The Journal of Economic Asymmetries 11 (2014) 30–45

Contents lists available at ScienceDirect

The Journal of Economic Asymmetries

www.elsevier.com/locate/jeca

Some new evidence on the determinants of money demand in

developing countries – A case study of Tunisia ✩
Ousama Ben-Salha a,b,∗ , Zied Jaidi c
a
National Engineering School of Gabes, Rue Omar Ibn-Elkhattab, 6029, University of Gabes, Tunisia
b
International Finance Group-Tunisia, University Tunis El Manar, Tunisia
c
Central Bank of Tunisia, Tunisia

a r t i c l e i n f o a b s t r a c t

Article history: The current article aims to estimate the money demand function in Tunisia. Unlike many
Received 18 February 2014 previous money demand studies, the major components of real income are considered.
Received in revised form 1 June 2014 Based on annual data ranging between 1979 and 2011 and the ARDL bounds testing
Accepted 3 June 2014
approach, results reveal evidence of cointegration between the broad money demand
Available online xxxx
and its determinants, namely the final consumption expenditure, the expenditure on
Keywords: investment goods, the export expenditure and the interest rate. The error correction model
Broad money demand shows that the demand for money is only affected by the interest rate and the expenditure
Expenditure components on investment goods in the short-run, while in the long-run the final consumption
ARDL modeling expenditure and the interest rate represent the major money demand determinants.
Structural change These findings are robust to a variety of alternative money demand specifications and
Tunisia estimation methods. The Saikkonen–Lütkepohl cointegration test with structural shift and
the Johansen–Mosconi–Nielsen structural break cointegration test are performed in order
to control for structural change. In addition, the stability of the relationship is checked
using the Chow stability test and the Hansen parameter instability test. In the light of the
study, we advance that monetary policy in Tunisia should be based on a broad definition
of money. Furthermore, the estimation of money demand functions must take into account
the different expenditure components of real income.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction

The conduct of monetary policy is a debated topic in the economic literature. The estimation of money demand functions
has particularly attracted the attention of economists. Generally, the objectives consist in presenting the main determinants
of the demand for money in closed and/or open economies and checking the stability of money demand functions. The
demand for various monetary aggregates is often linked to a scale variable measuring the economic activity such as the in-
come and a variable representing the opportunity cost of holding money such as the domestic interest rate. Econometrically,
techniques that allow distinguishing the short-run effects from those of the long-run, such as the error-correction modeling,
are usually employed. Empirical studies have approximately covered countries from all over the world, despite some regions

✩
The views and opinions expressed in the article are those of the authors and do not necessarily reflect the views or policies of the Central Bank of
Tunisia.
*
Corresponding author at: National Engineering School of Gabes, Rue Omar Ibn-Elkhattab, 6029, University of Gabes, Tunisia. Tel.: +216 21673706.
E-mail addresses: oussama.bensalha@isgs.rnu.tn (O. Ben-Salha), zied.jaidi@gmail.com (Z. Jaidi).

http://dx.doi.org/10.1016/j.jeca.2014.06.001
1703-4949/© 2014 Elsevier B.V. All rights reserved.
 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45 31

received more attention than others. With regards to this point, a review of the empirical literature shows that few studies
focused on the Tunisian economy. These include Simmons (1992), Treichel (1997) and Boughrara (2001). In line with the
majority of empirical investigations on the subject, the above studies adopted the traditional approach, where the demand
for a specified monetary aggregate depends on a scale variable and an opportunity cost measure. However, the different
components of income may differently affect the demand for money (Ziramba, 2007). Checking the impact of aggregate
income on money demand may be considered as a methodological limitation that may hide the impact of each expenditure
component. To be more accurate when addressing policy recommendations, it is important to consider the different impacts
of expenditure components in both the short- and long-run.
Throughout this article, we attempt to add some fresh empirical evidence to the debate. The present case study is differ-
ent from previous ones on money demand since it estimates the short- and long-run impacts of different macroeconomic
components of real income, as well as the interest rate, on the demand for M2 balances. To the best of our knowledge, no
previous empirical research estimated the effects of disaggregated real income on money demand in Tunisia.1 In addition,
we employ the ARDL bounds testing approach for cointegration proposed by Pesaran, Shin, and Smith (2001) given its su-
periority to other cointegration techniques, especially in the case of small sample studies, such as the present. Furthermore,
a special attention is given to the potential structural changes that might have occurred during the period and the stability
of money demand. This is motivated by the occurrence of economic reforms and events that might distort the equilibrium
relationship. The choice of Tunisia as a case study is motivated by many reasons. In fact, Tunisia has undertaken serious
economic reforms such as the opening-up of the economy to foreign investors, the creation of new financial instruments,
the partial openness of the capital account and the access to international financial markets. This study allows checking the
determinants of money demand and the stability of the money demand function in an unstable macroeconomic environ-
ment. In addition, the current study might guide the policymakers when conducting the monetary policy. In the light of
the study, one may determine factors on which the policymakers may act in order to reach the objectives of the monetary
policy.
The rest of the article is structured as follows. In Section 2, we survey the related empirical literature focusing on
Tunisia on the one hand, and splitting the real income into expenditure components on the other hand.2 Section 3 briefly
discusses the conduct of the monetary policy in Tunisia. The model specification, data, and econometric issues are discussed
in Section 4. Sections 5 and 6 present empirical findings and a number of robustness and sensitivity checks, respectively.
Finally, conclusions and some policy implications close the article.

2. Selected empirical literature

In spite of the boom in empirical works on the demand for money in developing countries during previous years, papers
focusing on the Tunisian case received a little attention.3 These few empirical studies are based on the conventional theory
of money demand relating the volume of the demanded money to a scale variable that reflects the level of transactions
in the economy (such as real income) and a variable that represents the opportunity cost of holding money (such as the
interest rate or the inflation rate). Simmons (1992) investigates the demand for narrow money (M1) in five African countries
(Congo, Côte d’Ivoire, Mauritius, Morocco and Tunisia) using an error-correction model. Based on annual data covering the
period 1962–1989, results associated with the Tunisian case show that the demand for money, the real income, the discount
rate and the price level converge to a long-run equilibrium relationship. The author concludes also that only real income
plays a statistically significant role in explaining the demand for real narrow money in the long-run. Finally, real income
and inflation rate are found to be significant money demand determinants in the short-run. The same issue has been
discussed by Treichel (1997) using both annual and monthly data between 1962 and 1995 and the Johansen cointegration
framework. The author concludes that real M2 is cointegrated with real income, but not with the money market rate or
the rediscount rate. This result is also supported by the error-correction model, since the error-correction term is negative
and statistically significant. The estimated income elasticity over the whole period is about 0.80. It has been also shown
that the cointegrating relationship has been stable, especially over the period 1962–1990. The income elasticity over that
period is twice higher than the one associated with the entire period (1962–1995). The author attributes these findings
to the reduced demand for M2 over the period 1990–1995, due essentially to the introduction of Treasury bills in 1990.
These results are confirmed econometrically using quarterly data over the period 1990–1995 since a long-run relationship
between the demand for M2, income and the Treasury bill rate is found. However, the income elasticity dramatically falls
and is lower than the one found over the whole period.
Arize and Shwiff (1998) estimate a money demand function in 25 developing countries using annual data covering the
period 1960–1990 in the case of Tunisia. Money demand functions were augmented by two variables representing the
exchange rate dynamics: the official exchange rate and the black market exchange rate. Empirical results suggest that real
broad money demand is cointegrated with the real income, the interest rate, the inflation rate and the exchange rate. The

1
Few authors disaggregated real income when estimating money demand functions. This has been especially done by Tang (2002, 2004, 2007) in several
times for some Asian economies and Ziramba (2007) for South Africa.
2
Giving the abundance of the literature on the subject, we do not report the theoretical background on the determinants of money demand. In addition,
we limit the literature survey to empirical studies focusing on Tunisia and on those splitting the real income.
3
Table A.2 in Appendix A summarizes some selected empirical works on the subject.
32 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45

same results are revealed for the case of narrow money. The dynamic OLS technique suggests that real income, interest
rate and either the official exchange rate or the black market exchange rate are the main long-run determinants of money
demand. In addition, the long-run income elasticity is greater than the unity in all cases. Finally, based on Fair (1987) and
Davidson and Mackinnon (1981) procedures, the authors conclude that the introduction of the black market exchange rate
is more relevant than the official exchange rate in developing countries.
Arize, Malindretos, and Shwiff (1999) focus on the same issue by estimating the money demand function in 12 devel-
oping countries including Tunisia. Besides the introduction of the traditional determinants of money demand, such as the
income and the interest rate, the authors use variables measuring the degree of openness of the economy, such as the ex-
change rate, the exchange rate variability and the foreign interest rate. Empirical findings reveal that a long-run equilibrium
relationship exists between either real M1 or real M2 balances and their determinants. With respect to real M2 demand, the
estimated long-run parameters are about 1.25, −0.01 and −0.03 for real income, interest rate and exchange rate variability,
respectively. Boughrara (2001) sheds light on the impact of structural reforms undertaken starting from the 1980s on the
broad money demand in Tunisia. The study employed quarterly data covering the period between the first quarter of 1987
and the second quarter of 1992. It has been shown that the demand for M2 monetary aggregate is cointegrated with real
income, the Treasury bill interest rate and the special deposits interest rate. In the long-run, all variables affect the demand
for money. The long-run income elasticity is found to be close to the unity and is higher than the one of the short-run.
The Chow test and the recursive regression method suggest that the money demand function in Tunisia is stable over the
sample period.
An important methodological shortcoming arising from the empirical literature presented above consists in the fact that
no study attempted to estimate the impact of expenditure components on money demand. In fact, it is crucial to check
the impact of each component of real income on the demand for money. This allows determining which of them affect
the demand for money. A survey of the empirical literature suggests that few papers made such a disaggregation when
estimating the demand for money function. Tang (2002) focuses on the determinants of money demand in Malaysia using
the ARDL bounds testing approach developed by Pesaran et al. (2001). The author finds evidence that the demand for M3
monetary aggregate and its determinants are cointegrated. The cumulative sum of recursive residuals (CUSUM) and the
cumulative sum of squares of recursive residuals (CUSUMSQ) tests show that the relationship has been stable over time.
The estimated long-run coefficients are about 0.98 for the final consumption expenditure, −0.48 for the expenditure on
investment goods, 0.94 for the expenditure on export of goods and services, −1.39 for the exchange rate and 0.03 for the
interest rate. In the short-run, the money demand is only affected by the exports expenditure and the exchange rate. The
author confirms that income components exert different impacts on the demand for money and concludes the bias of using
a single real income variable. Tang (2004) estimates a money demand function based on Japanese quarterly data covering
the period between the first quarter of 1973 and the second quarter of 2000. The author shows that the demand for broad
money, as defined by the sum of M2 monetary aggregate and certificates of deposit, has been stable during that period.
The study confirms the existence of a long-run relationship between the demand for broad money, the final consumption
expenditure, the expenditure on investment goods, the exports expenditure, the deposit rate and the government bond yield
rate. In the long-run, the expenditure on investment goods is the most important long-run determinant of the Japanese
broad money demand with elasticity equal to 1.07. However, in the short-run, all variables are statistically significant at the
10% significance level.
Ziramba (2007) estimates a money demand function by considering various components of real income in South Africa
over the period 1970–2005. The research is based on the demand for M1, M2 and M3 balances. Even if long-run results
depend on the used monetary aggregate, the interest rate and the expenditure on investment goods are found to be the
common determinants of the demand for all monetary aggregates. The last paper that disaggregated real income accord-
ing to expenditure components is the one of Tang (2007) studying the case of five Southeast Asian countries (Malaysia,
Philippines, Thailand, Indonesia and Singapore). The study is based on annual data with sizes ranging between 34 and 45
observations. Empirical results show that the demand for M2 balances is cointegrated with its determinants only for three
countries, that is, Malaysia, Philippines, and Singapore. For two of these countries, the final consumption expenditure and
the exports expenditure are found to be long-run determinants of money demand, while the interest rate exerts negative
and significant impact only in Philippines. In addition, the CUSUM and CUSUMSQ tests suggest the stability of the estimated
parameters. The author concludes that M2 is the right instrument to be targeted in order to conduct monetary policy in
Philippines, Singapore and Malaysia.

3. A brief overview of the monetary policy framework in Tunisia

Like many developing and emerging countries, Tunisia has undertaken a series of financial reforms in mid-1980s. These
reforms came just after the occurrence of the balance of payment crisis in 1986.4 Accordingly, Tunisia signed a Stand-by
Agreement with the International Monetary Fund in 1986 and adopted a multitude of structural adjustment and stabilization
programs. The objectives of such programs are the liberalization of interest rates, the improvement of banking supervision
and the development of new market-based instruments of monetary policy (Treichel, 1997).

4
For details on the main features of the economic crisis that occurred in 1986, please see Bechri and Naccache (2006).
 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45 33

The announced objectives of the monetary policy implemented by the Central Bank of Tunisia are to minimize the infla-
tion rate and to stabilize the real effective exchange rate. As pointed out by Boughrara (2001), the M2 monetary aggregate
is considered as the intermediate target of the monetary policy since the adoption of reforms. The intermediate objective
of the Central Bank of Tunisia is to link the growth of the money supply with the growth of the economic activity. The
aim is to keep the growth of M2 monetary aggregate at 2% below the projected nominal GDP growth. However, Treichel
(1997) highlights that the difference between the growth of money supply and the growth of nominal GDP has continually
fluctuated and rarely reached the target of 2%, due essentially to the occurrence of structural changes touching the economic
activity. According to Boughrara, Boughzala, and Moussa (2008), the most important instrument of money control on behalf
of the Central Bank of Tunisia consists in the injection or the withdrawal of the central bank money.5 The nature of the in-
tervention is determined according to the difference between the targeted level and observed level. The intervention of the
Central Bank of Tunisia in the money market has been especially intensified in 1997, just after the removal of restrictions
imposed on commercial banks (International Monetary Fund, 2005). Regarding the exchange rate policy, the aim has been
to stabilize the real effective exchange rate to preserve the external competitiveness of Tunisia. Fanizza, Laframboise, Martin,
Sab, and Karpowicz (2002) admit the efficacy of the constant real exchange rate rule followed by Tunisia during the 1990s,
since it allowed maintaining a competitive position relative to its main trade partners.

4. Empirical issues

4.1. Model specification and data

The demand for money is generally expressed in terms of a scale variable, such as the real income, and a variable
representing the opportunity cost of holding money, such as the interest rate or the inflation rate. In the money demand
function, the real domestic income and the opportunity cost of holding money are both important variables, because the
efficacy of the monetary policy is highly dependent on the responsiveness of their elasticities. The demand for money is
generally expressed as follows:
 
M
= f ( y, i) (1)
P
where real money demand (M / P ) depends on real income ( y) and a variable measuring the opportunity cost of holding
money (i). Following Tang (2002, 2004 and 2007) and Ziramba (2007), the current study splits real income into its compo-
nents, namely investment, consumption and export expenditures. The decomposition of real income and the application of
the natural logarithm give the following money demand function:

ln M2t = γ0 + γ1 ln FCEt + γ2 ln EIGt + γ3 ln EXPt + γ4 it + εt (2)

where M2, FCE, EIG, EXP and i stand for the demand for broad money, the final consumption expenditure, the expenditure
on investment goods, the expenditure on total exports of goods and services and the interest rate, respectively. εt is the
error term and ln is the natural logarithmic transformation. As in Narayan and Narayan (2008) and Avouyi-Dovi, Drumetz,
and Sahuc (2011), the interest rate is not in the logarithmic form. All variables introduced in Eq. (2) are in constant local
currency. Based on theory, it is expected that the signs of γ1 , γ2 and γ3 to be positive, while γ4 negative. The study is
based on annual data ranging between 1979 and 2011. The use of annual time series is essentially due to unavailability of
long period quarterly data, particularly for the various components of real domestic income. Full definitions and sources of
time series introduced in the empirical investigation are given in Table A.1 in Appendix A.
The choice of M2 monetary aggregate to carry out this study is not arbitrary. Bahmani-Oskooee and Techaratanachai
(2001) suggest that the use of the M2 monetary aggregate is more appropriate in formulating monetary policy. In addition,
as pointed out previously, the Central Bank of Tunisia primarily considers the M2 monetary aggregate as an intermediate
target when conducting the monetary policy (Boughrara, 2001; Mohamed Sghaier & Abida, 2013). Treichel (1997) indicates
that the broad money is a controllable and operational target in Tunisia and thus, it may be considered as an intermediate
target.6 Finally, as indicated in Table A.2, four studies out of five focusing on Tunisia used the M2 monetary aggregate as a
dependent variable when estimating the money demand function.

4.2. The econometric methodology

Compared to conventional cointegration techniques such as the Engle–Granger two-step technique (Engle & Granger,
1987) and the Johansen and Juselius technique (Johansen & Juselius, 1990), the ARDL bounds testing approach introduced
originally by Pesaran and Shin (1999) and then extended by Pesaran et al. (2001) presents some major advantages. First,
the conventional techniques of cointegration require that variables introduced in the regression are integrated of order one,

5
For a detailed analysis on the conduct of monetary policy in Tunisia, please see Boughrara (2007).
6
Arize and Shwiff (1998) conclude also that M2 is preferred to M1 when estimating the money demand function in 25 developing economies, including
Tunisia.
34 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45

while the ARDL bounds testing approach may be implemented regardless of the stationary properties of variables (integrated
of order zero, order one or fractionally integrated). Accordingly, this technique eliminates the uncertainty associated with
the order of integration (Ben-Salha, 2013). Second, it may be applied in small sample sizes, whereas the Engle–Granger
or the Johansen and Juselius procedure is not consistent for relatively small samples (Duasa, 2007; Akpan, 2011). The
current study is based on annual data ranging between 1979 and 2011, which makes the ARDL approach more suitable.
Bahmani-Oskooee and Gelan (2008) report that the ARDL bounds testing approach is the most appropriate technique for the
estimation of money demand functions in developing countries. Third, the approach provides unbiased long-run estimates
and valid t statistics even if some of regressors are endogenous (Paul, Salah Uddin, & Noman, 2011; Odhiambo, 2009).
Finally, Hammoudeh and Sari (2011) suggest that when using the ARDL approach, the derived error-correction model is
obtained through a simple linear transformation.
In order to implement of the ARDL bounds testing approach, the money demand function presented in Eq. (2) is trans-
formed as follows:

p

p

p

p
 ln M2t = α0 + ϕ j  ln M2t − j + ψk  ln FCEt −k + λm  ln EIGt −m + κn  ln EXPt −n
j =1 k =0 m =0 n =0


p
+ ηh it −h + δ1 ln M2t −1 + δ2 ln FCEt −1 + δ3 ln EIGt −1 + δ4 ln EXPt −1 + δ5 it −1 + εt (3)
h =0

where  is the first difference operator. To examine the evidence for a long-run relationship between ln M2t , ln FCEt , ln EIGt ,
ln E X t and i t , Pesaran et al. (2001) propose the bounds test conducted based on the Wald test (F -test). The F -test is a test
where the null hypothesis is the absence of cointegration among variables against the presence of cointegration as an
alternative hypothesis, both denoted as:

H 0: δ1 = δ2 = δ3 = δ4 = δ5 = 0 i.e., the absence of cointegrating relationships

H 1 : δ1 = 0; δ2 = 0; δ3 = 0; δ4 = 0; δ5 = 0 i.e., the existence of cointegrating relationships
It is important to note that the asymptotic distribution of the F -statistic is not standard under the null hypothesis of no
cointegration, irrespective of whether the explanatory variables are purely I (0) or I (1). According to Narayan and Narayan
(2005), the two critical bounds values (lower and upper) computed by Pesaran and Pesaran (1997) depend on three main
criteria: (i) The integration order of explanatory variables (I (0) or I (1)); (ii) The number of explanatory variables; and (iii)
The inclusion of an intercept and a trend or of only an intercept. The decision on the existence of cointegration is based on
the comparison between the computed F -statistic and the bounds critical values. For instance, for a given significance level,
if the computed F -statistic is higher than the upper critical bounds value then the null hypothesis for no cointegration is
rejected. When the F -statistic lies between the upper and lower critical values, no conclusive decision on the existence of
cointegration would be advanced. Finally, if the F -statistic falls below the lower critical bounds value, the null hypothesis of
no cointegration cannot be rejected. Once a cointegrating relationship between the demand for money and its determinants
is found, we estimate the long-run elasticities using the following equation:


u 
q

r 
s 
v
ln M2t = γ0 + γ1 ln M2t − j + γ2 ln FCEt − j + γ 3 ln EIGt − j + γ4 ln EXPt − j + γ5 it − j + νt (4)
j =1 j =0 j =0 j =0 j =0

Finally, the estimation of the short-run dynamics is done by estimating the error-correction model associated with the ARDL
model. It assumes the following form:


n 
n 
n
 ln M2t = γ0 + γ1  ln M2t − j + γ2  ln FCEt − j + γ 3  ln EIGt − j
j =1 j =0 j =0


n 
n
+ γ4  ln EXPt − j + γ5 it − j + γ6 εt −1 + νt (5)
j =0 j =0

where εt −1 is the one period lagged error-correction term. This term measures the speed of adjustments towards the
long-run equilibrium relationship if short-run shocks occurred. In addition, a negative and statistically error-correction term
confirms the results of the F -test concerning the existence of cointegration between variables (Bahmani-Oskooee & Rehman,
2005).
Despite the ARDL bounds testing approach is the main technique used to estimate the money demand function, a special
attention will be given to the potential existence of structural changes in the relationship between the demand for money
and its determinants. To do this, we will re-estimate the money demand function using the Saikkonen and Lütkepohl
cointegration test with structural shift (Saikkonen & Lütkepohl, 2000a, 2000b, 2000c) and the Johansen–Mosconi–Nielsen
structural break cointegration test (Johansen, Mosconi, & Nielsen, 2000). More details on these approaches will be presented
in Section 6.
 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45 35

Table 1
PP and DF-GLS unit root tests.

Variable Phillips–Perron Dickey–Fuller GLS

Intercept Trend and intercept Intercept Trend and intercept
Level
ln M2 0.877 −0.977 0.746 −1.519
ln FCE −1.145 −2.297 −0.121 −1.773
ln EIG −0.326 −1.743 −1.488 −2.275
ln EXP −0.674 −1.767 −0.135 −1.664
i −0.592 −2.543 −0.590 −1.743

First difference
ln M2 −4.093∗∗∗ −4.155∗∗ −4.018∗∗∗ −4.216∗∗∗
ln FCE −3.700∗∗∗ −3.649∗∗ −0.757 −1.755
ln EIG −3.498∗∗ −3.433∗ −4.163∗∗∗ −4.115∗∗∗
ln EXP −4.633∗∗∗ −4.577∗∗∗ −4.321∗∗∗ −4.690∗∗∗
i −4.479∗∗∗ −4.787∗∗∗ −4.479∗∗∗ −5.005∗∗∗

Critical values 1% −3.653 −4.273 −2.641 −3.770

5% −2.957 −3.557 −1.952 −3.190
10% −2.617 −3.212 −1.610 −2.890
Notes: The null hypothesis is the existence of unit root in variables. Critical values are from Mackinnon (1996). ∗ ∗ ∗ , ∗∗ and ∗ denote the rejection of the
null hypothesis at the 1%, 5% and 10% significance levels, respectively.

Table 2
Lumsdaine–Papell Two-Break unit root test.

Variable Intercept TB1/TB2 Trend and intercept TB1/TB2

Level
ln M2 −4.256 1985/1990 −4.611 1987/1996
ln FCE −4.941 1986/1999 −5.086 1992/2008
ln EIG −5.526 1993/2008 −6.546∗ 1992/1997
ln EXP −4.196 1987/1996 −4.853 1986/2006
i −4.459 1989/1992 −6.084 1989/2000

First difference
ln M2 −8.570∗∗∗ 1989/1998 −8.501∗∗∗ 1994/1998
ln FCE −6.180∗∗ 1989/1997 −6.912∗∗ 1989/2007
ln EIG −7.310∗∗∗ 1991/1997 −7.064∗∗ 1988/1996
ln EXP −6.163∗∗ 1986/1989 −7.033∗∗ 1989/2008
i −6.520∗∗ 1985/1992 −8.050∗∗∗ 1989/1995

Critical values 1% −6.740 −7.1900

5% −6.160 −6.750
10% −5.890 −6.480
Notes: TB1 and TB2 stand for the first and second time break, respectively. Critical values are from Lumsdaine and Papell (1997). The null hypothesis is the
existence of unit root in variables. ∗ ∗ ∗ , ∗∗ and ∗ denote the rejection of the null hypothesis at the 1%, 5% and 10% significance levels, respectively.

5. Empirical findings

5.1. Integration and cointegration analyses

Before testing the presence of long-run relationships between the demand for M2 balances and its determinants, we
have to study the stationary properties of variables introduced in the model. Even if the ARDL approach allows testing the
cointegration in the presence of variables characterized by different orders of integration, it is not possible to implement
it if some of them are integrated of order two or above.7 For this purpose, we apply two traditional unit root tests: the
Phillips–Perron and Dickey–Fuller GLS unit root tests. In addition, due to the possible existence of structural changes that
might have occurred in the period under study,8 we employ the unit root test proposed by Lumsdaine and Papell (1997)
which allows testing the stationarity status of variables in the presence of two endogenous structural breaks. The test may
be considered as an extension of the Zivot–Andrews unit root which only considers a single endogenous structural break. In
Table 1, we report results of the Phillips–Perron and Dickey–Fuller GLS unit root tests, while Table 2 summarizes results of
the Lumsdaine–Papell structural break unit root test. The unit root tests are performed at level and at first difference with
an intercept and with an intercept and a trend term.
As shown, the Phillips–Perron unit root test indicates that all variables are not stationary in level and stationary in
first difference. Results of the Dickey–Fuller GLS test almost confirm those of the Phillips–Perron test. With regards to the

7
This is due to the fact that there is no provision for I (2) in the critical values for bounds testing approach.
8
We thank a reviewer for bringing this point to our attention.
36 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45

Table 3
F -statistics for cointegration relationships.

Model F -statistic Asymptotic critical values

95% 99%
I (0) I (1) I (0) I (1)
F ln M2 (ln M2| ln FCE, ln EIG, ln EXP , i ) 6.474∗∗∗ 2.947 4.088 4.093 5.532
F ln FCE (ln FCE| ln M2, ln EIG, ln EXP , i ) 4.568∗∗ 2.947 4.088 4.093 5.532
F ln EIG (ln EIG| ln FCE, ln M2, ln EXP , i ) 3.811 2.947 4.088 4.093 5.532
F ln EXP (ln EXP | ln FCE, ln M2, ln EIG, i ) 3.050 2.947 4.088 4.093 5.532
F i (i | ln FCE, ln M2, ln EXP , ln EIG) 4.067 2.947 4.088 4.093 5.532

Notes: The Akaike information criterion is used to select the optimal lag length for the ARDL model. The critical values are obtained from Narayan (2005),
p. 1987, Case II, n = 35. ∗∗∗ and ∗∗ denote the rejection of the null hypothesis of no cointegration at the confidence level of the 1% and 5% levels,
respectively.

Lumsdaine–Papell structural break unit root test, Table 2 shows that variables are integrated of order one at 1% or 5%
significance levels. In addition, the test provides two break dates for each variable. These structural break dates provide
valuable information related to the occurrence of macroeconomic or policy events in the country.
The test reveals that almost all variables present a structural break in mid-1980s. This time period coincides with the
occurrence of the most harmful economic and financial crisis in Tunisia and the intervention of the International Monetary
Fund by implementing the structural adjustment program in 1986. To summarize, the unit root tests suggest that variables
used in the empirical investigation are all integrated of order one. Therefore, the ARDL bounds testing approach may be
applied to test the existence of cointegration relationships. In Table 3, we report F -statistics calculated when each variable
is taken as a dependent variable, which means that we have five different specifications. As mentioned previously, the
bounds testing approach to cointegration involves the comparison of the F -statistics against the tabulated critical value
bounds.
Results suggest that when the real money demand is considered as the dependent variable, the computed F -statistic
exceeds the Narayan’s (2005) upper bound critical value at the 1% significance level. On the other hand, when the rest of
variables are taken as dependent variables, only the F -statistics in the consumption model is higher than the 5% upper
critical bound value. Consequently, the use of the ARDL modeling confirms that the null hypothesis of no cointegration
cannot be rejected when the expenditure on investment goods, the expenditure on export of goods and services and the
interest rate are treated as dependent variables. However, as shown in the table, there is strong evidence that a cointe-
gration relationship between real broad money demand and its determinants exists. In other words, there is a long-run
equilibrium relationship between the demand for broad money on the one hand and the components of final expenditure
(final consumption expenditure, expenditure on investment goods and expenditure on exports) and the interest rate on the
other hand. It is important to mention that, given the short period on which this study is based, we used the critical values
generated by Narayan (2005) rather than those of Pesaran and Pesaran (1997). The next step consists in the estimation of
long-run relationships and short-run dynamics between the demand for money and its determinants.

5.2. Long- and short-run coefficients

Having confirmed the existence of a long-run relationship between the demand for broad money and its determinants,
long-run elasticities associated with variables of interest are estimated using Eq. (4). The lag structure is selected on the
basis of the Akaike information criterion. Results are reported in Table 4, Panel A.
The final consumption expenditure and the interest rate are found to be the only significant variables that explain the
demand for M2 in the long-run. The estimated elasticities of the determinants of money demand are 1.79 for the final
consumption expenditure and −0.07 for the interest rate. The elasticity of the final consumption expenditure is higher
than the unity, which is not surprising since final consumption expenditure represents the major component of GDP in
Tunisia. The demand for the M2 monetary aggregate is essentially due to private consumption expenditure and government
consumption expenditure. With regards to the interest rate, the statistical significance and the magnitude of its coefficient
point out that its effect on real money demand is weak. Finally, the expenditure on investment goods and the expenditure
on export of goods and services exert no impact on the demand for money in the long-run.
Given the existence of long-run relationship between the money demand and its determinants, we proceed to the esti-
mation of the short-run dynamics via the implementation of the error-correction model. The deviations from the long-run
relationship may be due to the occurrence of shocks in the short-run. The use of the error-correction model allows checking
the short-run elasticities and measuring the speed of adjustments to the long-run equilibrium via the error-correction term.
A negative and statistically significant coefficient on this term is an indicator of the existence of cointegration. Panel B of
Table 4 reports the ARDL based error-correction representation. As can be seen, the error-correction term is negative and
statistically significant at 5% level, confirming the results of the bounds testing procedure and indicating that the volume
of money demand and its determinants cannot diverge systematically from a long-run equilibrium position. In this context,
the value of the error-correction term is relatively low, indicating that nearly 25% of the disequilibria in the demand for
M2 monetary aggregate due to previous shocks adjust back to the long-run equilibrium in the current year. Regarding the
components of real income, only the expenditure on investment goods is found to influence the demand for M2 and its sign
 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45 37

Table 4
The ARDL estimates of long- and short-run relationships.

Coefficient Std. error p-value

Panel A: The long-run relationships
ln FCEt 1.790∗∗∗ 0.380 0.000
ln EIGt −0.337 0.245 0.183
ln EXPt −0.358 0.224 0.124
it −0.070∗∗ 0.030 0.032
Constant −2.503 3.813 0.518

Coefficient Std. error p-value

Panel B: The short-run dynamics
 ln FCEt 0.033 0.274 0.903
 ln EIGt 0.220∗∗∗ 0.078 0.009
 ln EXPt 0.070 0.112 0.539
i t −0.017∗∗∗ 0.003 0.000
ECMt −1 −0.247∗∗ 0.093 0.014
Constant −0.620 1.114 0.583

Test statistic p-value

Panel C: Diagnostic tests
LM test of residual serial correlation 1.639 0.200
Normality test 0.650 0.722
Ramsey Reset test 2.499 0.114
ARCH test 0.031 0.860

Notes: LM test is the Lagrange Multiplier test of residual serial correlation; Normality test is based
on skewness and kurtosis; Ramsey Reset test is the Ramsey test for omitted variables/functional form
and the ARCH test is the Engle autoregressive conditional heteroskedasticity test. ∗∗∗ , ∗∗ , ∗ denote the
statistical significance at the 1%, 5% and 10% levels, respectively.

is as expected with an estimated elasticity of about 0.220. The coefficient on the interest rate is also statistically significant
and negative. Panel C of the same table presents some diagnostic tests that aim at measuring the goodness of fit and the
adequacy of the estimated model. Results reveal that there is no serial correlation, normality or heteroskedasticity in the
residuals. The Ramsey RESET test also confirms the correct functional form of the model. Tang (2004) reports that the RE-
SET test allows detecting specification errors, such as simultaneous equation issues, omitted variables and serially correlated
disturbances. The R-squared indicates that about 68% of variations in the demand for M2 monetary aggregate are explained
by variables introduced in the specification.

6. Robustness checks

The current section aims at considering some robustness checks of our main analysis. It allows us gauging the sensitivity
of previous results with respect to many empirical issues such as the potential existence of structural change, the choice
of econometric techniques or the set of control variables. Four robustness and sensitivity checks are conducted. First, we
introduce additional control variables to our baseline specification and re-estimate the long-run parameters associated with
the modified money demand function. Second, we re-estimate the long-run parameters using the fully-modified ordinary
least squares (FMOLS) and the dynamic ordinary least squares (DOLS). Third, a special attention is being paid to the potential
existence of structural change in the relationship by incorporating a time trend variable and testing the existence of long-run
relationships based on the Saikkonen–Lütkepohl cointegration test with structural shift and the Johansen–Mosconi–Nielsen
structural break cointegration test. Finally, we check the stability of parameters using the chow stability test, the Hansen
parameter instability test and the CUSUM and CUSUMSQ tests.

6.1. Additional control variables

We start by incorporating other control variables that appear to be potential explanatory variables of the demand for
money. First, the nominal effective exchange rate is included in the money demand function (Panel A of Table 5) as a mea-
sure of currency substitution. As reported by Debson and Ramlogan (2001), the exchange rate is usually introduced in the
money demand function in order to capture the degree of openness of the economy. Second, as suggested by many authors,
the inflation rate could be employed as a second measure of the opportunity cost of holding money (Bahmani-Oskooee &
Gelan, 2008; Kumar, 2011). For example, Baharumshah, Mohd, and Masih (2009) advance that for the case of China, using
the interest rate or the inflation rate as a measure of the opportunity cost of holding money is a difficult task. Treichel
(1997) uses both the inflation rate and the interest rate when estimating the money demand function in Tunisia. In Panel B
of Table 5, the inflation rate as well as the interest rate are introduced, whereas in Panel C, only the inflation rate is
maintained. The estimation of long-run parameters is based on the ARDL approach.
Before estimating long-run coefficients, we checked the existence of cointegration relationships between broad money
demand, the different components of income, the interest rate and the newly introduced variables. The ARDL testing ap-
38 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45

Table 5
Long-run coefficients of the modified money demand function.

Coefficient Std. error p-value

Panel A: ln M2t = f (ln FCEt , ln EIGt , ln EXPt , it , ln NEERt )
ln FCEt 1.923∗∗∗ 0.503 0.001
ln EIGt −0.486 0.431 0.272
ln EXPt −0.232 0.360 0.526
it −0.066∗ 0.032 0.056
ln NEERt 0.262 0.586 0.659
Constant −6.421 9.203 0.493

Panel B: ln M2t = f (ln FCEt , ln EIGt , ln EXPt , it , πt )

ln FCEt 1.672∗∗∗ 0.338 0.000
ln EIGt −0.439∗∗ 0.208 0.047
ln EXPt −0.123 0.249 0.625
it −0.063∗∗ 0.024 0.016
πt 0.011 0.016 0.476
Constant −2.938 3.783 0.446

Panel C: ln M2t = f (ln FCEt , ln EIGt , ln EXPt , πt )

ln FCEt 2.164∗∗∗ 0.494 0.000
ln EIGt −0.288 0.250 0.261
ln EXPt −0.390 0.341 0.264
πt 0.002 0.018 0.896
Constant −12.100∗∗∗ 2.483 0.000

Notes: ∗∗∗ , ∗∗ , ∗ denote the statistical significance at the 1%, 5% and 10% levels, respectively.

proach suggests that for Panels A and B, real broad money demand and its determinants are cointegrated at 5% and 1%
statistical levels, respectively.9 When we use only the inflation rate as an opportunity cost measure, no cointegration re-
lationship is found, but we keep it in the analysis.10 Turning to the estimation of long-run coefficients, results reveal that
the introduction of the nominal effective exchange rate or the inflation rate does not affect our main conclusion given that
the final consumption expenditure and the interest rate are both the determinant of broad money demand.11 Coefficients
associated with final consumption expenditure are significant at 1% level in all cases and are close to those obtained in the
baseline specification. In addition, when introduced in the regression, the inflation rate exerts no effect on the demand for
money in the long-run. According to Boughrara (2001), the inflation rate has been stable in Tunisia over the past decades
and consequently he did not introduce it in the money demand function. Similar results are found for the nominal effective
exchange rate, which may be explained by the fact that the detention of foreign currencies is highly regulated in Tunisia.
The substitution between domestic and foreign currencies is not an easy operation in the case of an appreciation or a
depreciation of the exchange rate.

6.2. Further estimates of long-run parameters

As reported by Narayan and Narayan (2004), the FMOLS technique advocated by Phillips and Hansen (1990) has two
main advantages. On the one hand, it eliminates the sample bias. On the second hand, it corrects the endogeneity and
serial correlation effects that may exist. With regards to the DOLS technique developed by Stock and Watson (1993), it
essentially avoids to problems related to small sample bias and simultaneity.12 In Table 6, we present long-run results using
the dynamic OLS (Panel A) and fully modified OLS (Panel B).
Findings based on the DOLS technique are similar to those of the ARDL approach since the final consumption expen-
diture and the interest rate have positive and negative coefficients, respectively, and are statistically significant. The use of
the FMOLS technique partially corroborates those found previously. The consumption component of GDP is the only signif-
icant explanatory variable of the broad money demand. Finally, it is important to note that the elasticity associated with
consumption is higher than the unity (1.474 and 1.613 using the DOLS and the FMOLS, respectively, against 1.790 using the
ARDL). Consequently, a common feature that arises from the analysis is that final consumption and the interest rate are

9
F ln M2 (ln M2| ln FCE, ln EIG, ln EXP , i , ln NEER) is equal to 5.252 while the Narayan’s (2005) critical values at 5% statistical level are 2.804 and 4.013,
whereas F ln M2 (ln M2| ln FCE, ln EIG, ln EXP , i , π ) is about 6.018 while the critical values at 1% statistical level are 3.900 and 5.419.
10
The associated F -statistic is equal to 1.491. Bahmani-Oskooee and Rehman (2005) and Tang (2007) point out that the F -statistic remains preliminary
when testing the existence of cointegration and that the coefficient on the lagged error-correction term is a more efficient tool. Kremers, Ericson, and Dolado
(1992) advances that the error-correction term is more powerful when testing the existence of cointegration. The results of the error-correction model (not
reported here because of space constraints but are available upon request from the authors) suggest that the coefficient on the lagged error-correction term
is negative (−0.271) and statistically significant at 5% level.
11
Even when the real effective exchange rate is used instead of the nominal effective exchange rate, the demand for money and its determinants remain
cointegrated. In addition, the final consumption expenditure and the interest rate have both statistically significant coefficients with long-run elasticities of
1.923 and −0.066, respectively.
12
According to Stock and Watson (1993), the problem of simultaneity and small sample bias is resolved by regressing the dependent variable on explana-
tory variables in levels, leads and lags of the explanatory variables.
 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45 39

Table 6
Long-run coefficients of the money demand function – Further estimates.

Coefficient Std. error p-value

Panel A: Dynamic OLS estimates
ln FCEt 1.474∗∗∗ 0.329 0.000
ln EIGt −0.106 0.238 0.663
ln EXPt −0.082 0.166 0.628
it −0.022∗∗ 0.010 0.043
Constant −6.907∗∗∗ 1.988 0.004

Panel B: Fully Modified OLS estimates

ln FCEt 1.613∗∗∗ 0.219 0.000
ln EIGt 0.023 0.136 0.862
ln EXPt −0.177 0.133 0.195
it −0.002 0.011 0.850
Constant −11.790∗∗∗ 1.994 0.000

Notes: ∗∗∗ , ∗∗ , ∗ denote the statistical significance at the 1%, 5% and 10% levels, respectively.

the determinants of the demand for broad money in the long-run. Such results are similar to those found when the ARDL
approach is employed.

6.3. Controlling for structural change

This subsection addresses the concern regarding the existence of structural changes in the relationship between the
demand for broad money and its determinants. Structural breaks are generally resulting from various economic, finan-
cial, institutional and/or political measures and affecting the economic activity at both the macro and micro levels. Égert,
Jiménez-Rodríguez, Kočenda, and Morales-Zumaquero (2006) point out that structural changes in macroeconomic data often
correspond to the instability in the real economy. Valadkhani, Layton, and Pahlavani (2005) outline that not taking into con-
sideration the presence of structural breaks may generate biased empirical results. In addition, Kumar and Webber (2013)
outline that if the estimation of the money demand function do not control for structural changes, results may be biased.
In the current study, we control for structural changes in three different ways. First, we implement the most commonly
used approach by incorporating a dummy variable to capture the effects of any structural change in the baseline models
estimated in Section 5. In other words, we estimate modified versions of Eqs. (3), (4) and (5) in which we introduce the
dummy variable. Second, we employ the Saikkonen–Lütkepohl cointegration test with structural shift (Saikkonen & Lütke-
pohl, 2000a, 2000b, 2000c). It has been mentioned that the introduction of a shift dummy to capture structural changes
in economic relationships leads to a change in the asymptotic distributions of the Johansen type tests (Husein, 2009). To
overcome this shortcoming, Saikkonen and Lütkepohl (2000a, 2000b, 2000c) propose an approach to test the existence of
cointegrating relationships between variables in models with structural shifts. Third, we perform the Johansen–Mosconi–
Nielsen structural break cointegration test (Johansen et al., 2000). The test is a generalization of the multivariate likelihood
procedure of Johansen (1988) that identifies the existence of long-run cointegrating relationships in the presence of up to
two structural breaks.13
The selection of the date of occurrence of the structural break is essentially based on the Lumsdaine–Papell structural
break unit root test and on the stylized facts. Findings of the structural break unit root test suggest that time breaks occurred
in almost all variables in the second half of the 1980s. As pointed out previously, the Tunisian economy experienced a
severe financial and economic crisis in 1986 that has prompted the intervention of the International Monetary Fund by
implementing the structural adjustment program at the same year. One of the objectives of the program is to improve
the monetary and fiscal policies in Tunisia. Khedhiri and Boudhina (2005) state that the structural adjustment program
is considered as an important event in the conduct of the monetary policy in Tunisia. These economic and institutional
events are translated into a dummy variable taking the value of one in 1986 and zero otherwise. The dummy variable,
denoted D86, is then incorporated in the analysis in order to detect the effects of the economic crisis and the adoption
of the structural adjustment program on the relationship between the broad money demand and its determinants. Table 7
provides the ARDL-based estimation results of Eqs. (2), (3) and (4) in which D86 is incorporated.
The Panel A of Table 7 suggests the rejection of the null hypothesis of no cointegration at the 5% significance level. Ac-
cordingly, it confirms the existence of a long-run equilibrium relationship between the demand for money, the components
of income and the interest rate, even when we control for structural changes. In Panels B and C, we estimate the long- and
short-run coefficients, respectively. Findings are close to those found in Table 4 since the final consumption expenditure and
the interest rate affect the broad money demand in the long-run and the expenditure on investment goods and the interest
rate affect it in the short-run. The main theme is that the coefficient of the final consumption expenditure continues to
be huge and highly significant in the long-run, highlighting the importance of such an income component in boosting the
demand for money.

13
Since the Saikkonen–Lütkepohl cointegration test with structural shift and the Johansen–Mosconi–Nielsen structural break cointegration test are modi-
fied versions of the Johansen framework, variables should be integrated of order one. As shown in Tables 1 and 2, this condition is respected.
40 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45

Table 7
The ARDL estimates with dummy variable.

F -statistic Asymptotic critical values

95% 99%
I (0) I (1) I (0) I (1)
Panel A: The ARDL cointegration test
F ln M2 (ln M2| ln FCE, ln EIG, ln EXP , i , D86) = 5.171∗∗ 2.804 4.013 3.900 5.419

Coefficient Std. error p-value

Panel B: The long-run relationships
ln FCEt 1.863∗∗∗ 0.401 0.000
ln EIGt −0.373 0.259 0.165
ln EXPt −0.401 0.240 0.109
it −0.071∗∗ 0.031 0.037
D86 −0.128 0.146 0.390
Constant −2.437 3.940 0.542

Panel C: The short-run dynamics

 ln FCEt 0.065 0.277 0.816
 ln EIGt 0.184∗∗ 0.086 0.043
 ln EXPt 0.064 0.113 0.575
i t −0.017∗∗∗ 0.003 0.000
ECMt −1 −0.241∗∗ 0.094 0.017
 D86 −0.030 0.032 0.347
Constant −0.587 1.116 0.604

Notes: D86 is a shift dummy which takes the value one in 1986 and zero otherwise. For Panel A, the
Akaike information criterion is used to select the optimal lag length for the ARDL model. The criti-
cal values are obtained from Narayan (2005), p. 1987, Case II, n = 35. ∗∗∗ , ∗∗ , ∗ denote the statistical
significance at the 1%, 5% and 10% levels, respectively.

Table 8
Saikkonen–Lütkepohl cointegration test with structural shift.

Null hypothesis LR 5% critical value p-value

None (r = 0) 75.35 66.13 0.006
At most 1 (r ≤ 1) 41.01 45.32 0.129
At most 2 (r ≤ 2) 17.95 28.52 0.545
At most 3 (r ≤ 3) 3.42 15.76 0.967
At most 4 (r ≤ 4) 1.03 6.79 0.784

Notes: The model contains a dummy variable, D86. Critical values are from Trenkler (2004).

Table 9
Johansen–Mosconi–Nielsen structural break cointegration test.

Null hypothesis LR 5% critical value p-value

None (r = 0) 125.39 96.17 0.000
At most 1 (r ≤ 1) 80.10 69.99 0.005
At most 2 (r ≤ 2) 46.02 47.77 0.073
At most 3 (r ≤ 3) 24.89 29.36 0.157
At most 4 (r ≤ 4) 11.51 14.50 0.145

Notes: The test is performed with the year 1986 as the break date. Critical values are from Johansen et
al. (2000).

Tables 8 and 9 present results of the Saikkonen–Lütkepohl cointegration test with structural shift and the Johansen–
Mosconi–Nielsen structural break cointegration test, respectively. The findings show that both test statistics reject the null
hypothesis of no cointegration at 5% level of significance. While the former indicates that a unique cointegrating relation-
ship between money demand and its determinants exists, the latter suggests that there are two cointegrating relationships.
Therefore, the two tests reveal that a long-run equilibrium relationship between money demand and its determinants exists
in Tunisia, even when we control for structural changes. These results corroborate those found previously using the ARDL
modeling without and with the dummy variable, which confirms the robustness of findings.

6.4. Stability analysis

Based on results of the previous section, a long-run relationship between the demand for broad money and its deter-
minants exists. In addition, the final consumption expenditure and the interest rate are found to be the main long-run
determinants of broad money demand. The issue that arises is whether this long-run relationship is stable over time. This
is an important point since monetary policies may not be designed if the studied relationship is unstable (Dobnik, 2013;
Hansen & Kim, 1995). Treichel (1997) indicates that the effectiveness of the money demand function in terms of policy rec-
 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45 41

Table 10
Chow stability test.

Breakpoint/Forecast period F -statistic p-value

Chow breakpoint test 1986 0.530 0.750
Chow forecast test 1986–2011 19.339 0.050

Table 11
Hansen parameter instability test.

Test Statistic p-value

Lc test 0.564 >0.2

ommendations depends primary on its stability. The Chow stability test and the Hansen parameter instability test are both
used to check whether structural changes affect or not the stability of the money demand function. The use of the chow test
needs prior knowledge of the date of the structural change. As in the previous subsection, the year 1986 is defined to be
the date of the potential structure change. Thus, the Chow breakpoint test and Chow forecast test are used to check the ex-
istence of significant structural change in the relationship in 1986 and in the post-crisis and structural adjustment program
period (1986–2011), respectively. Tables 10 and 11 summarize results of the Chow test and the Hansen test, respectively.
From Table 10, it is evident that there is no support for the presence of significant structural changes in the long-run
relationship. The Chow stability test shows that the occurrence of the economic crisis and the adoption of the structural
adjustment program do not affect the long-run relationship between broad money demand and its determinants. In addition,
the Chow forecast test aims to check the constancy of parameters between two different time periods (before 1986 and
after 1986). Results suggest that there are no significant structural differences in the behavior of the model in the two time
periods.
The Hansen parameter instability test is also conducted to check the constancy of the cointegration space. Since our
objective is to test the stability of the model, we report the Hansen’s Lc statistic in Table 11. This test is based on the null of
cointegration and the alternative hypothesis of no cointegration. Any captured instability is interpreted as a sign of absence
of cointegration. From Table 11, the Lc statistic reveals a stable long-run cointegrating relationship between variables under
consideration since the probability is higher than 0.2.
A simple and efficient way to test long-run parameters stability is the use of the CUSUM and CUSUMSQ tests proposed
by Brown, Durbin, and Evans (1975). These two statistics are updated recursively and plotted against break points in the
studied relationship. The decision is generally made graphically. If plots of the CUSUM and CUSUMSQ tests do not cross
pairs of the 5% critical lines, the relationship is considered as stable. A graphical presentation of these two tests (based on
the ARDL estimates) is provided in Fig. 1.
As is evident from plots, there is no movement outside the critical lines, suggesting the absence of any structural insta-
bility in the estimated ARDL models during the investigated period.

7. Conclusion and policy implications

The main aim of this article is to examine the long- and short-run determinants of broad money demand Tunisia. Un-
like the majority of previous studies on the subject, the current one considers the components of real income. It splits the
domestic income into its components such as consumption, investment and export expenditures. Using a relatively recent
cointegration technique, the ARDL bounds testing approach, we found evidence of the existence of long-run cointegration
relationship between M2 balances, the various components of real income and the interest rate. It emerges also that the
final consumption expenditure and the interest rate are the only determinants of real money demand in the long-run. How-
ever, the impact of the final consumption expenditure is higher than the one of the interest rate. These findings are robust
to a variety of sensitivity checks. We start by introducing the nominal effective exchange rate and the inflation rate in the
money demand function, but our main conclusions remain valid. In addition, the fully modified OLS and the dynamic OLS
almost confirm the results of the ARDL approach. We also perform the Saikkonen–Lütkepohl cointegration test with struc-
tural shift and the Johansen–Mosconi–Nielsen structural break cointegration test in order to control for structural change.
Findings reveal that the final consumption expenditure and the interest rate are the main determinant of money demand
even when we control for structural change. The occurrence of the balance of payment crisis and the adoption of the struc-
tural adjustment program in 1986 seem not to affect the relationship between the demand for money and its determinants.
With respect to the stability of the money demand function, results based on the Chow stability test, the Hansen parameter
instability test and the CUSUM and CUSUMSQ tests suggest that the estimated parameters are stable over the studied period.
These findings have important implications on the policy formulation in Tunisia. An expenditure-switching or expenditure-
reducing policy on final consumption can be adopted in order to act on the demand for money. The negative elasticity on
the interest rate means that it can be used to influence monetary policy in Tunisia at both the long- and short-run, but the
effect is relatively small. This implies that if the demand for broad money is considered as a monetary target, it will take
quite a large change in the interest rate before inducing the desired change in the M2 demand.
42 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45

Fig. 1. Plots of (a) cumulative sum of recursive residuals (CUSUM) and (b) cumulative sum of squares of recursive residuals (CUSUMSQ) statistics.

Appendix A

Table A.1
Variables derivation and data sources.

Variable Description Source

Broad money (M2)
Final consumption expenditure (FCE)
Expenditure on investment goods (EIG)
Expenditure on exports of goods and services
(EXP)
Interest rate (i)
Inflation rate (π )
Nominal effective exchange rate (NEER)
Real effective exchange rate (REER)
Nominal values of M2 are deflated using the World Development Indicators Online, World
consumer price index (1990 = 100). Bank [accessed on January 2013]
Tunisian National Institute of Statistics
The sum of household consumption and World Development Indicators Online, World
general government consumption at constant Bank [accessed on January 2013]
prices 1990 (1990 = 100)
The gross fixed capital formation at constant World Development Indicators Online, World
prices 1990 (1990 = 100) Bank [accessed on January 2013]
– World Development Indicators Online, World
Bank [accessed on January 2013]
Money market interest rate Central Bank of Tunisia
– Tunisian National Institute of Statistics
– International Financial Statistics, IMF
– International Financial Statistics, IMF
Table A.2
Survey of selected empirical works.

Determinants of money demanda,b

Study Monetary aggregate Countries Period Technique
Short-run Long-run
First group: Studies focusing on Tunisia
Simmons (1992) M1 5 African countries, 1962–1989 Error-correction Income, inflation Real income
including Tunisia model

O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45

Treichel (1997) M2, M4 Tunisia 1962–1995 Johansen – Real income, treasury bill rate
Arize and Shwiff (1998) M1, M2 25 Developing countries, 1960–1990 Johansen – Real income, interest rate, black
including Tunisia market exchange rate, official
exchange rate.
Arize et al. (1999) M1, M2 12 Developing countries, 1964–1996 Johansen – Real income, exchange rate,
including Tunisia exchange rate variability,
inflation rate.
Boughrara (2001) M2 Tunisia 1987–1992 Engle–Granger, Real income Real income, treasury bills
Shin procedure interest rate, special deposits
interest rate

Second group: Studies decomposing real income

Tang (2002) M3 Malaysia 1973–1998 ARDL Exports expenditure, exchange rate. Final consumption expenditure,
expenditure on investment
goods, exports expenditure,
exchange rate, interest rate.
Tang (2004) M2 Japan 1973–2000 Johansen and Final consumption expenditure, Expenditure on investment
ARDL expenditure on investment goods, goods, deposit rate.
exports expenditure, government
bond yield rate, deposit rate.
Tang (2007) M2 Five Southeast Asian Varies according Final consumption expenditure (1), Final consumption expenditure
countries to the country exports expenditure (3), inflation rate (2), exports expenditure (2),
(1), exchange rate (1). inflation rate (1).
Ziramba (2007) M1, M2 and M3 South Africa 1970–2005 ARDL Final consumption expenditure, Expenditure on investment
interest rate. goods, final consumption
expenditure, exports
expenditure, interest rate,
government bond yield rate,
exchange rate.

Notes: a For the first group, if the study focuses on a set of countries, we only report the determinants of money demand in Tunisia. b
Figures in brackets indicate the number of countries in which the associated
regressor affects money demand.

43
44 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45

References

Akpan, U. F. (2011). Cointegration, causality and Wagner’s hypothesis: time series evidence for Nigeria, 1970–2008. Journal of Economic Research, 16(1),
59–84.
Arize, A. C., & Shwiff, S. S. (1998). The appropriate exchange-rate variable in the money demand of 25 countries: an empirical investigation. North American
Journal of Economics and Finance, 9(2), 169–185.
Arize, A. C., Malindretos, J., & Shwiff, S. S. (1999). Structural breaks, cointegration, and speed of adjustment Evidence from 12 LDCs money demand.
International Review of Economics and Finance, 8(4), 399–420.
Avouyi-Dovi, S., Drumetz, F., & Sahuc, J. G. (2011). The money demand function for the Euro area: some empirical evidence. Bulletin of Economic Research,
64(3), 377–392.
Baharumshah, A. Z., Mohd, S. H., & Masih, A. M. M. (2009). The stability of money demand in China: evidence from the ARDL model. Economic Systems,
33(3), 231–244.
Bahmani-Oskooee, M., & Gelan, A. (2008). How stable is the demand for money in African countries? Journal of Economic Studies, 36(3), 216–235.
Bahmani-Oskooee, M., & Rehman, H. (2005). Stability of the money demand function in Asian developing countries. Applied Economics, 37(7), 773–792.
Bahmani-Oskooee, M., & Techaratanachai, A. (2001). Currency substitution in Thailand. Journal of Policy Modeling, 23(2), 141–145.
Bechri, M. Z., & Naccache, S. (2006). The political economy of development policy in Tunisia. In J. B. Nugent, & M. Hashem Pesaran (Eds.), Explaining growth
in the Middle East. In Contributions to Economic Analysis, Vol. 278. Emerald Group Publishing Limited.
Ben-Salha, O. (2013). Economic globalization, wages and wage inequality in Tunisia: an ARDL bounds testing approach. Review of Middle East Economics and
Finance, 9(3), 321–356.
Boughrara, A. (2001). Money demand in Tunisia during the reform period. Savings and Development, 25(2), 117–137.
Boughrara, A. (2007). Can Tunisia move to inflation targeting? The Developing Economies, 45(1), 27–62.
Boughrara, A., Boughzala, M., & Moussa, H. (2008). The Tunisian financial system and inflation targeting. Paper presented at the Workshop on ‘Monetary
Policy and Inflation Targeting’. October 24th–25th 2008, Tunis, Tunisia.
Brown, R. L., Durbin, J., & Evans, J. M. (1975). Techniques for testing the constancy of regression relationships over time. Journal of the Royal Statistical Society,
Series B (Methodological), 37(2), 149–192.
Davidson, R., & Mackinnon, J. G. (1981). Several tests for model specification in the presence of alternative hypotheses. Econometrica, 49(3), 781–793.
Debson, S., & Ramlogan, C. (2001). Money demand and economic liberalization in a small open economy—Trinidad and Tobago. Open Economies Review,
12(3), 325–339.
Dobnik, F. (2013). Long-run money demand in OECD countries: what role do common factors play? Empirical Economics, 45(1), 89–113.
Duasa, J. (2007). Determinants of Malaysian trade balance: an ARDL bound testing approach. Journal of Economic Cooperation, 28(3), 21–40.
Égert, B., Jiménez-Rodríguez, R., Kočenda, E., & Morales-Zumaquero, A. (2006). Structural changes in Central and Eastern European economies: breaking
news or breaking the ice? Economic Change and Restructuring, 39(1), 85–103.
Engle, R. F., & Granger, C. W. J. (1987). Co-integration and error-correction: representation, estimation and testing. Econometrica, 55(2), 251–276.
Fair, R. C. (1987). International evidence on the demand for money. The Review of Economics and Statistics, 69(3), 473–480.
Fanizza, D., Laframboise, N., Martin, E., Sab, R., & Karpowicz, I. (2002). Tunisia’s experience with real exchange rate targeting and the transition to a flexible
exchange rate regime. IMF Working Paper No 02/190. International Monetary Fund.
Hammoudeh, S., & Sari, R. (2011). Financial CDS, stock market and interest rates: which drives which? The North American Journal of Economics and Finance,
22(3), 257–276.
Hansen, G., & Kim, J. R. (1995). The stability of German money demand: tests of the cointegration relation. Review of World Economics, 131(2), 286–301.
Husein, J. (2009). Export-led growth hypothesis: a multivariate cointegration and causality evidence for Jordan. The Journal of developing Areas, 42(2), 253–266.
International Monetary Fund (2005). Monetary policy implementation at different stages of market development. IMF Occasional Paper No 244. International
Monetary Fund.
Johansen, S. (1988). Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control, 12(2–3), 231–254.
Johansen, S., & Juselius, K. (1990). Maximum likelihood estimation and inference on cointegration – with applications to the demand for money. Oxford
Bulletin of Economics and Statistics, 52(2), 169–210.
Johansen, S., Mosconi, R., & Nielsen, B. (2000). Cointegration analysis in the presence of structural breaks in the deterministic trend. Econometrics Journal,
3(2), 216–249.
Khedhiri, S., & Boudhina, R. (2005). Application of non-linear error correction models to the demand for money in Tunisia. Middle East Business and Economic
Review, 17(2), 20–39.
Kremers, J. J., Ericson, N. R., & Dolado, J. J. (1992). The power of cointegration tests. Oxford Bulletin of Economics and Statistics, 54(3), 325–347.
Kumar, S. (2011). Financial reforms and money demand: evidence from 20 developing countries. Economic Systems, 35(3), 323–334.
Kumar, S., & Webber, D. J. (2013). Australasian money demand stability: application of structural break tests. Applied Economics, 45(8), 1011–1025.
Lumsdaine, R. L., & Papell, D. H. (1997). Multiple trend breaks and the unit root hypothesis. Review of Economics and Statistics, 79(2), 212–218.
MacKinnon, J. G. (1996). Numerical distribution functions for unit root and cointegration tests. Journal of Applied Econometrics, 11(6), 601–618.
Mohamed Sghaier, I., & Abida, Z. (2013). Monetary policy rules for a developing countries: evidence from Tunisia. The Review of Finance and Banking, 5(1),
35–46.
Narayan, P. K. (2005). The saving and investment nexus for China: evidence from cointegration tests. Applied Economics, 37(17), 1979–1990.
Narayan, P. K., & Narayan, S. (2008). Estimating the demand for money in an unstable open economy: the case of the Fiji Islands. Economic Issues, 13(1),
71–91.
Narayan, S., & Narayan, P. K. (2004). Determinants of demand for Fiji’s Exports: An empirical investigation. The Developing Economies, XLII(1), 95–112.
Narayan, S., & Narayan, P. K. (2005). An empirical analysis of Fiji’s import demand function. Journal of Economic Studies, 32(2), 158–168.
Odhiambo, N. M. (2009). Energy consumption and economic growth nexus in Tanzania: an ARDL bounds testing approach. Energy Policy, 37(2), 617–622.
Paul, B. P., Salah Uddin, M. G., & Noman, A. M. (2011). Remittances and output in Bangladesh: an ARDL bounds testing approach to cointegration. Interna-
tional Review of Economics, 58(2), 229–242.
Pesaran, M. H., & Pesaran, B. (1997). Working with Microfit 4.0: interactive economic analysis. Oxford: Oxford University Press.
Pesaran, M. H., & Shin, Y. (1999). An autoregressive distributed lag modelling approach to cointegration analysis. In S. Strom (Ed.), Econometrics and economic
theory in the 20th century: the Ragnar Frisch centennial symposium. Cambridge: Cambridge University Press.
Pesaran, M. H., Shin, Y., & Smith, R. (2001). Bound testing approaches to the analysis of level relationships. Journal of Applied Econometrics, 16(3), 289–326.
Phillips, P. C. B., & Hansen, B. E. (1990). Statistical inference in instrumental variables regression with I(1) processes. Review of Economic Studies, 57(1),
99–125.
Saikkonen, P., & Lütkepohl, H. (2000a). Testing for the cointegrating rank of a VAR process with an intercept. Econometric Theory, 16(3), 373–406.
Saikkonen, P., & Lütkepohl, H. (2000b). Testing for the cointegrating rank of a VAR process with structural shifts. Journal of Business & Economic Statistics,
18(4), 451–464.
Saikkonen, P., & Lütkepohl, H. (2000c). Trend adjustment prior to testing for the cointegrating rank of a vector autoregressive process. Journal of Time Series
Analysis, 21(4), 435–456.
 O. Ben-Salha, Z. Jaidi / The Journal of Economic Asymmetries 11 (2014) 30–45 45

Simmons, R. (1992). An error-correction approach to demand for money in five African developing countries. Journal of Economic Studies, 19(1), 29–47.
Stock, J. H., & Watson, M. W. (1993). A simple estimator of cointegrating vectors in higher order integrated systems. Econometrica, 61(4), 783–820.
Tang, T. C. (2002). Demand for M3 and expenditure components in Malaysia: assessment from bounds testing approach. Applied Economics Letters, 9(11),
721–725.
Tang, T. C. (2004). Demand for broad money and expenditure components in Japan: an empirical study. Japan and the World Economy, 16(4), 487–502.
Tang, T. C. (2007). Money demand function for Southeast Asian countries: an empirical view from expenditure components. Journal of Economic Studies,
34(6), 476–496.
Treichel, V. (1997). Broad money demand and monetary policy in Tunisia. IMF Working Paper No. 97/22. International Monetary Fund.
Trenkler, C. 2004. Determining p-values for systems cointegration tests with a prior adjustment for deterministic terms. Mimeo, Humboldt-Universität zu Berlin.
Valadkhani, A., Layton, A. P., & Pahlavani, M. (2005). Multiple structural breaks in Australia’s macroeconomic data: an application of the Lumsdaine and
Papell test. International Journal of Applied Econometrics and Quantitative Studies, 2(3), 31–44.
Ziramba, E. (2007). Demand for money and expenditure components in South Africa: assessment from unrestricted error-correction models. South African
Journal of Economics, 75(3), 412–424.