Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Econ Change Restruct (2011) 44:203–219

DOI 10.1007/s10644-010-9099-z

Economic growth and government spending


in Malaysia: a re-examination of Wagner
and Keynesian views

V. G. R. Chandran Govindaraju • Ramesh Rao •

Sajid Anwar

Received: 30 March 2010 / Accepted: 23 November 2010 / Published online: 4 December 2010
Ó Springer Science+Business Media, LLC. 2010

Abstract By making use of annual data from Malaysia for the period 1970 to
2006, this paper examines Wagner’s law and the Keynesian hypothesis concerning
the link between real government spending and real GDP. Unlike most existing
studies, we utilize both a bivariate and a multivariate model. In addition, we con-
sider two cases: one that focuses on the link between aggregate government
spending and GDP and the other where the link between government spending on
education and GDP is considered. The use of a multivariate model serves to reduce
the problem of serious misspecification which appears to have been ignored by most
existing studies. The presence of cointegration is investigated by means of Auto
Regressive Distributed Lag (ARDL) approach. This approach also allows one to
distinguish between the short and the long-run relationships. Within the context of a
bivariate model, our empirical analysis reveals that aggregate government spending
Granger causes the real GDP which supports Wagner’s law. However, in a multi-
variate framework, we found support for the Keynesian hypothesis suggesting that
omitted variables bias can significantly alter the validity of Wagner’s law.

Keywords ARDL bounds approach  Spending on education  GDP growth 


Malaysia  Government spending

V. G. R. Chandran Govindaraju (&)


Department of Economics, Universiti Teknologi Mara, 85009 Johor, Malaysia
e-mail: vgrchan@gmail.com

R. Rao
Department of Development Studies, Faculty of Economics and Administration,
University of Malaya, 50603 Kuala Lumpur, Malaysia

S. Anwar
Faculty of Business, University of the Sunshine Coast, Maroochydore DC, QLD 4558, Australia
e-mail: SAnwar@usc.edu.au

123
204 Econ Change Restruct (2011) 44:203–219

JEL Classifications H52  C22  O23  E62

1 Introduction

Wagner’s Law suggests that there is a long-run equilibrium relationship between


public spending and GDP. The law asserts that the causality runs from GDP to
public spending. In contrast, Keynesians view public spending as an exogenous
policy instrument that influences GDP growth.1 Both of these hypotheses have been
empirically tested for the developed as well as the developing countries yielding
mixed results.2 For example, Chow et al. (2002), Thornton (1999) and Ansari et al.
(1997) have found evidence in support of Wagner’s law. However, Ram (1986),
Afxentiou and Serletis (1996) and Wahab (2004) failed to find evidence in support
of Wagner’s law. Some studies have however, found evidence in support of both
Wagner and Keynesian hypotheses.3 For example, Devlin and Hansen (2001) and
Biswal et al. (1999) found evidence of bidirectional causality between the real GDP
and real public spending. It must also be noted that Huang (2006) among others has
found no evidence in support of either hypothesis. It is therefore clear that the
debate regarding Wagner’s law and the Keynesian hypothesis is far from settled.
It is also interesting to note that a large number of studies have merely attempted
to test the Wagner’s law and the Keynesian hypothesis by estimating a simple
regression equation involving only one independent variable. It is well known that a
large number of factor affect GDP growth. Government spending including
spending on education is also affected by a number of variables. It is clear that a
simple regression involving GDP and government spending is subject to serious
specification error (Chow et al. 2002). The main objective of this paper is to test
Wagner’s law and the Keynesian hypothesis in the Malaysian context by making
use of data for the period 1970–2006. We focus on the link between government
spending and the GDP. In addition, following other studies (Samudram et al. 2009;
Kolluri et al. 2000), we also consider the link between GDP and the components of
government spending. Specifically, we consider government spending on education
which in the context of our paper can be interpreted as a public input.
Our focus on government spending on education can be justified on a number of
grounds. First, governments around the world emphasize the importance of
education which has resulted in a steady increase in funding for education.
Education is found to be one of the key variables that have contributed to poverty
reduction (Grubb and Michelson 1974). Furthermore, increased global competition
has forced policymakers to pay more attention to funding for education in Malaysia.
Compared to other social services provided by the government, education receives
the largest allocation (See Table 1). Such a trend can also be noted in all the

1
See Al-Yousif (2008) for an interesting survey.
2
Studies that have considered the developed countries include Biswal et al. (1999) and Islam (2001).
Studies that have considered developing countries include Khan (1990) and Narayan, Nielsen and Smyth
(2008).
3
See Iyare and Lorde (2004) for an interesting survey.

123
Econ Change Restruct (2011) 44:203–219 205

Table 1 Federal government expenditure by sectors as percent of total expenditure


Sectors 1970 1980 1990 2000 2005 2006

Defense and internal security 23.1 19.1 13.6 11.0 13.2 13.8
Education 18.0 15.7 18.5 23.7 21.0 21.2
Transport 4.4 6.3 6.8 7.1 7.8 6.5
Health 6.1 4.4 5.0 6.4 6.6 6.4
Agriculture 8.7 8.7 6.6 3.0 3.3 4.1
Trade and industry 3.6 9.7 9.3 6.4 3.4 4.3
GDP (RM million) 11533 53308 119394 343215 495239 546343
Total expenditure (RM million) 2888 21163 35715 84488 128755 134748

Asian Development Bank, various issues

Malaysian development plans and blueprints. For instance, under the 9th Malaysian
Plan, 55% of the allocation for development of social sector has been set aside for
education and training (MOF 2006).
Secondly, as indicated earlier, spending on education has been used by several
studies as a measure of human capital (Romer 1990; J}urges and Schneider 2004). As
far as the developing countries are concerned, due to budget constraint, economic
growth may precede spending on education. As such understanding the causal link
between spending on education and GDP growth is important for policy reasons as
it has implications for government spending and other fiscal instruments. Thirdly,
although the relationship between education and growth is like a double edged
sword, without considering the causal directions of the link, many studies (e.g.,
Rehme 2007) have estimated econometric models by explicitly assuming that
education spending influences economic growth.4 Bilateral causality can be
examined by making use of appropriate statistical procedures, yet a number of
studies have made no attempt to investigate the direction of causality (Li and Liu
2005). This has implications for studies that have used public spending on education
as a proxy for human capital (Blankenau and Simpson 2004). Hence, examining the
issue of causality using Malaysia is likely to enhance our understanding of the
complex relationship between the two variables and provide lessons for developing
countries as well as for a substantial body of empirical and theoretical work that
fails to explicitly examine the linkage between the two variables. In examining the
link between total public spending as well as spending on education and GDP
growth, we turn our attention to two different schools of thought—Wagner and
Keynes. Unlike most existing studies, in order to minimize the specification error,
we include some control variables. We also make use of a relatively recently
developed estimation technique that allows one to examine the presence of a long-
run relationship between economic variables.

4
Following Romer (1990) a number of studies have used education expenditure as a proxy for human
capital. It has been argued that years of schooling and number of scientist and engineers are better
indicators of the level of human capital. However, due to unavailability of data many studies have used
spending on education as a proxy for human capital.

123
206 Econ Change Restruct (2011) 44:203–219

The rest of the paper is organized as follows. Section 2 provides a brief


background of the Malaysian economy and spending on education. Section 3
explains the methodology while Sect. 4 discusses the empirical results. Section 5
contains implication of the findings and concluding remarks.

2 Government spending, spending on education and GDP growth in Malaysia

The Malaysian economy registered an average annual growth rate of 6.7% from
1971 to 1990. The annual average GDP growth rate during 1990–2000 was 8.1,
which was higher than other ASEAN members. Prior to independence in 1957, the
Malaysian economy relied heavily on the primary and the agricultural sectors. The
transition to export-oriented industrialization started in the late 1960s. This shift has
made a significant contribution to economic growth in Malaysia. Government
spending as a percentage of GDP increased steadily from 25% in 1970 to 42% in
1980. The share of government spending in GDP over time has declined as a result
of increasing private sectors involvement in the economy. In 1990 and 2006,
respectively, the share of government spending in GDP was 30% and 25%.
However, the real government spending in Malaysia has increased from RM$4.7
billion to RM$120 billion in 2006. The operating expenditure accounts for more
than half of the total spending with the development spending accounting the rest.
Both of these spending have steadily increased over time.
The introduction of FDI promoting policies in the early 1970s resulted in the
growth of manufacturing exports. However, the Malaysian manufacturing sector is
now facing stiff competition from new low cost production locations in China and
Vietnam. Recognizing the important role that is played by human capital, the Ninth
Malaysia Plan stipulates additional spending on education and training. This policy
is aimed at increasing the number of skilled workers which is essential in enhancing
Malaysia’s competitiveness. In Malaysia, the two main sectors having the highest
allocation from the expanding fiscal expenditure includes defense and education.
However, relative trend over the years especially after 1990s indicates an increase in
allocation for education as compared to that of defense in percentage terms. In 2002,
the share of education spending in Malaysia as a percentage of GDP was
approximately 8%—well above its neighbors (4.2% in Thailand, 3.2% in the
Philippines and 0.9% in Indonesia).5 As a percentage of total government spending,
allocation for education was 18, 24 and 21.2%, respectively, in 1970, 2000 and 2006
(See Table 1). It is perhaps worth mentioning that the growth of the education
spending closely follows the trend of total government spending (see Fig. 1).
The importance of education in the economy can be seen through the role it plays
in the formation of human capital. Following growth literature that highlights
education as one of the primary components of human capital, a number of studies
(Ghailani and Khan 2004; Vinod and Kaushik 2007) have argued that education
plays a critical role in promoting economic growth. Heckman (2005) argues that the
Chinese government’s emphasis on investment in physical capital at the expense of

5
See World Bank (2009).

123
Econ Change Restruct (2011) 44:203–219 207

140000
Real Total Government Expenditure
120000 Real Total Education Expenditure

MYR Million
100000

80000

60000

40000

20000

0
1970 1975 1980 1985 1990 1995 2000 2005
Year

Fig. 1 Real government spending and real spending on education, 1970–2006

human capital could retard the economic development of the country. An increase in
spending on education in Malaysia can be viewed as an attempt by the government
to transform the Malaysian economy into a knowledge driven economy. After the
implementation of the New Economic Policy (NEP) in 1970, the government
intensified its efforts through fiscal expansion. The fiscal expenditure of the federal
government rose from RM2.8 billion in 1970 to RM 134.7 billion in 2006 (MOF,
various years). Prior to independence, education spending was considered a part of
consumption component in public accounts. In subsequent five year plans,
education spending has been viewed as a component of investment that leads to
human capital accumulation. For instance, the First Malaysia Plan (1966–70)
emphasized economic growth by mobilizing resources for education to improve the
human capital development. Similarly, the Second Malaysia Plan (1971–75)
emphasized education as a means to achieve social balance between societies.
Subsequently, rapid industrialization with the export orientation policy demanded
the Malaysia plans (until the recent Tenth Malaysia Plan) to emphasize on education
in order to significantly reduce the skill gaps at all levels.

3 An overview of the literature on public spending and economic growth

Several studies have attempted to establish the relationship between public spending
and GDP growth by making use of Wagner’s law and Keynesian views. These
studies vary considerably—from single country to those that use panel data on a
number of countries. However, the findings of these studies have been mixed.
Ansari et al. (1997) tested Wagner’s law for three African countries namely
Ghana, Kenya and South Africa. They applied both Granger and Holmes-Hutton
procedures to test the relationship and found evidence in support of Wagner’s law.
Another recent study for developing countries by Akitoby et al. (2006) examined
the relationship for 51 developing countries for the period 1970–2002. They found

123
208 Econ Change Restruct (2011) 44:203–219

evidence to support Wagner’s law at least in one of the spending component in 70%
of the countries. In contrast, Gandhi (1971) and Ram (1987) found no support of
Wagner’s law in developing countries. Studies using cross sectional data in
developing countries have also reported mixed results.
Recent studies focusing on developing countries using single country analysis
include Samudram et al. (2009) for Malaysia, and Narayan et al. (2008) for China.
Samudram et al. examine the role of public spending and economic growth in
Malaysia for the period 1970–2004 in the context of Wagner and Keynesian debate.
The results provide mixed evidence supporting Wagner and Keynesian views
depending on the components of spending under investigation. Similarly, Narayan
et al. (2008) tested Wagner’s law using a panel unit root, co integration and Granger
causality approach for Chinese provinces. As a whole, they found mixed evidence
supporting Wagner’s law for the central and western provinces but not for the
eastern provinces.
Similarly, Kolluri et al. (2000) investigated the relationship between public
spending and economic growth in G7 countries for the period of 1960–1993. Their
study supports Wagner’s law. Both the short and long-run effects of growth on
public spending were found to be significant. In addition, they also examined the
relationship between different components of government spending and growth.
Afxentiou and Serletis (1996), among others, examined Wagner’s law for the
European Union. They considered the long-run relationship between different
categories of government spending and GDP. In most cases they found no
significant link between government spendings and GDP growth. Additionally, they
also failed to detect causality from GDP to these spending categories thus rejecting
Wagner’s law. Others, rejecting Wagner’s law in developed countries include
Wahab (2004) and Ram (1986). Al-Yousif (2008) using the time-series data
examined the direction of relationship between education spending and economic
growth in the six GCC economies over the period 1977–2004. Their findings based
on Granger-causality analysis indicate mixed evidence on the relationship between
education spending and growth. The results were found to vary across countries and
proxies used for human capital. Al-Yousif (2008) suggests that future studies should
focus on the complex relationship between education and growth by considering the
countries that have similar policy and institutional environment.

4 Model specification, methodology and data

Most existing studies have attempted to test Wagner’s law by making use of the
methodology employed by (1) Peacock and Wiseman (1961) who used government
spending and GDP, or (2) Gupta (1967) who used government expenditure per
capita and GDP per capita or (3) Goffman (1968) who used government expenditure
as a share of GDP and GDP per capita. Since our study involves only one country,
the three methodologies do not produce significantly different results. Wagner’s
hypothesis can be tested by making use of the share of public expenditure in GDP
and income per capita (see Musgrave 1969). However, a large number of existing
studies have evaluated Wagner’s hypothesis by making use of government

123
Econ Change Restruct (2011) 44:203–219 209

expenditure and GDP (see Peacock and Scott 2000). The use of government
expenditure and GDP is preferred because it also allows one to estimate the long-run
elasticity of government expenditure with respect to GDP (Chow et al. 2002). In this
study, we employ Peacock and Wiseman’s version which leads to a bivariate model
employed in the literature as follows:
Gt ¼ b0 þ b1 Yt þ tt ð1Þ
where G, and Y, respectively are the natural logarithms of real government
expenditure, real GDP, and vt is the error term.
In addition to investigating the link between real GDP and real government
spending, we also separately consider the relationship between real government
spending on education (Et) and real GDP. In order to reduce the severity of the
omitted variables bias, we estimate Eq. 1 after including capital stock and labour as
control variables as follows6:
Gt ¼ a0 þ a1 Yt þ a2 Kt þ a3 Lt þ lt ð2Þ
where K and L, respectively are the natural logarithms of real capital stock and
labour and lt is the error term.
Equation 2 can be interpreted as the aggregate production function written in
terms of G [Y = f (K, L, G)]. The aggregate output (i.e., Y) depends on capital,
labour and real government spending. The real GDP can be viewed as a composite
public good/input or infrastructure which contributes to increased productivity. We
have also considered a specific part of government spending—i.e., real spending on
education (E) which is a proxy for human capital. The endogenous growth theory
has highlighted the important role that both human capital and government spending
play in the long-term economic development of a country (Barro 1990, 1991 and
Romer 1990).
Equation 2 is estimated by means of the ARDL approach which is based on a
system of equations. Consequently, the problem of serious multicollinearity
involving government spending (or spending on education), stock of capital and
labour can be mitigated. The ARDL approach employed in this study is also known
to yield consistent long-run estimates even when the right hand side variables are
endogenous (Inder 1993). Indeed, Pesaran and Shin (1999) have shown that by
using appropriate order of the ARDL model it is possible to simultaneously correct
for (1) serial correlation in residuals and (2) the problem of endogenous regressors.
Nevertheless, in order to check for robustness and to account for the omitted
variables bias, we have also estimated the bivariate model (i.e., Eq. 1).
The empirical analysis presented in this paper is based on the data collected from
the Asian Development Bank. Following Marwah and Tavakoli (2004), capital
stock is calculated by accumulating gross investment (using 5% rate of

6
Inclusion of control variables serves to reduce the problem of misspecification. Chow et al. (2002) have
argued that omitted variables in a bivariate analysis can lead to spurious conclusions. The equation that
we estimate in this paper is based on a production function where real output depends on capital, labour
and human capital (or public infrastructure/input). If we do not include labour and capital then we would
really be arguing that real output depends on human capital or public infrastructure/input alone and the
estimated model will surely suffer from omitted variables bias.

123
210 Econ Change Restruct (2011) 44:203–219

depreciation).7 All nominal variables such as the GDP, aggregate government


spending, spending on education and capital stock were converted into real values
by using the GDP deflator (2000 is the base year). We also considered the effect of
macroeconomics shocks by including dummy variables for recession in 1985, the
Asian financial crisis of 1997–1998 and recession in 2001. In addition, we also
estimated the model with linear trend. However, the estimated coefficients of these
dummies were insignificant and consequently were excluded from the estimation.8
In testing the Wagner’s law, the econometric methodology proceeds in three stages.
In stage one, we tested all time series variables for stationarity. It is well known that
failure to determine the stationary of the data could lead to spurious regression (Chang
2002). We used Augmented Dicky-Fuller (ADF) and Phillips-Perron (PP) unit root
testing procedures. The next step involves testing for cointegration. Engle and Granger
(1987) and Toda and Phillips (1993) have shown that ignoring the existence of
cointegration in the series could lead to serious model misspecification. There are two
ways of performing the cointegration test; the Johansen cointegration using
maximum-likelihood method (Johansen 1988) and the fairly new method known as
‘‘bounds testing approach’’ suggested by Pesaran et al. (2001) which has three major
advantages. First, the bounds test for cointegration was found to be super-consistent
for small sample studies. Second, it allows the variables to have different optimal lags.
Third, the approach allows one to include dummy variables that could account for any
macroeconomic shocks or institutional differences. Hence, in this study, owing to
small sample size, bounds test is preferred over Johansen’s approach.9
Peacock and Scott (2000) have suggested that cointegration approach alone was
enough to describe the Wagner’s view by examining the long-run relationship.
Specifically, in this paper, the model was tested for cointegration by making use of
the autoregressive distributive lag (ARDL) based bounds test. This involves testing
the following ARDL for Eq. 2:
DYt ¼ uy0 þ py1 Yt1 þ py2 Kt1 þ py3 Lt1 þ py4 Gt1
Xp Xp Xp
þ kiy DYti þ ciy DKti þ aiy DLti
i¼1 i¼0 i¼0 ð2:1Þ
X
p
þ biy DGti þ e1t
i¼0

7
Kt = (1 - .05) KDt-1?IDt, KDt is gross fixed capital formation and IDt is gross capital formation. In
the base year (i.e., 1969), IDt = KD0.
8
The estimated coefficient of the dummy variable for 1985 recession is 0.017 with a p-value of 0.356.
The estimated coefficient of the dummy for financial crisis of 1997–98 is 0.022 with a p-value of 0.425.
The estimated coefficient of the dummy for 2001 recession is 0.015 with a p-value of 0.321, whereas the
estimated coefficient of the linear trend is 0.127 with a p-value of 0.551. All dummy variables and the
linear trend are insignificant which is consistent with Samudram et al. (2009). Samudram et al. have found
no evidence of significant structural breaks arising from financial crisis on aggregate government
spending and spending on education.
9
A large number of studies involving relatively small sample size have utilised ARDL-bounds testing
procedure. For example, see Odhiambo (2010) and references therein.

123
Econ Change Restruct (2011) 44:203–219 211

DKt ¼ uk0 þ pk1 Kt1 þ pk2 Yt1 þ pk3 Lt1 þ pk4 Gt1
Xp X p
þ kik DKti þ cik DYti
i¼1 i¼0 ð2:2Þ
Xp Xp
þ aik DLti þ bik DGti þ e2t
i¼0 i¼0

DLt ¼ ul0 þ pl1 Lt1 þ pl2 Yt1 þ pl3 Kt1 þ pl4 Gt1
Xp Xp
þ kil DLti þ cil DYti
i¼1 i¼0 ð2:3Þ
Xp X p
þ ail DKti þ bil Gti þ e3t
i¼0 i¼0

DGt ¼ ue0 þ pe1 Gt1 þ pe2 Yt1 þ pe3 Kt1 þ pe4 Lt1
Xp X p
þ kie DGti þ cie DYti
i¼1 i¼0 ð2:4Þ
X
p X
p
þ aie DKti þ bie Lti þ e4t
i¼0 i¼0

where j = y, k, l, e; D is the first difference operator; uj0 is the constant; ps are the
long-run coefficients; k, c, a, b represent short-run dynamics and et is the random
variable which is assumed to be white noise. Similarly, the bivariate model can also
be tested using the same procedure without K and L.
The optimal lag structure under ARDL approach is determined by estimating
(p ? 1)k regressions for each equation, where p is the maximum number of lags and
k is the number of variables in the equation. The optimal lag structure is determined
by making use of Schwartz-Bayesian Criteria (SBC) or Akaike Information Criteria
(AIC). We used AIC to ensure that the residuals do not suffer from the problem of
significant serial correlation.
The asymptotic distributions of the test statistics are non-standard regardless of
whether the variables are I(0) or I(1). Two separate bounds test are employed to
examine the presence of long-run relationship among the variables of interest: an
F-test for the joint null hypothesis p1 = p2 = p3 = p4 = 0, and a t-test on the
lagged level dependent variable (so that H0: p1 = 0). Since the asymptotic
distribution of Wald or F statistics is non-standard, we use the critical bounds
provided by Pesaran et al. (2001). Pesaran et al. (2001) have computed two
asymptotic critical values—one when the variables are assumed to be I(0) and the
other when the variables are assumed to be I(1). These are respectively known as the
lower critical bound (LCB) and the upper critical bound (UCB). Following Pesaran
et al. (2001), if the test statistic exceeds the corresponding UCB then there is
evidence of a significant long-run relationship. Alternatively, if the test statistic is
below the LCB then the null hypothesis cannot be rejected. However, if the sample
test statistic falls between these two bounds then the result is inconclusive.

123
212 Econ Change Restruct (2011) 44:203–219

The long-run relationship can be estimated using the selected ARDL model. For
example, if variables are cointegrated in Eq. 2.1, where Y is used as the dependent
variable then there is a stable long-run level relationship among the variables, which
can be described as follows:
Yt ¼ U0 þ U1 Kt þ U2 L þ U3 Gt þ lt ð3Þ
where U0 ¼ uy0 =py1 ; U1 ¼ py2 =py1 ; U2 ¼ py3 =py1 ; U3 ¼ py4 =py1 and lt is
the usual error term.
These long-run coefficients are estimated by the ARDL approach to cointegra-
tion. The same process can be used when other variables are used as a dependent
variable. Given the existence of long-run relationship among variables, an error
correction representation can be developed as follows10:
2 3 2 3 2 3
Y a1 b11i b12i b13i b14i
6 7 6 7 6 7
6 K 7 6 a2 7 X p 6 b21i b22i b23i b24i 7
6 7 6 7 6 7
ð1  LÞ6 7 ¼ 6 7 þ ð1  LÞ6 7
6 L 7 6 a3 7 i¼1 6 b31i b32i b33i b34i 7
4 5 4 5 4 5
G a4 b41i b42i b43i b44i
2 3 ð4Þ
2 3 e1t
h 6 7
6 7 6 e2t 7
þ6 # 7ECTt1 þ 66 7
7
4 5 6 e3t 7
4 5
u
e4t
where (1 - L) is the difference operator; ECTt-1 is the lagged error-correction term
derived from the long-run cointegrating relationship; and e1t, e2t, e3t and e4t are
serially independent random errors with mean zero and finite covariance matrix.
The order of integration of each variable was determined by making use of the
augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) methods. Following
Pesaran et al. (2001), in order to conduct the bounds test for the null hypothesis of
no cointegration, we employed a VAR(p) model augmented with deterministic
variables, such as the intercept and a time trend. The lag length p of the underlying
VAR was chosen by AIC and equals 3 and 2, respectively for the bivariate and
multivariate models. This ensures non-existence of significant residual serial
correlation and hence the conditional ECTt-1 is not unduly over-parameterized.
The presence of a significant relationship in first differences of the variables
provides evidence on the direction of the short-run causation while a significant
t-statistic pertaining to the error correction term (ECT) suggests the presence
significant long-run causation. However, it should be kept in mind that the results of
the statistical testing can only be interpreted in a predictive rather than in the
deterministic sense. In other words, the causality has to be interpreted in the Granger
sense.

10
If cointegration is not detected, the causality test is performed without an error correction term (ECT).

123
Econ Change Restruct (2011) 44:203–219 213

5 Empirical findings

Table 2 shows the results of the Augmented Dicky-Fuller (ADF) and Phillips-
Perron (PP) tests. It is clear that the order of integration of the variables involved in
our model is mixed. Some variables are I(0) whereas others are I(1) and hence the
use of ARDL bounds testing approach is justified. The Phillips-Perron (PP) test
without trend indicates that Y, K, and L are I(0). In addition, the ADF test without
trend indicates that L is I(0). Whereas G can only be characterized as an I(1)
process. However, in first differences, all variables are stationary. The cointegration
results are reported in Table 3. The results for the bivariate model show that long-
run relationships exist when G and E are treated as endogenous variables at the 5
and 10% level of significance. In other words, Y is found to be the long-run forcing
variable for both G and E. This supports the Wagner’s law and the result is
consistant with Samudram et al. (2009). In contrast, in the multivariate model, it is
evident that a long-run relationship exists only when Y is used as the dependent
variable. The reported F-statistic of 6.74 (5.33), when G (E) is the exogenous
variable, exceeds the UCB value of 4.35, suggesting a significant co-movement
among variables at the 5% level of significance.11 Our empirical analysis therefore
seems to support the Keynesian hypothesis. The results clearly suggest that
inferences regarding long-run relationship can be quite sensitive to omitted variable
bias which undermines the validity of Wagner’s law.
It has been argued that the critical values computed by Pesaran et al. (2001) may
not be valid for a small sample study. Accordingly, we also used the critical values
that can be computed by making use of the response surface procedure developed by
Turner (2006). These critical values are considered to be more reliable when sample
size is small. In the present case the sample size is 37. By making use of the procedure
outlined by Turner, we computed the critical values of the upper and the lower bounds
(for the multivariate models) for the F-test at the 1% level of significance as 5.1314
and 6.8933. The values at the 5% level were computed as 3.5670 and 4.9176 and the
critical values at the 10% level of significance were computed as 3.5171 and 4.1053.
Based on these critical values it is clear that a cointegration relationship exists at the
5% level of significance when Y is the dependent variable suggesting that our
conclusions based on by Pesaran et al. critical values are valid.
Based on the results of the bounds test, it is possible to argue that GDP is
significantly affected by labour, capital stock and government spending both total
spending and spending on education. There is a clear evidence to suggest that a
long-run relationship exists among the GDP, stock of capital, labour and
government spending (including education spending) and that this relationship
has been ignored by several studies that have attempted to test Wagner’s hypothesis
by simply considering the link between GDP and government spending. Future
research should consider Wagner’s hypothesis after including appropriate control
variables. Panel studies must also include appropriate country specific variables.

11
Note that it is also found that cointegration exist when K serves as endogenous variable. Therefore, it
is also possible to argue that the capital stock is significantly affected by GDP, labour and government
spending on education.

123
214 Econ Change Restruct (2011) 44:203–219

Table 2 Testing for unit roots (ADF and PP Tests)


Variables Levels First differences
in Logs
Without Without With With Without Without With trend With trend
trend (PP) trend Trend Trend trend (PP) trend (PP) (ADF)
(ADF) (PP) (ADF) (ADF)

Y -2.742* -1.599 -1.404 -1.472 -5.789*** -5.788*** -8.007*** -6.035***


K -2.709* -1.995 -0.901 -1.215 -3.967*** -3.967*** -4.288*** -4.368***
L -3.277** -3.272** -0.853 -1.153 -6.849*** -6.755*** -8.618*** -6.633***
G -2.315 -1.966 -1.759 -1.614 -4.939*** -3.537** -5.172*** -3.745**
E -3.661*** -2.064 -2.208 -2.408 -5.442*** -5.291*** -7.343*** -5.471***

***, ** and * denotes significant at the 1, 5 and 10% level, respectively

Since the bounds test confirms the existence of long-run relationship between
capital, labour, government spending (or spending on education) and GDP in a
multivariate framework, it is possible to estimate the long-run impact of G and E on
Y using the ARDL (Pesaran et al. 2001). Although, cointegration alone is enough to
test the Wagner’s hypothesis (Peacock and Scott 2000), we proceed to estimate the
elasticity to identify the magnitude of the impact of the aggregate government
spending in Malaysia in the long-run. We have also separately considered the
impact of government spending on education in the long-run. Given that all
variables are in natural logarithm the estimated coefficients can be interpreted as
elasticities. The estimated long-run relationships are as follows:
Yt ¼ 8:799 þ 1:832Lt þ 0:300Kt þ 0:108Gt ð5Þ
ð0:000Þ ð0:000Þ ð0:000Þ ð0:002Þ

Yt ¼ 7:315 þ 1:615Lt þ 0:316Kt þ 0:153Et ð6Þ


ð0:000Þ ð0:000Þ ð0:000Þ ð0:032Þ

Equation 5 includes government spending as an exogenous variable whereas Eq. 6


includes government spending on education as an exogenous variable. All levels
estimates are highly significant and have the expected signs. The results imply that a
1% increase in G and E leads to a 0.11 and 0.15% increase in Y. The magnitude of
impact seems to be fairly small in this case.
The existence of a cointegration relationship suggests that there must be Granger
causality in at least one direction; however, it does not indicate the direction of
temporal causality between the variables. Next, we proceed to test the direction of
causation in the short-run and long-run. The direction of causality is shown in
Table 4. In the short-run, Y significantly Granger causes G, whereas a bi-directional
causality exists between E and Y. Examining the short-run causality in the
mutlivariate model, it is found that Y is insignificant at the 5% level when either
G or E is the dependent variable. This implies that (1) GDP does not Granger cause
government spending and (2) GDP also does not Granger cause spending on
education. Similarly, G and E are also found to be insignificant in the short-run.
Thus, we conclude that aggregate government spending and GDP are independent—
this same applies to government spending on education and GDP. This result holds

123
Econ Change Restruct (2011) 44:203–219 215

Table 3 Results of bounds test


Model F-Statistics Lag 5% 10%

Bivariate model with G Bounds critical values


Y|G 2.713 3 LCB = 3.79 LCB = 3.17
G|Y 6.323** 3 UCB = 4.85 UCB = 4.14
Bivariate model with E
Y|E 2.512 3
E|Y 4.652* 3
Multivariate model with G
Y|K,L, G 6.741** 2 LCB = 3.23 LCB = 2.72
K|Y,L, G 3.723 2 UCB = 4.35 UCB = 3.77
L|Y, K, G 3.566 2
G|Y, K, L 1.677 2
Multivariate model with E
Y|K,L, E 5.335** 2
K|Y,L, E 5.455** 2
L|Y, K, E 2.354 2
E|Y, K, L 3.077 2

** and * denotes 5 and 10% significant level. The Bounds critical values were obtained from Table CI
(iii) Case III: unrestricted intercept and no trend (Pesaran et al. 2001). The underlying equations pass
series of the diagnostic test (e.g., the residuals are homoskedastic, have no serial correlation and satisfy
the normality assumption). The CUSUM plot and CUSUMSQ plot showed that the test statistic was
within the 5% significant level suggesting the absence of any significant structural instability

only when labour and capital are included as control variables suggesting that a
bivariate model can lead to misleading results.
The long-run impact is captured by the error correction term (ECT). The ECT’s
coefficient was found to be statistically significant at 5% significance level. This
suggests that in the long-run G (or E), K and L Granger-Cause Y. This means that
causality runs interactively through the ECT from total spending (including
spending on education), capital stock and labour to GDP. The magnitude of the ECT
term suggests that a deviation from the equilibrium level of GDP during the current
period will be corrected by 62 and 80%, respectively in the next period to restore
equilibrium when there is a shock to the steady-state relationship. The speed of
adjustment is considered relatively high for a developing country. The results of the
long-run causality imply that an increase in aggregate government spending or
spending on education can have the desired effect in the long-run.

6 Concluding remarks

A number of studies have examined the validity of Wagner’s law and the Keynesian
hypothesis concerning the link between government spending and the GDP. This
paper examines Wagner’s law and the Keynesian hypothesis by making use of data

123
216

Table 4 Granger causality test

123
Dependent variables F-value (P-value) Coefficient [t-statsitic]

DYt DKt DLt DEt DGt ECTt-1

Bivariate model (with G)


DYt – – – – 1.040 (0.336) –
DGt 8.378*** (0.000) – – – – -0.234 [-3.018]**
Bivariate model (with E)
DYt – – – 4.0358 (0.012)** – –
DEt 4.481 (0.011)** – – – – -0.473 [3.131]**
Multivariate model (with G)
DYt – 32.261*** (0.000) 12.785*** (0.000) – 2.471 (0.130) -0.618 [-6.333]***
DKt 16.379*** (0.000) – 14.907*** (0.000) – 7.0907** (0.002) –
DLt 1.579 (0.223) 0.985 (0.418) – – 0.680 (0.573) –
DGt 2.185 (0.118) 3.201** (0.043) – – 0.296 (0.827) –
Multivariate model (with E)
DYt – 28.937*** (0.000) 4.509** (0.021) 0.568 (0.573) – -0.803 [-4.587]***
DKt 31.937*** (0.000) – 6.999*** (0.004) 1.524 (0.237) – -0.341 [-4.555]***
DLt 0.168 (0.846) 0.0141 (0.986) – 0.235 (0.791) – –
DEt 0.062 (0.939) 1.880 (0.172) 0.149 (0.862) – – –

Selection of lag terms is based on the AIC


*** and ** denotes significance at the 1% and the 5% level, respectively. The CUSUM plot and CUSUMSQ plot showed that the test statistics was within the 5%
significant level. Hence there is no evidence of any significant structural instability. The above equations are estimated using the least squares and when cointegration
exists ECTt-1 has been included as an additional independent variable
Econ Change Restruct (2011) 44:203–219
Econ Change Restruct (2011) 44:203–219 217

on government spending as well as education spending and the GDP. We focus on


Malaysia because the level of government spending increases over the years with
education spending being one of the highest. In addition, Malaysia has registered
strong economic growth over a significant period of time. Unlike a number of
existing studies that merely consider the relationship between GDP and government
spending, we test Wagner’s law and the Keynesian hypothesis after including
capital stock and labour as control variables which helps to reduce the problem of
serious misspecification.12 The empirical analysis is conducted by making use of a
relatively recently developed estimation technique (namely the ARDL bounds
testing approach) which is deemed to be robust in small sample. In addition, we also
consider the direction of causality, both in the short and the long-run.
The empirical analysis presented in this paper shows that within a multivariate
framework, Wagner’s law does not hold in the long-run in the case of Malaysia.
Instead, our analysis supports the Keynesian view which contradicts some existing
results based on a bivariate analysis (e.g., see Samudram et al. 2009). Studies based
on a two variable model suffer from serious misspecification which can render the
estimated results to be invalid.
In overall terms, the Keynesian premise that government spending precedes GDP
growth, in contrast to Wagner, is evident in Malaysia. Our analysis suggests that in
the case of Malaysia, government spending (including education) is an important
determinant of GDP growth. This suggests that in developing countries, at least in
Malaysia, efforts to promote education can lead to GDP growth. Consequently,
policy makers and international agencies should focus on promoting education and
training in Malaysia. However, it should be kept in mind that such investment can
yield benefits only in the long-run as shown by our analysis. In addition, our analysis
suggests that an examination of the GDP and government spending nexus without
considering control variables and the direction of causality has its limitation.

Acknowledgement We appreciate valuable comments from anonymous referees of this journal.

References

Afxentiou PC, Serletis A (1996) Government expenditure in the European Union: do they converge or
follow Wagner’s Law? Int Econ J 10:33–47
Akitoby B, Clements B, Gupta S, Inchauste G (2006) Public spending, voracity and Wagner’s law in
developing countries. Eur J Polit Econ 22(4):908–924
Al-Yousif YK (2008) Education expenditure and economic growth: some empirical evidence from the
GCC Countries. J Dev Areas 42(1):69–80
Ansari MI, Gordon DV, Akuamoah C (1997) Keynes versus Wagner: public expenditure and national
income for three African countries. Appl Econ 29(4):543–550
Barro R (1990) Government spending in a simple model of endogenous growth. J Polit Econ 98:S103–
S125
Barro R (1991) Economic growth in a crosssection of countries. Quart J Econ 106:407–443
Biswal B, Dhawan U, Hooi-Yean L (1999) Testing Wagner versus Keynes using disaggregated public
expenditure data for Canada. Appl Econ 31(10):1283–1291

12
The problems associated with the use of a bivariate model have also been highlighted by Wolde-
Rufael (2009).

123
218 Econ Change Restruct (2011) 44:203–219

Blankenau W, Simpson NB (2004) Public education expenditures and growth. J Dev Econ 73(2):583–605
Chang T (2002) An econometric test of Wagner’s law for six countries based on cointegration and error-
correction modelling techniques. Appl Econ 34(9):1157–1169
Chow Y-F, Cotsomitist JA, Kwan ACC (2002) Multivariate cointegration and causality tests of Wagner’s
hypothesis: evidence from the UK. Appl Econ 34(13):1671–1677
Devlin N, Hansen P (2001) Health care spending and economic output: granger causality. Appl Econ Lett
8(8):561–564
MOF (various issue) Economic report (1970–2007). Retrieved 18 Dec 2009, from http://www.treasury.
gov.my
Engle RF, Granger CWJ (1987) Cointegration and error correction: representation, estimation, testing.
Econometrica 55:251–276
Gandhi VP (1971) Wagner’s law of public expenditure: do recent cross section studies confirm it? Public
Financ 26:44–56
Ghailani JS, Khan SA (2004) Quality of secondary education and labour market requirement. J Serv Res
4(1):161–172
Goffman IJ (1968) On the empirical testing of Wagner’s law: a technical note. Public Financ 23:359–364
Grubb WN, Michelson S (1974) States and schools: the political economy of public school finance.
Lexington Books, Toronto
Gupta SP (1967) Public expenditure and economic growth: a time series analysis. Public Financ
22:423–466
Heckman JJ (2005) China’s human capital investment. China Econ Rev 16(1):50–70
Huang C-J (2006) Testing Wagner’s law using bounds test and a new granger non-causality test: evidence
for Taiwan. J Am Acad Bus 8(2):86–90
Inder (1993) Estimating long-run relationships in economics: a comparison of different approaches.
J Econom 57:53–68
Islam AM (2001) Wagner’s Law revisited: cointegration and exogeneity test for the USA. Appl Econ Lett
8:509–515
Iyare SO, Lorde T (2004) Co-integration, causality and Wagner’s law: tests for selected Caribbean
countries. Appl Econ Lett 11(13):815–825
Johansen S (1988) Statistical analysis of cointegration vectors. J Econ Dyn Control 12(2–3):231–254
Jürges H, Schneider K (2004) International differences in student achievement: an economic perspective.
German Econ Rev 5(3):357–380
Khan AH (1990) Wagner’s Law and the developing economy: time series evidence from Pakistan. Indian
J Econ 38:115–123
Kolluri BR, Panik MJ, Wahab M (2000) Government expenditure and economic growth: evidence from
G7 countries. Appl Econ 32:1059–1068
Li X, Liu X (2005) Foreign direct investment and economic growth: an increasing endogenous
relationship. World Dev 33(3):393–407
Marwah K, Tavakoli A (2004) The effect of foreign capital and imports on economic growth further
evidence from four Asian countries (1970–1998). J Asian Econ 15:399–413
Musgrave RA (1969) Fiscal systems. Yale University Press, New Haven and London
Narayan PK, Nielsen I, Smyth R (2008) Panel data, cointegration, causality and Wagner’s law: empirical
evidence from Chinese provinces. China Econ Rev 19:297–307
Odhiambo NM (2010) Finance-investment-growth nexus in South Africa: an ARDL-bounds testing
procedure. Econ Change Restruct. doi 10.1007/s10644-010-9085-5
Peacock A, Scott A (2000) The curious attraction of Wagner’s law. Public Choice 102(1/2):1–17
Peacock AT, Wiseman J (1961) The growth of public expenditure in the United Kingdom. Princeton
University Press, Princeton
Pesaran MH, Shin Y (1999) An autoregressive distributed lag modelling approach to cointegration
analysis. In: Strom S (ed) Econometrics and economic theory in the 20th century: the Ragnar Frisch
Centennial Symposium, Cambridge University Press, Cambridge
Pesaran MH, Shin Y, Smith RJ (2001) Bounds testing approaches to the analysis of level relationships.
J Appl Econom 16(3):289–326
Ram R (1986) Causality between income and government expenditure: a broad international perspective.
Public Financ 41:393–414
Ram R (1987) Wagner’s hypothesis in time-series and cross-section perspectives: evidence from ‘‘Real’’
data for 115 countries. Rev Econ Stat 69(2):194–204

123
Econ Change Restruct (2011) 44:203–219 219

Rehme G (2007) Education, economic growth and measured income inequality. Economica 74(295):
493–514
Romer P (1990) Endogenous technological change. J Polit Econ 98:71–102
Samudram M, Nair M, Vaithilingam S (2009) Keynes and Wagner on government expenditures and
economic development: the case of a developing economy. Empir Econ 36(3):697–712
Thornton J (1999) Cointegration, causality and Wagner’s law in 19th century Europe. Appl Econ Lett
6(7):413–416
Toda HY, Phillips PCB (1993) Vector autoregressions and causality. Econometrica 61:1367–1393
Turner P (2006) Response surfaces for an F-test for cointegration. Appl Econ Lett 13:479–482
Vinod HD, Kaushik SK (2007) Human capital and economic growth: evidence from developing
countries. Am Econ 51(1):29–39
Wahab M (2004) Economic growth and government expenditure: evidence from a new test specification.
Appl Econ 36:2125–2135
Wolde-Rufael Y (2009) Re-examining the financial development and economic growth nexus in Kenya.
Econ Model 26:1140–1146
World Bank (2009) World Development Indicators 2009. World Bank, Washington DC

123

You might also like