Hussein KA, Mello Jr LR (1999) International Capital Mobility in Developing Countries: Theory and Evidence. Journal of International Money and Finance 18: 367-381.
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24]
International Research Journal of Finance and Economics - Issue 11 (2007) [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45]
122
Isaksson A (2001) Financial Liberalization, Foreign Aid, and Capital Mobility: Evidence from 90 Developing Countries. Journal of International Financial Markets, Institutions and Money 11: 309-338. Im KS, Pesaran MH, Shin Y (1997) Testing for Unit Roots in Heterogeneous Panels. University of Cambridge DAE Working Paper No.9526. Im KS, Pesaran MH, Shin Y (2003) Testing for Unit Roots in Heterogeneous Panels. Journal of Econometrics 115: 53-74. Jansen WJ, (2000) International Capital Mobility: Evidence from Panel Data. Journal of International Money and Finance 19: 507-511. Johansen S (1995) Likelihood Based Inference in Cointegrated Vector Autoregressive Models. Oxford University Press, Oxford. Johanse S (1988) Statistical Analysis of Cointegration Vectors. Journal of Economic Dynamics and Control 12: 231-254. Karlsson S, Lothgren M (2000) On the Power and Interpretation of Panel Unit Root Tests. Economics Letters 66: 249-255. Kim H, Oh KY, Jeong CW (2005) Panel Cointegration Results on International Capital Mobility in Asian Economies. Journal of International Money and Finance 24: 71-82. Larsson R, Lyhagen J, Lothgren M (2001) Likelihood-based Cointegration Tests in Heterogeneous Panels. Econometric Journal 4: 109-142. Levin A. Lin CF,Chu CSJ (2002) Unit Root Tests in Panel Data: Asymptotic and Finite Sample Properties. Journal of Econometrics 108: 1-24. Levy D (2000) Investment-Saving Comovement and Capital Mobility: Evidence from Century Long U.S. Time Series. Review of Economic Dynamics 3: 100-136. Maddala GS, Wu S (1999) A Comparative Study of Unit Root Tests with Panel Data and a New Simple Test. Oxford Bulletin of Economics and Statistics Special Issue: 631-652. Montiel P (1994) Capital Mobility in Developing Countries: Some Measurement Issues and Empirical Estimates. World Bank Economic Review 8: 311-350. Murphy RG (1984) Capital Mobility and the Relationship between Saving and Investment in OECD Countries. Journal of International Money and Finance 3: 327-342. Murthy NRV (2005) Capital Mobility in African Countries: Evidence from Panel Data Cointegration Tests. Indian Journal Economics and Business 4: 257-266. Obstfeld M, Rogoff K (2000) The Six Puzzles in International Macroeconomics: Is there a Common Cause. NBER Working Paper 7777. Payne JE, Kumazawa R (2005) Capital Mobility, Foreign Aid, and Openness: Further Panel Data Evidence from Sub-Saharan Africa. Journal of Economics and Finance 29: 122-126. Pedroni P (2004) Panel Cointegration: Asymptotic and Finite Sample Properties of Pooled Time Series Tests with an Application to the PPP Hypothesis. Econometric Theory 20: 597625. Rocha F (2000) Capital Mobility in Developing Countries: Evidence from Panel Data. Universidad Sao Paulo, Department of Economics Working Paper, 1-15. Sinha D (2004) The Mother of All Puzzles Would Not Go Away. Economics Letters 82: 259267. World Development Indicators CD-Rom, World Bank, Washington, DC, 2005.
Does Inflation Have an Impact on Conditional Stock Market Volatility?: Evidence from Turkey and Canada
Fatma Sonmez Saryal P.h. D Candidate, Rotman Management School Department of Finance, University of Toronto, Canada E-mail: fatma.sonmezsaryal04@rotman.utoronto.ca Abstract I use Generalized AutoRegressive Conditional Heteroskedasticity (GARCH) models to estimate conditional stock market volatility using monthly data from Turkey and Canada. Following Engle and Rangel (2005), I look at the impact of inflation and find that the rate of inflation has high predictive power for stock market volatility in Turkey, whereas it is weaker, but still significant for Canada. The findings suggest that the higher the rate of inflation, the greater the stock market volatility, that is, higher rates of inflation are coincident with greater stock market risk. These results are also supported when the change in the rate of inflation is used as explanatory variable for predicting conditional volatility.
Introduction
The volatility of security prices has been studied for many years and a number of stylized facts have been presented in the literature. One of the most prominent stylized facts is of volatility clustering; that is, large (small) shocks tend to follow similar large (small) shocks. One reason for this might be that stock market volatility depends on the overall health of the economy, and real economic variables which tend to display persistence. Therefore, an interesting question in finance is: what derives stock market volatility? Understanding the nature of stock market volatility gives some important implications for policy makers, economic forecasters and investors. The classic paper in the literature is that by Schwert (1989), which looked at the relationship between stock market volatility and the volatility of real and nominal macroeconomic variables. He examined the impact of the level of economic activity, financial leverage, and stock trading. He found that macroeconomic volatility as measured by movements in inflation and real output have weak predictive power for stock market volatility and returns. In particular, inflation volatility predicts stock market volatility only for the sub-period 1953-1987. His results point to a positive link between macroeconomic volatility and stock market volatility, with the direction of causality being stronger from the stock market volatility to macro economic variables. Davis and Kutan (2003) extended Schwerts study by accounting for volatility persistence in an international setting. Their results are in line with the findings in Schwerts paper in the sense that the variability of inflation and output growth rate has weak predictive power for conditional stock market volatility. On the other hand, existing studies in the literature, for example Engle and Rangel (2005) provide evidence for the impact of the overall health of the economy on unconditional stock market volatility. By using Spline-Garch model they find that volatility in macroeconomic factors such as GDP growth, inflation and short term interest rates are important explanatory variables that increase volatility. They observed positive relations among long term market volatilities and each of the following variables: emerging markets, inflation growth, and macroeconomic volatilities. In particular, they observe that emerging markets including Turkey show larger expected volatility compared to developed markets including Canada; countries with high
124
inflation rates experience larger expected volatilities than those with more stable prices. They compare Spline-GARCH model results with the results of annual realized volatility as an alternative measure of unconditional volatility. Inflation variables are no longer good predictors of annual realized volatilities. They claim that changes in significance due to the fact that realized volatility is a noisier measure of unconditional volatility. Other recent research has examined the information effects on stock market volatility by looking at the release of news of key macroeconomic variables such as the rate of inflation, GDP growth, etc.1 Earlier studies on the impact of inflation on stock market behavior do not account for this stylized fact of volatility clustering, e.g. Schwert (1989), Huang and Kracaw (1984), Kaul (1987), while Hamilton and Lin (1996) address some of these issue for the US market. Moreover, many of these studies are empirically tested on data from developed markets. Engle and Rangel (2005) studied emerging markets (included Turkey) as well as developed markets by accounting for volatility clustering and found that countries with high rates of inflation tend to have high stock market volatility. Studying the impact of macro economic factors such as the rate of inflation on conditional stock market volatility also has important implications for investors and policymakers. In many ways the performance of the economy influences the success of the stock market and vice versa. Irving Fisher found that real interest rates were equal to nominal interest rates minus expected inflation. This macroeconomic relationship is known as the Fisher Effect (Mankiw, 1997). So, the understanding of impact of Fisher effect on stock market index through stock volatility can broaden our understanding of such risks allowing it to be priced more efficiently. In this study, I examine the impact of inflation on stock market volatility. Specifically, I examined following questions: How does inflation rate impact conditional stock market volatility estimated by using nominal stock return series? and Does this relationship (if any) differ between countries with different rates of inflation? I estimate this relationship with data from Turkey and Canada. These two countries are selected for comparison on the basis of their inflation level: Turkey is an emerging market country with a high inflation rate and Canada a developed country with a low inflation rate.
Methodology
Since the introduction of ARCH and GARCH models by Engle (1982) and Bollerslev (1986), respectively, there has been an explosion of research looking for the dynamics of conditional stock market volatility. Although the standard GARCH (1,1) model captures the stylized fact of stock return volatility in terms of volatility clustering, it does not capture the asymmetric effect of information shocks on volatility. Therefore, it is necessary to search for the appropriate type of GARCH specification to model the dynamics of stock market volatility. The specification procedure for GARCH models includes determining time-varying volatility behavior as well as searching for the asymmetric effects of shocks on volatility.2 In this research, I first apply the standard linear GARCH (1,1) model and then test for nonlinearities in the effect of information using Quadratic GARCH model, (QGARCH) introduced by Sentana (1995). I estimate these models using monthly price data from the Istanbul (ISE) and Toronto (TSE) Stock Exchange. The ISE100 index and TSE300 index both consist of the largest firms in Turkey and Canada and are regarded as most accurately reflecting stock market behavior in each country. For the linear model, define the information set t of monthly returns to be {rt , rt q ,..., r1}. The jointly estimated GARCH (1, 1) model introduced by Bollerslev (1986) is then t = t z t and z t i.i.d. (0,1) rt = + t ,
t2 = + t21 + t21
1
Aksoy and Kutan examined CPI releases on stock market returns and volatility for Istanbul Stock Exchange (ISE) and found that stock market responses significantly to the information arrival. To test the time-varying volatility, the LM (Lagrange Multiplier) test developed by Engle (1982) is applied for ARCH (1) to the monthly ISE 100 index (Turkey) and TSE 300 index (Canada) returns. The LM test against ARCH (1) is the same as one for GARCH (1, 1).
125
where t2 is measurable with respect to t 1 and > 0, > 0 , 0 and + < 1 , such that the model is covariance-stationary that is first two moments of the unconditional distribution of the return series is time invariant . To estimate the impact of asymmetric effect of shocks on volatility, the above specification is replaced with Sentanas QGARCH (1, 1) model 3 t = t z t and z t i.i.d. (0,1) rt = + t ,
t2 = + t21 + t21 + t 1 The additional term t 1 makes it possible for positive and negative shocks to have different
effects on conditional volatility. The condition for covariance-stationary is the same as standard GARCH (1,1). Once the appropriate model is determined the estimation of the impact of inflation on stock market volatility can be investigated. To do so, I use two different model specifications. Model 1: rt = + t
t2 = + t21 + t21 + (inflation) t 1 The first model is an extension of the standard GARCH (1, 1) model to estimate the impact of the previous period inflation rate.
Model 2: rt = + t
t2 = + t21 + t21 + (inflation) t 1 Here, the standard GARCH (1, 1) model is extended by including the impact of changing of inflation rates. Whenever there is evidence that an asymmetric GARCH specification is suitable for conditional volatility estimation then the above models should be replaced with the appropriate conditional variance specification given by QGARCH.
Empirical Results
Appropriate Data
Monthly data on the stock price indices of the Turkish stock market (ISE100 index), and the Consumer Price Index (CPI) are obtained from the Istanbul Stock Exchange database. Stock market data for Turkey is available from January 1986 to September 2005. For Canada, the same frequency data on CPI and stock market price index (TSE 300 index) are obtained from CansimII database. The data used is from January 1961 to December 2005. Variables at time t are calculated as follows: Stock indext 4 CPI t Inflation t = ln( ) Nominal stock return t = ln( ) Stock indext 1 CPI t 1 Table 1 shows the descriptive statistics for both countries. The results indicate that the average monthly return is 4.4% and 0.54% for Turkey and Canada respectively. The difference between these two returns is caused by the high rate inflation in Turkey, which averages 3.7% per month while it is 0.22% per month for Canada. This suggests that the higher the rate of inflation, the higher the nominal stock returns consistent with the simple Fisher effect.
There are many asymmetric GARCH models in the literature. I apply QGARCH model simply because apart from the asymmetry, it is very similar to the standard GARCH model. For example, the unconditional variance of as implied by QGARCH (1,1) model is the same as that implied by the GARCH(1,1) model. I will use ``return`` instead of continuously compounded nominal return for the sake of simplicity
126
The standard deviation of the return series is 16% standard deviation per month for Turkey or 55% annually. In contrast for Canada, it is only 15.58% per year. Clearly stock market returns are of the order of four times more variable in Turkey than Canada. The standard deviation of the annual inflation rate is 8.6% and 1.38% per year for Turkey and Canada respectively. Clearly, Turkey is highly volatile in terms of inflation due to economic and political instability. Given these results, one can expect to see high conditional stock market volatility for Turkey due to high inflation and high inflation variability. This can be explained by the observation that high rates of inflation are associated with high inflation variability. The figures 1.a and 1.b. that show the stock return and inflation can be found in Appendix. Descriptive Statistics
Table 1. Descriptive statistics for Turkey and Canada
Turkey Nominal Stock Returns 01/1986-09/2005 +0.0441552 +0.1688146 +0.5804150 +5.2858684 -0.4942963 +0.8193392 Canada Inflation +0.0373163 +0.0252511 +1.7743652 +13.5285001 -0.0090407 +0.2207407 Nominal Stock Returns 01/1986-09/2005 +0.0054888 +0.0450914 -1.5465318 +9.7553251 -0.2565690 +0.1118717 Inflation +0.0022063 +0.0032978 +1.3615257 +14.0459449 -0.0085985 +0.0259485
Sample period Mean Std deviation Skewness Kurtosis Min Max
Evidence for time-Varying Volatility The first step is to find the appropriate model for stock return volatility for ISE100 index and TSE300 index. Table 2.a- Table 2.b reports the estimated coefficients for standard GARCH (1,1) model.5 Associated figures can be seen in Appendix given by Figures 1.a and 1.b. As seen from Table 2.a, this months stock return volatility can be explained by approximately 90% of previous month return volatility. This is highly significant for Turkey. However, neither the constant term and nor the coefficient for return innovations are statistically significant. Together these imply that the new information arriving to stock market last month does not have predict power for this months volatility. With these results, the persistence parameter + is 0.92 and to zero, which indicates that the impact of shocks on the conditional variance diminishes very slowly, that is the impact of shocks last for a long time period. The constant term for the return series is significant and it is an unbiased estimator of the mean of the stock return.
The Ljung-Box Q-statistic results for autocorrelation of return series for Turkey and Canada is given in the Appendix by Table A1. The results show that the monthly return series for both Turkey and Canada are not serially correlated.
127
Table 2.a: The estimated coefficients for the GARCH (1, 1) volatility model for return series, Turkey. Since there is no serial correlation in nominal stock return series in Turkey, GARCH (1, 1) is
rt = +
2 t
t 2 t 1
= +
2 t 1
Unbiased sample variance: +0.02850 Parameters: Estimated Coef.: Standard errors: Estimated sample variance (
+0.0447004 +0.0115585 ): +0.02842
+0.0023361 +0.0037482
(ARCH)
+0.0215967 +0.0305705
(GARCH) +0.8962170 +0.1548529
On the other hand, results for Canada show that the persistence parameter + =0.80000353 and =0.0004465, where the all the coefficients are significantly different than zero. Contrary to the results for Turkey, new information arrival has an impact on predicting the next months stock market volatility in Canada. The constant term for the return series is significant, too. Using these results we can conclude that there is significant time varying volatility for both countries stock market returns during the sample periods.
Table 2.b: The estimated coefficients for the GARCH (1, 1) volatility model for return series, Canada.
rt = + t
2 t
= +
2 t 1
2 t 1
Unbiased sample variance: +0.00203 Parameters:
+0.0004465 +0.0002680
(ARCH)
Estimated Coef.: +0.0071805 Standard errors: +0.0034185 Estimated sample variance:+0.00223
+0.1384334 +0.0558986
(GARCH) +0.6616019 +0.1530623
The standard GARCH (1, 1) model does not take into account the asymmetric effect of shocks on stock market volatility. Negative innovations might have different effects on volatility than positive innovations. For Canada, as seen from Table 3, in line with our expectations, bad news has a larger impact on stock market volatility than good news. This is a significant result. Although all the other coefficient estimates are significant including the constant term in the return series, the estimated coefficient > 0 is too small to economically explain conditional volatility. However, for Turkey, asymmetric GARCH specification could not give sensible results. Therefore, I prefer to use the standard GARCH (1, 1) model specification for both countries in this paper.
Table 3: The estimated coefficients for the Asymmetric GARCH (1, 1) volatility model for nominal return series, Canada Model:
rt = + t ,
2 t
t = t z t and z t i.i.d. (0,1)
= +
2 t 1
+ t21 + t 1
Unbiased sample variance: +0.00203 Parameters: Estimated Coef.: Standard errors: +0.0066434 +0.0030745
+0.0010931 +0.0002197
+0.1402399 +0.0740163
+0.3423112 +0.0943333
-0.0162566 +0.0027319
128
The impact of inflation on conditional stock market volatility The impact of the inflation rate on stock market volatility is investigated separately by estimating the coefficients of the models given by Tables 5.a-.b and 6.a-b for Turkey and Canada respectively. First, I look at the prediction power of the previous periods inflation rate for conditional stock market volatility by using the GARCH (1,1) model. As seen from the results in Table 5.a, the estimated coefficient on the inflation rate is +0.2597451 which is both statistically and economically significant. A 1% of increase in inflation causes a 0.25% increase in conditional volatility in Turkey. The estimated GARCH coefficient is +0.6590645, which is less than the 0.89 where inflation is not taken into account. This suggests that the inflation rate itself has strong predictive power in Turkey. The estimated ARCH coefficient is insignificant as in the standard GARCH (1,1) model without inflation, which leads to the conclusion that information does not impact the conditional stock market volatility. Secondly, the impact of inflation variability as measured by the movement of inflation, on conditional stock market volatility is investigated. As seen from Table 5.b, the estimated coefficient +0.3324536 is highly significant. It has a bigger impact on volatility than the inflation rate itself. The ARCH coefficient is still insignificant. If we take into account that inflation is highly autocorrelated, it is better to look at the three month average inflation rates impact on conditional volatility.6
Table 5a: The estimated coefficients for the GARCH (1,1) volatility model for nominal return series with inflation, Turkey Model:
rt = +
2 t
= + t2 1 + t2 1 + t 1 where t 1 is previous period inflation level.

Unbiased sample variance:+0.02850 Parameters Estimated Coef.: Standard errors: +0.0430587 +0.0104024
+0.0000000 +0.0000000
+0.0480063 +0.0322875
+0.6590645 +0.0952243
+0.2597451 +0.0910932
Table5b:
The estimated coefficients for the GARCH (1,1) volatility model for nominal return series with change in inflation, Turkey
t2 = + t21 + t21 + (change in inflation) t 1 t 1 as change in inflation t = t t 1

Unbiased sample variance:+0.02854 Parameters Estimated Coef.: Standard errors: +0.0430612 0.0111948
Model:
rt = + t
where change in inflation is computed
+0.0006090 +0.0010106
+0.0580487 +0.0449653
+0.5587142 +0.1000748
+0.3324536 +0.0449338
For Canada, surprisingly, a decrease in inflation in the previous period increases conditional volatility this month. Although the results show that inflation has a negative impact on conditional volatility, its impact is weaker than Turkey. Since last months conditional volatility is insignificant Table 6.a shows that this months conditional volatility can be predicted mostly by using last months returns innovations. Contrary to the results for Turkey, the estimated coefficient on inflation is less
6
Table 5.c in Appendix shows that the estimated coefficients for ARCH, GARCH and average three month inflation are all significant, but now some of the predictive power of inflation is taken by the ARCH term which indicates that past return innovations now impact on predicting todays conditional volatility.
129
statistically significant, but more important, less significant economically as the estimated coefficient is one tenth size of that Turkey.
Table 6a: The estimated coefficients for the GARCH (1,1) volatility model for nominal return series with inflation, Canada
t2 = + t21 + t21 + where t 1 is previous period inflation rate.

Unbiased sample variance: +0.00203 Parameters Estimated Coef: Standard errors:
Model:
rt = + t
t 1
+0.0076936 + 0.0034360
+0.0012927 +0.0007416
+0.1359401 +0.0614234
+0.2856400 +0.3889716
-0.0443826 +0.0216767
Table 6b: The estimated coefficients for the GARCH (1,1) volatility model for nominal return series with change in inflation, Canada Model:
rt = + t
t2 = + t21 + t21 + (change in inflation) t 1

where change in inflation is computed as change in inflation t = t t 1
+0.0095680 +0.0026015
+0.0020048 +0.0003110
+0.1537629 +0.0677385
+0.0099732 +0.1490025
-0.0910035 +0.0613717
Conclusion
Does the economy actually have a significant influence on the performance of the stock market? If so, how can investors benefit from significant relationships between the economy and the stock market? Results from this study show that the rate of inflation is one of the underlying determinants of conditional stock market volatility particularly in a highly inflated country like Turkey. I also show that the variability in the inflation rate as measured by the change in the rate of inflation has a stronger impact in predicting stock market volatility in Turkey than in Canada. These results can be discussed in the context of whether or not a higher rate of inflation leads to greater variability around the rate of inflation which also affects conditional stock market volatility. The results can also support Fisher effect in international stock market as well as providing the tools to investors to make good portfolio decisions given their knowledge about past of the economy and expectations for the future of the economy.
130
Appendix
Figure 1a: Nominal return series and inflation level for Turkey
Figure 1b: Nominal return series and inflation level for Canada
Table A1: H= 0 indicates acceptance of the null hypothesis that the model fit is adequate (no serial correlation at the corresponding element of 1 Lag). 1 indicates rejection of the null hypothesis.
Results: Turkey Nominal returns: Inflation: Canada Nominal returns: Inflation: H +0.0000000 +1.0000000 +0.0000000 +0.0000000 pValue +0.4815966 +0.0000000 +0.1435734 +0.1732355 Qstat +0.4952458 +343.5915071 +2.1392361 +1.8547196 CriticalValue +3.8414588 +31.4104328 +3.8414588 +3.8414588
131
Figure 2a: The conditional variance for return series computed by the standard GARCH (1,1) model for Turkey
Figure 2b: The conditional variance for return series computed by the standard GARCH(1,1) model for Canada
132
Table 5c: The estimated coefficients for the GARCH (1, 1) volatility model for stock return series with average three months lag inflation level, Turkey Model:
rt = + t
in which average three month at t is calculated as

Unbiased sample variance: +0.02864 Parameters: Estimated Coef.: Standard errors:
t2 = + t21 + t21 + (average inflation) t 1 t 1 + t 2 + t 3

3
+0.0391029 +0.0107254
+0.0001238 +0.0009565
+0.0860368 +0.0396189
+0.6627557 +0.1586708
+0.2156172 +0.1176060
Table 6c: The estimated coefficients for the GARCH (1,1) volatility model for nominal return series with average three lag inflation level, Canada Model:
rt = + t
in which average three month at t is calculated as t-k.

t2 = + t21 + t21 + (average inflation)t 1 t 1 + t 2 + t 3

3
where xt k is inflation rate at
-0.0077261 +0.0578790
+0.0444710 +0.0332992
+0.0004849 +1.0400235
+0.0171295 +0.8283962
-1.7447750 +0.6300648
133
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] Istanbul Stock Exchange database: www.ise.org International Monetary Fund (IMF) database: www.imf.org CANSIM II at CHASS, University of Toronto: datacentre2.chass.utoronto.ca/cansim2 Aksoy, Tansu and Kutan, Ali, M., 2003, Public Information Arrival and the Fisher Effect in Emerging Markets: Evidence from Stock and Bond Markets in Turkey, Journal of Financial Services Research, 23, 225-239 Beltratti, A., Morana, C., 2006, Breaks and persistency: Macroeconomic causes of stock market volatility, Journal of Econometrics, Vol.131, Issue 1-2. Bollerslev, T., 1986, Generalized conditional heteroskedasticity, Journal of Econometrics 31, 307-327. Davis, Nicolas and Kutan, Ali, M., 2003, Inflation and output as predictors of stock returns and volatility: International evidence, Applied Financial Economics 13, 693-700. Engle, Robert, F., 2001, What good is a volatility model?, Quantitative Finance Volume 1, 237245. Engle, Robert, F., 1982, Autoregressive conditional heteroskedasticity with estimates of the variance of United Kingdom inflation, Econometrica 50, 987-1007. Engle, Robert, F. and Rangel, Jose, G., 2005, The Spline GARCH Model for Unconditional Volatility and its Global Macroeconomic Causes (The latest version August 12, 2005). Hamilton, J.D., Lin, 1996, Stock market volatility and the business cycle, Journal of Applied Econometrics 11, 573593. Huang, R. D., and Kracaw, W. A., 1984, Stock market returns and real activity: A note, Journal of Finance 39, 267-72. Kaul, G., 1987, Stock returns and inflation: The role of monetary sector, Journal of Financial economics 18, 253-276. Sadorsky, P., 2003, The macroeconomic determinants of technology and stock price volatility, Review of Financial Economics 12, 191205. Schwert, William, G., 1989, Why does stock market volatility change over time?, The Journal of Finance 44, 1115-1153. Sentana, E., 1995, Quadratic ARCH models, Review of Economic Studies, 62, 639-661.
Impacts of Higher Crude Oil Prices and Changing Macroeconomic Conditions on Output Growth in Germany
Yu Hsing Department of General Business, College of Business Southeastern Louisiana University, Hammond, LA 70402, USA E-mail: yhsing@selu.edu Tel.: 985-549-2086; Fax: 985-549-2881 Abstract Applying the monetary policy function (Romer, 2000, 2006; Taylor, 1993, 1999, 2001) to an open economy, this paper finds that an increased real crude oil price may or may not reduce the German output, depending upon whether it has reached a critical value and that there is a nonlinear relationship between real output and the real crude oil price. In addition, a lower ratio of government spending to GDP, a higher real stock price, a lower real world interest rate, and a lower expected inflation rate would raise real output for Germany. Keywords: Critical value of the crude oil price, monetary policy function, stock price, world interest rate Jel Classification Codes: Q43, E52
1. Introduction
Rapid rises of the crude oil price in recent years have led to continual study of its impacts on real output for major industrialized countries such as Germany. According to the U.S. Energy Information Administration, Germany consumed 2.630 million barrels of crude oil per day in 2006 and ranked 5th in the world in the demand for crude oil. German crude oil consumption has continued to decline from a record high of 2.922 million barrels of crude oil per day in 1998. Germany produced 67,600 barrels of crude oil per day in 2006, which was only 2.57% of total crude oil consumption. To meet its demand, Germany imported 2.483 million barrels of crude oil per day in 2006 and is 4th largest net oilimporting country with major sources coming from Russia (34%), Norway (16%), the U.K. (12%), and Libya (12%). To save energy cost and develop renewable energy technologies, Germany is the world largest producer of biodiesel and generator of electricity from wind and provided tax incentives to encourage consumers to blend biodiesel with conventional diesel fuel. The Oil and Gas Journal reported that as of January 2006, Germany had an estimated 367 million barrels of proven oil reserves. This study has several focuses. First, the monetary policy function (Romer, 2000, 2006; Taylor, 1993, 1999, 2001) is employed along with an open economy IS function in order to indicate that the Deutsche Bundesbank has pursued inflation targeting and conducts monetary policy to maintain price stability and output growth. The choice of the monetary policy function is consistent with the notion that the central bank determines the short-term interest rate, which is expected to affect other interest rates, bank loans, household consumption spending, investment spending, etc. Second, this study would like to find if there may exist a critical value of the real crude oil price below which the oil price-output relationship is positive and above which the oil price-output relationship is negative.
135
Third, in addition to the real crude oil price, other relevant macroeconomic variables such as fiscal policy, the real exchange rate, the real financial stock price, the real world interest rate, etc. are also included in the model to estimate their respective impacts on the equilibrium real output. A simple regression analysis including real GDP and the real crude oil price may imply that the real oil price is the only explanatory variable for the variation in real GDP without controlling for other possible factors. Fourth, comparative-static analysis is applied to determine the possible impact of a change in one of the exogenous variables on the equilibrium real output. Several recent articles addressed the related subjects for Germany and other countries. Applying the Chow stability test and Granger causality test, Hooker (1996) found that there is a breakdown of the negative oil price-macroeconomy relationship after 1973.Q3 whereas Hamilton (1996) maintained that the negative oil price-macroeconomy relationship continues to hold after 1973.Q3. Abeysinghe (2001) revealed that higher oil prices have negative impacts on output growth even for oil-exporting countries like Indonesia and Malaysia. Based on a sample of fourteen European countries, Cunado and Peres de Gracia (2003) showed that oil price shocks cause short-term asymmetric negative impacts on production growth, permanent effects on the inflation rate, and different impacts for these countries. Huntington (2004) that when the crude oil price doubles, the German output would drop 1.7% based on the initial output gap and decline 1.3% based on the oil share, which is less than the 2.3% decrease for the U.S. Jimenez-Rodriguez and Sanchez (2005) showed that the negative impact of higher oil prices on the German output growth is smaller than the U.S. and the euro area, that the negative impacts for Germany are short-term and nonlinear, and that the impact of higher oil prices on output growth is positive for Japan, and negative for the oil-exporting U.K. and positive for the oil-exporting Norway. Kilian (2005) indicated that the decrease in output due to an oil shock is temporary and occurs in the second year, that the increase in inflation due to an oil shock is not permanent and would reach the peak after three to four quarters, and that there is strong evidence of stagflation for Germany, Canada, and Japan.
2. The Model
Higher oil prices are expected to affect a macroeconomy through several channels. For a net oilimporting country like Germany, higher real crude oil prices are expected to reduce net exports and cause real depreciation of the currency (Chen and Chen, 2007) if the amount of oil imports and other factors remain the same. It is possible that some countries may reduce oil consumption to offset the crude oil price increase and may reduce total expenditure on oil consumption. A higher crude oil price is expected to raise the gasoline price, petroleum-product price, transport cost, electricity bill, etc. and lead to a higher price level or inflation rate, which would reduce real wealth and consumption spending. A higher energy cost may cause households to reduce consumption spending on other goods and services and business firms to generate less profit. Suppose that aggregate spending is a function of real output, the real interest rate, real government spending, real government revenues, the real stock price, the real effective exchange rate, and the real crude oil price, that the Deutsche Bundesbank affects the real short-term interest rate based on the inflation gap, the output gap, the exchange rate gap, and the real world interest rate, and that the inflation rate is determined by the expected inflation rate, the output gap, the real effective exchange rate, and the real crude oil price. Extending Romer (2000, 2006), Taylor (1993, 1999, 2001), and other previous studies, the macroeconomic model for Germany can be expressed as: Y = S (Y , R, G , T , F , , P ) (1) * R = R( , Y , , R ) (2)
= e + (Y ) + P
where Y R = real GDP for Germany, = the real interest rate,
(3)
International Research Journal of Finance and Economics - Issue 11 (2007) G T F = real government spending, = real government revenues, = the real stock price, = the real effective exchange rate, = the real crude oil price per barrel, = the inflation rate, = the target inflation rate, = potential output, = the target real effective exchange rate, = the real world interest rate,
136

R*
e = the expected inflation rate, and , , = positive parameters. Note that equation (1) is an open economy IS function, equation (2) is a monetary policy function, and equation (3) is an augmented Phillips curve. Let 0 < S Y < 1, S R < 0, S G > 0, S T < 0, S F > 0, S < 0, S P > or < 0, R > 0, RY > 0, R < 0, RR > 0, Y = > 0, = < 0, P = > 0.
*
Note that the sign of S P is unclear for Germany because of possible reduction in the quantity of oil imports or consumption due to a higher oil price and energy-saving measures. Solving for the three endogenous variables of real output, the real interest rate, and the inflation rate, we have the equilibrium real GDP: Y = Y ( P, G, T , F , , R * , e ; , , , , , ). (4)
(5) As the real crude oil price rises, aggregate spending may or may not decline, the inflation rate is expected to increase, and the Deutsche Bundesbank would raise the real interest rate, which would lower aggregate spending: Y / P = ( S P + S R R ) / J > or < 0. (6) Theoretically, government deficit spending is expected to raise real output in the short run: Y / G Y / T = ( S G S T ) / J > 0. (7) However, the impact of deficit-financed government spending on real output may be unclear or neutral in the long run partly due to the crowding-out effect, implementation lags, impact lags, and the Ricardian equivalence theory (Barro, 1989; Taylor, 2000). A higher real stock price is expected to cause households to increase consumption spending because of the wealth effect and business firms to increase investment spending owing to the balancesheet effect (Mishkin, 1995): Y / F = S F / J > 0. (8) Real exchange rate appreciation may or may not raise real output, depending upon whether the negative impact of the appreciation would be greater or less than the positive impact of a lower real interest rate caused by a lower inflation rate due to exchange rate appreciation: Y / = ( S S R R + S R R ) / J > or < 0. (9) Bahmani-Oskooee and Miteza (2003) indicated that the effect of currency devaluations is unclear, depending upon the theoretical models, the stages of economic development and growth, countries under study, time periods used in empirical work, regression methods employed in estimating parameters and testing the hypothesis, and other related factors.
The Jacobian for the endogenous variables has a positive sign: J = (1 S Y ) S R R S R RY > 0.
137
The Deutsche Bundesbank may consider adjusting its real interest rate based on the real world interest rate in order to make the interest rate comparable and avoid international outflows of the capital. Thus, the impact of a higher real world interest rate on the German output would be negative: Y / R * = S R RR* / J < 0. (10) To test for a possible nonlinear relationship between real output and the real oil price, a quadratic function for the real oil price will be used in empirical work. If the coefficient of the linear term is positive and significant and if the coefficient of the quadratic term is negative and significant, real output and the real oil price exhibit a bell-shaped relationship. It indicates an initial positive relationship when the real oil price is less than the critical value and a negative relationship when the real oil price is greater than the critical value.
The data were collected from the International Financial Statistics, which is published by the International Monetary Fund. Real GDP is measured in billion German marks at the 1995 price. The real crude oil price per barrel in U.S. dollar is equal to the nominal world crude oil price per barrel divided by the consumer price index in the U.S. Due to lack of complete date for real government spending and revenues, the ratio of government consumption spending to nominal GDP is used to represent fiscal policy. The real effective exchange rate is a trade-weighted exchange rate adjusted for relative prices. An increase in the real effective exchange rate means a real appreciation. The real stock price is measured by the nominal share price divided by the consumer price index. The real world interest rate is represented by the U.S. federal funds rate minus the inflation rate in the U.S. The inflation rate is derived from the consumer price index, and the expected inflation rate is the lagged inflation rate. Except for negative values and the real oil price, variables are measured in the logarithmic scale. The sample ranges from 1991.Q3 to 2006.Q4 with a total of 62 observations. The data for the consumer price index and real GDP before 1991 are not available. According to the ADF unit root test with the critical values of -3.540, -2.909, and -2.592 at the 1%, 5%, and 10% levels, respectively, these variables have unit roots in levels and are stationary in first difference at the 5% level. According to the Johansen test, the null hypothesis that these variables have one cointegrating relationship cannot be rejected at the 5% level because the trace test statistic of 258.768 is greater than the critical value of 159.530. Table 1 presents the estimated regression and related statistics. First difference is not used due to potential loss of valuable information and obscure outcomes (Greene, 2003). To correct for possible autocorrelation and heteroskedasticity, the Newey-West (1987) method is employed to generate consistent estimates when their forms are unknown. As shown, the value of R2 suggests that 80.8% of the variation in real GDP can be explained by the six right-hand side variables. Except for the real effective exchange rate, the coefficients of other variables are significant at the 1% or 5% level. Real GDP is positively associated with the real crude oil price and the real stock price and negatively affected by the squared real crude oil price, the ratio of government consumption spending to GDP, the real world interest rate, and the expected inflation rate.

Table 1: Estimated Double-Log Regression of Real GDP for Germany
138
Dependent Variable: Log(Real GDP) Method: Least Squares Sample (adjusted): 1991.Q3 2006.Q4 Included observations: 62 after adjustments Newey-West HAC Standard Errors & Covariance (lag truncation=3) Variable Coefficient Std. Error t-Statistic Constant 10.15567 1.118581 9.079066 Real oil price 0.007110 0.002104 3.379681 (Real oil price)^2 -7.68E-05 2.75E-05 -2.796212 Log(Government spending/GDP) -1.293203 0.320008 -4.041159 Log(Real stock price) 0.069740 0.030078 2.318617 Log(Real effective exchange rate) 0.032234 0.154042 0.209256 Real federal funds rate -0.005046 0.002417 -2.087357 Expected inflation rate -0.035787 0.011081 -3.229470 R-squared 0.807552 Mean dependent var Adjusted R-squared 0.782606 S.D. dependent var S.E. of regression 0.031432 Akaike info criterion Sum squared resid 0.053351 Schwarz criterion Log likelihood 130.8239 F-statistic Durbin-Watson stat 1.056110 Prob(F-statistic)
Prob. 0.0000 0.0014 0.0071 0.0002 0.0242 0.8350 0.0416 0.0021 6.862307 0.067414 -3.962062 -3.687593 32.37085 0.000000
Because the real crude oil price has a positive coefficient and the squared real crude oil price has a negative coefficient, it suggests that the relationship between real output and the real crude oil price has a bell shape. In other words, when the real crude oil price is relative small and less than the critical value, real output and the real crude oil price have a positive relationship whereas when the real crude oil price is relatively high and greater than the critical value, the relationship becomes negative. Based on the estimated coefficients, the critical value for the real crude oil price is calculated to be $46.29. Thus, the real crude oil price of $47.04 in 2005 is so close to the critical value that its negative impact would be minimal whereas the real crude oil price of $54.90 in 2006 would have a negative impact on real output. Several comments can be made. An increased real crude oil price may affect the German output positively or negatively depending upon whether the real crude oil price is greater or less than the critical value. Therefore, even for a large oil-importing country like Germany, the impact is not always negative as the conventional wisdom may suggest. The negative significant coefficient of the ratio of government consumption spending to GDP suggests that expansionary fiscal policy needs to be pursued with caution and that continual expansion of the government may not be conducive to economic growth. A healthy stock market is important as a rising real stock price is expected to increase aggregate spending and real output due to the wealth effect for households and the balancesheet effect for business firms (Mishkin, 1995). The insignificant coefficient of the real effective exchange rate may indicate that maintaining the stability and a fair market value of the euro dollar may serve the euro area well. The interdependence between central banks implies that monetary tightening by the Federal Reserve Bank may cause the Deutsche Bundesbank to respond positively, leading to a higher real interest rate and a lower real output. Several different versions are considered. When the linear form is used in empirical estimation, the value of R2 is estimated to be 0.801, which is slightly lower than the value of the double-log regression (Table 2). Except for the real effective exchange rate, other variables have significant coefficients at the 1%, 5% or 10% level and the same signs as the double-log form. The critical value of the real crude oil price is estimated to be $48.02, which is slightly greater than $46.29 calculated from the coefficients in the double-log form.
139
Table 2:

Estimated Linear Regression of Real GDP for Germany
Dependent Variable: Real GDP Method: Least Squares Sample (adjusted): 1991.Q3 2006.Q4 Included observations: 62 after adjustments Newey-West HAC Standard Errors & Covariance (lag truncation=3) Variable Coefficient Std. Error t-Statistic Constant 2177.558 353.3107 6.163295 Real oil price 6.196733 1.959227 3.162846 (Real oil price)^2 -0.064529 0.025459 -2.534586 Government spending/GDP -71.37332 17.78313 -4.013541 Real stock price 0.838671 0.465450 1.801850 Real effective exchange rate 0.163784 1.408506 0.116282 Real federal funds rate -4.262705 2.355049 -1.810029 Expected inflation rate -35.81924 10.22777 -3.502154 R-squared 0.801150 Mean dependent var Adjusted R-squared 0.775373 S.D. dependent var S.E. of regression 30.37214 Akaike info criterion Sum squared resid 49813.20 Schwarz criterion Log likelihood -295.3301 F-statistic Durbin-Watson stat 1.077629 Prob(F-statistic)
Prob. 0.0000 0.0026 0.0142 0.0002 0.0772 0.9079 0.0759 0.0009 957.6971 64.08323 9.784842 10.05931 31.08020 0.000000
An attempt was made to include the lagged dependent variable to test the partial adjustment model. Its coefficient is estimated to be 0.965 and significant at the 1% level. However, except for the real stock price, other coefficients become insignificant at the 10% level, largely due to a high degree of multicollinearity.
4. Summary and Conclusions

This paper has examined the impacts of a rising crude oil price and changing macroeconomic conditions on real output in Germany. A macroeconomic model including the monetary policy function is considered. The results show that a higher real crude oil price may or may not cause a negative impact on the German output, depending upon whether the real crude oil price is greater or less than the critical value of $46.29 per barrel estimated from the double-log from and $48.02 per barrel estimated from the linear form. As the trend of the world crude oil price has been on the rise, its potential negative impact on real output needs to be considered in determining the economic growth rate for Germany. In addition, a lower ratio of government consumption spending to GDP, a higher real stock price, a lower real world interest rate, and a lower expected inflation rate would cause real output to rise. There may be areas for future research. The expected inflation rate may be constructed by more sophisticated methods to determine whether different results would be generated. If the data are available, the debt/GDP ratio may be considered to determine its impact on real output in the long run. In the specification of the model, a money demand function may be considered to replace the monetary policy function. That being the case, the relationship between real money demand and the real stock price may be positive due to the wealth effect or negative due to substitution effect (Friedman, 1988), and an increased real stock price may affect real output negatively or positively.
140
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] Abeysinghe, T. (2001) Estimation of Direct and Indirect Impact of Oil Price on Growth, Economics Letters, 73(2), 147-153. Bahmani-Oskooee, M. and I. Miteza (2003) Are Devaluations Expansionary or Contractionary? A Survey Article, Economic Issues, 8(2), 1-28. Barro, R. J. (1989) The Ricardian Approach to Budget Deficits, Journal of Economic Perspectives, 3(2), 37-54. Chen, S.-S. and H.-C. Chen (2007) Oil Prices and Real Exchange Rates, Energy Economics, 29(3), 390404. Cunado, J. and F. Perez de Gracia (2003) Do Oil Price Shocks Matter? Evidence for Some European Countries, Energy Economics, 25(2), 137-154. Friedman, M., 1988. Money and the Stock Market. Journal of Political Economy, 96 (2), 221245 Greene, W. (2003) Econometric Analysis. Upper Saddle River, New Jersey: Prentice Hall. Hamilton, J. D. (1996) This is What Happened to the Oil PriceMacroeconomy Relationship, Journal of Monetary Economics, 38(2), 215220. Hooker, M. (1996) What Happened to the Oil PriceMacroeconomy Relationship? Journal of Monetary Economics, 38(2), 195213. Huang, B.-N., M. J. Hwang, and H.-P. Peng (2005) The Asymmetry of the Impact of Oil Price Shocks on Economic Activities: An Application of the Multivariate Threshold Model Energy Economics, 27(3), 455-476. Huntington, H. G. (2004) Shares, Gaps and the Economy's Response to Oil Disruptions, Energy Economics, 26(3), 415-424. Jimenez-Rodriguez, R. and M. Sanchez (2005) Oil Price Shocks and Real GDP Growth: Empirical Evidence for Some OECD Countries, Applied Economics, 37(2), 201-228. Kilian, L. (2005) The Effects of Exogenous Oil Supply Shocks on Output and Inflation: Evidence from the G7 Countries, CEPR Discussion Papers: 5404. Mishkin, F. S. (1995) Symposium on the Monetary Transmission Mechanism, Journal of Economic Perspectives, 9(4), 3-10. Newey, W. K. and K. D. West (1987) A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix, Econometrica, 55(3), 703-708. Romer, D. (2000) Keynesian Macroeconomics without the LM Curve, Journal of Economic Perspectives, 14(2), 149-169. Romer, D. (2006) Advanced Macroeconomics. 3rd editon, Boston: McGraw-Hill/Irwin. Taylor, J. B. (1993) Discretion versus Policy Rules in Practice, Carnegie-Rochester Conference Series on Public Policy, 59(0), 195-214. Taylor, J. B. (1999) Monetary Policy Rules: Introduction, In Monetary Policy Rules, ed. John B. Taylor, pp. 1-14, NBER Conference Report series. Chicago and London: University of Chicago Press. Taylor, J. B. (2000) Reassessing Discretionary Fiscal Policy, Journal of Economic Perspectives, 14(3), 21-36. Taylor, J. B. (2001) The role of the Exchange Rate in Monetary-Policy Rules, American Economic Review, 91(2), Papers and Proceedings of the 113th Annual Meeting of the American Economic Association, May, 263-267.
The Regulation of the Credit Card Market in Turkey

Ahmet Faruk Aysan Boazii University, Department of Economics 34342 Bebek, Istanbul, Turkey E-mail: ahmet.aysan@boun.edu.tr Tel: 90-212-359 76 39; Fax:90-212-287 24 53 Lerzan Yldz Boazii University, Department of Economics 34342 Bebek, Istanbul, Turkey Tel: 90-212-359 76 39; Fax:90-212-287 24 53 Abstract The rapid growth in Turkish credit card market brought together new issues. Card holders and consumer unions complain about the high interest rates, economists complain about the default rates and banks complain about the amnesties. After all of these complaints coinciding with the accelerating suicide incidences due to credit card debts, regulation has been enacted in the credit card market in Turkey. In 2003, credit cards had been taken into the scope of the Consumer Protection Law. This was the first legal arrangement on the credit cards. However it was not satisfying. It was criticized for bringing out temporary solutions. In 2005, a more comprehensive credit card law came into effect. With this regulation, Central Bank of Turkey has put a ceiling on the credit card interest rates and clarified some issues that were left untouched. In this paper; reasons, advantages, disadvantages of this regulation are discussed along with a quick glance on the development of credit card market in Turkey. The regulation and amnesties in 2003 and 2005 are examined and their effects are exposed from the point of view of parties involved: banks, customers and government. Keywords: Credit Cards, Regulation, Amnesty, Interest Rates, Default, Banks, Consumer Unions Jel Classifications Codes: L500, L510, G280, G380
1. Introduction
Three years...Thirty four cases and forty one deaths...1 These numbers do not show the consequences of a tragic car accident or a disaster. They show the number of people who committed suicide or killed their acquaintances because of their defaulted credit card debt between 2003 and 2006 in Turkey. These alarming statistics subsequently drew all the attention in the public and speeded up the process in the formation of the regulation in the credit card market in Turkey. Despite the belated regulation for the credit cards, Turkey had encountered the first credit card in 1968 which did not have a feature in crediting. Later, big banks started to introduce their credit cards in 1988. Since then the number of
1
These statistics are taken from the web site of Consumers Union, for further information see http://www.tuketiciler.org
142
credit cards outstanding increases day by day and credit cards started to replace cash in consumers pockets. Credit cards are plastic cards bearing an account number assigned to a cardholder with a credit limit that can be used to purchase goods and services and interest is charged on the outstanding balance. With this definition credit cards may seem to be a good and harmless alternative for payment. But in fact what makes credit cards so popular in Turkey is not its ease of use or availability but the complaints about it in recent years. Card holders complain about the high interest rates that the banks charge on unpaid credit card debts. Consumer Unions have started campaigns to intensify pressure on the government to lower the interest rates. After all these campaigns and complaints, together with the extent of credit card volume and its spread of usage especially after the accelerating suicide cases due to credit card debt, Turkish government started to regulate the credit card market. With the new regulation, the Central Bank of Turkey puts a ceiling on the credit card interest rates every three months and the banks are not allowed to charge beyond this rate. Card holders complain that the interest rates are still high. Banks complain that they are harmed financially and economists argue that the regulation is contrary to competition. As can be understood from this brief summary about the reactions of various parties, nobody is satisfied with what regulation has brought. However, above all these arguments, there is an interesting fact that Turkey is one of the fastest growing credit card markets in Europe despite these complaints. Turkey is ranked the third with respect to the number of credit cards in Europe and it is the tenth in the credit card usage. In the early 1990s, the number of credit cards was just 556,000. It became 13,408,477 in 2000, and then 13,996,806 in 2001, 15,705,370 in 2002, 19,863,167 in 2003, 26,681,128 in 2004, 29 million in 2005 and now it has reached 31,289,000.2 As the number of card holders increases, the number of problematic cards increases in proportion. The total number of people in the black list has become 481,261 by July 2006.
Table 1:
Years 2000 2001 2002 2003 2004 2005
Source :
Black List
No of cards 13,408,477 13,996,806 15,705,370 19,863,167 26,681,128 29,050,403 Defaulted card no 17,266 111,072 41,586 43,480 77,972 198,373 Late payment no 788 9,490 7,567 19,048 43,847 35,580 Default ratio(%) 0.13 0.86 0.31 0.31 0.46 0.80 Debt per person($) 379 260 323 425 649 772 Unpaid debt per person($) 7,173 1,461 3,519 3,290 3,404 6,002
Data from Milliyet newspaper
This electronic payment tool may even become the reason for depression, divorce or even the death of its holder. Hence, in order to prevent such undesirable outcomes and with the pressure of Consumer Unions, Turkish government put forward amnesties in 2003 and 2005. The amnesties in 2003 and 2005 aim at helping the card holders to pay out their debt to banks. In order to do that, a lower interest rate was announced and the maturity date was extended. Also in 2003, an amnesty to clear the records of holders who had defaulted on their credit card debt until that time was enacted. The only condition to get use of that amnesty was to pay back the defaulted debt before the amnesty or in three months following the amnesty. While the government is trying to find solutions to credit card problems, the actions taken with their disclosed intention are not appreciated by some parties. Particularly, banks are the first ones complaining about these amnesties because they earn profit from the credit card holders. Hence, the governments ceiling on the interest rates and the amnesties are to the disadvantage of the banks. The paper is organized as follows. In the next section, we give a brief overview of the development of credit card market in Turkey. The Section 3 explores the regulation and the amnesties in 2003 and 2005. In this section, characteristics, effects, advantages and disadvantages of the
2
The number of credit cards with respect to years has been taken from the Interbank Card Center
143
regulations and the amnesties are exposed from the point of view of parties involved: banks, customers and government. The last section is relegated for the conclusions.
2. The Development of Credit Card Market in Turkey

Credit cards have become one of the indispensable tools of daily lives in Turkey. At first, the number of credit cards did not even exceed 10,000 but nowadays it has reached a number of 30 million. Turkish consumers skipped the personal check-account stage altogether and moved directly from cash to plastic in the mid-1990s. Turkey first met with the credit cards in 1968 with Diners Club. Diners Club cards were charge cards that required to pay the whole balance as of final payment date. The card limit was assigned as the blocked cash amount. It was perceived as a prestige element, so it was just given to the rich and creditable customers. Because of that reason, the number of Diners Club holders was very few. It did not exceed 10,000 until 1975. It could only be used in Turkey. Instead of todays POS machines a mechanical device called Imprinter was used. The number of shops that had the Imprinter devices was very low. Hence the usage of Diners Club Card was very limited. The technology was old comparing to todays technology that provision was taken by telephone. After Diners Club, Turkish Express Aviation and Tourism entered the market with American Express Cards. These two cards continued their business unrivalled until 1975. Starting from 1980, the credit cards that belong to Visa, Master Card branded banks started to enter into the market. In the beginning of 1980s, Turkish population was one of the largest in Europe, its investments were low and it was developing. Hence its credit card market carried a high potential. Then, Turkish banks started to distribute credit cards to their customers. In that period different products like gold, classic cards started to take place. In 1984, Visa opened an office in Turkey and that accelerated the development of Turkish credit card market. In 1990, Interbank Card Center (ICC) was established with the association of thirteen private and public banks. The purpose of establishing ICC was to execute authorization procedures between different banks credit cards and also to execute interchange procedures. Also in 1990, the application of bonus points started to take place. This application is still popular and consumers get points as they spend with their credit cards and these points can be spent in shopping ahead. In 1993, first electronic POS terminal was used and in that year, EuroPay/MasterCard opened its first office in Turkey which increased the credit card numbers. In 1999, for the first time Turkey witnessed the payment installments in credit cards. Credit card usage turnover has a portion of 25 percent of GDP in Turkey. In the first six months of 2006, consumers spent 50 billion 171 million YTL with their credit cards. As of July 2006, there are 31,289,000 credit cards and according to ICCs estimation each customer has 1.8 cards. Hence, there are approximately 17,382,778 credit card holders. In Turkey, among 20 million salaried employees, there are 11-12 million credit card holders which generate approximately 20 million credit cards3. Whereas in England, 32 million employees have 59 million credit cards. Comparing to developed countries population-credit card number in the market ratio, it can be concluded that number of credit cards in Turkey are still to increase. This situation shows that Turkish credit card market carries a high potential. Turkish credit card market consists of 22 banks currently. These 22 banks compete to increase their market share by launching different credit cards for various needs of different customers. For that reason, the market is comprised of various types of credit cards that emphasize their installment advantages or bonus points or low interest rates, etc. The credit card market is growing day by day and card issuers are trying to get higher shares from this growth. Hence they are using all of their financial and marketing power to attract new customers. In fact the situation was not alike in the late 1980s and in the beginning of 1990s. In that period, what was important for a card holder was just the usage area of the cards. The important criteria for choosing a credit card was whether it could be used in many stores. But today the situation has changed. Now almost every store has the POS machines which enable the customers to shop with whatever card they want. POS machine, which
3
From an interview of Prof. Dr. Ahmet Ulusoy, economy professor in Karadeniz Technical University
144
stands for Point of Sale, is a machine that is used in the shopping done by credit cards and it sends the credit card information to the bank through telephone line. In Turkey, first POS system was launched by IBM in 1973 and in 1991 first POS terminal started to be used. The number of POS machines has reached an amount of 1,105,482 as of end of year 2005; this number shows a 800,000 increase in five years. But recently, the banks started to act collectively about POS machines and in order to reduce the operational costs some banks agreed to serve from the joint POS machines.
Table 3 :
POS machine 2000 299,636 2001 364,636 2002 495,718 2003 662,429 2004 912,118 2005 1,105,482 2006(1st quarter) 1,165,590
Source : Obtained from the web site of Banking Regulation and Supervision Agency
Table 2:
1 2 3 4 5 6 7 8 9 10
Source :
How did "TOP TEN" change in credit card market in 9 years? (based on credit card numbers)
1997 Yap Kredi Pamukbank Bank Vakfbank Akbank Garanti Ziraat Bank Kobank Dbank Denizbank 2000 Yap Kredi Bank Garanti Vakfbank Pamukbank Ziraat Bank Akbank Citibank Finansbank Kobank 2003 Yap Kredi Garanti Bank Akbank Vakfbank HSBC Finansbank Pamukbank Dbank Denizbank 2005 Yap Kredi Garanti Bank Akbank Finansbank Vakfbank HSBC Denizbank Halkbank Ziraat Bank
Capital Monthly Economy and Business Journal 4
Since 1980s, there have been banks that went out of the market or entered the credit card market. The credit card league has not much changed considering the leader and its followers in recent years (See Table 2). Yap Kredi Bank has been the market leader with its 17 percent market share in the number of cards and 25 percent market share in the card volume5. The banks known World Card has also featured versions (like Worlduniversity, adakart, Kontroll Worldcard, etc.) in order to fulfill their customers different needs. Garanti Bank which is a close follower of Yap Kredi Bank released Turkeys first credit card with chip in 2000 that is called Bonus Card. As of September 2005, Garanti Bank has 15 percent market share in the number of cards and 21.05 percent market share in the card volume6. The third important player in this competitive market is bank. banks Maximum Card has 11 percent market share in the number of credit cards and 15 percent market share in the card volume7.
3. The Regulations & The Amnesties in 2003 and 2005

The number of credit cards has increased 275 percent between 1998 and 2004. This rapid increase brought about the complaints and dissatisfaction in return. Most of the complaints are about the usury. Some card holders get into difficulties to pay their card debts and when they cannot pay, their debt increases rapidly with high interest rates. Hence some finally default. Especially after the economic crisis in 2000 and 2001, the number of defaults skyrocketed. The number of people who defaulted on their debt in 2001 reached a peak of 111,072. However recently, even though there is no crisis default numbers still increase rapidly. The main reason for that is banks wrong marketing tactics. In order to get more market share and increase their profits, banks attempt to increase their
4 5 6 7
http://www.capital.com.tr/haber.aspx?HBR_KOD=3331 Taken from the web site of Yap Kredi Bank, http://www.ykb.com Taken from the web site of Garanti Bank, http://www.garanti.com.tr Taken from the web site of bank, http://www.isbank.com.tr
145
credit card numbers. However, marketing competition in the credit card market has become so intense that banks give credit cards to everyone without looking at their credit history; they even distribute cards on the streets or send cards to the addresses without request. Hence, this situation causes the defaulted credit card numbers to go up. Chakravorti (2003) categorizes the consumers under two groups according to their usage of credit cards. The convenience users pay the credit card bill on due date and the revolvers use the credit feature of the cards. The convenience users are not as profitable as the revolvers for the issuers given that they just use the credit cards as the payment instruments. In Turkish credit card market, 76 percent of the cardholders pay their bill on due date8. Hence, a revolver subsidizes the cost of three convenience users. Extremely high interest rates in Turkish credit card market are likely to stem from these consumer characteristics. Since this low share of the revolving customers are also separated as illiquid and higher risk customers, the interest rate charged to the customers appear to be an increasing function of risk. Indeed, this mechanism generates a vicious circle. High interest rates lead to fewer revolvers and more convenience users. High share of convenience users in turn increases costs and thereby leads to high card rates (Aysan and Muslim, 2005). As more people default on their debt, Consumer Unions react more aggressively and they blame the government for not taking any action. Consumer Unions organize campaigns and gather signatures to protest the banks. On the other side, banks claim that they get harmed from the regulation. In fact their claim is true because defaulted debts are the parts where banks get the most profit from. As interests bear on unpaid debts, banks gain large amounts in short period. Hence putting a ceiling on the interest rates hurts banks profitability. Also some economists argue that regulating the market and putting a ceiling on the interest rates are contrary to competition. They comment that at that point consumers should be conscious and consumers should choose the bank with lower interest rate, so that other banks will also lower their rates to compete. Ausubel (1991) categorizes existing explanations for this failure of competition under three clusters: consumer irrationality, search costs and switch costs. In addition to these explanations, Ausubel provides his own explanation as well. The consumers who are not willing to borrow at the beginning are insensitive to changes in the interest rates. However, the consumers who plan to borrow are very sensitive to the changes in the credit card interest rates. Hence, when an issuer decreases the credit card rates, the issuer can only attract these risky consumers. Moreover, the consumers searching for a lower rate would also be the ones who carry out more interest payment. The low risk customers pay the bill on due date. Hence, the return from searching a lower rate would be higher than the search cost. Since the benefit from searching is higher for consumers with high balances, a credit card issuer by offering a lower rate will again only attract high risk consumer (Ausubel, 1991). Except the consumers complaints, suicides due to credit card debt accelerated the process of regulation. In three years, forty one people killed themselves or others because of their credit card debts. As a result, after all these campaigns and complaints, with the growing credit card volume and its spread of usage and mainly with the accelerating role of credit card debt suicides, Turkish government started to regulate the credit card market. The process of regulation has started in 2003. Even though Bank cards and credit cards law draft was being prepared, this draft somehow could not progress one step ahead and become a law until 2005. However with the 4822 numbered law, which was legalized in 2003, credit cards had been taken into the scope of the 4077 numbered Consumer Protection Law. By adding an article that is related to credit cards -named 10/A- to the law, the first legal arrangement was made on the credit cards. The text that Consumers Union proposed was put exactly as the temporary article. With this temporary article, it was aimed to end the credit card problems by providing twelve month installments with a low interest rate for the defaulted card holders. The important question now is who was going to get benefit from this temporary law. The ones who were already taken or that were going to be taken to the court because of their unpaid card debts and the ones who did not pay their card debts even two
8
See: http://www.bkm.com.tr/images/basinodasi/06082005_dunya.jpg
146
consecutive months were going to utilize from the law. In this temporary law, credit card debts were stated to be paid in twelve equal installments with a maximum annual 50 percent interest rate to be applied on the capital. The most important part of the amnesty in 2003 is that it made installments on the principal without processing interest on it. Later on, the government would be protested for 2005 amnesty because of falling short of 2003 amnesty. The details of this topic will be argued in the following sections. There are still question marks about the benefits of amnesties. Do really people apply to configure their debt, to get benefit from the amnesty? According to Ersin zince9, the easiness in payments of credit card debts and partial amnesties do not attain the target. The number of people that was expected to benefit from 2003 amnesty was between 500,000 and 1 million, but only 150,000 people applied. In fact, banks do not get benefit from the amnesties as they may seem so. Because, among the credit card operations that they make, banks get the most of their profit from their defaulted customers. Accumulated debts make banks get much money in short period of time; hence they are against the amnesty. Ausubel (1991) in estimating the profitability of the credit card issuers uncovers abnormal returns and asymmetric power in the credit card market. The return from the regular banking operations was around 20 percent per annum. However, the credit card branches of the banks accrued 60-100 percent profits. On the other hand, the general idea among economists is not much different. Abdurrahman Yldrm10 argues that people do not apply to consolidate their debt, firstly because they do not have enough money to pay the consolidated debt and secondly and most importantly they believe that some time later another amnesty will be released. Hence, they postpone paying their debts. Because of that reason, he claims that amnesties whip defaulting on the debt and provoke the regular payers not to pay. However from the consumers unions view, the amnesties are insufficient and they claim that amnesty adjusted- interest rate is still high. Apparently, compromise seems to be far away and discussions seem to last for a long time. In 2003, the number of cards was 19,863,167 and money that was spent for shopping reached an amount of 40,3 billion YTL by increasing 15 billion YTL compared to previous year. Of this amount, 5.3 percent was financed with credit cards. In 2005, Turkish consumers financed 15.8 percent of their expenditures with credit cards. As credit card usage volume increases, default rates increase as well. But also there is a more important factor that increases default rates which is amnesty. After the temporary law was accepted, an amnesty on the bad credit records was also legislated. The record clearing amnesty was legislated in December 2003. According to that amnesty, if the legal and real people who defaulted on their debt (debt should be due before December 2003) pay or consolidate their debt before the amnesty or in three months following the legislation of amnesty, then their bad credit histories will be erased and they will not be taken into account in further banking transactions. For those who are not in the scope of amnesty; the Central Bank of Turkey removes people from the black list three years later if they pay their debts and if there is no information whether the debt was paid or not, then they are removed from the list after five years. In the end of December 2003, in which the results of record amnesty were reflected to the black list, the number in the black list decreased by 324,316 and fell to 500,486. 2003 amnesty did not go further from being a partial and temporary solution to the credit card problems. In order to regulate the market more efficiently studies continued and finally from the beginning of March 2006 on, Bank and Credit Cards Law got into effect. This new law has been the most comprehensive credit card law. There are several subjects in the law that get attention. The organizations that want to establish credit card system and make agreements with member stores are required to get permission from BRSA11. Card issuers cannot distribute credit cards without the permission or request of consumer. If the minimum payment is not paid within three months following
9 10
11
Taken from a report that Ersin zince sent to deputies. Ersin zince is the president of Turkey Banks Union Abdurrahman Yldrm is an economist in Sabah newspaper, for further information http://www.sabah.com.tr/2005/12/28/yildirim.html BRSA is responsible for the regulation and supervision of banking system in Turkey.
about
that
news
see:
147
the final payment, then the card issuer sends a notice to the card holder for his debt. If the debt is still not paid after one month of this notice, then the card issuer cancels the credit cards that it has given to the card holder and do not give a new card until the debt is paid. Before the law, consumers were complaining about credit card limits. They were stating that their credit card limits were being increased without their consent. Credit card issuers were used to increase the card limit if the holder made his payments regularly. With this new law, card issuers will not be able to increase credit card limit without the consent of the holder. A card holders total credit card limit of his all credit cards, for the first year, cannot exceed two times of his monthly net average income. For the second year, it cannot exceed four times of his monthly net average income. 2003s temporary article gave the right to card holders to object their debt summary in thirty days following the final payment day. But in 2005 law, this period has been decreased to ten days. In 2003, card issuers had to respond to consumers objections in fifteen days; but after the law in 2005 they have to respond in twenty days. This situation supports the claims of consumers union because they claim that Bank and Credit Cards Law that was legalized in 2005 is in favor of banks. According to them, this law has been published not to help consumers but to help banks. The decrease in the number of objection days to the debt summary is one of their evidence in supporting their viewpoint. With this law a new application has started and so called card insurance application was aimed to prevent unjustifiable usage of credit cards. According to that, if a card holders credit card is stolen or his credit card information is lost and if he informs relevant authorities; his responsibility from illegal usage of his credit card prior to his declaration has been limited to 150 YTL. Another feature that aims to protect consumers rights is about concealing the consumers information. Consumer information cannot be sold or shared with third parties. Of course the most important part of this regulation is about the interest rates. The Central Bank of Turkey is the only authority to determine credit card interest rates and it announces these rates every three months. Banks should not pass beyond this ceiling rate.12 As of the beginning of July 2006, the Central Bank of Turkey announced the monthly maximum conventional interest rate for YTL as 5.72 percent -it was 5.75 percent after the law was legalized in March 2006- and 2.68 percent for US $ and 2.5 percent for . The monthly maximum default interest rate was announced for YTL as 6.44 percent -it was 6.88 percent after the law was legalized in March 2006- and 2.98 percent for US $ and 3.05 percent for . There are still objections and complaints about these rates. Banks claim that they are in loss with this configuration. On the other hand, Consumer Unions claim that CB determined the interest rates by taking into account banks requests. They state that country realities and consumer demands are not taken into account and request that the ceiling that was announced as 5.72 percent to be reduced to 3 percent. On the other hand, Turkish National Assembly European Union Conformity Commission suggested old card debts to be consolidated using a rate that takes producer price index as the base, but the government and Turkey Banks Union object their suggestion. The government members state that after the amnesty in 2005, only 266,840 people consolidated their debt. 35 percent of 1.5 billion YTL card debt was just consolidated. Hence they state that after the regulation the credit card problem does not occur because of the system but because of the ones who did not get benefit from the amnesty. The president of BRSA, Tevfik Bilgin made an invitation to banks to help card holders reconfigure their debt. He stated that they were criticized for their works during the arrangement of credit cards and for acting like Consumer Unions. However he also notes that the new arrangements that have been brought together with the law required banks to make self-sacrifice. These arrangements have provided fewer advantages than the funding costs to the banks.
12
Banks maximum monthly interest rates are stated in the next page. The rates are taken from the web site of Turkish Republic Central Bank, http://www.tcmb.gov.tr
148
Table 4:

Banks' Maximum Monthly Interest Rates
YTL (as of July 2006) (in percentages) Maximum monthly conventional interest rate Maximum monthly default interest rate 5.72 6.44 5.70 6.44 5.70 6.44 5.70 6.44 5.70 6.43 5.70 6.43 5.65 6.44 5.65 6.44 5.65 6.44 5.65 6.44 5.65 6.39 5.62 6.44 5.45 6.44 5.00 5.00 4.75 6.18 4.75 4.75 4.72 6.14 4.25 5.52 3.75 4.88 2.99 3.89 2.90 3.77 2.75 3.58
Banks Citibank Kobank Tekstil BANK Yapi VE Kredi Bank Finansbank Hsbc Bank Akbank Denizbank Fortis Bank Garanti Bank ekerbank Anadolubank Trk Ekonomi Bank Tekfenbank Trkiye i Bank Trkiye Vakiflar Bank Bankeuropa Oyak Bank Turkish Bank Trkiye Halk Bank T.C.Ziraat Bank MNG Bank
Table 5:
Banks' Maximum Monthly Interest Rates

US $ (as of July 2006) (in percentages) Maximum monthly conventional interest rate Maximum monthly default interest rate 2.59 2.50 2.97 2.50 2.50 2.50 2.50 2.00 2.60 2.00 2.60 2.00 2.00 2.00 1.90 1.90 1.90 -
Banks ekerbank Hsbc bank Trk.vakiflar bank Mng bank Tekstil bank Trkiye i bank Akbank Finansbank Turkish bank Kobank Oyak bank
Table 6:
Banks' maximum monthly interest rates

Euro (as of july 2006) (in percentages) Maximum monthly conventional interest rate Maximum monthly default interest rate 2.50 2.50 2.00 2.60
Banks Trk.vakiflar bank Trkiye i bank
Bank and Credit Cards Law came along with the amnesty in 2005. This amnesty was added to the law at the last minute because the government refrained from consumers reactions. After 2003 amnesty, consumers and Consumer Unions complaints went on, number of defaulted cards increased and number of suicides because of credit card debts also continued to increase. Hence the amnesty in 2003 was criticized for bringing out temporary solutions. Hence, same problems continued and finally with the legislation on credit cards a new amnesty was released. However this new amnesty was not only criticized again for bringing out temporary solutions, but also for falling short of 2003 amnesty. The main reason for that criticism is that with the new amnesty, only the ones whose debts
149
accumulated with high interests and who were taken to the court would get benefit. Consumer Unions wanted the unpaid debts to be taken under the scope of the amnesty. However, their request was not taken into account. The card holders that defaulted until the end of January 2006 or the ones that were taken to the court because of their card debts got benefit from this amnesty. The debt was to be paid in 18 equal payments with an annual interest rate of 18 percent.
Table 7: Card Statement
958.948 1.572.000.000 YTL. 266.840 548.055.000 YTL.
No of people that had card debt Debt Amount No of people who took advantage of the law Configured Debt Amount
Source: Data was taken from Sabah newspapers 14.05.2006 edition
From 958,948 people, number of those who had card debt, only 266,840 people applied to take advantage of the amnesty in 2005 (see Table 7). The number of people who had credit card debt has reached an amount of nearly one million. One of the reasons for that increase is that some people pay their credit card debt with other credit cards. Hence, it becomes a vicious circle. Banks unreasonable card distributions without checking the consumers credit history is the main cause of this circle. According to the president of BRSA, Tevfik Bilgin 13; credit cards are perceived as the way of the continuation of lives not the way of easing the lives. Blent Deniz14 states that in a country where average civil servant salary is about 800 YTL, there are examples that a teacher has 21 credit cards and he pays 2800 YTL interest of these cards every month. The teacher admits that he is paying his card debts by borrowing from his acquaintances and when there will be nobody left to borrow from, he thinks that suicide will be only solution for him. There have been 41 deaths up to now because of card problems and this serious problem still continues as long as this vicious circle is not broken apart. According to a research named Consumer Index made by AC Nielsen in 2005; in the world, 40 percent of consumers prefer to save spare cash left over their essential needs. Europeans prefer to utilize 38 percent of spare cash after covering essential needs with buying new dresses and going to amusements. People in Turkey prefer to utilize 44 percent of spare cash after covering essential needs with credit card debts and other debts payments. The new law promises stricter standards for the evaluation of credit card applications, tighter alignment of account limits with regular income, and a reduction in the ceilings on penalties for late payments. This is good news considering Turkeys problem of card debt. However, Turkish people have had a love-hate relationship with plastic money. Chronically high inflation, though currently under control, and reciprocally high nominal and real interest rates, combined with a low GDP per capita, have proven to be disastrous for many consumers taking on more credit card debt than they can afford. However, banks are also to blame, at least in part, for an inordinate rise in the stock of debt associated with credit card transactions. They have generally maintained rather loose standards on ability to pay when issuing credit cards, peddling them through stands at malls and supermarkets, making them available to anyone able to complete an application form, and sending out pre-approved credit cards. Data released by the Interbank Card Center (ICC) and the Central Bank of Turkey (CBT) provide a good indication of the basic scenario. According to the ICC, the number of credit cards in circulation in Turkey reached 29 million as of the end of September 2005. Cumulative CBT figures up to the same point in time show that 570,000 credit cards were in default, with debts amounting to 1,17 billion YTL, the equivalent of $875million. The total volume of debt outstanding in connection with credit card usage amounted to 16 billion YTL or $12billion, according to the CBT report. However, the ICC and CBT data shows that almost 2 percent of credit cards in circulation as of the end of the third quarter of 2005 were in default, or, from a positive perspective, that slightly more than 98 percent were
13 14
Taken from a report in Sabah newspaper in 20.02.2006 Blent Deniz is the president of Consumers Union
150
in good standing. On a less positive note regardless of the perspective, the underlying monetary values indicate that the volume of debt associated with accounts in default represented 7.3 percent of the total stock of debt on credit cards. This seems to paint a more somber picture of the default rate than does the ratio indexed to the number of cards. Especially as the 7.3 percent rate prevailing in September 2005 reflects an increase of three percentage points over the corresponding figure at the end of 2004. It also represents the highest rate on record since a peak of 8.9 percent registered at the end of 2001, the year Turkey experienced a major economic downturn. While the current default rate is slightly lower than the global average, estimated to be somewhere between 8 and 10 percent, the pace that Turkey is catching up with international norms is particularly alarming. The only mitigating factor is that the problem seems to be localized, as suggested by CBT figures anchoring the entire amount in default to 570,000 cards and an assessment by the Banks Association of Turkey connecting these cards with 250,000 people. Regulation has been done to gain some degree of control over a persistent albeit isolated tendency to finance rather than earn a living, and a system providing the means to do so. Together with the disadvantages Bank and Credit Cards Law has also some deficiencies according to banks and Consumer Unions. According to the president of Bank Asia Payment Systems, Faruk Nurdoan,15 the biggest problem is about when the new arrangement will be applied. In the past, the credit card limit of a person who has 10 credit cards could reach 10-20 times of his income. However the new law gives permission to two times of income for the first year as the limit and four times of income for the second year. This new law is not clear about the way how the document about the annual income of a person will be obtained. Not the old card holders, but the ones that request for new cards will be investigated. But what if the card holder previously owns other credit cards? How its limit will be learned? These issues are still not clear with the new law. Another problem is the standardization in limits. For example, a self-employed persons tax statement sometimes can be very low or it can even show loss on it. Only the income of employees with payroll can be determined. Hence this new law may bring the problems about the documentation of income. A solution offered for that problem is the application of Turkish Republic Identity Number. If the consumers all credit cards information, both previously owned ones and the ones that will be owned in the future, are collected in a pool relating to identity numbers, supervision of credit card market will be easier. In the past banks were used to keep the limits high. But with the law, the limit will be determined according to income and it will not be increased without the consent of the card holder. This situation will increase the bureaucracy and also it will increase the banks burden. Another point in the new law is about the joint bail. The guarantor that signed in the credit card purchase was legally responsible in the past. According to new regulation, no operation will be made for the guarantor without completing the legal procedure. Faruk Nurdoan interprets this as the removal of joint bail and adds that by that way, competition among the banks will become fierce and big banks will get advantage from this regulation. The president of Consumer Association Engin Baaran16 claims that the law is in favor of credit card victims but not in general. He asks that: In order to be a credit card victim is it necessary to default, disappear or even commit suicide? The law does not include the consumers who make the minimum payments, do not miss the final payment. Hence the law is argued to be contrary to equality principle. Consumers Unions claim that the law will be modified in the future because it will not meet the needs. However, according to the general manager of Industry and Trade Ministry Consumers and Competition Authority, zcan Pekta: The law is comprehensive. I wish it was not too late then there would not be so much problems. The release of this law should make consumers pleased. Contrary to his statement, consumers complain mostly from the credit card interests. From those who applied to the Application Center, which was established in the structure of Consumers Union for helping consumers solving their problems, 37.41 percent complained about high credit card interest rates. The development of credit card market has accelerated after 2001. In 2001, because of the economic crisis that Turkey experienced, 111,072 people defaulted on their credit card debt. After that
15 16
Taken from a report in Aksiyon magazine in 06.03.2006 Taken from a report in Aksiyon magazine in 06.03.2006
151
crisis, as the governments need for debt decreased, banks inclined to increase their consumer credits. That is why credit cards are marketed everywhere, in the streets, malls, etc. According to a questionnaire made in 2003, the answers to the question of how the consumers that joined the questionnaire possessed their credit cards reveals that; 43.9 percent reported that they applied to the bank, 29.2 percent answered that they possessed elsewhere than the bank, 26.9 percent told that credit card was sent without their consent. Banks are trying to get as more individual customers as possible to get more share of this market which will grow very much in the following 5-10 years. While doing that, they are not taking into account the financial situation of their customers. Nowadays, banks make cooperation in order to obtain wide spread usage in a short period of time. With these cooperation, one banks credit card customer can get benefit from other banks credit card opportunities. Another marketing tactic that the leader of Turkish credit card market realizes is the controlled credit cards. With the controlled credit cards, the bank aims to attract the consumers who are in the black list and who cannot document their income. With this new card, no legal procedure will be done for the collection of debt. Controlled card holders just need to open an account worth of 750 YTL-15,000 YTL. The limit will be equal to 70 percent of this amount. If the card holder defaults, then the amount will be taken from the customers account. This is a good penetration tactic for banks in an environment where two of every three people do not have a credit card among adult population in Turkey.
Table 8:
1 2 3 4 5 6 7 8 9 10 11 12 13
Source:
Performance Analysis of 13 Banks in 6 Years (card numbers and their volumes)

Banks Fortisbank Denizbank Finansbank Akbank Garantibank Kobank bank Yap Kredi Vakfbank ekerbank Ziraatbank Citibank Halkbank Total 2000 39,000 97,000 261,000 723,000 1,205,000 229,000 1,504,000 2,574,000 1,219,000 159,000 747,000 399,000 1,325,000 13,338,000 2005 894,652 1,394,419 2,235,198 2,879,629 4,750,387 692,266 3,205,210 5,374,170 2,022,147 219,344 962,569 493,724 1,102,047 29,263,906 Change(%) 2194% 1337.5% 756.4% 298.3% 294.2% 202.3% 113.1% 108.8% 65.9% 38% 28.9% 23.7% -17% 119.4 Trade Amount(million YTL) 122 152 369 958 1,457 121 891 1,613 228 37 98 57 95 6,198
Capital Monthly Economy and Business Journal
The most recent news about the regulation of credit card market in Turkey is about the minimum payment. From the beginning of September 2006 on, the minimum payment limit will increase to 20 percent. In fact, it was not a surprise because this decision was stated in the Bank and Credit Cards Law. According to the law, minimum payment cannot be less than 10 percent of term debt following six months of legislation of the law; and it cannot be less than 20 percent of term debt after these six moths. In the case of partial payment of term debt, interest will be calculated on the remaining account balance. To the remaining account balance, if the payment is made above the minimum payment, conventional interest will be applied and if the payment is made below the minimum payment, default interest will be applied. It is now expected to bring more financial straits. As Tevfik Bilgin, the president of BRSA, stated that nowhere in the world has such an application called minimum payment in the credit cards17. He told that during the preparation studies of the law, BRSA offered to prohibit minimum payment application, but their offer was not taken into account. Consumer Unions argue that as long as the minimum payment application is not prohibited, the number of credit card victims will continue to increase. After this application, what will happen to
17
http://www.sabah.com.tr/2006/08/28/eko124.html
152
700,000 people who can just pay the minimum payment amount? This is the most recent question Consumer Unions are asking to the government and banks. Only a few banks heard of their calls and provided some opportunities for paying out their debt in fixed installments. According to Vicepresident of Consumer Rights Center, Faruk Haner,18 the biggest problem for credit card holders is the minimum payment amount. He claims that if a legal arrangement is made in order to abolish this application, then the number of credit card victims will be decreased by 80 percent. It can be understood from that statement is that credit card holders pay the minimum payment amount but do not pay the remaining balance. If the minimum payment rate is increased from 10 percent to 20 percent, then the number of those who pay the minimum payment amount will decrease sharply. The credit card usage prevents the informal economy, but on the other hand it sometimes causes illegal operations. Are usurers working with POS machines? This is the question that is being asked recently. According to Consumer Unions, usurers have started to settle credit card holders debts through POS machines that they took from the banks in the name of companies that they established. They are using these POS machines in order to collect the debt that they have lent to consumers. The consumers who have difficulties in paying out credit card debt have also difficulties in borrowing consumer credits because of banks deliberation in order to avoid bad loans. Hence the consumers who cannot take consumer credit borrow money from the unofficial usurers in order to pay their credit card debts who lend money with interest illegally. Usurers collect the debt by making installments from the consumers same credit card that they have closed out the debt. While determining the interest, usurers also take the maturity date into account. They apply a monthly interest rate of 2.5 percent. For example, a person who wants to close out his 10,000 YTL credit card debt in 12 months pays around 13,500 YTL to the usurer. As the maturity date extends, the payment amount increases. If the same amount of money is wanted to be withdrawn from the bank through the credit card, by taking into account monthly average 5.15 percent interest rate, 18,300 YTL is paid in 12 months. Hence, the banks collect almost 5,000 YTL more from the consumer compared to the usurers.19 According to Consumer Unions minimum payment application should be removed totally. By that way the consumer should be directed to learn expending considering his earnings when the day of payment comes. Because minimum payment does not go further from causing debt increase with accumulated interest. Beside, banks and BRSA should investigate better when giving POS machines to the companies. Turkey first met with credit cards that make installments in 1999. From that time on, Turkish people loved making installments for their expenditures in shopping. According to a questionnaire that was done about the credit cards, from the participant card holders, 63 percent told that the most important advantage that a credit card can offer is the option to increase the installment numbers.20 Making installments for the expenditures using credit cards is a widespread application. In European countries installments are made with credit cards as well, but the maximum installment amount is 6, whereas in Turkey there are installments up to 24 months. In the regulation of credit card market, there is not even a word about the installment application. However one of the most discussed topics of this regulation is installments to credit cards. The credit amount that is obtained from the shopping done through using installments in credit cards is equal to an amount of 3.8 million YTL21. Even the money of fast consumed goods like bread, cheese was paid in installments. This situation was bothering the banks because consumers were paying the money of the consumed goods even 6 or 12 months later. Finally in February 2005, in the leadership of Banks Union, banks made an agreement among themselves and agreed to stop making installments in fast consumed goods and fuel oil. In fact, this situation was contrary to competition. The banks who joined the agreement could even get a punishment for the violation of competition law. Nonetheless as of April 2005, with the request of BRSA, the application of installments in the fast consumed goods was abolished legally. The reason of
18 19 20 21
Taken from a report in the web site of Haber Kenti, http://www.haberkenti.com.tr Taken from the web site of http://www.haber3.com/haber.php?haber_id=148199 Results of MasterCards MasterIndex Research as of July 2006 http://www.milliyet.com/2005/03/05/yazar/tamer.html
153
BRSA for their request is that shopping in installments is expending unearned money. People expend more if their expenditure is made in installments. As they expend more, the probability to default increases. As default number increases, credit cards continue to be a big problem. The card holders are frequently warned about the possible adverse consequences of extreme and unconscious usage of credit cards. It is even offered to put warning notes on the credit cards to prevent the increase in the number of defaulted credit cards. Most of the defaulted card holders have small amount of credit card limits. 75 percent of those who are in the legal pursuit are the ones that have credit card limits of 0-500 YTL. In Turkey, for those who do not pay their credit card debt for 120 days, legal pursuit is started. Among those who have card limits of 0-250 YTL 433,000 people could not pay their debt back. There are 387,000 people who cannot pay their debt back and whose card limit is between 250-500 YTL. Hence, 820,000 people of whom credit card limit are between 0-500 YTL were not able to pay their card debt. The authorities claim that after the regulation in credit card market, card holders should become more conscious and should not undertake more debt than they can afford. Despite the fact, if the credit card problems continue and default numbers go on to increase, then card holders are to blame not the system. Then the question comes to minds: Will this situation last forever? Everybody answers this question from a different point of view, some authorities claim that the problem will never end. On the other hand some others claim that the problem will be solved soon as the results of the regulation come into effect. Contrary to an expenditure amount of 11,9 billion $, a default amount of 875 million $ has occurred and the default rate has increased to 7.4 percent as of July 2006. This rate even passed the maximum level in 2000 which was 5.5 percent. The number of unpaid cards in the first seven months of 2005 was 176,215. The average monthly number of people who had troubles in paying out their credit card debts was 5,640, but this number has increased to 13,985 in the first seven months of this year. Hence, every month nearly 14,000 people get into trouble in paying out their card debts. At first, this seems to be the problem of defaulters and the banks only. But if we think the situation from a macroeconomic perspective, we see that this problem also means the cut off of 14,000 peoples expenditures. These 14,000 people will stop buying most of the things they buy in order to pay their debts. Hence, the credit card problems also affect tradesmen and cause stagnancy in their business. This stagnancy will last until the card debts are paid. In fact, the beginning year of the problem is 2004. In 2004, growth reached a level of 9.9 percent and consumer demand was high, habit of making installments in shopping using credit cards was very widespread. Now, both the expenditure side and payments system side are trying to digest the effects of the credit card debts occurred as a consequence of this extensive and unconscious usage. System will be able to clean up itself probably in 2-3 years. The return of credit cards will make new expenditures possible starting from 2007. Although this process gives an advantage to control the inflation, it also causes some disadvantages in the growth in the domestic market. Hence credit cards bring many disadvantages together with its advantages. There are many issues that are being talked about credit cards; banks and consumers; both sides have important roles in the process of gathering solutions for credit card problems. However the first and the most important issue is that consumers should be conscious while making expenditures with their credit cards.
4. Conclusion
Turkish credit card market is a growing market that also carries a high potential. In this growing market, the competition is so fierce that banks try every possible tools to get more market share. Distributing cards in the streets, malls, etc is one way of this. But this extensive and unconscious distribution puts the credit card market into trouble. Card holders and Consumer Unions complain about the high interest rates, economists complain about the default rates and banks complain about the amnesties. After all of these complaints together with the accelerating role of credit card debt suicides, regulation has been made in the credit card market. With this regulation, Central Bank of Turkey has
154
put a ceiling on the credit card interest rates and clarified some points that were complained about. In this paper; reasons, advantages, disadvantages of this regulation are discussed. The regulation and amnesties in 2003 and 2005 are examined and the effects of them are presented from the point of view of both sides. Banks claim that they are hurt from the amnesties, consumers cry out about the interest rates. None of the sides seem to be pleased with the regulations or with the amnesties. This situation is also obvious from the results of the number of people who got benefit from the amnesties. Few people apply to consolidate their defaulted debt. Hence it is argued that no further amnesty is needed since it does not reach the target. With the Bank and Credit Cards Law, Central Bank of Turkey announces the credit card interest rates every three months and banks rates should be below this ceiling rate. This paper compares the default numbers before and after the regulation and then looks at the effect of the regulation from this perspective.
References
[1] [2] [3] [4] [5] [6] [7] Aysan, Ahmet F. and Muslim, Nusret A. The Failure of Competition in the Credit Card Market in Turkey: The New Emprical Evidence, 2005 Ausubel, Lawrence M. The Failure of Competition in the Credit Card Market. American Economic Review, March 1991, pp.50-81 Ayadi, Felix O. Adverse Selection, Search Costs and Sticky Credit Card Rates. Financial Services Review, 1997,Vol 6/1, pp.53-67 Stakelbeck, Frederick W. The Role of Credit Cards in an Increasingly Indebted World Economy. Retrieved from the World Wide Web: http://www.philadelphiafed.org/src/srcinsights/srcinsights/q1si3_05.html Gans, Joshua S. and King, Stephen P. A Theoretical Analysis of Credit Card Regulation. Economic Record, Vol.79, No.247, December 2003, pp.462-472. Muris, Timothy J., "Payment Card Regulation and the (Mis)Application of the Economics of Two-Sided Markets". Columbia Business Law Review, Vol. 2005, No. 3, pp. 515-550, 2006 Retrieved from the World Wide Web: http://ssrn.com/abstract=901649 Chakravorti, Sujit and To, Ted Why Do Merchants Accept Credit Card? Federal Reserve Bank of Chicago Research Paper, September 1997
Non-Linear Dynamics and Chaos: "The Case of the Price Indicator at the Athens Stock Exchange"
Mike P. Hanias TEI of Chalkis, Greece E-mail: mhanias@teihal.gr Panayiotis G. Curtis TEI of Chalkis, Greece E-mail: pcurtis@teihal.gr John E. Thalassinos Universitiy of Piraeus Greece E-mail: thalassinos@altecnet.gr Abstract Time series analysis using daily data is used to analyze the Athens Stock Exchange Indica-tor by the method proposed by Grassberger and Procaccia. The correlation and minimum embedding dimension and mmin respectively were calculated. Also the corresponding Kol-mogorov entropy was calculated. With these parameters as inputs and using the chaotic dy-namic multistep ahead prediction achieved. The result is that the stock exchange system is governed by a low dimensional strange attractor. Keywords: Time series analysis, stock exchange. Kolmogorov entropy, chaos theory.
1. Introduction
Fractal behaviour and long-range dependence have been observed in many phenomena, basically in the field of fluctuations in physical systems such as diffusion, financial time series, tele-traffic etc. In this paper, we characterize the dynamics of Athens stock exchange indicator. We have applied the method proposed by Grassberger and Procaccia to evaluate the invariant parameters of the system such as the correlation and minimum embedding dimensions, and the Kolmogorov entropy. We have applied the results from the non linear analysis to predict the corresponding time series.
2. Athens Stock Exchange Indicator Time Series

The Athens stock exchange indicator is presented as a signal x=x(t) as it shown at fig 1. It covers data from 1998 to 2005. The sampling rate was t=1 day.
156

Figure 1: Time Series of Athens Stock Index.
7000 6000
5000
S tock Inde x
4000 3000
2000 1000 0 500 1000 1500 2000
Time units (Days)
3. Non Linear Signal Analysis

In order to evaluate the afore mentioned time series we have used the method proposed by Grassberger and Procaccia [1, 2] and successfully applied in similar cases [3-5]. According to Takens theory [6] the measured time series was used to reconstruct the original phase space. For this purpose we calculated the correlation integral, for the recorded signal, defined by the following relation [7], for lim r 0 and N , N 1 (1) C m (r ) = H r Xi X J N pairs i =1,
j =i + w
Where W N pairs
is the number of points, is the Theiler window is the Heaviside function, 2 = ( N m + 1)( N m + W + 1) ,
m is the embedding dimension The summation counts the number of pairs ( X i , X j ) for which the distance, (Euclidean norm), X i X J is less than r, in an m dimensional Euclidean space. In the above equation N is the number of the experimental points (stock index data), here N=2047, X i is a vector in the m dimensional phase space given by the following relation [6] (2) X i = {xi,xi-,xi-2,..xi+(m-1)} The vector X i = {xi,xi-,xi-2,..xi+(m-1)}, represents a point to the m dimensional phase space in which the attractor is embedded each time, where is the time delay =it determined by the first minimum of the mutual information. In our case =50 time steps as shown at fig - 2.

Figure 2: Mutual Information i vs time delay .
4.0 3.5 3.0 2.5
157
I
2.0 1.5 1.0 0.5 0 50 100 150 200 250 300
We used this value =50 as an optimum delay time. The parameter W is the Theiler window. As Theiler pointed out if temporally correlated points are not neglected, spuriously low dimension estimate may be obtained [7, 9]. Not concrete rule exists for choosing W although suggestions are given by Kantz and Schreiber [7]. According to Kantz and Schreiber [7] we took the W as the first crossing zero value of correlation function ACF() in other words we use the correlation length as starting value of W [9]. We calculate the correlation length c=325 so the W= c=325 time lags as shown at fig - 3.
Figure 3: Autocorrelation function vs time delay
1.0
0.5
AC()
0.0
-0.5
-1.0 0 100 200 300 400 500
From the same figure we see that there is a weak correlation between the data so the past states affect the present state of the system. We used these values for the reconstruction of phase space. With (1) dividing this space into hypercubes with a linear dimension r we count all points with mutual distance less than r. It has been proven [7-8] that if our attractor is a strange one, the correlation integral is promotional to r where v is a measure of the dimen-sion of the attractor, called the correlation dimension. The correlation integral C(r), by defini-tion is the limit of correlation sum of eq-
158
1, has been numerically calculated as a function of r from formula (1), for embedding dimensions m=110 and shown in fig 4. In fig 5 the slopes v of the lower linear parts of these double logarithmic curves give information characterizing the attractor.
Figure 4: The correlation integral, logCm(r) vs logr, for different embedding dimensions m
log(C(r))
0.1
10
1 00
1000
log(r)
Figure 5: The corresponding slopes and scaling region of Fig.4
15
d(logC(r))/dr)
10
0 10 100 1000
logr

Figure 6: The Correlation dimension v for different embedding dimension m
10
159
4 2 0 0 2 4 6 8 10
In fig - 6 the corresponding average slopes v are given as a function of the embedding dimension m It is obvious from these curves that v tends to saturate, for higher ms at non in-teger value v=2.09. With this value of v the minimum embedding dimension could be mmin=3 [7]. So the minimum embedding dimension of the attractor for one to one embedding is 3. In order to get more precise measurements of the strength of the chaos present in the time series we have introduce the Kolmogorov entropy. According to [7] the method described above also gives an estimate of the Kolmogorov entropy, i.e. the correlation integral C (r) scales with the embedding dimension m according to the following relation (3) Where K2 is a lower bound to the Kolomogorov entropy. From the plateau of fig 7 we estimate K2=0.1 bit/s
Figure 7: The Kolmogorov entropy vs embedding
C (r ) ~ e mK2
K2 (bit/s)
0 10 100 1000
logr
160
4. Time Series Prediction

Figure 8: Actual and predicted time series for k=1, (a), 5 (b), 15 (c), 30 (d) time steps ahead.
7000
Actual va lues predicted values wi th step =1, day ahead
7000
6000
6000
5000
Actual val ues Predicted values wi th step =5, days ahead
Stock index
5000
4000
Stock i ndex
0 50 0 1 00 0 15 00 20 00
4000 3000
3000
2000
2000 1000 0 5 00 100 0 15 00 200 0
1000
T im e u nits (Days)
Time units (Days)
(a)
7 000
(b)
6 000
5 000
Actula values Predicted values with step =15,days ahead

St ock in dex
7 00 0
6 00 0
Act ual value s Predi cted val ues with step = 30,days a head
Stock index
5 00 0
4 000
4 00 0 3 00 0
3 000
2 000
2 00 0
1 000 0 5 00 10 00 15 0 0 20 00
1 00 0 0 5 00 1 00 0 1 500 2 0 00
Ti me uni ts (Days)
Ti me un its (Days)
(c)
(d)
This simple prediction model works locally on the reconstructed deterministic data in the state space. It is described fully in [7] and [10, 11] from the signal of equation (2); we can construct an mdimensional signal {xim } = {( xi , x i + , x i + 2 ,...., xi + ( m 1) )} R m The reconstructed m-dimensional signal projected into the state space can exhibit a range of trajectories, some of which have structures or patterns that can be used for system prediction and modeling. Essentially, in order to predict k steps into the future from the last m-dimensional vector m m point {x N } , we have to find all the nearest neighbors {x NN } in the -neighbourhood of this point. To be
m m more specific, let B ( x N ) be the set of points within of {x N } (i.e. the -ball). Thus any point in m m m B ( x N ) is closer to the {x N } than . All these points {x NN } come from the previous trajectories of the m system and hence we can follow their evolution k-steps into the future {x NN + k } . The final prediction for m the point {x N } is obtained by averaging over all neighbours projections k-steps into the future. The algorithm can be written as

{x m +k } = N 1 B ( x
m NN
161
m NN + k m m X NN B ( x NN )
(4)
where
m denotes the number of nearest neighbours in the neighbourhood of the point {x N } [7]. As an example we suppose that we want t predict k=2 steps ahead .The next step in the algom rithm is to check that the projections, one and two steps into the past, of the points in {x NN } are also
B ( x
m NN
m m nearest neighbours of the two previous readings {x N 1 } and {x N 2 } respec-tively. This criterion m excludes unrelated trajectories that enter and leave the -neighbourhood of {x N } but do not track m m back to - neighbourhoods of {x N 1 } and {x N 2 } , thus making them unsuitable for prediction Assuming that any nearest neighbours have been found and checked using the criterion dem tailed previously, we project their trajectories into the future and average them to get results for {x N +1 } m and {x N + 2 } . We used the values of and m from our previous analysis so the appropriate time delay was chosen to be =50. The embedding dimension m = 3 follows from the Takens embedding theorem and was used for the state space reconstruction. The number of nearest neighbours was found to be 100 for optimum results. Also we exclude 325 points according to Theiler window. We apply our predictor with continually increasing number of prediction steps ahead from k=1 to k=30 was performed. Actual and predicted time series for k=1,5,15,30 time steps ahead are presented at Fig 8 (a),(b),(c),(d) respectively.The prediction error for establishing the quality of the fit was chosen to be the classical root mean square error (RMSE) .Fig.9 shows a growing error with increasing prediction horizon from k=1 to k=15 steps ahead. This fits very well with a property of nonlinear dynamical systems, namely sensitive dependence on initial conditions. The more steps in the future we want to predict, the greater (indeed exponentially greater) the prediction error will be.
Figure 9: Root Mean squared error of our predictor
0.30
0.25
0.20
RMSE
0.15
0.10
0.05
10
12
14
16
K time st eps (days) ahead
5. Discussion
Applying the methods of non linear analysis in these time series we found that there is evi-dence of deterministic chaos in these data with a strange attractor with correlation dimension v=2.09 This is also evident from the multistep s ahead prediction with the use of the correspondence strange attractor invariants as input parameters.
162
We can support that the data set is not simply stochastic, because of the saturation at m plot thus, there is a deterministic component of dynamics which is must examine further.
6. Conclusion
In this preliminary analysis, evidence of nonlinearity was discovered in these data, and the analysis presented here examines this question further. This study poses that nonlinear events and a strange attractor governs the econometric system although this hypothesis would have to be investigated further [12, 14]. The appearance of nonlinear structures is im-portant to the question of multi-step prediction. The discovery of non-linear structures then might be a sign that developing a predictive model is promising.
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] P. Grassberger and I. Procaccia, Phys. Rev Lett., 50,346-349 (1983) P. Grassberger and I. Procaccia, Physica D 9,189 (1983) M.P. Hanias,J.A. Kalomiros,Ch. Karakotsou,A.N. Anagnostopoulos, J. Spyridelis, Phys. Rev B. 49,16994 (1994) M.P. Hanias, G. Giannaris, A.Spyridakis,A.Rigas, Chaos Solitons and Fractals, v.27, #2, 569573, 2006 M.P. Hanias,J.A.N. Anagnostopoulos, Phys. Rev B. 47,4261 (1993) F. Takens, Lecture Notes in Mathematics 898 (1981) H. Kantz and T.Schreiber, Nonlinear Time Series Analysis, Cambridge University Press, Cambridge (1997) T. Aasen, D. Kugiumtzis, S.H.G. Nordahl, Computers and Biomedical Research 30, 95-116 (1997) Fraser AM, Swinney HL, Phys Rev A 33, 1134 (1986) Ott, E.; Sauer, T.; Yorke, J.A.: Coping with chaos. Wiley Interscience Publication, New York, 1994. Abarbanel, H.D.I.: Analysis of Observed Chaotic Data. Springer-Verlag, 1995. Peters, E.E.: Chaos and order in the capital markets. Wiley Finance Editions, New York, 1991 Hongchun Wang a, b, Guanrong Chen b, Jinhu LPhysics Letters A 333 (2004) 246255 Harrison, R., Yu, D., Oxley, L., Lu, W., & D. George, Non-linear noise reduction and detecting chaos: Some evidence from the S&P Composite Price Index, Mathematics and Computers in Simulation, 48, 497-502 1999.
Does Free Cash Flow Anomaly Exist in an Emerging Market? Evidence from the Istanbul Stock Exchange
zgr Arslan Hacettepe University Department of Business Administration Beytepe 06800, Ankara, TURKEY E-mail: arslan@hacettepe.edu.tr Tel: 00 90 312 297 87 00 Ext: 129; Fax: 00 90 312 299 20 55 Mehmet Baha Karan Hacettepe University, Department of Business Administration Beytepe 06800, Ankara, TURKEY E-mail: mbkaran@hacettepe.edu.tr Tel: 00 90 312 297 87 00 Ext: 147; Fax: 00 90 312 299 20 55 Abstract The objective of this paper is to investigate if the previously documented free cash flow anomaly can be generalised to emerging markets. Therefore we examine the performance of an investment strategy based on free cash flow using financial statement data of Turkish firms during the period 1999-2005. We identify large-capitalisation companies with positive free cash flows, low free cash flow multiples, and low financial leverage. Firstly, our results fail to provide evidence that a portfolio of these companies outperform the market index in general. However, our findings suggest that a free cash flow portfolio earns higher returns than market portfolio during market downturns. Consequently, our results present a partial support for the existence of free cash flow anomaly in Turkey. Keywords: Free cash flow, anomaly, portfolio management, investment strategy Jel Classification Codes: G11, G12
1. Introduction
It is previously shown by Livnat and Zarovin (1990) and Ali (1994) that cash flows have incremental information superiority over earnings for explaining the cross sectional variations in stock returns. Previous studies1 also identify that investors can benefit a trading strategy based on information derived from the financial statement to earn abnormal returns. In accord with these findings, the studies of Hackel, Rivnat and Rai (1994, 2000) and Hackel and Livnat (1995) for U.S and Jokipii and Vahama (2006) for Finland report that an investment strategy based on free cash flows can outperform the market portfolio. Moreover, this free cash flow anomaly is found to be unrelated to the previously acknowledged cross-sectional anomalies.
See; Ball (1992) for a review and Bernard, Thomas and Wahlen (1997); Penman and Zhang (2002) and Kothari (2001)
164
This paper analyzes the previously identified free cash flow anomaly from the perspective of an emerging market. Our main question is; does an investment strategy based on free cash flows can outperform the market portfolio in every emerging markets in the world? The purpose of this study is to discuss this question using stocks of the Istanbul Stock Exchange (ISE) as a laboratory. We examine the returns on portfolios to see whether investors in an emerging market relying on the information extracted from the developed capital markets based on the free cash flow would earn higher returns. In conclusion, this paper provides an appropriate framework to evaluate whether the empirical findings obtained for the developed markets; namely US and Finland, can be generalized. Studies on anomalies for emerging markets and specifically on Turkey is not novel in the investment literature. There has been an intensive research on the investment anomalies in ISE during the last decade. Particularly, these studies 2 document the impacts of previously acknowledged factors on returns of stocks traded in ISE and generalise these impacts to other emerging markets. These factors are; day of the week effect, growth stocks versus value stocks, turn of the month effect and even the impact of cloudy days on stock returns; to name the some. Our research on the free cash flow effect as an asset pricing anomaly contributes to these previous studies generalised to emerging markets through its uniqueness and originality. Last but not the least the empirical findings of Jokipii and Vahamaa (2006) demonstrate that free cash flow strategy is particularly attractive in declining markets. Turkey has experienced severe financial crises in November 2000 and February 2001 3 which have resulted in sharp declines in ISE particularly in 2001 and 2002 4. Therefore our analysis provides considerable verification for this finding. Our results for the period 1999 - 2005 concerning publicly traded firms on the ISE are obtained through constructing a portfolio of large capitalization companies with free cash flows, low free cash flow multiples and low financial leverage. Our findings provide partial support for the existence of free cash flow anomaly in Turkey. A portfolio of these companies is not found to outperform the market index in general for the entire sample period. However, the results presented in this paper suggest that investors can earn abnormal returns with investment strategies based on free cash flows only in market downturns, in other words, bear markets. The last finding for Turkish firms aligns with that of Jokipii and Vahamaa (2006) for Finland firms that the impact of free cash flow anomaly is magnified in the bear markets rather than the bull markets. The remainder of this paper is organized as follows. The next section presents the data used in our empirical analysis and the portfolio selection rules. The third section reports results on empirical findings about depict the performance of the free cash flow strategy. Finally the fourth section concludes our work.
2. Data
Our work is conducted on the listed firms in ISE for the period of 1999 and 2005. The financial statement data is obtained directly from the ISE web-site 5 while the monthly investment data is gathered from Datastream. Having the ability of revealing the price developments of stocks listed on the ISE, ISE-100 index is used as a benchmark portfolio in our analysis 6. Free cash flow is defined as the cash, which is generated by operations, can be distributed to shareholders as dividends without jeopardizing the current level of firm growth. More specifically, Jensen (1986), who has highlighted this issue in the corporate finance literature, defines the free cash flow as cash flow in excess of that required to fund all projects that have net present values when discounted at the relevant cost of capital. Therefore there are three alternative usage of the free cash
2 3 4 5 6
See respectively; Gonenc and Karan (2003), Balaban (1995), Demirer and Karan (2001), Oguzsoy and Guven (2006), Tufan and Hamarat (2004). See Ozkan (2005) for the details of the Turkeys last consecutive financial crises See Arslan, Florackis and Ozkan (2006) www.imkb.gov.tr ISE-100 is extensively used as a benchmark portfolios in studies on Turkey such as in Berument and Ince (2005) and Dogan and Yalcin (2007) to name a few
165
flow; it can be used to take advantage of new business opportunities, paying back debt or distributed to shareholders, providing that market value of the firm is unaffected. Free cash flow is measured as the net cash flow from operating activities minus capital expenditures. The net cash flow from operating activities is defined as the sum of net income, all non-cash charges and credits (e.g. depreciation, amortization of intangibles and deferred taxes.), net change in working capital and extraordinary items. Following Jokipii and Vahamaa (2006), a more simplified portfolio selection criteria than those of Hackel et al. (1994 and 2000) and Hackel and Livnat (1995) are applied for this paper. Specifically, a portfolio of; large capitalization companies with free cash flows, low free cash flow multiples and relatively low financial leverage is constructed. The portfolio selection criteria are described in detail below; First Criterion: Free Cash Flowt > $ 0 Second Criterion: 5 < Market Valuet / Free Cash Flowt-1 < 30 Third Criterion: Total Debtt / Free Cash Flowt < 10 Fourth Criterion: Market Valuet > $ 100 million. Aligning with Jokipii and Vahamaa (2006), due to the limitations in ISE, we do not assume neither any growth in free cash flows nor positive four-year average free cash flows as in Hackel et al. (2000). The second criterion indicated above is called as free cash flow multiple and it is estimated as the market value of equity to the free cash measure in the previous year. This criterion is applied in our analyses to guarantee that relative to their free cash flow status, firms are priced at a reasonable level. Although the free cash flow multiple interval in Hackel et al. (2005) is between 5 and 20, due to the fact that free cash flow multiples are generally low or negative ISE we have adopted 30 as the upper bound of the interval as in Jokipii and Vahamaa (2006) 7. It would be inconvenient to select firms having high free cash flows that are highly leveraged with a disadvantageous debt capacity. Therefore we apply the third criterion called as the debt multiple which is estimated by the ratio of total debt to free cash flow. The fourth an the last criterion is applied so as to ensure that the free cash flow portfolio is comprised by the firms that are relatively large and traded sufficiently. In accord with the previous studies 8, we have chosen this upper bound arbitrarily so as to match up with the market capitalisation mean in ISE, which is around $ 116 million, and in turn ensure that the portfolio contains a reasonable number of companies. Aligning with Gonenc and Karan (2003) the beginning of July is picked in each year to form our free cash flow portfolio. In specific, after forming the portfolio we assume a 12-month buy-andhold strategy from July 1, t to June 30, t+1. The average number of companies fulfilling the criteria on average is 13 which represent 5% of the companies publicly traded in ISE. The percentage of the companies in the free cash flow portfolio in the previous studies is %10 and %1 of the publicly traded companies in Jokipii and Vahama (2006) for Finnish firms and Hackel et al. (2000) for US firms respectively. Compared with those in the previous studies, the number of companies included in the portfolio may be considered moderate relative to the number of publicly traded firms in the ISE, which totals to 300 on average 9. The number of companies in the free cash flow portfolio is 12, 14, 6, 7, 13, 17 and 19 for the years 1999, 2000, 2001, 2002, 2003, 2004 and 2005 respectively. As expected, number of firms in the portfolio decrease dramatically during the crises period, namely the years 2001 and 2002. Table 1 presents the descriptive statistics of the selection criteria for the free cash flow portfolio where median market values (in million dollars) and the free cash flow and debt multiples for the companies selected in the free cash flow portfolio and the corresponding medians for all listed
7
8 9
We have also assigned 20 as the upper limit of the interval as in Hackel et al. (2000) however number of firms staying in the interval was too low for our analyses Hackel et al. (2000) and Jokipii and Vahamaa (2006). As indicated in Jokipii and Vahama (2006) on account of the low number of eligible companies in the protfolio, an investment strategy based on free cash flow portfolio is anticipated to have a superior advantage for foreign investors who wish to invest in some stocks in thr ISE rather than domestic investors who aim to form a well-diversified portfolio
166
companies on the ISE-100, for the years between 1999 and 2005. The median market values of the companies in the free cash flow portfolio are lower than those of the market portfolio except for the crisis period, namely the years 2001 and 2002. During the crisis period market capitalisations of free cash flow portfolio exceed those of market portfolio. Consequently, the companies in the constructed portfolio may not be regarded as large capitalization firms relative to the market, except for the crisis period. The results in the table on the median free cash flow multiple for the both free cash flow portfolio and the ISE-100 show that the median values of this variable display an identical comovement in accord with the recession and expansion in the economy. The median value of free cash flow multiple is negative for the market portfolio while decreases almost 40 percent during the crisis period. The similar co-movement pattern in both the free cash flow portfolio and the ISE-100 is also observed for the debt multiple. The table reports that the debt multiple of the free cash flow portfolio is always above than that of the ISE-100 however they all decrease during the crisis period but rise again right after. Finally Table 1 shows beta coefficients for the free cash flow portfolio. The reported betas are estimated through employing the market model with monthly returns for the previous 36 months and using the ISE-100 index as a proxy portfolio for the market return. The betas of the free cash flow portfolio range from 0.52 to 0.80 reflecting that companies in the portfolio may be considered to have relatively low systematic risk 10.
Table 1:
Year 1999 2000 2001 2002 2003 2004 2005
Notes:
Descriptive Statistics
Market Value FCF ISE100 133,543 164,362 228,435 301,432 105,320 94,241 116,582 106,843 137,760 154,876 149,873 273,421 164,931 295,321 FCF Multiple FCF ISE100 10.53 1.43 11.84 2.94 6.22 -2.54 7.31 -1.94 10.26 -0.03 14.59 4.72 17.43 6.43 Debt Multiple FCF ISE 100 4.41 2.08 5.26 1.43 1.81 -1.65 1.54 -0.87 3.42 0.21 4.81 2.51 6.05 3.21 Beta FCF 0.58 0.80 0.52 0.61 0.69 0.74 0.59
This table reports medians of the free cash flow (FCF) portfolio selection criteria and the corresponding medians for all companies listed on the ISE100. The free cash flow multiple is represented as the FCF Multiple and estimated as the ratio of market value of equity to free cash flow. The Debt Multiple is the ratio of total debt to free cash flow. The market values are reported in million dollars. Beta is estimated using the market model with monthly returns from the previous 36 months. The ISE100 portfolio index is used as a proxy for the market return
3. Results
Equally weighted average of returns for the individual stocks included in the portfolio is calculated for the buy-and-hold return for the free cash flow portfolio. We use two different measures of abnormal returns (M1 and M2) in order to compare the performance of the free cash flow investment strategy to the ISE-100 index. The first measure, M1, is the conventional market adjusted return and it is defined as; M1 = RFCF RISE100 (1) In equation 1, RFCF denotes the return on the portfolio of free cash flow whereas RISE100 is the return on the ISE-100 index. Following Gonenc and Karan (2003) and Jokipii and Vahamaa (2006) we employ Fama and French (1993) 11 three-factor model adjusted return as the next measure in our analyses, M2. Fama and French (1996) state that this three-factor asset pricing model, defined in the equation (2), has ability to explain most cross-sectional anomalies in the finance literature. M2 = RFCF RF (RISE100 RF) SMB - HML (2) In equation 2, RF represents the risk free rate, SMB is the difference between the return on a portfolio of small stocks and the return on a portfolio of large stocks, and HML is the difference
10
11
This finding aligns with that of Jokippi and Vahama (2006) for the Finnish portfolio however contradicts with that of Hackel et al. (1994, 2000) who find for the US companies that the betas for the potfolio are around 1 Henceforth this will be called as the Fama-French three-factor model
167
between the return on a portfolio of high book-to-market stocks and the return on a portfolio of low book-to-market stocks. Through using monthly stock returns for the previous 36-month period, parameters for both the market model and the Fama-French three-factor model are estimated. We apply conventional ttests to test if the calculated abnormal returns are statistically significant. 3. a. Performance of the Free Cash Flow Investment Strategy Table 2 demonstrates the comparison of the annual buy-and-hold returns for the free cash flow portfolio with the returns for the ISE-100 index. Firstly, judging into the both mean (median) annual free cash flow portfolio and the ISE-100 index returns, which are 27.4% (30.9%) and 18.8 (31.2%) respectively, it is unclear to verify which portfolio outperforms the other. On the one hand, the free cash flow portfolio yields positive returns for five periods whereas the annual return for the ISE-100 portfolio is positive for only four periods. On the other hand ISE-100 portfolio outperforms the free cash flow portfolio in four out of six periods. These four periods includes the years which are characterised as bull markets and this result aligns with that of Jokipii and Vahamaa (2006). It should be noted that during the period involving the years 2001 and 2002, in which economic crisis reflect its impact on the Turkish stock market, the free cash flow portfolio considerably outperforms the ISE100 index.
Table 2: Annual Returns
FCF 0.573 0.385 0.342 0.123 -0.054 0.276 0.274 (0.071) 0.309 -0.054 0.573 5 6 ISE100 0.632 -0.362 -0.134 0.275 0.365 0.349 0.188 (0.142) 0.312 -0.362 0.632 4 6 M1 -0.059 0.747 0.476 -0.152 -0.419 -0.073 0.087 (0.097) -0.066 -0.419 0.747 2 6 M2 0.401 0.206 0.143 -0.109 -0.221 0.076 0.083 (0.052) 0.110 -0.221 0.401 4 6
7/99 - 6/00 7/00 - 6/01 7/01 - 6/02 7/02 - 6/03 7/03 - 6/04 7/04 - 6/05 Mean p-value Median Minimum Maximum Number of positive periods Number of observations
Notes:
The table reports 12-month buy-and-hold returns for the free cash flow portfolio (FCF) and for the ISE100 index. The buy-and-hold returns are calculated as the equally weighted average of returns for the individual stocks in the portfolio. M1 is the market adjusted return, calculated as M1 = RFCF RISE100 where RFCF denotes the return on the free cash flow portfolio and RISE100 is the return on the ISE100 index. M2 is the Fama and French (1993) adjusted return, calculated as M2 = RFCF RF (RISE100 RF) SMB - HML, RF where is the risk free return, SMB is the difference between the return on a small stocks and the return on a portfolio of large stocks, and HML is the difference between the return on a portfolio of high book-to-market stocks and the return on a portfolio of low book-to-market stocks. The parameters for the market model and for the Fama-French three-factor model are estimated with monthly returns for the previous 36 months. The p-value for the null hypothesis of zero mean is based on a conventional t-test.
In order to see the bigger picture we refer to the results for the annual market adjusted returns and that of the Fama-French three-factor model. Firstly, value of the mean (median) annual market adjusted return is 8.7% (9.7%), and the mean (median) Fama-French adjusted return is 8.3% (5.2%). Furthermore, judging into the returns for the intervals for both July 2000 June 2001 and July 2001 and June 2002, which are overlapped by the crises period, the returns of the both market adjusted and the Fama-French models are positive. The results from both of these latter models reinforce the idea that free cash flow portfolio has beaten the market portfolio during the financial crises period. To be more specific, we report summary statistics of the monthly returns for the free cash flow portfolio and for the ISE-100 portfolio in Table 3. The table shows that the mean (median) monthly return for the free cash flow portfolio is 0.7% (0.9%), while the mean (median) monthly return for the ISE portfolio index is 0.8% (0.8%). However, the mean monthly return of the market index is statistically insignificant. Moreover, minimum and maximum returns for the both portfolios do not
168
present an obvious result, since the maximum return of the free cash flow portfolio is higher than that of the ISE-100 portfolio while the minimum return of the portfolio is lower than that of the ISE-100 portfolio. Standard deviation of the monthly returns for the free cash flow portfolio is somewhat below than that of the ISE-100 index as in Jokipii and Vahama (2006). The authors comment for this result by suggesting that the free cash flow portfolio opposes the fundamental mean-variance theorem. Nonetheless, the return for the free cash flow portfolio was positive for 47 out of 72 months while the return for the ISE-100 portfolio was positive for only 38 months.
Table 3: Monthly Returns
FCF 0.007 (0.000) 0.009 -0.287 0.372 0.069 47 72 ISE100 0.008 (0.106) 0.008 -0.321 0.316 0.071 38 72 M1 -0.001 (0.000) -0.001 -0.202 0.287 0.010 35 72 M2 0.005 (0.000) 0.001 -0.197 0.265 0.005 37 72
Mean p-value Median Minimum Maximum Standard deviation Number of positive periods Number of observations
Notes:
The table reports monthly returns for the free cash flow portfolio (FCF) and for the ISE100 index. The buy-and-hold returns are calculated as the equally weighted average of returns for the individual stocks in the portfolio. M1 is the market adjusted return, calculated as M1 = RFCF RISE100 where RFCF denotes the return on the free cash flow portfolio and RISE100 is the return on the ISE100 index. M2 is the Fama and French (1993) adjusted return, calculated as M2 = RFCF RF (RISE100 RF) SMB - HML, RF where is the risk free return, SMB is the difference between the return on a small stocks and the return on a portfolio of large stocks, and HML is the difference between the return on a portfolio of high book-to-market stocks and the return on a portfolio of low book-to-market stocks. The parameters for the market model and for the Fama-French three-factor model are estimated with monthly returns for the previous 36 months. The p-value for the null hypothesis of zero mean is based on a conventional t-test.
The results in Table 2 and Table 3, which report the annual and monthly returns of the both free cash flow strategy and the market index respectively, do not provide an evidence on the outperformance of the free cash flow strategy over the market portfolio. However the results on Table 2 specify a clue that a free cash flow portfolio is a better investment strategy during the crisis period over the market portfolio. The objective of our further analysis is to substantiate evidence if investors can earn abnormal returns with investment strategies based on free cash flows during bear markets. Table 4 reports the performance of free cash flow portfolio when the ISE-100 portfolio fell by more than 5%. 14 months are identified that fulfil the stipulation and they are called as the market downturns. Majority of these months coincide with the crisis period in Turkey. Firstly all the mean return values of the models are statistically significant except for the Fama-French adjusted returns, which are indistinguishable from zero. The most striking result from the table is that, mean (median) value of returns belonging to the free cash flow portfolio is around 1% (0%), while the mean (median) value of the market portfolio is around -11% (-0.9%). This result clearly suggests that free cash flow portfolio outperforms the market portfolio during market downturns. Furthermore, the maximum (minimum) values of returns of these portfolios, which are around 0.4% (-0.8%) and 0.3% (-1%) respectively, is considerably higher (lower) than maximum (minimum) value of the ISE-100 portfolio, which are around -0.5% (-4%). Finally while the returns from the market portfolio are negative during the selected 23 months, the returns for the free cash flow portfolio and the market adjusted models are positive for nine and four periods respectively. Lakonishok, Shleifer and Vishny (1994) suggest that unknown risk factors are more strongly pronounced during the periods of market downturns. The results from Table 4 suggest that the superior performance of the free cash flow strategy may be attributed to unknown risk factors that are generally highlighted during market downturns. Therefore free cash flow investment strategy is advantageous over market portfolio during bear markets.

Table 4: Monthly Returns During Market Downturns
FCF 0.001 (0.021) 0.000 -0.075 0.041 0.036 9 14 ISE100 -0.107 (0.034) -0.094 -0.381 -0.051 0.029 0 14 M1 0.108 (0.088) -0.000 -0.096 0.078 0.016 5 14 M2 0.000 (0.271) 0.000 -0.062 0.062 0.011 6 14
169
Mean p-value Median Minimum Maximum Standard deviation Number of positive periods Number of observations
Notes:
The table reports monthly returns for the free cash flow portfolio (FCF) and for the ISE100 index. The buy-and-hold returns are calculated as the equally weighted average of returns for the individual stocks in the portfolio. M1 is the market adjusted return, calculated as M1 = RFCF RISE100 where RFCF denotes the return on the free cash flow portfolio and RISE100 is the return on the ISE100 index. M2 is the Fama and French (1993) adjusted return, calculated as M2 = RFCF RF (RISE100 RF) SMB - HML, RF where is the risk free return, SMB is the difference between the return on a small stocks and the return on a portfolio of large stocks, and HML is the difference between the return on a portfolio of high book-to-market stocks and the return on a portfolio of low book-to-market stocks. The parameters for the market model and for the Fama-French three-factor model are estimated with monthly returns for the previous 36 months. The p-value for the null hypothesis of zero mean is based on a conventional t-test.
4. Conclusion
Empirical findings on free cash flow investment anomaly was previously reported by Hackel et al (1994, 2000), Hackel and Livnat (1995) for US companies and Jokipii and Vahamaa (2006) for Finnish companies. This paper investigates if the free cash flow investment anomaly can be generalised to emerging markets through examining the Turkish firms. Therefore, we use annual financial statement data of Turkish companies to identify large-capitalization companies with positive free cash flows, low free cash flow multiples, and low financial leverage. We firstly compare annual and monthly returns of the free cash flow portfolio and those of the market portfolio. We also include market adjusted and Fama and French (1993) adjusted returns of the free cash flow portfolio for the comparison. Our results do not provide sufficient evidence that a portfolio of large capitalization companies with positive free cash flows, low free cash flow multiples and low financial leverage outperforms the market portfolio. Therefore our findings differ from those of the previous work on this anomaly. Nonetheless, after receiving some clue from the annual returns we refer to monthly returns during market downturns to verify if the free cash flow investment strategy is more beneficial than investment in market portfolio during bear markets. Jokipii and Vahamaa (2006) report for Finland firms that free cash flow anomaly is more pronounced in the bull markets rather than the bear markets. Our findings aligns with the finding of the authors in this sense since it is suggested that investors can earn abnormal returns with investment strategies based on free cash flows only in the market downturns. Thus, our findings provide partial support for the existence of free cash flow anomaly in Turkey. Our suggestion for further research is firstly portioning firms according to their industry groups while forming a free cash flow portfolio. Murray (2005) underlines that nature of the business can influence the amount of free cash flow firms can generate since different businesses have different capital expenditure requirements. In other words, capital intensive businesses tend to generate lower amounts of free cash flow relative to businesses operating with high level of intellectual capital. Due to the low number of companies listed on ISE, our sample size do not allow for the industry based partition. Furthermore, testing the free cash flow anomaly for other emerging market countries, which were subject to severe economic crisis during the last decade, not only enables a generalisation for the performance of the free cash flow portfolio but also verifies if the anomaly is more pronounced during crisis period.
170
International Research Journal of Finance and Economics - Issue 11 (2007) ALI, A. (1994) The Incremental Information Content of Earnings, Working Capital from Operations, and Cash Flows, Journal of Accounting Research, Vol. 32, pp. 6176. ARSLAN, ., FLORACKIS, C. and OZKAN, A. (2006) The Role of Cash Holdings on Investment-Cash Flow Sensitivity: Evidence from a Financial Crisis Period in an Emerging Market. Emerging Markets Review, 7 (4), 320-338. BALABAN, E. (1995) Day of the Week Effect: New Evidence from an Emerging Stock Market, Applied Economic Letters, 2, 139-143. BALL, R. (1992), The Earnings-price Anomaly, Journal of Accounting and Economics, Vol. 15, pp. 31947. BERUMENT, H. and INCE, O. (2005) Effect of S&P500s Return on Emerging Markets: Turkish Experience, Applied Financial Economic Letters, 1 (1), 59 - 64. BERNARD, V.L., J. THOMAS and WAHLEN, J. (1997), Accounting-Based Stock Price Anomalies: Separating Market Inefficiencies from Risk, Contemporary Accounting Research, Vol. 14, pp. 89136. DEMIRER, R. and M. B. KARAN (2001), An Investigation of the Day of the Week Effect on Stock Returns in Turkey, Journal of Russian and East European Finance and Trade, November- December, 47 -77. DOGAN, N. and Y. YALCIN (2007) The Effects of the Exchange Rate Movements on the Istanbul Stock Exchange, Applied Financial Economic Letters, 3 (1), 39 - 46. FAMA, E. and K. FRENCH (1993) The Cross-section of Expected Stock Returns, Journal of Finance, 47, 427-465. FAMA, E. and K. FRENCH (1996) Multifactor Explanations of Asset Pricing Anomalies, Journal of Finance, 51, 55-84. GONENC, H. and M. B. KARAN (2003) Do Value Stocks Earn Higher than Growth Stocks in an Emerging Market: Evidence from Istanbul Stock Exchange Journal of International Financial Management and Accounting, 14, 1-25. HACKEL, K.S. and J. LIVNAT (1995), International Investments Based on Free Cash Flow: A Practical Approach, Journal of Financial Statement Analysis, Vol. 1, pp. 110. HACKEL, K.S., J. LIVNAT and A. RAI (1994), The Free Cash Flow/Small-cap Anomaly, Financial Analysts Journal, Vol. 50, pp. 3342. HACKEL, K.S., J. LIVNAT and A. RAI (2000), A Free Cash Flow Investment Anomaly, Journal of Accounting, Auditing and Finance, Vol. 15, pp. 124. JENSEN, M. C. (1986) Agency Costs of Free Cash Flow, Corporate Finance and Takeovers, American Economic Review, 76 (2), 323- 329. JOKIPII, A. and S. VAHAMAA (2006), The Free Cash Flow Anomaly Revisited: Finnish Evidence, Journal of Business Finance and Accounting, Vol. 33, pp. 1-18. KOTHARI, S. (2001), Capital Markets Research in Accounting, Journal of Accounting & Economics, Vol. 31, pp.105-231. LAKONISHOK, J., A. SHLEIFER and R. W. VISHNY (1994), Contrarian Investment, Extrapolation and Risk, Journal of Finance, 99, 1541-1578. LIVNAT, J. and P. ZAROWIN (1990), The Incremental Information Content of Cash-flow Components, Journal of Accounting and Economics, Vol. 13, pp. 2546. MURRAY, N. (2005), Wealth Manager: Free Cash Flow Stocks Making Market Return. The London Free Press, September 12, http://www.fyilondon.com OGUZSOY, C. B. and S. GUVEN (2006), Turn of the Month and Turn of the Month Surrounding Days Effect in Istanbul Stock Exchange, Journal of Emerging Market Finance, 5 (1), 1-13. OZKAN F.G. (2005), Currency and Financial Crises in Turkey 2000-2001: Bad Fundamentals or Bad Luck? World Economy, Vol. 28; pp. 541-572.
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22]
International Research Journal of Finance and Economics - Issue 11 (2007) [23] [24]
171
PENMAN, S. H. and X. ZHANG (2002), Accounting Conservatism, the Quality of Earnings and Stock Returns, The Accounting Review, Vol. 77, pp. 237-264. TUFAN, E. and B. HAMARAT (2004), Do Cloudy Days Affect Stock Exchange Returns: Evidence from Istanbul Stock Exchange Journal of Naval Science and Engineering, 2 (1), 117 126.
Export Diversification in the Export Led Growth Process of Turkey

Filiz Ozag The Department of Economics, The Faculty of Economics and Administration Gazi University, Beevler/ANKARA E-mail: ozfiliz@gazi.edu.tr Suleyman Degirmen The Department of Economics, The Faculty of Economics and Administration Mersin University, Ciftlik Koy Campus, Mezitli, 33342 Mersin E-mail: sdegirmen@mersin.edu.tr Tel: 90-537-288 93 56 Abstract In the 1970s, the export oriented growth model, recommended to less developed countries, was targeting to open the economies of these countries to the international markets. Turkey was involved in this conjuncture with the 24th January 1980 economic structural package. The aim of this study is to analyze changes in the degree of export diversification in regard of the process. As a regard of the beginning period of the export led growth model, the 1980-2004 interval is selected and tested. Calculations in the test are based on the SITC classification for a country group (i.e., EU countries), having big share in foreign trade. The results suggest the link between the hypothesis of diversification-led development and outward-oriented growth model are not directly related to one another. Keywords: xport Diversification, Outward-oriented growth model. Jel Classification Codes: 14, F43, O40
1. Introduction
The notion of development in the agenda of the post-WWII era has also brought a discussion of why some countries have differed from each other or have become similar to each other. Along with itself, among criterion determining economical differences and similarities, there have been income level, physical and human capital, geographic stand, industrial structure, and political independency. In this classification, economies in the group under the title of the Less Developed Countries (hereinafter, LDC), which have had inadequate income level, low productivity-growth rate, incremental unemployment rate, and export of mostly agricultural products, have taken place. While import substitution growth model until the 1970s had transformed mentioned indicators to positive for LDC, and hence, the model had got these countries closer to developed countries, (hereinafter, DC), afterwards export-leading growth model has taken place, instead. In this new model, which is also named as outward oriented growth model, it has been recommended that LDCs open their economies to international market and apply liberal economies. This prospectus will enable diversification in export and thus, will change the structure of countrys export simultaneously.
173
The aim of this study is to specify the influence of outward oriented growth model on export diversification, which Turkey has applied parallel to the World conjuncture. In this respect, after the introduction section, next section surveys related literature. The third section provides statistical methodology; fourth one provides information about data and method of the study. Last section comments the test results.
2. Theoretical Survey
Diversifying a countrys export is to increase the ranking of export specialization, and to escape a countrys export from dependency on only one good. Main factor behind this diversification is willingness to increase export incomes and hence to grow [econgomy] rapidly (Stanley and Baunnagi (2001), Al-Marhubi (2000), Pineres and Ferrantino (1997), Koekkoek (1992), Balabanis (2001)). Of studies examining the relationship between foreign trade and growth, the first one belongs to Prebish (1950) and Singer (1950, 1952). These authors studied this issue over developing countries and stated that import oriented growth model has enabled a diversification in export ( Pineres and Ferrantino [1997], p. 376) During 1950s, 1960s, and 1970s, large number of development economists defended the import substitution growth model. However, there was abundant evidence suggesting that more open and outward oriented economies had outperformed in countries pursuing import substitution growth model. Even though outward orientation policy was unpopular and slowly gained followers among academies in 1960s and 1970s, in the 1980s, economists dealing with poorer nations began to recommend outward orientation: (i) development strategies based on market oriented reforms that included the reduction of trade barriers as a fundamental component and (ii) the opening of international trade to foreign competition. Moreover, the World Bank, the International Monetary Fund, and other multilateral institutions routinely required the developing countries to embark on trade liberalization and to open up their external sector as a condition for receiving financial assistance (Edwards [1993], pp.1358-1359). [Recent] empirical work on trade policy and [economic] growth can be classified into two broad and distinct categories: (a) large scale multi-country studies that have investigated in detail the experiences of a group of countries with trade policy reform (b) lower case econometric studies that have investigated, on broad cross-country data, the relationship between the pace of exports expansion and aggregate economics growth (Edwards [1993], pp.1358-1359). In recent years, studies on export-led growth in developing countries have increased. Along with hypothesis that outward-oriented countries grow more rapidly, the other hypothesis suggests that there is a relation between the pattern of economic development and structural change in exports and increased export diversification. However, the link between the hypothesis of diversification-led development and outward-oriented growth model is not directly related to each other (see for details, Rosenstein-Rodan [1943], Prebisch [1950], Singer [1950, 1952], Nurkse [1953]). On the other hand, according to Vernon (1966), Krugman (1979), Dollar (1986), Segerstrom et al., (1990), Grossman and Helpman (1991). The other models of trade and growth also imply a link between export diversification and the development process, which innovative activity leads to an increasing diversity of products while imitative activity leads to an increasing diversity of products being produced and exported from low-wage locations. Thus, we say there is a link between export growth and diversification (Pineres, Ferrantino [1997], p. 376-377). While developing countries export undifferentiated goods based on factor advantages, developed countries export differentiated products. Presumably, the process of graduation from Least Developed Countries (LDC) to Developed Countries (DC) status experienced by some countries should be accompanied by structural changes of export in direction of diversity (Pineres and Ferrantino, 1997, p. 376). Thus, since the 1980s openness, trade liberalization and outward-oriented policies have become popular prescriptions among economists and policy makers for achieving rapid economic growth.
174
However, parallel to the outward-orientation paradigm, another hypothesis related to structural changes in exports and increased diversification of exports gained even greater popularity in the literature... It has been argued that a more diversified export mix may enable a country to accomplish stability-and growth-oriented policy goals... a diversified export portfolio would be expected to minimize the volatility in export earnings and to foster economic growth. It has also been suggested that export diversification initiatives be undertaken within a broad policy approach, whereby national governments should design and support a coherent macroeconomic policy framework consistent with export promotion strategies. While export diversification programs should be implemented primarily by the private sector, the role of the government should be to prevent distortions and create an environment which promotes diversification (United Nations, 2004, p.2-3).
3. Statistical Methodology1
The methodology used in this study based on Dalum and Villumsen (1996), and Laursen (1998) published by Danish Research Unit for Industrial Dynamics. Reveal Comparative Advantage IndexRCA developed by Balassa (1965) in Taylor (2003) has been mostly used in the studies. Therefore, following the formula of index measures changes in the competition power of a country, Xij / Xij i RCA = Xij / Xij j ij (1) Xij indicates export of sector i from country j. The numerator of this formula represents the percentage share of a given sector in national export. The denominator of the RCA index formula represents the percentage share of a given sector in the export of a selected country or the world. Thus, the RCA index contains a comparison of national export structure (the numerator) with the country or world country structure (denominator). When RCA is one for a given sector of a given country, percentage share of that sector is identical to the group average. The value of RCA changes between zero and positive infinite. Where RCA is above 1 the country is said to be specialized in that sector and just the opposite where RCA is below one. However, there can be some problems in the computation of the index. First, when export for a given sector for a given country (i.e., national export) is zero, the index is indefinite. Second, the inherent skewness of this measure casts doubt on the normality of its distribution. For this reason, Laursen (2000) suggests a simple adjustment to the index to yield the following formula: RSCAij = (RCAij -1) / (RCAij+ 1) (2) The values of the Revealed Symmetric Comparative Advantage, (RSCA), changes between +1 and -1. When RSCA>1 implies specialization, RSCA 1 implies non-specialization. Using these definitions, two elements of diversification can be assessed. The first relates to the degree of structural change in a countrys entire export portfolio. Simply stated, if the range of products exported increases or there are substantial changes in the rankings of exports, some degree of diversification occurs. However, a narrower, and perhaps more critical element of diversification relates to changes in the structure of exports revealed to be specialized. One area is investigating the extent to which the portfolio of exports revealed to be specialized broadens and changes in terms of product rankings. The use of the RSCA measure permits the assessment of both aspects of diversification (Taylor, 2003, p.108). Computation of the RSCA in the study uses this broad definition, and export diversification is defined as the increase in ranking of export specialization. Computation of RSCA index for a given country or country group is done by two periods--past and current periodand to assess these changes,
1
This section is mostly acknowledged by Keld Laursen (2000) 4.Structural change in OECD Export Specialization Patterns: De-specialization and Stickness, in Keld Laursen, 2000, Trade Specialization, Technology and Economic Growth, Edward Elgar: UK, pp.53-68, and Taylor (2003).
175
a Galtonian regression is used to compare the distribution of the RSCA for each country at two points in time (past period t0 and current period t1) (see for details, Laursen [1998] [2000], Dalum and Villumsen [1996]). Used linear regression is as follows: (3) RSCAij ,1= j + j RSCAij, 0 + i Where t1, and t0 refer to terminal and base time periods of analysis respectively, and i is assumed to be a normally distributed disturbance term with mean 0, constant variance is independent of RSCAij,0. It should be emphasized that period 1 and 0 refer to reference periods and not necessarily successive time periods. In essence, Eq. (3) measures changes in export structure by comparing the distribution of the RSCA for country j at two points in time.
Figure 1: Galtonian Regression
If the expected value of RSCA1 is equal to that of RSCA 0 and is equal to one (a), the relative pattern of export specialization does not change in the rankings. This implies that there have been no changes in the rankings or, equivalently, in the relative pattern of export specialization. In the case depicted by line (c), the value of is greater than one. In this situation, exports in which the country is specialized become more so, and exports in which the country is non-specialized become more nonspecialized. Line (b) illustrates the case where 0 < < 1. Under this scenario, the pattern of export specialization demonstrated by a given country moves toward the group average. Exports in which the country is specialized become less so, and those in which the country has low levels of specialization increase their values (i.e., they become less non-specialized). This is what is termed as regression towards the mean. Although not depicted in Figure 1, it is also possible for < 0 to occur. In this situation there is a reversal in the pattern of specialization. Industries demonstrating export specialization switch to being non-specialized and those initially demonstrating export nonspecialization become specialized. In assessing the degree and nature of export diversification, these measures must be interpreted with care. Estimated values of < 1 may suggest that there are changes in the pattern of export specialization demonstrated. However, diversification also entails changes in the rankings of industries demonstrating export specialization or non-specialization. As Eq. (3) essentially compares the distribution of the structure of export specialization in the base and terminal periods of analysis, the coefficient of determination (R2) provides insight into this issue. If R2 = 1, then there is no change in the ranking of export specialization revealed by a countrys industries. In contrast, low values of R2 suggest that considerable change has occurred in the ranking of industries that exhibit specialization.
176
As R2 and hence R must be less than one, an estimated value of > 1, necessarily yields ||/|R|>1 suggesting that there has been an increase in specialization in terms of both the magnitude of the estimated RSCA and the narrowing of the range of products in which specialization occurs2 .
4. Data and Method

In this study, the effect of the procedure of [trade] liberalizationto open Turkish economy to international marketon export diversification is tested. To do this, there should be four steps; 1. Analysis Period of the study has begun with January 24th 1980 Economic Stability Package referring outward oriented growth model and it covers the period of 1980-2003. In the study, 1981 is defined as starting year because of (i) transformation in the foreign trade regimes and (ii) removal of quota on foreign trade. To limit the period, it has been considered the given/selected country group for the studyespecially, to end the period at the end of 2003. This group includes EU countries having big share in Turkish export. To extend the period of the study by 2004 brings influences of new 10 members in the expansion procedure of EU to the study. Because of ending the addition procedure of 10 new members into EU in 2005 and the unclear influence of this recent addition to the study, we pull out 10 new members from the analysis, and the period is limited by 2003. 2. Country group of this study includes EU countries, because EU countries have a big share in Turkish export. The ratio of share has increased from 31% in 1981 to 54% in 2002 and recently to 72% in 2003. Data of export by countries is provided from State Statistic Institute (SSI). Standard International Trade Code (hereinafter, SITC) is used to classify goods. Forming the EU group, we used the data of countries and aggregated the data in regard of the expansion period of the EU. 3. For structural change, the years of trade liberalization has been considered for the aim of the study. The adaptation procedure of Turkish foreign trade to free market economy started in 1981. Import pre-1980 in Turkey was executed in the extent of three different lists: (i) the list of liberalization, (ii) the list of preferential import goods, and (iii) the quota list of agreed countries. Importing any goods and services out of this list was prohibited. In the list of liberalization, there were industrial and intermediate goods that are required to be imported. The quota list of agreed countries had the imports of goods and services released by countries. The list of preferential import goods (i.e., quota list) mostly had consumption goods. According to the agenda of developments in the World trade, the practice of quota was ended in 1981. Then, 1984 was an important year for the liberalization of import. Goods were classified into three categories such as prohibited, permitted, and liberalized (i.e., free) to import. And different lists for goods to be imported by paying related tax (i.e., sort of a fund) were applied. In the same year, the value added tax and the support fund of resource usage were applied on imported goods. Thus, the regime of foreign trade got closer to the conditions of free market view. In 1990, the import of all goods which left out of the goods prohibited to import was freed. The practice of guarantee and goods on permission was ended and then, custom duty rates and fund aggregated in one list. Therefore, this was the second big structural change in the foreign trade regime after 1984. In 1992, all stamp duty and other taxes that were left out of custom duty rate and collective housing fund, in the framework of harmonize system and single tax systematic, were removed from the system. Nominal tariff rates were decreased about 50 % for sectors. The Decree, numbered 95 and having been applied since 1996, enabled an agreement on Custom Union with European Union. According to the agreement of Custom Union, import was
Please see Appendix-1 for the stages of formulizing after the Equation 3 to (||/|R|>1).
177
operated in terms of seven different lists. The reductions required by the agreement of Custom Union started levying after 2002. In short, foreign trade regime was progressively accompanied with the outward oriented growth model during the period of 1981-2003. The aim of mentioned liberalization policies was to unite domestic and international market on the base of [free] market economic view. In the practice of the period, the important years of this transmission emerged as 1984, 1990, 1992, and 1996. 4. Test Results: Following table provides empirical results.
Table 1: Regression Estimates for EU Countries (Selected Years)
-0,00030 0,00060 0,00020 -0,00009 -0,00010 R2 0,0030 0,0150 0,0029 0,0003 0,0006 R 0,054772 0,122474 0,053852 0,017321 0,024495 (||/|R|) 0,00548 0,00489 0,00371 0,00520 0,00408
Periods 1981-1984 1981-1990 1981-1992 1981-1996 1981-2003
According to the test results on Table 1, ||/|R| is smaller than 1 for the period of 1981-1984. This refers that the practices pre-1984 liberalization did not have any effects on diversification in export. Similar results arose for the periods of 1981-1990, 1981-1992, 1981-1996 and 1991-2003. For all selected periods, ||/|R| is less than one. In brief, the result on the Table 1 shows that the process of trade liberalization of Turkish economy does not have any influence on diversification in export to EU countries.
5. Conclusion
The study analyzes if the process of trade liberalization of Turkish economy starting with 24 January 1980 Economic Stability Package has any influence on diversification in export to EU countries. In the determination of country group of EU countries, the large share of the group in total export of Turkish economy, increasing from 30% in outset year to 70% in 2003, become a reason to select EU countries. Data period covers the period of 1981-2003. The period is divided into five sub-periods in regard of structural reforms for foreign trade policies to open the economy to international market. In the extent of new growth model embraced after 1980, import was liberalized and export was supported with export promotion strategies. In this content, the practice of quota was ended in 1981; foreign trade with a new import regime got closer to [free] market economy; in 1990, all goods that left out of goods prohibited to import were freed to import; in 1992, all taxes levied on import was decreased and tarifs were declined. The year of 1996 is also important due to signing the Custom Union Agrement with EU. Therefore, determination of periods for this study is as such: 1981-1984, 1981-1990, 19811992, 1981-1996 and 1991-2003. In calculations, the study uses 66 sectors in two-digits that Turkish economy exported EU in the classification of Standard International Trade Code, SITC. When looking at the share of sectors in export of Turkey to EU, we see that textile and food have alltogether 75 percent for 1980-2003. These two sectors have been traditional competitive sectors. Export figures in dollar indicates that this situation did not change much in time. However, in recent years, two different developments have taken in consideration. First, production in textile has decreased because of a decrease in the usage of full capacity and unfulfilled new investments. In this process, technological innovations were not also brought out. This situation states that a country can still be competitive in extent of the World economy; however, decreases in its economic performance can still be seen. This negative developments has gradually continued after the Custom Union Agreement. If government does not take any measurements to amend textile sector, Turkey will likely loose its competitiveness in the sector. Therefore, Turkey needs to succeed the diversification in export not to loose its competitiveness against the EU and OECD countries.
178
Second development is an rapid increase in export of electric devices and otomotive especially in 2000s. When looking at export figures in dollar, this situation can be interpreted as addition of a new good. These two sectors, however, have not still had large share in the total exporti.e., the rate of increae in percentage in time is 8.96. The share of transformation and electirical devices in total export increased from 20% in 2000 to 28.96 in 2004. In conclusion, according to the test results, the study states that trade liberalization policies does not have any effects on diversification of its exports to EU countries. The ratio of (||/|R|) is less than one for all period. This finding means that there is a change in the structure of specialization of export when we compare the base year of 1981 with the current (or terminal) years (1984,1990,1992, and 1996) which are selected in terms of structural fragility in foreign trade policies. The result of the analysis of the study supports the findings of Rosenstein-Rodan (1943), Prebish (1950), Singer (1950,1952) ve Nurkse (1953). The process of the liberalization of Turkish trade did not have any effects on diversification in export to EU countries.
Appendix-1
Decomposing the variance of Eq. (3) has done by following in Taylor (2003). More specifically, let (4) denotes the variance of the RSCA in period k = 0, 1. Using the well known result Where that the coefficient of determination for the regression in Eq. (3) can be written as: (5) which after appropriate substitution it may be written as (6) This may be equivalently stated as (7) The expression in Eq. (7) provides a measure of the degree to which diversification (specialization) has occurred between time period 1 and 0.
179
International Research Journal of Finance and Economics - Issue 11 (2007) Al-Marhubi, F. Export Diversification and Growth: An Empirical Investigation, Applied Economics Letters, 7, 559-562, 2000. Balabanis, G., L., The relations and Performance in Export Intermadiary Firms, British Journal of Management, Vol: 12, 67-84, 2001 Dalum, B., Villumsen, G., Are OECD Export Specialization Patterns Sticky? Relations to the Converge-Divergence Debate, DRUID Working Paper No: 96-3, 1996 Edwards, Sebastian, Openness, Trade Liberalization, and Growth in Developing Countries, Journal of Economic Literature, Vol.31, Issue 3,, pp.1358-1393, September 1993. Koekkoek, A. Exports of Developing Countries: Differentiation and Comparative Advantage, The International Trade Journal, Volume VI, No:3, Spring 1992 Laursen, K., Trade Specialization, Technology and Economic Growth: Theory and Evidence from Advanced Countries, USA, 2000 Revealed Comparative Advantage and the Alternative as Measures of International Specialization, DRUID Working Paper No: 98-30, 1998. Pineres, S.A.G., Ferrantino, M. Export Diversification and Structural Dynamics in Growth Process: The case of Chile, Journal of Development Economics,Vol: 52, 375-391, 1997. Prebish, R., The economic Development of Latin America and its Principal Prices, (United Nations) New York, 1950. Singer, H., The distribution of gains between investing and borrowing countries, American Economic Review 40(2), p. 473-85, 1950. Stanley, D.L., Bunnagi, S., A New Look At the Benefits of Diversification: Lessons from Central America, Applied Economics, 33, 1369-1383, 2001. Taylor, T. Export Diversification in Latin America and The Caribbean, The International Trade Journal, Vol: XVII, No: 2, pp. 101-128, Summer 2003. United Nations, Export Diversification and Economic Growth: The Experience of Selected Least Developed Countries, Economic and Social Commission for Asia and the Pacific (ESCAP) Unit, Development Papers No. 24, United Nations Publication,, New York, the US, pp.1-7, 2004.
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13]
RASM: A Risk-Based Projects Auditing Selection Methodology for Large Scale Programs
Othonas Zacharias School of Electrical & Computer Engineering National Technical University of Athens 9 Iroon Politechniou Str., Zografou, 15773 Athens, Greece E-mail: ozach@epu.ntua.gr John Mylonakis 10 Nikiforou Str., Glyfada, 166 75 Athens, Greece E-mail: imylonakis@vodafone.net.gr Dimitrios Th. Askounis School of Electrical & Computer Engineering National Technical University of Athens 9 Iroon Politechniou Str., Zografou, 15773 Athens, Greece E-mail: askous@epu.ntua.gr Abstract This paper proposes RASM, a Risk-based projects Auditing Selection Methodology for large scale programs. RASM is based on risk factor analysis, identifying a large number of risk factors, in the context of large scale programs. RASM is not static, i.e. it can be used throughout a program life cycle, suggesting different samples in different time periods, regarding projects progress and probabilities fluctuation of risk appearance. The paper covers all aspects of projects auditing sampling, proposing a project risk ranking and a sampling methodology, incorporating feedback from previous audits with a structured and simplified way. Finally, this paper presents a RASM application case study in real data of the Operational Program Information Society of Community Support Framework III in Greece. Keywords: Risk Management, Program Management, Projects Auditing, Risk Ranking, Projects Selection, Large Scale Programs
1. Introduction
Project auditing is a fundamental procedure in the context of large scale programs, for which in most cases, multi-level independent auditing structures are created. A large scale program is defined in this paper as a program, that is constituted of a large number of projects (more than 100), which projects even though they differ considerably (on the objectives, the included activities, the complexity, the technology used, the available resources, etc), nevertheless, they serve the same strategic goals in a specific sector or region. It is worth mentioning that in the majority of large scale programs, despite their size and importance for the national and international economies, still no commonly accepted risk management practices and techniques have been established.
181
Projects within large scale programs are funded by one central institution; however, they are usually implemented by different local project implementing bodies. As a result, project auditing actually constitutes the only way for the funding institution to obtain a reasonable assurance that the funds were indeed efficiently used as they were intended, achieving the set objectives, with the principles of economy and sound management. Given the significant importance of project auditing, some of the most important risks a large scale program managing authority faces are related to the efficient implementation of these audits. Bearing in mind that its not possible (at an acceptable cost basis) to perform on-site audits to all the projects of a large scale program, it is clear that one of the most important risks is the wrong project selection. Auditors are very familiar with the statistical sampling method, which allows conclusions to be drawn from the results of audits of a sample on the overall expenditure from which the sample was taken. This is absolutely rational, when audit objectives are limited to providing reasonable reassurance about program implementation progress. However, statistical sampling methods often require significantly big samples, in order to allow conclusions to be made with a sufficient confidence level. On the other hand, program managing authorities although undoubtedly seek reasonable reassurance for the program implementation progress, they also use audits as an efficient risk management tool, i.e. managing authorities should audit the most risky projects in programs as soon as possible, so that they can reduce risk level and propose specific controls and contingency plans. From this perspective, program audit sampling objective should not be the estimation of a representative sample, but the estimation of a sample, that would provide the maximum reassurance that all risky projects would be audited, thus the risks threatening projects implementation would be under control. Towards this direction, this paper proposes a Risk-based projects Auditing Selection Methodology (RASM) for large-scale programs. RASM is covering all aspects of projects auditing selection, proposing a project risk ranking and a sampling methodology, incorporating feedback from previous audits with a structured and simplified way. Given the size of large scale programs, which are under consideration, RASM is fairly simple and user friendly, so that it can be applied in a short time period. It is based in a semi-quantitative approach, using the projects data that usually are collected by program managing authorities anyway. Finally, this paper examines a case study, in which RASM was implemented in the Greek Operational Program Information Society (OPIS).
2. Program Risk Management and Program Risk Based Auditing

2.1 Program Risk Management Risk management is a rapidly developing discipline and many attempts have been made during last years towards standardization (Standards Australia/ Standards New Zealand, 2004; AIRMIC, ALARM, IRM, 2002; HM TREASURY, 2004). These efforts however are focused mainly on project risk management or in generic risk management, so there are very few references in program risk management. Actually, there are very few reports dedicated in program management in general, most of them using contradictory terminology. Additionally, the rapid development of the new project portfolio discipline often evokes confusion between the various terms. However, PMI has recently (2006) published two new standards, one for program management and one for portfolio management, contributing significantly to the clarification and standardization of the two disciplines. As for program risk management, the PMI (2006a) program management standard adopts a quite generic approach, allowing the flexibility of being applied in almost every activity / project and in every different type of organization, but on the other hand limiting the standardized approach. In any case, risk has different meanings to different people, and the concept of risk varies according to viewpoint, attitudes, and experience. Engineers, designers, and contractors often view risk from a technical perspective, while owners and developers tend to view it from the economic and financial side. Similarly, safety and health professionals take a more hazard perspective. Risk is therefore generally seen as an abstract concept where measurement is very difficult (Raftery 1999). In
182
program risk management case, the risk concept is considered even more abstract, since programs are usually governed by multi-level hierarchic structures, which consider risk from completely different perspectives. 2.2 Risk Based Auditing During the last years, there has been a rapid development of audit methodologies, due to explosion of public consciousness in recent scandals of late 90s (Enron, Parmalat, etc.). We are, for example, accustomed to hearing of medical audit, efficiency audit, effectiveness audit, performance audit, valuefor-money audit and environmental audit (Pentland, 2000). One of the most important developments in audit profession was the introduction of risk based auditing, which is about the incorporation of business risk into the assessment and planning of audit risk. In their majority, risk based approaches are driven by the simple conclusion that Business risk drives audit risks (Eilifsen et al., 2001). Stated simply, anything that had the potential to increase the risk that organization would not meet its objectives was also seen as a source of increased risk. The concept of risk was not new to auditing profession; however it was only with the development of broader perceptions of risk such as enunciated in the Committee of Sponsoring Organizations (1992) Framework, that auditors adopted a broader focus on risk management. Here, for the first time, auditors and accountants were presented with a view of risk and internal control that reflected something other than accounting errors. Similar initiatives later came out in Canada, Guidance on Control issued by the CICAs Criteria of Control Committee (1995), and the UK, the socalled Turnbull Code, issued by Institute of Chartered Accountants in England and Wales, 1999. Furthermore, the Committee of Sponsoring Organizations (2004) went a step forward by introducing the COSO Enterprise Risk Management Integrated Framework, which incorporated the internal control framework and provided a more robust and extensive focus on the broader subject of enterprise risk management. 2.3. Risk Ranking in Auditing Given the Risk Based Auditing development, there has been a corresponding development in risk ranking methodologies. For the implementation of risk ranking in an organizations activities / departments / projects, a risk assessment should be used, so that the whole population would be ranked accordingly to risk. According to the Institute of Internal Auditors (1991), the elements of the risk assessment process are as following: 1. Identification of the auditable activities; 2. Assignment of risk factors; 3. Risk assessment. In this paper, the examination is about large scale programs, thus auditable activities are actually all the projects activities. As a result, all funded projects are identified as auditable activities. For the assignment of risk factors and risk assessment, many approaches have been proposed. The more widespread include the traditional techniques, according to Ziegenfuss (1995) classification, the Wilson and Ranson (1971), the Patton et al. (1982), the Siers and Blyskal (1987), as well as the CERN (2006) techniques. All these techniques were developed mainly for auditing in business units and financial auditing. In the case of large scale program auditing, the proposed techniques include the ESF, ERDF (2005) model, the technique developed by Canada Western Economic Diversification (2006) and the Baccarini and Archer (2001) methodology. All the above mentioned techniques include risk factors identification, risk factor weighting and risk factor assessment, according to objective or subjective criteria, or according to a combination of objective and subjective criteria. Risk factor is a definitely different term from risk. According to AIRMIC, ALARM, IRM (2002), risk can de defined as the combination of the probability of an event and its consequences. Risk factors on the other hand, as used in the above mentioned
183
techniques and this paper, are factors (internal or external) that increase the probability of appearance or the impact of risks. Thus risk factors do not refer to events that could have consequences on the organisations objectives achievable, but they rather refer to circumstances / characteristics / attitudes that increase the probability of appearance or the impact of such events. Risk factor identification and weighting is quite different in the above mentioned techniques, depending on the author. For the risk factors assessment, simple numerical scales are used. The scoring model comprising weighted key criteria technique, using simple qualitative scales, such as 1-5, is quite renowned for project ranking; a similar model is also proposed by the new PMI (2006b) standard for a different ranking other than risk, but using exactly the same logic. All the above mentioned techniques are quite widespread with many practical applications; however they all use a limited set of risk factors, ranging from 4 to 13, resulting often to very close rankings. The biggest disadvantage however of all these techniques is that most of them are actually static, i.e. they would result in exactly the same rankings for exactly the same projects, every time they would be performed, regardless of the projects implementation progress. Only the ESF, ERDF (2005) model is an exception, since 2 out of the 13 risk factors vary depending the expenditure progress, but still the projects life cycle in not taken into consideration.
3. Proposed Risk-Based Projects Selection for Auditing in Large-Scale Programs

3.1 Defining the context The proposed methodology exploits risk management from the programs managing authorities perspective. Managing authorities are responsible for the efficient and effective implementation and management of programs. Among their most important responsibilities is the monitoring and auditing of the executed projects. Project monitoring and auditing activities include on-site auditing, as well as remote examinations and cross checking of projects important data. Such data include implementation or expenditure reports, transactions, tendering of documents, etc., which are often gathered by managing authorities. Auditing on the projects site, however, is of great importance, since it contributes to: Effective audit of the implementation and the expenditure progress, in relation both to the planning and to the data reported to the managing authority; Reasonable assurance that all expenditure funded is eligible and all transactions are denoted appropriately in financial reports; Reasonable assurance that all project risks are recorded and that effective controls are in place for their treatment; Reasonable assurance that emergency scenarios exist for a proper response to a major risk; Effective program risk management, since through project auditing the managing authorities have a better idea about the risks faced by projects or the program in general; thus, they have the opportunity to make specific proposals for improvements in any handicaps identified on the above issues. Given the significant importance of project auditing, many attempts have been made towards standardisation. In EU context, emphasis is given in statistical sampling (Commission of the European Communities, 2001; Commission of the European Communities, 2006), while in Canada a more enhanced Risk Based Audit Framework has been developed (Treasury Board of Canada Secretariat, 2003). By making a step forward, this paper proposes a Risk-based projects Auditing Selection Methodology (RASM) for large-scale programs. RASM goes through the following phases: 1. Risk Identification; 2. Risk Estimation; 3. Risk-based Auditing Feedback;
184
4. Risk-based Project Ranking; 5. Project Sampling. RASM was developed initially for Greek CSF III Programs; however it is our belief that the developed methodology can be applied in any large scale programs, with the following characteristics: The program organization chart includes a multi-level structure, in which the managing authority has the general coordination and management, whereas projects are implemented by different locally distributed bodies. Managing authorities collect, control and cross-examine projects progress data in a regular time basis. Managing authorities use a management information system, in which they enter all basic collected projects data. Project implementing bodies belong, regarding risk maturity, in the two last categories, according to Institute of Internal Auditors UK and Ireland (2003) classification, i.e. they belong either in the risk nave category, so there is no formal approach developed for risk management, or they belong to risk aware category and consequently there is only a scattered silo based approach to risk management. 3.2. Risk Identification This paper proposes a risk factor classification in 2 major categories; probability and impact factors. Probability factors are actually the circumstances/characteristics/ attitudes, which affect the probability of risk appearance. On the other hand, impact factors are related to the impact/consequences, which would occur, provided risks appear. In RASM only one impact factor is considered, which is corresponding to project entire budget. Probability factors could further be classified in 4 groups; project inherent risk factors, recipient risk factors, management risk factors and feedback risk factors. Project inherent risk factors are the factors existing in projects due to their nature, type or complexity, regardless of the recipient and its management adequacy. Recipient risk factors are related to risk elements associated with the recipient, such as entity type or recipient sufficiency, regardless of the nature of the project under consideration. The recipient risk should remain constant across all active projects with a given recipient. While recipient risk should be re-evaluated periodically, it is unlikely to change as frequently as management risk factors do. Management risk factors are associated with the recipients compliance to the terms and conditions of the contribution agreement. These risk elements capture risks associated with the progress and financial performance of the project and must be continually reassessed throughout the lifespan of the project. Feedback risk factors constitute a special category, referring to the RASM feedback with implication data. Additionally, risk factors could be classified in 2 extra categories, regarding the assessment procedure; the objective analysis risk factors and the subjective analysis risk factors. Objective analysis risk factors are associated with the general projects characteristics, thus they can be assessed objectively, regardless of the specific characteristics or the nature of the projects under consideration. For instance, risk factor Project Type is assessed with exactly the same way for all the projects characterized as new, regardless of their specific characteristics. Subjective analysis risk factors cant be assessed without deep knowledge of projects specific characteristics; thus, risk assessment can only be subjective in this case, based on experts experience and/or the conclusions of previous audits. Subjective risk analysis requires much more effort and a higher level of risk maturity than objective risk analysis and this is the reason why in the proposed methodology the use of subjective risk analysis is quite limited. In particular, two out of the twenty proposed risk factors are dealt with subjective risk analysis; Project
185
International Research Journal of Finance and Economics - Issue 11 (2007) Implementing Body sufficiency and Risk assessment during previous audit, which were selected due to their importance and easy manipulation.
Figure 1: Risk factor hierarchy in large scale programs
ANALYSIS Objective Subjective
PROJECT RISK FACTORS Project nature Project Type Project complexity: Project size in relation to Recipients other Project Inherent projects size Sub-projects included Sub-projects type Construction site ownership Technology used Recipient entity type PROBABILITY RISK FACTORS
Recipient related
Recipient sufficiency Number of projects implemented by the same Body Submission in the regular set time basis of progress reports and other required data Modifications of projects initial planning
Management related
Number of different entities involved throughout the project life cycle Procurement Procedures Project Performance: Achievement of milestones Time schedule delays Cost increases Time from previous audit
Feedback related
It should be noted that the above mentioned risk factors are generic and can be identified in all large scale programs; however, their exact values depend on the Program under examination.
International Research Journal of Finance and Economics - Issue 11 (2007) 3.3 Risk Estimation
186
Risk Estimation is the stage in which a project risk exposure measure can be estimated, combining the probabilities / impact of all risks. The risk estimation phase includes the following steps: Step 1: Risk Factor Weight Estimation Step 2: Risk Factor Assessment Step 3: Project Risk Exposure Estimation Step 1 During the first step, the risk factors weights are estimated. The weights are proposed to be different for the different phases of projects life cycle. This differentiation is especially useful, since it is clear that projects face different risks in different project life cycle phases. For instance, the identified risk factor construction site ownership is completely irrelevant for a project close to its termination, whereas the same risk factor is definitely essential for projects during their first phases. Considering that the projects under examination have already been approved for funding within a large scale program, this paper proposes a slightly different approach for the project life cycle definition, from the classic analysis proposed by Adams and Barndt (1988) or Chapman and Ward (1997). For instance, given that RASM does not deal with risks occurring during the projects approval procedures, the project conceptualization phase is out of the examination context. Respectively, all the planning phase tasks, which are usually considered as a prerequisite for the project approval within a program, are not examined. Furthermore, considering project life cycle, it should be noted that in large scale programs projects usually include many sub-projects, which in many cases could be considered as projects themselves. In that case, the sub-projects do not progress in parallel. Besides, the exact phase a project or a subproject goes through is not always information entered in the MIS used by managing authorities. Instead, managing authorities tend to put emphasis on economic data rather than technical details, through their MIS. In order to get over this complexity, RASM considers a project life cycle phases discrimination, simple enough and objectively derived from expenditure rates, as follows: Activation Phase. During this phase, the project has just been approved for the program funding, so it is either in a tendering phase or in the first stages of execution. The phase is objectively determined by 10%, in terms of expenditure rate, and includes the planning phase and the first stages of execution phase, as described by Adams and Barndt (1988). Execution Phase. This phase is defined in a quite similar way with the Chapman and Ward (1997) definitions. The start of this phase signals the start of order-of-magnitude increases in effort and expenditure. The phase is objectively determined, in terms of rate of expenditure, by 10% as the lower limit and 90%, as the upper limit. Termination Phase. This phase actually includes the deliver, review and support stages. In this phase, basic deliverables have to be verified and evaluated, the whole project has to be reviewed, and finally the operation entity has to undertake project deliverables and start operating, developing basic maintenance and support schedule. The phase is objectively determined, in terms of rate of expenditure, by 90% as the lower limit, or it can be defined regardless of expenditure rate, by the project implementing bodys project completion denotation. This phase may include the last stages of execution phase, as well as all the stages of termination phase, as described by Adams and Barndt (1988). Step 2 During the second step of risk estimation stage, RASM includes: Assessment of probability of risks appearance, due to each probability risk factor;
187
Assessment of impact of risks appearance, which actually involves assessment of the impact risk factor. For practical purposes, both probability and impact risk factors assessment is proposed to be conducted in a 1-5 qualitative scale, in which 1 denotes a minor risk and 5 a maximum risk. The fivepoint scale is proposed, since it is simple to comprehend, easy to use in a workshop environment and provides adequate discrimination. Besides, as Cooper et al (2005) note, the three-point and four-point scales do not always allow critical distinctions to be made, and scales with more than five points are often cumbersome to use in practice. Step 3 During the third step, the total Project Risk Exposure is estimated. More particularly, for project x Risk Exposure R x is denoted by: Rx =
(w
i
A ix )
w i stands for the weight of risk factor i; A ix stands for the assessment of risk factor i in project x. In case that the identified risk factors are applicable only in sub-project/contract level e.g. the risk factor sub-projects type, A ix assessment is denoted by: A ix = (( B xk / B x ) AS ixk )
k
B xk stands for the budget of sub-project/contract k of project x; B x stands for the total budget of project x; S ixk stands for the assessment of risk factor i in sub-project/contract k of project x.
3.4 Risk-based Auditing Feedback
Risk based auditing feedback is the phase in which the audit conclusions are used as a feedback to project risk exposure estimations. Auditing feedback is a very crucial procedure, since its practically the only way for all the assessments to be reviewed and refined, both for risk factors and weights assessments. Two audit feedback risk factors are considered in RASM. Time from previous audit is assessed objectively, while risk assessment during previous audit is assessed subjectively. Thats why the latter is proposed to be conducted in a clearly structured way, in which all auditors will assess in a 0-5 scale all the impacts of risks occurred during the project, according to the following criteria: The percentage of non eligible expenditure, in relevance to total expenditure; The delays percentage, in relevance to the initial schedule; The percentage of deliverables not acceptable or under question; The percentage of expenditure over the approved budget; The number of irregularities. Projects will be assessed accordingly with the criterion assessed with the higher number. In case that a project has not been audited, risk assessment during previous audit is proposed to be assessed as a high risk situation, since in that case managing authority is not aware if project is progressing as stated. Audit feedback factors are proposed to be assessed with different weights during the different projects life cycle phases. More specifically, it is proposed that audit feedback risk factors should not be included in the analysis during projects activation phase, since it is very unlikely for a project to have been audited during activation phase. On the contrary, during the two other phases, audit feedback risk factors are very important, and it is proposed that they should be assessed with a predefined weight of 20%, so that in case that audit feedback is very different than estimated project risk
188
exposure, project classification will be influenced radically. For example, in a case in which an audited project is considered to face practically no risks, it is assessed with a value equal to 0 in all criteria. Thus, the next time, after its audit, RASM is performed, it will be ranked in a risk category lower than the category it was originally ranked, assuming that all other risk factors will be exactly the same. As a result, besides reviewing assessments, feedback is also used as a means towards a more proper classification. 3.5. Risk-based Project Ranking During audit risk evaluation, all projects of a program are ranked according to their risk exposure. Considering that all estimations are conducted in a semi quantitative basis in a 1-5 scale, project ranking is proposed to be conducted in a corresponding 1-5 scale i.e. in 4 risk categories. For this purpose, risk factor weights should be normalized so that wi = 1
i
In this context, risk categories are defined according to risk exposure estimations, as follows:
Table 1: Risk Exposure Categories
Risk Exposure 1.00 2.00 2.01 3.00 3.01 4.00 4.01 5.00 Risk Category Low Risk Exposure Medium Risk Exposure High Risk Exposure Very High Risk Exposure Code L H VH
3.6 Risk-based Project Sampling Managing authorities should reassure through audits that no systematic risks / problems occur anywhere throughout their program. Consequently, an approach imposing audits only in projects ranked highly risky is not suggested, since it could result in undiscovered systematic faults in low risk projects. Besides, risk factors assessments were conducted as objectively as possible, but still they involve in remarkable extent subjectivity. As a result, auditing exclusively projects of high risk ranking, might exclude from auditing a project (or a whole project category) with some particular characteristics, which were underestimated either during the identification phase, or during the risk estimation phase. Consequently, it is proposed that managing authorities include in auditing samples projects of all risk categories. However, this does not imply that audits should be equally appointed to projects from all risk categories. On the contrary, risk-based project ranking is quite useful for audit effort allocation, i.e. the higher the risk ranking is, the higher the audit effort should be. Considering these facts, the audit effort is proposed to be appointed among risk categories, as follows: Projects of VH category should be all audited, since few projects are supposed to belong in this category. Furthermore, the ranking of a project in this category should be an alert. The effort available for audits, excluding the audit effort devoted to VH category, is proposed to be split throughout the rest of the categories. More particularly, 50% of that remaining available audit effort is proposed to be devoted to H category, 35% to M category and finally 15% to L category. It should be obvious, that if there are not many projects ranked in a specific category, so as to exploit all the provided effort, according to the above mentioned allocation, the remaining effort of this category should be allocated to the next critical one. Finally, RASM is depicted in Figure 2.
189

Figure 2: RASM: Risk-based projects Auditing Selection Methodology for large scale programs
4. Projects Risk Based Auditing Selection a Case Study in Greek Operational Program Information Society
Greek Operational Program Information Society (OPIS) is a Greek CSF III Program, with a very large number of incorporated projects. OPISs objectives include the full transition of Greece in the new emerging era of Information Society, through the exploitation of Information and Communications Technologies (ICT), without discriminations and with the protection of citizens rights (Hellenic Ministry of Economy and Finance, 2000). According to CSF III framework, OPIS is managed by a Managing Authority, especially founded for this purpose, under a Special Secretariat of the Hellenic Ministry of Economy and Finance. OPIS involves a large number of Recipients of almost all sectors throughout Greek territory, incorporating projects implemented by Ministries, Prefectures, Cities, Hospitals, Schools, Universities, Libraries and a large number of other public or private entities. The management of the projects is actually carried out by Recipients, whereas OPIS Managing Authoritys involvement is rather indirect, with the supervision of the projects progress, through the continuous contact and cooperation with the representatives of the Implementation Institutions. Through this process, OPIS Managing Authority verifies projects data, especially projects eligible expenditure, and under this verification, undertakes projects financing. Thus, one of the most important Managing Authoritys responsibilities is projects auditing, in order to ensure that the national and community resources are used correctly, under current legal framework, and efficiently for the implementation of the projects-subprojects.
190
OPIS incorporated projects up to now be exceeding 5,000, with a total expenditure of more than 2,400 million Euros. OPIS is structured in five Action Lines, whereas each Action Line is further structured in Measures. For proper project and program monitoring, a Management Information System (MIS), called Ergorama is used, which is supported by the Hellenic Ministry of Economy and Finance. Ergorama was specially designed for CSF III framework and it is used by Managing Authorities of all Operational Programs in Greece. As a result, Ergorama was not developed to meet the special needs of any of these Programs, but it was developed centrally for the management and administration of CSF III. All data necessary to risk factors assessments, as developed in paragraph 3, can be retrieved from Ergorama. For risk factor weighting and assessment, three workshops were conducted, in which managers, experts and consultants of OPIS Managing Authority participated. In the first workshop the methodology was presented and all the participants were asked to fill up questionnaires, about risk factors weighting and assessment. The participants were asked to pay attention to the risk factors differentiation, regarding project life cycle. In the second workshop, an extensive dialogue was coordinated amongst all the participants, for the review of the first round assessments. As a result, all assessments were finalized for the methodology initialization. Finally, during the third workshop, the conclusions from the implementation of methodology were discussed. In this study, we used an AHP approach for assigning risk factors weights. AHP is a widely used method for analyzing complex discrete alternative decision problems with multiple qualitative criteria (Saaty, 1988). In AHP, the decision problem is decomposed into a tree-like hierarchical structure, with the overall goal at the top and the discrete alternatives at the bottom. The intermediate levels of the hierarchy represent lower level criteria which contribute to the overall goal. In this process the decision-maker carries out only pairwise comparative judgments which are then used to develop overall priorities for ranking the alternatives. AHP has been applied to several auditing problems (Arrington et al., 1984; Bagranoff, 1989). In this study, we used AHP to establish the relative weights of risk factors as indicated by the nine participants of workshops. The nine experts were asked to indicate the relative importance of risk factors and risk factors categories, throughout projects life cycle. The pairwise comparisons were used to construct fifteen matrices which were used to compute weights which measure (1) the relative importance of each risk factor within each of the four probability risk factors groups; and (2) the relative importance of each risk factor group, including impact risk factor, to the overall goal of assessment of risk exposure. Each of the above mentioned matrices was constructed three times, for the weights computation throughout projects life cycle. Table 2 summarizes the average results of the AHP judgment model in terms of the mean weights with which risk factors were judged, throughout projects life cycle. The overall consistency ratio was also calculated for each respondent. In case that an experts judgment would have an inconsistency index greater than 0.1, that would mean that the specific expert would have to be excluded. In our case, there was not found an inconsistency ratio greater than 0.04, so all experts judgments participated in the final results.
191
Table 2:

Risk Factors Weights per Project Life Cycle
Activation Phase 0.291 0.051 0.056 0.061 0.032 0.014 0.046 0.031 0.149 0.026 0.081 0.042 0.358 0.019 0.021 0.032 0.106 0.083 0.074 0.023 0.000 0.000 0.000 0.202 0.202 Risk Factors Weights Execution Phase Termination Phase 0.198 0.170 0.046 0.047 0.049 0.050 0.023 0.013 0.028 0.018 0.024 0.021 0.000 0.000 0.028 0.021 0.136 0.136 0.025 0.025 0.078 0.078 0.033 0.033 0.269 0.293 0.017 0.019 0.021 0.019 0.032 0.043 0.043 0.036 0.079 0.081 0.036 0.016 0.041 0.079 0.200 0.200 0.047 0.047 0.153 0.153 0.197 0.201 0.197 0.201
Risk Factors Project Inherent Risk Factors Project nature Project Type Project size Sub-projects included in project Sub-projects / contracts type Construction site ownership Technology used Recipient Risk Factors Project Implementing Body entity type Project Implementing Body sufficiency Number of projects implemented by the same Body Management Risk Factors Regular Submission of reports Modifications of projects initial planning Number of different entities involved Procurement Procedures Achievement of milestones Time schedule delays Cost increases Feedback Risk Factors Time from previous audit Risk assessment during previous audit Impact Risk Factors Project Budget
The assessments of two indicative probability risk factors and the impact risk factor are depicted in Table 3.
Table 3:
Risk Factor Project Type
Indicative Risk Factors Assessment

Risk Factor Values New Extension Enhancement Cost Not Increased Cost Increased up to 5% Cost Increased up to 10% Cost Increased up to 20% Cost Increased more than 20% Budget belongs to (0, 100,000.00 ] Budget belongs to (100,000.00 , 500,000.00 ] Budget belongs to (500,000.00 , 1,000,000.00 ] Budget belongs to (1,000,000.00 , 5,000,000.00 ] Budget belongs to (5,000,000.00, ) Risk Assessment 5 3 2 1 2 3 4 5 1 2 3 4 5
Cost Increases
Project Budget
For the methodology implementation, an information system was developed, in which the appropriate data of Ergorama is loaded automatically, with an interface, and the auditors can enter the conclusions of all audits. Project ranking summary results for auditing purposes, as turned up through this process, are depicted in Table 4.

Table 4: Risk Ranking Summary
No of Projects Ranked 33 179 1187 3622
192
Risk Exposure Category VH: Very High Risk Exposure H: High Risk Exposure M: Medium Risk Exposure L: Low Risk Exposure
According to RASM, audits should be conducted in all projects of VH category. In projects of all other categories, sampling will be applied. Even in this case, sampling should not be statistical, since a representative sample should be intended for projects within the same risk category. Representativeness in this case is defined in terms of OPIS Action Lines and Measures. More particularly, samplings aim, regarding representativeness, was the participation in the sample of projects of all Measures and Action Lines, if possible. Each Measure or Action Line was decided to participate in the sample according to their expenditure weight in relevance to the total OPIS expenditure. Due to risk rankings and other limitations, a fully representative sample in such a way cannot be achieved; however the inclusion in the sample of projects of all Measures and Action Lines, in the most possible representative way, is quite useful. Finally, during the methodology results review conducted in the third workshop, all participants declared their satisfaction, denoting that the proposed sample was covering, according to their perception, all Programs risk areas. All participants agreed in the most risky projects, whereas some of them (two out of nine) had some objections about ranking in H and M category, considering that seven projects in total were mis-ranked.
5. Conclusions
This paper proposes RASM, an integrated risk based projects auditing selection methodology for large scale programs. RASM was initially developed for Operational Programs of CSF III in Greece; nevertheless it could be easily adapted to any large scale program with similar structure and characteristics. Most important, it gives enough flexibility in assessing risks in large scale programs, since it affectively remains insensitive to changes in risk factors identification. RASM is not static, i.e. it can be used throughout a Program life cycle, suggesting different samples in different time periods, regarding projects progress and probabilities fluctuation of risk appearance. Furthermore, by recognising the great importance of project life cycle in risk analysis, this paper proposes a differentiation in risk factor weighting, regarding projects life cycle status. For practical purposes, project life cycle is proposed to be characterized slightly differently from classical approaches, with 3 phases; Activation, Execution, Termination. RASM is covering all aspects of auditing sampling, proposing a risk ranking, and a sampling methodology, incorporating feedback from previous audits with a structured and simplified way. The conclusion derived from the implementation of RASM in real data of OPIS Program of CSF III in Greece is that RASM results are reasonable at first place. During a workshop held for this purpose, all participants agreed at the proposed sample and furthermore, most of them agreed totally with the entire projects risk classification. The only disadvantage of the proposed methodology was defined as the significant initial set up time, in which the initial weights and risk factors assessments have to be defined. On the other hand, this initial set up time is very important, since it allows a greater flexibility, for a more customized approach in risk factors relevant significance throughout projects life cycle. As a result, RASM offers a valuable tool to large scale program managing authorities, which, if used properly, can be a significant help towards their efforts for efficient project auditing. RASM offers a simple, user friendly context, in which a full computer aided risk analysis can be held in an extremely large number of projects, using a large number of risk factors.
193
Acknowledgement
This work is financially supported by the research project "Herakleitos" of EPEAEK II program. The project HERAKLEITOS is co-funded by the European Social Fund (75%) and Greek National Resources (25%).
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] Adams, J. R. and Barndt, S. E. (1988), Behavioral implications of the project life cycle, Chapter 10 in Cleland, D. I. and King, W. R. (eds), Project Management Handbook, Second edition. New York: Von Nostrand Reinold AIRMIC, ALARM, IRM (2002), A Risk Management Standard, London: AIRMIC (Association of Insurance and Risk Managers), ALARM (National Forum for Risk Management in the Public Sector) and IRM (Institute of Risk Management) Arrington, C.E., Hillison, W. and Jensen, R.E. (1984), An application of analytic hierarchy process to model expert judgment in analytic review procedures, Journal of Accounting Research, 22, pp: 298-312 Baccarini D. and Archer R. (2001), The risk ranking of projects: a methodology, International Journal of Project Management, 19 pp. 139-145 Bagranoff, N.A. (1989), Using analytic hierarchy approach to design internal control systems, Journal of Accounting and EDP, 4(4), pp: 37-41 CANADA Western Economic Diversification (2006), Report on the Audit of the Infrastructure Canada Program as at March 31, 2006, retrieved June 2007 at http://www.wd.gc.ca/rpts/audit/mgmt_phys_records/default_e.asp CERN (2006), Audit risk assessment methodology, retrieved June 2007 at http://internalaudit.web.cern.ch/internal-audit/method/risk_assessment. htm#objective Chapman, C. and Ward S. (1997), Project Risk Management: Processes, Techniques and Insights, Chichester: John Wiley & Sons Ltd Commission of the European Communities (2001), Commission Regulation (EC) No 438/2001 of 2 March 2001, Official Journal of the European Union, retrieved January 2007 at http://eur-lex.europa.eu/LexUriServ/site/en/oj/2001/ l_063/l_06320010303en 00210043.pdf Commission of the European Communities (2006), Commission Regulation (EC) No 1828/2006 of 8 December 2006, Official Journal of the European Union, retrieved May 2007 at http://eur-lex.europa.eu/LexUriServ/site/en/oj/ 2006/l_371/l_37120061227en00010163.pdf Committee of Sponsoring Organizations, (1992), Internal Control Integrated Framework, Jersey City: AICPA Committee of Sponsoring Organizations (2004), Enterprise Risk Management Integrated Framework, Jersey City: AICPA Cooper, D., Grey, S., Raymond, G., Walker, P. (2005), Project Risk Management Guidelines Managing Risk in Large Projects and Complex Procurements, Chichester: John Wiley & Sons Ltd Criteria of Control Committee (1995), Guidance on Internal Control, CICA Eilifsen, A., Knechel, W. R., & Wallage, P. (2001), Application of the Business Risk Audit Model: A field study. Accounting Horizons, Vol. 15 No.3, pp.193-207 ESF, ERDF (2005), Project Monitoring Model of Excellence retrieved March 2007 at http://www.erdf.communities.gov.uk/Repository/PDFs/MOEguide Hellenic Ministry of Economy and Finance (2000), Operational Program Information Society, retrieved December 2006 at http://www.infosoc.gr/infosoc/enUK/epktp/priority_actions/default111.htm HM TREASURY (2004), Orange Book: Management of Risk - Principles and Concepts, London: HMSO publications
[11] [12] [13] [14] [15] [16] [17] [18]
International Research Journal of Finance and Economics - Issue 11 (2007) [19] [20] [21]
194
[22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32]
Institute of Chartered Accountants in England and Wales (1999), Internal Control: Guidance for Directors on the Combined Code Institute of Internal Auditors (1991), Statement on Internal Auditing Standards No. 9, Risk Assessment, Altamonte Springs, FL: the Institute of Internal Auditors Institute of Internal Auditors, UK and Ireland (2003), Position Statement Risk Based Internal Auditing, The Institute of Internal Auditors UK and Ireland, London, retrieved at April 2007 from http://www.iia.org.uk/cms/IIA/uploads/ 48dc2e62-f2a7bd939a-7cef/RiskBasedInternalAuditing.pdf Patton, J.M., Evans, J.H. and Lewis, B.L. (1982), A Framework for Evaluating Internal Audit Risk, Altamonte Springs, FL: Institute of Internal Auditors Pentland, B. (2000), Will auditors take over the world? Program, technique and the verification of everything, Accounting, Organisations and Society, 25(3), pp. 307312 PMI, (2006a), The Standard for Program Management, Pennsylvania: Project Management Institute Inc PMI, (2006b), The Standard for Portfolio Management, Pennsylvania: Project Management Institute Inc. Raftery, J. (1999), Risk analysis in project management, London: E&FN SPON Saaty, T. L. (1988), Multicriteria Decision Making: the Analytic Hierarchy Process, Pittsburgh, PA: RWS Publications Siers, H., and Blyskal, J. (1987), Risk management of the internal audit function, Management Accounting, pp. 29-35 Standards Australia - Standards New Zealand (2004), AS/NZS 4360:2004 Risk management Standard, Standards Australia Ltd, Standards New Zealand Treasury Board of Canada Secretariat (2003), Risk-Based Audit Framework Guide, Version 4.3.1, retrieved March 2007 at http://www.tbs-sct.gc.ca/ia-vi/policies-politiques/rbafcvar/guide_e.pdf Wilson, D. and Ranson, R. (1971), Internal audit scheduling a mathematical model, The Internal Auditor, pp. 42-50 Ziegenfuss . D. (1995), The state of the art in internal auditing risk assessment techniques, Managerial Auditing Journal, Vol. 10 No. 4, pp. 3-11

1

Cargado por

Información del documento

Derechos de autor

Formatos disponibles

Compartir este documento

Compartir o incrustar documentos

Opciones para compartir

¿Le pareció útil este documento?

¿Este contenido es inapropiado?

Copyright:

Formatos disponibles

1

Cargado por

Copyright:

Formatos disponibles

INTERNATIONAL RESEARCH JOURNAL OF FINANCE AND ECONOMICS ISSN: 1450-2887

Issue 11 September, 2007

INTERNATIONAL RESEARCH JOURNAL OF FINANCE AND ECONOMICS

International Research Journal of Finance and Economics

163-170 172-178 180-193

On the Effect of Index Futures Trading on Stock Market Volatility

International Research Journal of Finance and Economics - Issue 11 (2007)

2. The Derivatives Market of the Athens Exchange

International Research Journal of Finance and Economics - Issue 11 (2007)

3. Methodology and Theoretical Approach

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

4. Description of Data and Preliminary Statistics

FTSE-20 FTSE-20 FTSE-20

International Research Journal of Finance and Economics - Issue 11 (2007)

FTSE-20 S&P FTSE DAX DJ

FTSE-20 FTSE-20 FTSE-20

FTSE-20 FTSE-20 FTSE-20

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

3. Data Selection and Descriptive Statistics

DIVt Et DIVt-1 FDIVt-1 ADIVt-1 DIVt Et DIVt-1 FDIVt-1 ADIVt-1

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

-0.012 0.281 0.011

0.185 0.000 0.016

International Research Journal of Finance and Economics - Issue 11 (2007)

0.003 0.934 0.041

0.279 0.085 0.155

International Research Journal of Finance and Economics - Issue 11 (2007)

Finance Growth Nexus: Evidence from Turkey

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

2. Financial Development and Growth

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

3. Model Specification, Data and Methodology

(2) (3) (4)

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

-0.13 -2.49 -5.91 -2.84 -3.51 -2.89

-1.29 -2.83 -2.60 -1.95

yt 1 . F-test statistics, 1 and 3 , are to test joint hypotheses

= a0 = 0 and = a2 = 0 , respectively. k is number of lags.

Cointegrating Vectors r=0 r1 r=0 r1 r=0 r1 r=0 r1

International Research Journal of Finance and Economics - Issue 11 (2007)

40 [49] [50] [51]

Volatility Clustering in the Greek Futures Market: Curse or Blessing?

International Research Journal of Finance and Economics - Issue 11 (2007)

International Research Journal of Finance and Economics - Issue 11 (2007)

2. Data and Methodology

International Research Journal of Finance and Economics - Issue 11 (2007)

n 1 where the mean daily return stands for R = (1 n ) R t

International Research Journal of Finance and Economics - Issue 11 (2007)

Mean Median Maximum Minimum Standard deviation Skewness Kurtosis JB Q(12)

*Statistically significant at the 5% level

International Research Journal of Finance and Economics - Issue 11 (2007)