Está en la página 1de 5

The relationship between

energy and GNP

Further results

Eden S. H. Yu and Been-Kwei Hwang

This paper reexamines the causality between GNP and energy consumption by
using updated US data for the period 1947-1979. As a secondary contribution,
we investigate the causal relationship between energy consumption and employ-
ment. Applying Sims’ technique, we ftnd no causal relationship between GNP
and energy consumption. We find further that there is a slight unidirectional
flow running from employment to energy consumption. Economic interpretations
of the empirical results are also presented.

Keywords: Causality; GNP; Energy consumption

The important impact of energy consumption on the a feasible policy option without impairing economic
operation of an economy has been increasingly recog activity. However, the unidirectional causality from
nized by economists since the worldwide economic GNP to energy was questioned by Akarca and LongP
recessions following the Arab oil embargo in 1973. By shortening the data sample of Kraft and Kraft by
The relationships between energy consumption (EC) two years, Akarca and Long argue that ‘the results
and employment and gross national product (GNP) obtained by K and K appear to be spurious and to arise
have been the subject of intense research and heated solely from the inclusion of two additional years in the
debate. Earlier work in this area published in the 1970s data sample’s
includes the studies by Allen,’ Hitch,’ Khazzoom,3 and In another paper Akarca and Long’ examined the
Long and SchipperP among others. Recently, Kraft and relation between US domestic energy consumption and
Kraft’ and Akarca and Long6 have analysed the causal total employment for the period 1973-1978. Using
relationship between gross energy consumption and sophisticated Box--Jenkins type time-series procedures
GNP. Using Sims’ technique? Kraft and Kraft’ found and Granger’s criterion, they established a unidirectional
unidirectional causality running only from GNP to gross causality running from energy to employment. Further-
energy inputs (CEI) for the postwar period (1947- more, the relationship between the response of employ
1974). Their results imply that energy conservation is ment to a change in the energy input is negative. This is
an interesting result in its own right, because it implies
that total employment would slightly increase as a result
Eden Yu is with the Division of Economics, University of energy conservation.
of Oklahoma, Norman, OK 73019, USA, and the Depart- The purpose of this paper is to reexamine the causality
ment of Economics, Louisiana State University, Baton between GNP and energy consumption (EC) by using
Rouge, LA 70803, USA. Been-Kwei Hwang is with the
updated US data for the period 1947-1979.* As a
Division of Economics, University of Oklahoma.
secondary contribution, we investigate the causal rela-
Research support from the Energy Resource Center at the tionships between EC and employment (EMP). Applying
University of Oklahoma is gratefully acknowledged. J. Y. Choi
performed competent research assistance. The authors are
Sims’ technique we find no causal relationship between
indebted to Charlie Turner and an anonymous referee for very GNP and energy consumption. This result lends support
useful comments on an earlier draft. However. the usual caveat to the earlier finding of Akarca and Long.6 We also
applies.
find that there is a slight unidirectional flow running
Final manuscript received 14 November 1983. from employment to energy consumption.

186 0140-9883/84/030186-05 303.00 @ 1984 Butterworth & Co (Publishers) Ltd


The relationship between energy and GNP: E. S. H. Yu and B-K. Hwang

The methodology and models that we employ in the To alleviate the problem of time-series autocorrelation,
causality tests are briefly presented in the next section. all the regressions are transformed into quasi first-
Further sections report results obtained from both difference form.? Since both Kraft and Kraft’ and
Granger and Sims tests for a variety of relationships, and Akarca and Long6 employed a lag structure of 3 past
present econometric remarks and economic interpreta- and 4 future in the right-hand side of the regressions,
tions of the empirical analysis. we adopt, for purpose of comparison, the same lag
scheme. A constant term, current, 3 past and 4 future
lags of the independent variables are specified on the
Methodology and model right-hand side of the unrestricted regressions for testing
Both Sims’ and Grange?’ have developed two econo- of causality.
metric tests of causality. According to Sims ‘the test An alternative way of testing the direction of causation
for unidirectional causality between X and Y is to is proposed by Granger.” According to Granger, X, is
regress Y on past and future values of X. Then if causing Y, if we are better able to predict Y, using all
causality runs from X to Y only, future values of X in available information than if the information apart from
the regression should have coefficients insignificantly the X, had been used. Suppose U, is the set of all avail-
different from zero as a group’.” able information including X,. Then X, is causing Y, if
To test via Sims’ technique the causality between and only if Var( Yr I U,) < Var( Yt I(U, - X,)). Granger
GNP and EC, the following four regression equations tests of causality between GNP and EC, for example,
are formulated: can be performed via the following set of regressions:

GNP= f(EC, m past lags and k future


lags of EC) GNP = f(m past GNP, k past EC) (5)
(1) GNP = f(m past GNP) (6)
GNP = f (EC, m past lags of EC) (2) EC = f(m past EC, k past GNP)
EC = f(GNP, m past lags and k future lags of GNP) EC = f(m past EC) I_:;
We use the quasi first-difference transformation and
EC = f(GNP, m past lags of GNP)
assume m = k = 4 in the lag structure.
Equations (1) and (3) are in unrestricted forms;
Equations (2) and (4) are in restricted form. F statistics
are calculated to test whether the coefficients of future Findings and results
values can be zero as a whole. Note that As noted earlier, Kraft and Kraft found a unidirec-
tional flow from GNP to gross energy input without
(RRSS - URSS)/df,
F(dfi , dfd = feedback for the period 1947-1974. Using an updated
URSSjdf, data sample from 1947 to 1979, our results, obtained
from both Sims and Granger tests, reveal no causal
where dfi is the difference of the degree of freedom of
relationship between energy consumption and GNP.
the restricted regression and that of .the unrestricted
The empirical results obtained from Sims’ test of
regression and df2 is the degree of freedom of the
causality between energy and GNP are presented in
unrestricted regression. RRSS is the residual sum of
Table 1. The findings reported earlier by Kraft and Kraft
squares of the restricted regression and URSS is the
are reproduced, for purpose of comparison, in the
residual sum of squares of the unrestricted regression.
second row of Table l.$ Our results obtained from an
updated data sample for 1947. -1979 are reported in the
*The data sources are as follows: Annual GNP figures are fifth row of the table. It is notable that the F statistics
obtained from the National Income and Product Account of
are insignificant for both our equations relating GNP to
the US, 1929-1974. Department of Commerce, 1976, and
Survey of Current Business, (1975, 1977, 1981 issues), Depart-
energy consumption and vice versa. We infer that there
ment of Commerce. Energy consumption in 8tu are obtained appears to be no causal relationship between energy
from Statistical Abstract of the United States (annual issue) for and GNP. Our results strengthen the Akarca and Long
the data covering 1947-1978. The energy consumption data argument6 that the causal order reported by Kraft and
for 1979 are obtained from Annual Report to the Congress,
1980. The sources for annual employment and population are
Kraft is a spurious one.
Business Statistics, 1977. Department of Commerce and Survey The empirical results obtained from replicating the
of Current Business (1978-1980 issues), Department of Com- study by Kraft and Kraft for the same sample period
merce. Quarterly data on GNP are available from Survey of of 1947-1974 are presented in the third row of Table 1.
Current Business. Quarterly gross energy consumption for
The results indicate a unidirectional causality running
1973-Ql to 1976--Q4 are obtained from Monthly Energy
Review, Federal Energy’ Administration and for 1977~Ql to from GNP to energy consumption, consistent with the
1981--01 are obtained from Energy Information, Report to findings of Kraft and Kraft, who relate GNP to gross
Congress, 1981. energy input.
We replicated the causality test for the period 1947-
tA quasi first-difference transformation can be formulated as
follows: Y, = a + @Xt + ut and ut = put-1 + It. A two-step
1972, the period studied by Akarca and Long. The
procedure suggested by Durbin “ involves running the regression,
Yt = cx(l -A) + AYr_1 + PXl - @Xt_1 + It, to obtain the
coefficient for Y,_l. The next step is to run fy, - AYr_1) -$The value of A, the type of regression R’ and DW statistics
against (X, - AXt__l). were not reported by Kraft and Kraft.

ENERGY ECONOMICS July 1984 187


The relationship between energy and GNP: E. S. H. Yu and B-K. Hwang

Table 1. Causality batwean energy and GNP: Sims test.

Degree of
Period Regmssion F freedom Type R2 DW

1947-74 GNP on GE/ a.077= (4.12)


(K&K’s result)* GEI on GNP 0.769 (4.12)

1947-74 GNP on EC 0.723 6.552= (4.11) Unrestricted 0.964 1.436


Restricted 0.947 0.833
EC on GNP 1.108 2.299 (4.11) Unrestricted 0.863 1.743
Restricted 0.748 1.751

1947-72 GNP on EC 0.626 2.815 (4.9) Unrestricted 0.992 1.596


Restricted 0.983 1.357

EC on GNP 0.667 2.957 (4.9) Unrestricted 0.986 1.799


Restricted 0.969 1.623
1947-79 GNP on EC 0.625 2.290 (4.16) Unrestricted 0.996 1.313
Restricted 0.994 1.057
EC on GNP 0.595 1.309 (4.16) Unrestricted 0.985 1.416
Restricted 0.980 1.226

1973-Ql GNP on EC 0.842 3.134b (5.12) Unrestricted 0.735 1.781


-81 -cQc: .&%tris%er! .C.X@ ‘?&lMc
(quariter’1.v bad,
-&LB?? \‘S 117’ btitlliti uiHJ” i.Ni
Restricted 0.315

8 Reproduced from Kraft and Kraft Ref 5. p 403.


bSigniElc3nt ilr%ii’~v&.
c Sigrriifxzari~B~V%‘rav&.

results, presented in the fourth row of Table 1, is com- March 1978. They found a unidirectional causality from
patible with the finding of Akarca and Long. Thus the energy to employment without feedback. Using a sample
caus& O&Y B.z~.&& by IdraFt 3& XX& disappa7ed OF X?.Z&W?&%+ hala Fur XQ??*?&I 19%1939, VW?$Wti
in the study sample extended for five more years or via Sims’s test that there is, however, a slightly uni-
shortened by two years, 1973-1974, during which directional flow from employment to energy. Para-
struciura( sfiicc may well kave occrmed. da&Xlly , @Xc &anger’s test W@Zals Ra WZk catrsal
Chow procedures were utilized by Akarca and Long relationship between energy and employment. The
to tt?$. <e&Z%!IG&@5nJf i&Z GYZ%G?iZXG&Z%?<G +%2- &X+-&Z& fm*&! @I!&G& &Z&c&X.-e ~WZ&%! &_ T&!&Z
197@ and 395D -IlPbB perio&. We ut%zea Bow tests 3. X Q$i%carit sIruc‘tur& Werence for Ihe two sui-
to determine possible slnsctual change &.uring the peri& t948--1973 and 1934-s979 was also detected
year 1973. The results, summarized in Table 2, indi- via the Chow test.+ The observed structural shift which
cate that there is a significant structural shift, which occurred during 1973 provides a partial explanation as
occurred around 1973.8 to why our empirical results conflict with those of
The second contribution of this note is to report Akarca and Long.
our results on the causality between energy consump-
tion and employment. Akarca and Long’ applied
Grare_rS crr?enbn co murfWy XY darrr~tic crrem
Some econometric and economic interpretations
and employment data for the period January 1973- Our empirical results clearly indicate a strong statistical
relationship between GNP and energy consumption.
That there was a significant correlation between the
Table 2. Chow tast - mgrass GNP on EC.
two variables is evident from the high R2 values reported
Period SSR R2 DW F (2.291 in Table 1. However, the causality tests reveal no uni-
directional causal linkage between GNP and energy.
1947-72 19203.6 0.983 0.352 Intuitively, energy should be correlated with GNP.
1973-79 8534.9 0.873 1.473 9.669=
However, causal relationships between energy and
1947-79 46234.8 0.983 0.332
GNP are not self-evident and warrant rigorous investi-
a Sigriiiilcant *Y%‘I~~+I. u@&n..
Several techniques are available to ascertain patterns
of causality among economic variables. Regression
gin view of this structural shift, we also fitted quarterly data
1973-01 to 1981-Ql to the regressions and found a slightly
analysis and spectral analysis have been used, for example,
unidirectional causality running from GNP to energy consump- to assess major determinants of common stock prices.1
tion. The results are presented at the bottom of Table 1. This
finding, consistenr w’lth the result ot Krait and KraSt, is in no *The F va\ue of Chow’s test is 175.13.
way definitive, because of the numerous contradicton/ empirical
resu Its ubraimd kum a &m-gmZ darabaze. *Far &%ails, se.? H.WnkW~er an& xoztinL3 a& CDWz+~!4

188 ENERGY ECONOMICS July 1984


The relationship between energy and GNP: E. S. H. Yu and B-K. Hwang

Table 3. Causality between EC and EMP.

Sims test Granger test

Regression Type F(3.15P R7 DW F(4.18) R’ DW

Unrestricted 0.721 1.155 0.412 2.120


EC on EMP 1.125 2.148
Restricted 0.658 1.262 0.131 2.019

Unrestricted 0.989 1.925 0.379 1.992


EMP on EC 3.560= 1.174
Restricted 0.981 1.749 0.218 2.102

aSignificant at 5% level.
bCritical F(3.15) value at 5% level = 3.29;at 1% = 5.42.

Procedures to test causality were originally developed no such non-spurious causal linkage for the USA. The
by Strotz and Weld:’ they proposed that a diagonal results are extremely sensitive to samples, both in terms
covariance matrix could be recursive and causal. of sample length and the period of the data.
Basemann16 designed a statistical test of explicit causal
chains. Sims” provided operational interpretations of
Conclusions
Granger causal orderings associated with exogeneity.
The techniques proposed by Sims and Granger delineated Applying Sims’ test of causality to an updated data
above are efficient methods for testing causality, in sample, we found no causal relationship between GNP
an ordering sense, between any pair of time series. and energy consumption. This finding is supportive of
Since the main purpose of thii paper is to produce Akarca and Long’s argument that the causal order
additional empirical evidence so as to shed light on the suggested by Kraft and Kraft from GNP to energy is
mild debate between Kraft and Kraft’ and Akarca and a spurious one. We also observed a slightly unidirec-
Long6 regarding the causal relationships between energy tional causality running from employment to energy,
and GNP, thus energy was regressed on GNP only. In a which is inconsistent with the findings of Akarca
broader framework, one may, of course, posit that energy and Long.
is a function of GNP, some structural characteristics of We also tested causal relationships between energy
GNP, energy price and other relevant variables. In such and employment structure represented by a ratio of
a model, test for causality can be done for the dependent agricultural employment to non-agricultural employ-
variable, ie energy, and any one of its determinants. ment. The empirical results of Sims tests indicate a
Essentially, four regressions are required for each of the slightly unidirectional flow from employment ratio
determinants of energy. They are: (1) regress energy on to energy. Our test failed to establish any causal relation-
current and lagged values of the selected determinant; ship between per capita energy and per capita GNP.
(2) regress energy on current, lagged and future values
of determinants; regressions (3) and (4) are identical to
(1) and (2), except the determinant becomes the depen- References
dent variables and energy becomes the independent
variable. The recent time-series study on the determinants E. L. Allen, US Energy and Economic Growth,
1975-2010, Institute for Energy Analysis, Report
of common stock prices by Kraft and Kraft i8 and the
ORAU/IEA 76-7. 1976.
analysis by Ortmeyer” on the causal relationship C. J. ‘Hitch, Energy Conservation and Economic
between income and saving constitute two excellent Growth, American Association for the Advancement
examples of the application of Sims and &anger tech- of Science, Selected Symposium 22, 1978.
niques to test for causality in a multivariable framework. J. D. Khazzoom, Proceedings of the Workshop on
Modelling the Interrelationships Between the
Berndt and Wood” and Griffin” investigated the Energy Sector and the General Economy, Special
issue of gross and net complementarity between energy Report 45, Electric Power Research Institute,
and capital. An aggregate production function relating Palo Alto, 1976.
GNP (Y) to energy (f?), capital (K), labour (L) and T. V. Long and L. Schipper, Study Series on US
Economic Growth from 1975 to 1985: Prospects,
material (M) is specified as follows: Y = F(f:‘, K, L, M).
Problems and Patterns, US GPO, Washington, DC,
Assuming weak separability of I:‘, K and L vi.&-vis 1976, with a modified version published in Energy,
materials, the production function may be further March 1978, pp 63-82.
expressed as Y = F( I’, M) where V = u(E’, K, L). In this J. Kraft and A. Kraft, ‘On the relationship between
formulation, energy as a practical matter is hypothesized, energy and GNP’, Journal of Energy and Develop-
ment, Vo13, 1978, pp 401-403.
along with capital and labour, as a major input in the A. T. Akarca and T. V. Long, ‘On the,relationship
production process. between energy and GNP: A reexamination’, Journal
The hypothesized production function itself and even of Energy and Development, 1980, pp 326-331.
the strong statistical correlations obtained are, however, C. A. Sims, ‘Money, income, and causality’,American
Economic Review, 1972, pp 540-552.
inadequate to indicate a unidirectional causal relation- Akarca and Long, op tit, Ref 6, p 330.
ship. The causal linkage between GNP and energy is not A. T. Akarca and T. V. Long, ‘Energy and employ-
self-evident, and our empirical analysis actually suggests ment, a time series analysis of the causal relation-

ENERGY ECONOMICS July 1984 189


The relationship between energy and GNP: E. S. H. Yu and B-K. Hwang

ship’, Resources and Energy, Vol 2, 1979, pp 151- of ‘interdependent’ models’, ‘Journal of American
162. .Q&tical Association, November 1965, pp 1083-
10 C. W. J. Granger, ‘Investigating causal relations by
econometric model and cross spectral methods’, 17 C. A: Sims. ‘Macroeconomics and reality’, Econo-
Econometrica, July 1969, pp 424-438. metrica, 1960, pp l-35.
11 Sims, op tit, Ref 7, p 543.. _ 18 J. Kraft and A. Kraft ‘Determinants of common
12 J. Durbin, ‘Estimation of parameters in time-series stock prices: a time ‘series analysis’, Journal of
regression models’, Journal of Royal Statistics Finance, May 1977, pp 417-425.
Society, B series, 1960, pp 139-153. 19 D. L. Ortmever. ‘Does income drive saving? Exo-
13 M. J. Hamburger and L. A. Kochin, ‘Money and geneity testing ‘applied to Asian data’, Journal
stock prices: the channels of influence’, Jourkl of of Development Economics, September 1980,
Finance, May 1972, pp 231-249. pp 397-407.
14 R. V L. Cooper, ‘Efficient capital markets and 20 E. R. Bemdt and D. 0. Wood, ‘Engineering and
quantity theory of money’, Journal of Finance, econometric interpretations of energy-capital com-
June 1974, pp 887-908. plementarity’, American Economic Review, June
1.5 R. H. Strotz and H. 0. A. Wold, ‘Recursive vs. 1979, pp 342-354.
non-recursive systems’, Econometrica, November 21 J. M. Griffin, ‘Engineering and econometric inter-
1960, pp 417-427. pretations of energy-capital complementarity:
16 R. L. Basemann? ‘A note on the statistical test- comment’, American Economic Review, December
ability of ‘exphcrt causal chains’ against the class 1981, pp 1100-1104.

ENERGY ECONOMICS July 1984

También podría gustarte