Professional Documents
Culture Documents
Narayan 2006
Narayan 2006
Applied Economics
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/raec20
To cite this article: Paresh Kumar Narayan & Seema Narayan (2006) Government revenue and government expenditure nexus:
evidence from developing countries, Applied Economics, 38:3, 285-291, DOI: 10.1080/00036840500369209
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Applied Economics, 2006, 38, 285–291
Moreover, using more recent developments in applied in the final section we conclude with some policy
time series econometric techniques, Baghestani and implications.
McNown (1994) found no evidence of any causality
between revenue and expenditure. Joulfaian and
Mookerjee (1991) studied a group of industrialized II. Methodology
countries and found that revenue Granger caused
expenditure in most of the countries except for KPSS unit root test
Canada, Iceland and Japan, and expenditure caused
revenue for all countries except for France, Greece The KPSS test for unit root differs from the ADF
and Ireland. In the case of studies on developing and the Phillip and Perron test in that the series Yt
countries, Baffes and Shah (1994) found bidirectional is assumed to be (trend-) stationary under the null.
causality between government revenue and govern- Put differently, the KPSS test reverses the null and
ment expenditure for Argentina and Mexico, while the alternative hypothesis. The KPSS statistic is based
for Brazil they found unidirectional causality. In a on the residuals from the OLS regression, which takes
more recent study, Fasano and Wang (2002) inves- the following form:
tigated the relationship for oil-dependent Gulf Yt ¼ t þ bt þ "t ð1Þ
Cooperation Council countries and found bidirec-
tional causality for Kuwait, Qatar and Saudi Arabia where t is a linear deterministic trend, "t is a station-
and unidirectional causality running from revenue ary error, and bt is a random walk; bt ¼ bt1 þ t ,
to expenditure in Bahrain, the United Arab Emirates where t are i.i.d. ð0, 2 Þ. The initial value of bo is
and Oman. treated as fixed and is interpreted as an intercept.
Given the theoretical and policy implications The test is conducted by first regressing Yt on a
of studying the relationship between government constant and a trend (t), allowing one to obtain the
revenue and government expenditure, we were residuals. The KPSS statistic is defined as:
X
encouraged to launch an investigation of this ðuÞ ¼ T2 S2t =S2 ðkÞ, ð2Þ
relationship for 12 developing countries. However, Pt
we do not do this in a bivariate framework as is the where St ¼ ei ði ¼ 1, 2, . . . , tÞ, ei is the partial sum
fashion; rather, given the limitations of modelling of the residuals, S2 ðkÞ is a consistent non-parametric
bivariate relationships, we model the relationship estimate of the disturbance variance and T is the
within a multivariate framework by including the sample size. Kwiatkowski et al. (1992) show that
real gross domestic product (GDP) variable. GDP the statistic ðuÞ has a nonstandard distribution, and
is an important variable, for government’s revenue critical values are provided therein. If the calcu-
and expenditure are both contingent on the level of lated value of ðuÞ is larger than the critical value,
economic activity. A growing economy, for instance, then the null of stationarity for the KPSS test is
will both demand increased spending on rejected.
1
Payne (1997), to the best of our knowledge, is the only study that models government revenue and government expenditure
together with GDP.
Government revenue and government expenditure nexus 287
Granger causality III. Empirical Results
The Granger causality test is based on the procedure
Data
suggested by Toda and Yamamoto (1995), which
allows for causal inference based on an augmented All data are from the World Bank World Tables and
level vector autoregressive (VAR) model with inte- the IMF publication International Financial Statistics
grated and cointegrated processes. The method is CD-ROM, and the sample period differs country by
useful because it does not require the practitioner country and is contingent on data availability.
to conduct tests for cointegration. The Toda and For instance, for Mauritius data is for the period
Yamamoto (1995) procedure uses a modified Wald 1966–2000, for South Africa data is for the period
test to conduct restrictions on the parameters of 1960–2000, for Peru data is for the period 1970–2000,
the VAR(k) model. The test has an asymptotic for Guyana data is for the 1961–1966, for Haiti
chi-squared distribution with k degrees of freedom data is for the period 1967–1997, for Chile data is for
in the limit when a VAR[k þ d(max)] is estimated the period 1973–1996, for Uruguay data is for
(where, d(max) is the optimal order of integration the period 1969–1996, for Venezuela and Ecuador
for the series in the system). The test essentially data is for the period 1950–1996, for El Salvador data
involves two stages. The first stage is where we need is for the period 1954–1996, for Guatemala data is
Downloaded by [Stony Brook University] at 21:24 31 October 2014
to determine the optimal lag length k and the for the period 1958–1996, and for Paraguay data
maximum order of integration d of the variables is for the period 1958–1993.
in the system. To achieve the optimal lag length, we
use the Schwarz Bayesian Criterion (SBC) and to Unit root tests
establish the order of integration, we use the ADF
and the KPSS tests. Once a VAR(k) is selected and The starting point for our empirical analysis is an
the order of integration d(max) obtained, a VAR in investigation of the unit root property of the three
levels is then estimated with the optimal lag length variables in our proposed model (see Equations 3–5).
of p ¼ [k þ d(max)]. The second stage is where we We find that for the natural logs of government
apply the Wald test to the first k VAR coefficient revenue, government expenditure and real GDP, the
matrix to test for Granger causality. These discus- null hypothesis of a unit root cannot be rejected
sions can be represented arithmetically as follows: at conventional levels of significance. However, when
we apply the ADF test to the first difference of these
X
kþdmax X
kþdmax
three series, we are able to reject the null hypothesis
ln GRt ¼ 0 þ i ln GRti þ i ln GEti
i¼1 i¼1
of a unit root at the 5% level of significance or better.
These results imply that government revenue, govern-
X
kþd max
þ i ln Yti þ "1t ð3Þ ment expenditure and the GDP series, for all the
i¼1 12 countries in our sample, are integrated of order
X
kþdmax X
kþd max one, Ið1Þ. The results are not reported here to
ln GEt ¼ 0 þ i ln GEti þ i ln GRti conserve space but are available from the authors
i¼1 i¼1 upon request. However, given the generally low
X
kþd max power of the ADF test, we also examine the unit
þ i ln Yti þ "2t ð4Þ root property using the KPSS test. We report the
i¼1 results in Table 1. The results generally suggest
X
kþd max X
kþd max that government revenue, government expenditure
ln Yt ¼ 0 þ i ln Yti þ i ln GEti and real GDP are non-stationary in their levels.
i¼1 i¼1
These findings corroborate those obtained from the
X
kþd max
ADF test.2
þ $i ln GRti þ "1t ð5Þ
i¼1
Granger causality results
Here, ln GR is the natural log of the government
revenue, ln GE is the natural log of the government The Granger causality test results are presented in
expenditure, ln Y is the natural log of the real gross Table 2. We find that there is unidirectional causality
domestic product, and "1t , "2t and "3t are serially running from revenue to expenditure for Mauritius,
independent random errors with a mean of zero and El Salvador, Haiti, Chile and Venezuela; hence,
a finite covariance matrix. there is support for the tax-and-spend hypothesis.
2
The Im et al. (2003) panel unit root test was also undertaken for the 3 series. We found that all the three variables were
non-stationary at the 10 per cent level or better. Detailed results are available from the authors upon request.
288
Downloaded by [Stony Brook University] at 21:24 31 October 2014
ln GRt 0.4709 [3] 0.3528 [4] 0.7649 [5] 0.5864 [5] 0.3725 [3] 0.3550 [2] 0.3830 [4] 0.5612 [3] 0.5853 [4] 0.3758 [4] 0.6442 [5] 0.3976 [4]
ln GEt 0.3562 [4] 0.6002 [4] 0.7215 [5] 0.4761 [5] 0.3609 [4] 0.3579 [4] 0.3863 [4] 0.6256 [3] 0.5767 [3] 0.3635 [4] 0.5533 [3] 0.3941 [3]
ln Yt 0.6879 [5] 0.6208 [4] 0.7446 [5] 0.3665 [4] 0.6212 [4] 0.7010 [5] 0.4276 [4] 0.6742 [3] 0.5995 [4] 0.6946 [4] 0.6970 [5] 0.6656 [5]
Notes: Critical values at the 1%, 5%, and 10% levels are 0.7390, 0.4630, and 0.3470, respectively.
Y ) GR 5.2509 1.8311 3.2035 9.1674** 1.9133 5.9274** 16.0101*** 4.7203 8.6832** 2.2862 7.6236** 1.8624
[0.1543] [0.6082] [0.3613] [0.0271] [0.5906] [0.0516] [0.0011] [0.1935] [0.0338] [0.5152] [0.0545] [0.3941]
Y ) GE 5.6313 8.8094** 3.3776 9.0880** 11.5467*** 0.0676 8.3956** 5.6201 2.4675 1.7087 5.6393 6.8830**
[0.1310] [0.0319] [0.3370] [0.0281] [0.0091] [0.9668] [0.0385] [0.1316] [0.4812] [0.6350] [0.1305] [0.0320]
GR ) GE 15.0536*** 0.7001 0.1481 1.9034 6.2983* 4.4276 10.3919** 11.1197** 1.9494 11.5857** 1.6631 15.4327***
[0.0018] [0.8732] [0.9855] [0.5927] [0.0980] [0.1093] [0.0155] [0.0111] [0.5830] [0.0089] [0.6452] [0.0004]
GE ) GR 2.6683 0.8501 3.2035 3.7455 4.9781 2.3063 14.9415*** 4.7968 2.4093 1.8642 14.5969 3.3147
[0.4456] [0.8374] [0.4836] [0.2903] [0.1734] [0.3156] [0.0019] [0.1873] [0.4919] [0.6011] [0.0022] [0.1906]
GE ) Y 3.5246 5.8894 11.1942** 12.1765*** 1.3746 14.0482*** 3.2627 1.4101 6.1838* 11.6684 2.4614 2.0336
[0.1376] [0.1171] [0.0107] [0.0068] [0.7115] [0.0009] [0.3527] [0.7031] [0.1030] [0.0086] [0.4823] [0.3618]
GR ) Y 5.3144 13.7302*** 5.8346 8.2734** 12.1325*** 9.2267*** 0.9912 5.6486 6.8355* 7.5098 2.8490 3.8294
[0.1502] [0.0033] [0.1199] [0.0407] [0.0069] [0.0099] [0.8034] [0.1300] [0.0773] [0.0573] [0.4823] [0.1474]
Notes: ***, **, * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
289
290 P. K. Narayan and S. Narayan
Meanwhile, for Haiti causality runs from government the signal is that government first spends and then,
expenditure to government revenue, supporting later, to pay for this expenditure, it raises taxes.
the spend-and-tax hypothesis. For the remaining Potential investors may construe this government
countries, namely Peru, South Africa, Guatemala, behaviour negatively – that is, investment decisions
Uruguay and Ecuador, there is neutrality between may take into account the possibilities of paying
revenue and expenditure, inconsistent with the higher taxes in future. In the light of this, it is safe
‘fiscal synchronization hypothesis’. This implies that to conclude that in countries where government
revenue and expenditure decisions are made expenditure causes government revenues, private
independently. investments may be hampered.
In terms of the causal relationship between GDP
and government’s revenue and expenditure, we find
some interesting results. First, we find that there is
bidirectional causality between GDP and government Acknowledgements
expenditure in the case of Guyana, and unidirectional We would like to thank Professor Mark Taylor and a
causality running from GDP to government expen- referee of this journal for helpful comments and
diture in the case of Peru, El Salvador, Haiti and suggestions on earlier versions of this paper.
Paraguay. Second, we find that there is bidirectional
Downloaded by [Stony Brook University] at 21:24 31 October 2014
Manage, N. and Marlow, M. (1986) The causal relationship the analysis of government expenditure growth,
between federal expenditure and receipts, Southern Public Finance Quarterly, 7, 3–23.
Economic Journal, 52, 617–29. Toda, H. Y. and Yamamoto, T. (1995) Statistical
Meltzer, A. H. and Richard, S. F. (1981) A rational theory inference in vector autoregressions with possibly
of the size of the government, Journal of Political integrated processes, Journal of Econometrics, 66,
Economy, 89, 914–27. 225–50.
Miller, S. and Russek, F. (1990) Cointegration and error Von Furstenberg, G. M., Green, R. J. and
correction models: the temporal causality between Jeong, J.-H. (1986) Tax and spend or spend
government taxes and spending, Southern Economic and tax, Review of Economics and Statistics, 68,
Journal, 57, 221–9. 179–88.