You are on page 1of 16

Empir Econ (2009) 36:697712 DOI 10.

1007/s00181-008-0214-1 ORIGINAL PAPER

Keynes and Wagner on government expenditures and economic development: the case of a developing economy
Muthi Samudram Mahendhiran Nair Santha Vaithilingam

Received: 15 November 2006 / Accepted: 15 February 2008 / Published online: 9 July 2008 Springer-Verlag 2008

Abstract This paper investigates the Keynesian view and the Wagners Law on the role of public expenditure on economic growth for Malaysia (19702004). The empirical results using the Auto-Regression Distributed Lag (ARDL) model and the bounds test (Pesaran et al. in J Appl Econ 16:289326, 2001) showed evidence of a long run relationship between total expenditures (including expenditures on defense, education, development and agriculture) and Gross National Product. The results also show that with the structural break in 1998, the long run causality is bi-directional for GNP and expenditures on administration and health, supporting both Keynes view and Wagners Law. For all other expenditure categories the long run causality runs from GNP to the expenditures, which supports Wagners Law. Keywords Wagners Law Keynesian view Public expenditures ARDL and Bounds test JEL Classication 1 Introduction Over the last two decades, many of the developing countries have experienced structural adjustment problems, tures and improve their revenue positions. An important H50 C22 O23 E62

M. Samudram (B M. Nair S. Vaithilingam ) School of Business, Monash University, Sunway Campus, Jalan Lagoon Selatan, 46150 Bandar Sunway, Selangor Darul Ehsan, Malaysia e-mail: Muthi.samudram@buseco.monash.edu.my M. Nair e-mail: Mahendhiran.nair@buseco.monash.edu.my S. Vaithilingam e-mail: Santha.vaithilingam@buseco.monash.edu.my

123

698

M. Samudram et al.

aspect of the adjustment process for these countries is the need to focus on scal adjustments and restructuring of the public sector. It has generally been accepted that structural imbalances were the result of high scal decits. The problem was further compounded by an expanding and inefcient public sector. Due to the linkage between scal decits and economic performance in the developing world, the growth of public expenditures as a proportion of GNP has received considerable attention from economists (Gupta 1967; Rubinson 1977; Landau 1983). Several studies in the literature used the cross-sectional, cross-country analysis (e.g. Ram 1987) and time series models (e.g. Khan 1990) to capture the dynamics between government expenditure and economic growth. Some studies have shown a positive inuence of government expenditure on economic growth (Holmes and Hutton 1990). There have also been a few studies that have shown the negative impact of government expenditure on economic growth (Landau 1983; Barth et al. 1990). In the time series models in the literature, the mere recognition of an association between government expenditure and economic growth is insufcient for establishing the direction of causality between these two attributes. An important drawback of the traditional time series is that these studies assume that the data generation process is stationary. Estimating a model when the series are non-stationary using ordinary least squares (OLS) can give rise to spurious regression results thus making the estimates unreliable (Chang et al. 2004). More recent studies have used error-correction models and cointegration analysis to capture the dynamics between government expenditures and economic growth. A study by Abu-Bader and Abu-Qarn (2003) argued that country specic factors are likely to inuence the causal nature of the relationship between government expenditure and economic growth. Further, Quah (1991) suggested that the underlying assumption of homogeneity of all economies in relation to growth path is unrealistic in crosssectional and cross-country studies due to the varying policies and implementation patterns across countries. To this extent, to overcome the problems cited above, more recent studies have applied the Granger causality test (Granger 1969) and cointegration methods in determining the direction of causality between these attributes. These methods have gained popularity over the years. For example, Ahsan et al. (1989), Ghali (1999) and Bharat et al. (2000) among many others used the Granger causality test and error-correction model (ECM) proposed in Engle and Granger (1987) to assess the dynamic longrun equilibrium relationship between government spending and economic growth for developed countries. In this context, two schools of thought exists namely the Wagners and the Keynesian school of thought. The fundamental argument for these two schools of thought centers on the direction of causality between government expenditure and economic growth, that is, whether government expenditure is a consequence of a growing economy or vice versa. Wagner in 1883 argued (which is referred in the literature as the Wagners Law) that during the industrialization process, as the per capita income of the nation increases, the share of the public expenditures increases as well, implying that causality runs from economic growth to government expenditure. However, the second school of thought, the Keynesian view postulates that government expenditure is a component of scal policy and can be used as a policy

123

Keynes and Wagner on government expenditures

699

instrument to inuence growth. Hence, the causality runs from government expenditure to economic growth. Some studies such as Gandhi (1971), Gupta (1967), Hondroyiannis et al. (1995) and Dritsakis and Adamopoulos (2004) found that Wagners law does hold for some of the components of expenditures. Ansari et al. (1997) showed that Keynesian view is not supported for Ghana, Kenya and South Africa. In addition, a bi-directional causality between these two attributes is also possible (refer to the study by Abu-Bader and Abu-Qarn 2003). As such determining this causal link is vital as it has implications for government spending as a scal policy instrument for the growth of the economy. From the foregoing discussion it can be noted that the direction of causality between government expenditure and economic growth has been debated intensely in the literature. However, most of the discussions in the literature were conned to developed countries (Bharat et al. 2000 and Ghali 1999, among others) and there are very few studies devoted to developing countries (Abu-Bader and Abu-Qarn 2003). The objective of this study is to empirically examine if the Keynesian view or Wagners Law hold for a developing country, namely, Malaysia for the period from 1970 to 2004. A two-step procedure was used to assess the above-mentioned hypothesis. In the rst step, the orders of integration of variables were assessed using the Philip-Perron unit-root test. Since this unit root tests do allow for the possibility of structural break in the series, the GregoryHansen procedure (Gregory and Hansen (1996a,b)) for testing of cointegration with endogenous structural breaks was used in this study. Secondly, the ARDL framework and the bounds test proposed in Pesaran et al. (2001) were used to examine whether economic growth and public expenditures are co-moving in the long run. The ARDL framework used in this study has two important advantages over the error correction models and cointegration methods proposed by Engle and Granger (1987), Johansen (1988) and Johansen and Juselius (1990). First, the order of integration of the explanatory variables in the ARDL framework proposed by Pesaran et al. (2001) can be a mixture of zero or one (I (0) or I (1), respectively). Second, the bounds test for cointegration was found to be super-consistent for small sample. Many of the above studies used annual data, where the sample period is small. Mah (2000) stated that the traditional cointegration methods, namely the EngleGranger and JohensenJuselius methods are not reliable for studies with small sample. Hence, the bounds test provides more robust results for studies with small or limited data, especially for studies on developing countries. The rest of the paper is organized in the following way. In Sect. 2, a brief review of the Malaysian economy is provided. In Sect. 3, the methodology and data used in this study is discussed. The empirical analysis is reported in Sect. 4. Policy implications are discussed in Sect. 5 and concluding remarks are given in Sect. 6.

2 A brief review of the Malaysian economy In this section, the evolution of the Malaysian economy from 1970 to 2004 is provided. Here, key developments that have had an impact on the Malaysian economy are discussed.

123

700

M. Samudram et al.

Prior to independence in 1957, the economic development of Malaya (Malaysia since 1963) was conned to tin-mining, rice and rubber cultivation. Since independence, economic growth in Malaysia was sustained at 5.4% (19661970) and an average growth rate of 8.3% during the period 19711980. The agriculture sectors share of real GDP stood at 40.2% in 1955 and steadily declined to 21% by the end of the 1980 (data were obtained from various Economic Reports, Ministry of Finance Malaysia, Department of Statistics Malaysia 19702004). The overdependence on the primary commodities such as rubber and tin provided the country with reasonable growth rates, but has subjected the economy to the instability arising from wide uctuations in commodity prices. Therefore the government began to not only diversify the agriculture sector to include oil palm, cocoa, timber and high yielding varieties of rice cultivation but also expanded the manufacturing sector. Besides providing incentives to promote foreign direct investment, the role of government in the economy has steadily expanded. In the 70 s with the introduction of the New Economic Policy, the government was directly and heavily involved in the economic activity of the nation. The share of the public and private sector investment as a percentage of real GDP rose from 20% in 1970 to 30% at the end of 1980. The expanding share reected the governments role in sustaining economic growth. As such the public expenditures (current and development) began to increase. The development expenditures increased from RM4.2 billion during 19661970 to RM27.5 billion during 19761980. In terms of the ratio of GDP, development expenditures were 25.1% in 1970 and gradually increased to a high of 42% in 1980 (data were obtained from various Economic Reports, Ministry of Finance Malaysia and Annual Reports of Bank Negara Malaysia 19702004). In terms of GDP, the ratio was 6.3 and 14.0%, respectively. There was a shift in strategy for nancing growth from the public sector to the private sector. The shift in policy was the result of a large public sector decit incurred during the 1980s. A national policy of Malaysia Incorporated was introduced in 1983 to enable the private sector to participate extensively in the economy with the government providing favorable regulatory framework, infrastructure facilities and other scal incentives. This approach reduced drastically the need for the government to borrow extensively to nance government projects. As a result of the private sector participation, total federal government spending declined from about 33.3% of GDP in 1980 to 24.6% in 2000 but subsequently increased to 26.7% in 2004 (Table 1). The share of development expenditures in total government expenditures declined from 42.1% in 1980 to 33.1% in 2000 and further declined to 24% in 2004. The rapid economic growth during the 1980 s and 1990 s generated increased revenues resulting in scal surplus since 1993. The extensive participation of the private sector in the economy also led to different sectoral allocations of government spending. The share of agriculture in total expenditures reduced drastically from 8.7% in 1980 to 6.6% in 1990 and to 2.7% in 2002. This was the result of the policy shift that led to rapid industrialization and increased employment in the manufacturing sector. However, from 2003 onwards, the share of agriculture in total expenditure witnessed a slight increase. This increase is attributed to improved productivity and efciency due to increased mechanization, ICT adoption and innovation in the agriculture sector.

123

Keynes and Wagner on government expenditures Table 1 Fiscal expenditures of the federal government by sector 1970 Total expenditures/GDP (%) 25.0 As a percent of total expenditure Defense and internal security Education Transport Health Agriculture Trade and industry General administration Current expenditures As a percent of GDP Defense and internal security Education Transport Health Agriculture Trade and industry General administration Current expenditures Development expenditures Total expenditures GDP (RM million) Total expenditure (RM million) 5.8 4.5 1.1 1.5 2.2 0.9 2.5 18.7 6.3 25.0 2888 6.4 5.2 2.1 1.58 2.9 3.2 2.6 19.3 14.0 33.3 4.1 5.5 2.0 1.5 2.0 2.8 2.4 21.0 8.9 29.9 2.7 5.8 1.8 1.6 0.7 1.6 3.3 16.5 8.1 24.6 84488 3.5 7.4 2.0 1.9 0.8 2.0 3.7 19.0 10.5 29.5 98992 3.7 8.1 2.1 1.8 0.8 1.5 2.6 19.0 9.9 28.9 4.2 7.4 2.5 2.3 0.9 1.4 2.6 19.0 9.9 28.9 23.1 18.0 4.4 6.1 8.7 3.6 10.1 74.9 19.1 15.7 6.3 4.4 8.7 9.7 7.8 57.9 42.1 13.6 18.5 6.8 5.0 6.6 9.3 7.9 70.1 29.9 11.0 23.7 7.1 6.4 3.0 6.4 13.4 66.9 33.1 11.7 25.0 6.8 6.3 2.8 6.8 12.6 64.4 35.69 12.8 28.1 7.1 6.4 2.7 5.1 9.0 65.6 34.4 14.3 25.5 8.5 7.8 2.9 4.8 9.0 65.7 34.3 1980 33.3 1990 29.9 2000 24.6 2001 29.6 2002 28.9 2003 29.0

701

2004 26.7 13.6 21.5 7.7 7.8 3.8 2.9 12.7 76.0 24.0 3.6 5.7 2.1 2.1 1.0 0.8 3.4 20.3 6.4 26.7

Development expenditures 25.1

11533 53308 119394 3432166 334406 362012 395017 449609 17762 35715 104676 114577 120162

Source: Economic Report, Ministry of Finance, various years

The allocations for the transport sector as a percentage of total expenditure has been increasing from 6.3% in 1980 to 7.7% in 2004, reecting governmentprivate sector collaborations in infrastructure development. The government expenditures on education reached a high of 28.1% in 2002 due to the construction of more schools and universities. Government had allocated RM112.7 million to implement the Electronic Government Flagship Project in the year 2002, of which RM72.3 million was allocated for Smart Schools. Apart from this, an additional amount of RM205.5 million was allocated for the introduction of information and communication technology (ICT) in schools. The general administration expenditures were maintained at an average share of 10% throughout the sample period of study. From 2000, higher allocations were provided for ICT infrastructure for the various ministries and for the implementation of the government electronic delivery system (electronic government initiative).

123

702

M. Samudram et al.

3 The empirical methodology There are a number of models used to test the Wagners Law and Keynesian view (Dritsakis and Adamopoulos 2004; Murthy 1993; Musgrave and Musgrave 1988; Gupta 1967; Mann 1980). All these models can be described by the following economic relationship: yt = f (xt ) + u t , (1)

where in the case of Wagners Law, yt = lr totex pt and xt = lrgnpt . In the case of Keynesian perspective, yt = lrgnpt and xt = lr totex pt .lr totex pt is the logarithm of real total government expenditures, lrgnpt is the logarithm of the real gross national product and u t is the residuals at period t. In this paper, prior to estimating the relationship between public expenditures and real gross national income, we conduct two important statistical tests. First, we determine the order of integration of variables using the Philip-Perron unit-root test (PP-test). If the variables were stationary, the ordinary least squares (OLS) method will be used to estimate the relationship between the variables. However, if the series are non-stationary, using OLS estimation will lead to spurious estimates. To overcome this problem, the ARDL model proposed by Pesaran et al. (2001) will be used to capture the dynamics between real total government expenditure and gross national product. If the series has structural breaks, then the dynamics between the variables can change. We therefore describe below the method used to determine structural breaks. Equation 1 was estimated with the assumption that yt and xt series did not undergo any structural change during the period of estimation. Structural breaks may have occurred due to the Asian Financial crisis in 1997/1998 or the recession of 1985, which resulted in negative growth rates of GDP during both instances. In this situation, the unit root hypothesis may have been rejected in favour of the series being I(0). The dynamics between economic growth and public expenditure were estimated using a two-step procedure. The procedure is discussed below. 3.1 GregoryHansen structural break test Malaysia was one of the regional economies that was severely impacted by the 1997 Asian Financial Crisis, where the country posted negative GDP growth rates and serious currency devaluation. Hence, in this step, we assess whether the economic have undergone structural change due to the nancial crisis. The Gregory and Hansen (1996a,b) test was used to ascertain if the structural change have an impact on the cointegration relationship between total government expenditures, including its components (lrtotexp) and economic growth (lrgnp). The null and alternate hypothesis is as follows: HO : No cointegration with structural break (1998 Asian Financial Crisis) H A : Have cointegration with structural break (1998 Asian Financial Crisis) Four models were estimated with alternate assumptions of the structural breaks and they are given as follows:

123

Keynes and Wagner on government expenditures

703

Model 1(Level shift): yt = 1 + 2 Dt + 1 xt + et , Model 2 (Level shift with trend): yt = 1 + 2 Dt + 1 Trend + 1 xt + et , Model 3 (Regime shift where intercept and slope coefcient change): yt = 1 + 2 Dt + 1 Trend + 1 xt + 2 (Dt xt ) + et , Model 4 (Regime Shift where intercept, slope coefcient and trend change) yt = 1 + 2 Dt + 1 Trend + 2 (Dt Trend) + 1 xt + 2 (Dt xt ) + et , (5) where, Dt is the dummy variable capturing the structural periodpre and post 1997 Asian Financial period (Dt = 0 if t 1997 and Dt = 1 if t > 1997). The trend variable is denoted by Trend. The GregoryHansen (GH) method estimates the cointegration relationship for the four models above by computing the absolute value of the AugmentedDickey Fuller test statistic (this is referred to as the GH-test statistic). The GH-test statistic is compared with the critical values given in Gregory and Hansen (1996a,b). If the GH-test statistic value is below the GH-critical value, do not reject the null hypothesis. In this case, the structural break does not have a signicant impact on the long run stability of (1). On the other hand, if the GH-test statistic value is above the GH-critical value, reject the null hypothesis. In this case, the structural break has a signicant impact on the long run stability of (1). 3.2 The ARDL estimation and the bounds test Next we describe the ARDL model to capture the dynamics between xt and yt . The bounds test will be used to test if expenditure and real gross national product are co-moving. The bounds test will also be used to test if lrtotexpt is a long run forcing variable on lrgnpt or vice-versa. Assume that the data generation process (DGP) for z t is given by the following specication:
p

(2)

(3)

(4)

zt = +
j=1

j zt j + t ,

(6)

where, zt = [yt , xt ] , is the vector of constant terms = y , x and j is a matrix of VAR parameters for lag j. The two series yt and xt can either be I (0) or I (1). Note that the series in xt can also be of different order of integration. The error term t is white noise.

123

704

M. Samudram et al.

If the structural break does not have a signicant impact on the data generation process (DGP), the DGP for yt in (6) can be re-written as the following auto-regressive distributed lag ARDL [ p, q] model:
p1 q1

yt = + yt1 + xt1 +
j=1

y, j yt j +
j=1

x, j xt j + u t ,

(7)

To test the cointegration relationship between yt and xt , we use the bounds test proposed in Pesaran et al. (2001). The test is conducted as follows, we estimate the DGP in (7) using the ordinary least square (OLS) method, and test the absence of a long-run relationship between yt and xt . The test is similar to the Wald-type test (F-statistics), where the null is the absence of long-run relationship between yt and xt . The alternative hypothesis is the case where long-run relationship between yt and xt exists. More formally, the null and alternative hypotheses are as follows: H0 : = = 0, H1 : = 0 and = 0. The asymptotic distribution of this test statistic is non-standard under the null hypothesis of no cointegration relationship between yt and xt , regardless whether the order of integration of the explanatory variables is I (0) or I (1). We compare the computed F-statistics under the null hypothesis with the critical values given in Pesaran et al. (2001). If the computed F-statistics (denoted as F (y|x)) is above the Upper Critical Bound (UCB), we conclude that yt is cointegrated with xt . We also conclude that xt is the long run forcing variable on yt . In this case, the stable long-run relationship between yt and xt can be described as follows: yt = 0 + 1 xt + vt (8)

where 0 = , 1 = and vt is a mean zero stationary process. Note that 1 is the long-run elasticity (Bardsen 1989). If F (y|x) is below the Lower Critical Bound (LCB), we conclude that yt is not cointegrated with xt . In this case, we also conclude that xt is not a long run forcing variable on yt . On the other hand, if F (y|x) is between LCB and UCB a conclusive inference cannot be made without examining the unit root properties of the series. The lag lengths for the models were determined using the Akaike Information Criterion (AIC) (Akaike 1973). The residuals of the model were tested for heteroskedaticity and serial correlation. The JarqueBera test was conducted to ascertain if the residuals were normally distributed. The Ramsey test was conducted to determine if the models were correctly specied. If the structural break does have a signicant impact on yt for Model-1, then estimation in (7) will include the level shift term (Dt ). Similar approach is taken if the structural break has signicant impact for Model-2 (inclusion of Dt and Trend), Model3 (inclusion of Dt , Trend and Dt xt ), and Model-4 (inclusion of Dt , Trend, Dt xt and Dt Trend).

123

Keynes and Wagner on government expenditures

705

The dynamics between lrgnp(is the log of real GNP at 1987 prices) and the disaggregated government expenditures (lradminthe logarithm of real government administrative expenditures, lreducationlogarithm of real government expenditures on education, lrdefenselogarithm of real government expenditures on defense, lrcurrent logarithm of current government expenditures, lrdeveloplogarithm of development expenditures, lrhealthlogarithm of expenditures on health, and lragri logarithm of expenditures on agriculture development) were also estimated using the empirical method outlined above. The sample period for this study is from 1970 to 2004. The variables are dened as above. All the nominal expenditure variables were deated using the gross domestic product deator. The data were obtained from the various issues of the Economic Report from the Ministry of Finance, Bank Negara Annual Reports and Year Book of Statistics Malaysia (Department of Statistics 19702004). 4 Empirical ndings The empirical results are reported in this section of the paper. The order of integration of all the series using the PP-test (Phillips and Perron 1988) is reported in Table 2. As it is evident from the results, the PP test fails to reject the null hypothesis of unit root in all the variables except lardmin and lreducation at 1 and 5% level of signicance respectively. Table 3 provides the results of cointegration tests for one structural break (1998). The results are cointegration tests based on models given in Eqs. (2)(5). The results
Table 2 Unit root test on the variables

Variables

Philip-Perron Test Statistic Level 1st difference Level 1st difference Level 1st difference Level 1st difference Level Level 1st difference Level 1st difference Level 1st difference Level 1st difference 1.8903 4.3188 2.7536 9.8600 2.9713 5.4589 3.9316 8.7394 14.810 2.8975 17.4973 2.7923 6.8948 1.9057 5.8797 2.6541 4.9972

order of integration

lrgnp lrcurrent lrdefence lreducation lradmin lrhealth lrtotalexp lpubcons


, and are signicant at the

I (1) I (1) I (1) I (1) I (0) I (1) I (1) I (1) I (1)

lrdevelop

1 and 5% signicance levels, respectively

123

706

M. Samudram et al.

Table 3 GregoryHansen test for structural break with break in 1998 (expenditures as dependant variable) Variables GH statistic Model-1 lrtotexp lreducation lradmin lrdefense lrhealth lragri lrdevelop lrcurrent 2.3153 3.6058 5.1245 2.7063 3.4902 1.6574 2.4442 2.1335 Model-2 2.9708 3.8284 6.0542 2.6605 3.4574 1.8341 2.4533 2.6364 Model-3 2.4451 3.9500 6.2434 2.9318 5.4471 1.9936 2.4233 2.7057 Model-4 2.5251 4.3206 6.2241 2.7377 5.8714 1.9468 2.3835 2.7014

The 1 and 5% critical values are denoted as and , respectively. Model 1s 1 and 5% critical values are 5.13 and 4.61, respectively. Model 2s 1 and 5% critical values are 5.45 and 4.99, respectively. Model 3s 1 and 5% critical values are 5.47 and 4.95, respectively. Model 4s 1 and 5% critical values are 5.97 and 5.50. The critical values were obtained form Gregory and Hansen (1996b) Table 4 GregoryHansen structural break test (for 1998) with GNP as the dependant variable Variables GH statistic Model-1 Lrgnp = f(lrtotexp) Lrgnp = f(lreducation) Lrgnp = f(lradmin) Lrgnp = f(lrdefense) Lrgnp = f(lrhealth) Lrgnp = f(lragri) Lrgnp = f(lrdevelop) Lrgnp = f(lrcurrent) 2.0658 3.1531 4.7262** 2.1329 3.0836 1.6622 1.8450 1.9894 Model-2 3.5630 3.3211 3.5456 3.0120 3.2347 3.2377 3.2204 3.0709 Model-3 2.9349 2.8880 3.2973 2.3433 2.5940 3.0576 2.5082 3.0659 Model-4 2.9258 2.9339 3.0845 2.2993 2.6041 3.2049 2.6597 2.9629

The 1 and 5% critical values are denoted as and , respectively. Model 1s 1 and 5% critical values are 5.13 and 4.61, respectively. Model 2s 1 and 5% critical values are 5.45 and 4.99, respectively. Model 3s 1 and 5% critical values are 5.47 and 4.95, respectively. Model 4s 1 and 5% critical values are 5.97 and 5.50. The critical values were obtained form Gregory and Hansen (1996b)

show that none of the variables other than lradmin and lrhealth exhibited any structural breaks, implying that there is cointegration between total government expenditures and economic growth (lrgnp). Only in the case of lradmin (expenditures on administration) and lrhealth (expenditures on health), the results indicate the structural break in 1998 is signicant and hence the cointegration analysis were conducted including the structural break terms in the equations. Table 4 provides the results of cointegration tests between lrgnp and total government expenditures (including all the components of government expenditures). The results show that the structural break of 1998 did not have signicant impact on the cointegration relationship when lrgnp is used as an endogenous variable.

123

Keynes and Wagner on government expenditures Table 5 Bounds test without structural break y Lrtotexp Lreducation Lrdefense Current exp Lrdevelop Lragri Lradmin x Lrgnp Lrgnp Lrgnp Lrgnp Lrgnp Lrgnp Lrgnp F (y|x) 17.67 14.54 13.95 2.88 12.03 11.30 5.02 y Lrgnp Lrgnp Lrgnp Lrgnp Lrgnp Lrgnp x Totexp Lreducation Lrdefense Current exp Lrdevelop Lragri

707

F (y|x) 1.011 0.950 0.924 0.873 1.777 0.765

The 1 and 5% critical values are denoted as and , respectively. The 1% [LCB, UCB] = [6.84, 7.84]; 5% [LCB, UCB] = [4.94, 5.73]; 10% [LCB, UCB] = [4.04, 4.78]. These critical values were obtained from Pesaran et al. (2001), Table CI(III): Unrestricted intercept and restricted trend, with one regressor (k = 1)

To select the appropriate model for the different categories of government expenditures, we estimated the ARDL [ p, q] using different lag lengths for p and q, i.e., p = {1 to 6} and q = {1 to 6}. The estimations were conducted using the statistical software MICROFIT4. The diagnostic test on the residuals (heteroskedasticity, serial correlation, JarqueBera Normality) and the model specication test (Ramsey test) was conducted. The estimated residuals were found to satisfy the standard regularity conditions. Based on the model specication test, the estimated specication was correctly specied (the diagnostic tests and model specication test are available from the authors). The bounds test was employed to determine the direction of causality between the variables (that is to identify the long-run forcing variables). The results for the bounds test (where the structural break in 1998 did not have signicant impact on the cointegration relation) are reported in Table 5. Based on the bounds test statistic, F (y|x) and the critical bounds, we observe thatlrgnpis the long run forcing variable on the following: lrtotexp, lreduc, lrdef, lrdevel and lragri at the 1% signicance level. None of the other variables were found to be long run forcing variables. It can also be noted that there was no evidence of reverse direction of causality of the variables that is lrtotexp, lreduc, lrdef, lrdevel and lragri were not long run forcing variables on lrgnp. The bounds test for capturing the direction of causality between lradmin & lrgnp and lrhealth & lrgnp are given in Table 6 (models where 1998 structural break had a signicant impact on the data generation process). In the case of lradmin-lrgnp, Model 1 suggests that direction of causality is unidirectional, that is lrgnp long-run forcing lradmin and not the reverse. Model 2 and Model 3 suggest that the direction of causality is bi-directional. However, Model 4 suggest that neither one of the variables long run force each other. As for the direction of causality between lrgnp and lrhealth, Model 1 and Model 2 are bi-directional. However, Model 3 and Model 4 suggest a unidirectional causal relationship running from lrgnp to lrhealth. The estimated long-run elasticities (without structural breaks) are given in Table 7. In the long run, one of the ve expenditure categories (that is, education) registered elasticities greater than one. This implies that as the economy grew, these expenditures

123

708 Table 6 Bounds test for cointegration with structural break Model y Lradmin 1 2 3 4 Lrhealth 1 2 3 4 Lrgnp 6.551 8.111 21.175 23.739 x Lrgnp 27.7842 14.639 5.3776 4.6008 Lrgnp F(y|x) y Lrgnp x

M. Samudram et al.

F(y|x)

Lradmin 0.61092 8.583 6.2305 3.6429 Lrhealth 5.2588 8.0754 2.5114 2.1876

The 1, 5 and 10% critical values are denoted as , and , respectively. The 1% [LCB, UCB] = [6.84, 7.84]; 5% [LCB, UCB] = [4.94, 5.73]; 10% [LCB, UCB] = [4.04, 4.78]. These critical values were obtained from Pesaran et al. (2001), Table CI(III): unrestricted intercept and restricted trend, with one regressor (k = 1) Table 7 Long run elasticities of government expenditures in Malaysia (without structural breaks)

Dependent Variable Lrtotexp Lrdefence Lreducation Lrdevelop Lragri Lradmin

Determinant Lrgnp Lrgnp Lrgnp Lrgnp Lrgnp Lrgnp

Elasticity 0.812 0.658 1.032 0.783 0.234 1.21

, and are signicant at the

1 and 10% signicance levels, respectively

Table 8 Long run elasticities of government expenditures in Malaysia (with structural breaks) Dependent variable Lradmin Lrhealth Determinant Lrgnp Lrgnp Elasticity 1.1177 (Model 1) 0.7685 (Model 4)

, and are signicant at the 1 and 10% signicance levels, respectively

grew faster than the growth of economy. In the case of total expenditures, defense expenditures, development expenditures and agriculture expenditures, the elasticities were less than one, implying that the impact on growth of the economy by these expenditures was less than the growth of economy. However, agriculture expenditures were found to be statistically insignicant at the 10% level. The long run elasticity of lrgnp on lradmin and lrhealth is given in Table 8. The long run elasticity for expenditures on administration (lradmin) is 1.117 (model 1), signicant at the 1% level and is very close to 1.21 (without structural break). With regards to health expenditures, the long run elasticity is 0.7685 (signicant at 10%)

123

Keynes and Wagner on government expenditures

709

implying that Model 4 is the only model relevant to explain the long run relationship between expenditures and lrgnp.

5 Policy implications Total government expenditures consist of current and development expenditures. The current expenditures of the government are to maintain basic services that the government provides to the public including the payment of salaries to the civil servants (this is the largest component of current expenditures). Therefore, these expenditures not only facilitate growth prospects but in part promote growth as well. The development expenditures on the other hand are allocated sectorally to provide basic infrastructures (such as roads, railways, bridges, ports and airports) to facilitate movement of goods and services across the country. During the early stages of economic growth, these expenditures increased in tandem with economic growth. However, in the 90 s, the development expenditures in particular increased signicantly to generate the growth of the economy. The Government of Malaysia has begun to recognize that economic growth needed to be nanced not only by the public sector but also the private sector. In the 1980s, efforts were made to promote private sector as an important engine of economic growth. This was largely due to the problems of ination and public sector decits that the government faced in the 1980s. The government expenditures during this period were focused mainly in improving the public sector delivery system. Expenditures on General administration rose from 10% of the total expenditures in 1970 to 13.4% in 2000 and declined marginally in the interim to rise to 12.7% in 2004. As a percentage of GDP the ratio recorded 3.4% in 2004. Given the output elasticity of 1.12, economic growth has signicant impact on government expenditures. The long run output elasticity of 1.12 indicated that economic growth generated greater demand for public services. This resulted in an increase in expenditures on general administration. For enhancing the efciency in the delivery system, the government embarked on the use of ICT into various aspects of the government administration. Under the Electronic Government Applications, the e-government, e-procurement and e-services were introduced to improve the efciency and the productivity of the government. The defense expenditures as a percentage of total expenditures declined from 23% in 1970 to 11% in 2000 but subsequently witnessed an increase during the next four years to reach 13.6% in 2004. In terms of GDP, the ratio stood at an average of 3.6% during the last decade. The rise of defense expenditures is not only the result of the growth of the economy but is also in response to the needs of regional security in the face of competing territorial claims. The expenditures were also in line with the continuous upgrading and modernization of the armed forces, the police and other security services due to increasing sophistication and challenges in combating crimes and elements that threaten national and regional security. Though the output elasticity is less than one (0.658), the causality runs from economic growth to defense expenditures. Economic growth is the long run forcing variable and the F statistic is highly signicant at 1%. The Malaysian government over the last two decades has been investing in the education sector with the expectations of raising the competitiveness of the workforce.

123

710

M. Samudram et al.

With a relatively smaller workforce compared to other regional economies such as Indonesia, China and India, Malaysian workers would not be able to continue to compete in labour-intensive sectors. Further a robust economy in the 1990s, had a signicant upward spiral effect on wages, making Malaysia less competitive in the region. To move up the value chain, the government increased its investments in the knowledge economy to technical know-how. More schools and universities were established to increase the diffusion of these new technologies among the new generation of Malaysians. This is one reason why that the long run elasticity of real gross national product on education is greater than unity (1.034) and was statistically signicant and the causality runs from lrgnp to lreducation. The ratio of development expenditures as a percentage of total expenditures increased from 25.1% in 1970 to a high of 36% in 2001 and declined to 24% in 2004. Development expenditure is the second largest component of total expenditures. In a developing economy such as Malaysia, these expenditures play a vital role in the overall development of the country. Therefore a large portion of these expenditures go into the provision and expansion of economic services such as rural development, transport, communications, maintenance and improvement of public utilities. Economic growth therefore has direct impact on these expenditures. Though the output elasticity is less than one, development expenditures continue to rise with the development of various sectors of the economy. Lrgnp is the long run forcing variable and the F statistic is signicant at 1%. Therefore the causality is uni-directional running from lrgnp to lrdevelop(development expenditures). The empirical results also showed that lrgnp is a long run forcing variable for expenditures on the agriculture sector (lragri). However, the long run elasticity of expenditures on agriculture with respect to output was not statistically signicant. This is consistent with the directions taken in the late 1980s, where the agriculture sector was given less priority over other value-added sectors such as manufacturing and services. Hence, during the economic boom in the 1990s, less resource were allocated to the agriculture sector and the contribution of this sector to the overall economy continued to decline. Further, low investments in the agriculture sector have resulted, in this sector still remaining labour intensive and losing its competitiveness to other larger agriculture based economies such as Indonesia, Thailand and Philippines. In the more recent government plan (ninth Malaysian Plan), the agriculture sector is seen as an important engine of economic growth from 2006 to 2010. An important emphasis of the 9th Malaysian Plan is to transform the agriculture sector to be more knowledge-driven and be globally competitive. The development of the agriculture sector is expected to have positive externalities for less developed areas of the country which includes improving the infrastructure (electricity, water, health-care, telecommunication and schools) and employment opportunities. From the above discussions, the empirical analysis supports the Wagners Law for total expenditures, expenditures on education, defense, development, administration, health and agriculture. However, for expenditures on administration and health services, both Wagners Law and Keynesian View of role of government expenditures are relevant in that the causality is bi-directional. For these two categories of government expenditures, structural break in 1998 was signicant and among the four

123

Keynes and Wagner on government expenditures

711

different models estimated, models 2 and 3 for lradminand models 1 and 2 for lrhealth are relevant and statistically signicant. 6 Conclusions There are two schools of thought on the direction of causality between government expenditures and economic growth. The Keynesian view argues that causality runs from government expenditure to economic growth. On the other hand, the Wagners theory suggests that the economic growth impacts government expenditure. In this paper, the two hypotheses were empirically tested for a developing country, namely Malaysia from 1970 to 2004. In this study, the ARDL and the bounds test proposed by Pesaran et al. (2001) were used to capture the dynamics between economic growth and government expenditures. We also tested empirically for any structural breaks in the data and found that the structural break in 1998 as a result of the Asian Financial Crisis was relevant in the cointegration relationship of government expenditures on administration (lradmin), health (lrhealth) and real GNP (lrgnp). Without structural break, the results indicate that lrgnp is the long run forcing variable of the expenditures on education, defense, development and agriculture and causality is uni-directional running from lrgnp to the expenditure categories. With the structural break included in the equation, lrgnp is still the long run forcing variable of lradmin and lrhealth On the other hand both these expenditure variables are also the long run forcing variables of lrgnp, indicating bi-directional causality. These results support both the Wagners law and Keynesian view of the role of government expenditures. Therefore, from the policy perspective, it is necessary to estimate long run elasticities for expenditures on administration and health with respect to real GNP with the relevant models (as discussed in Sect. 4).
Acknowledgements The authors thank Monash University Malaysia for nancial support (MUM Research Grant No. B-2.11-05) to conduct this study. The authors would also like to thank the anonymous referees for very valuable comments, which improved the paper signicantly. The standard caveat applies.

References
Abu-Bader S, Abu-Qarn A (2003) Government expenditures, military spending and economic growth: causality evidence from Egypt, Israel and Syria. J Policy Model 25:567583 Ahsan S, Kwan A, Sahni B (1989) Causality between government consumption expenditures and national income: OECD countries. Public Financ 44(2):204224 Akaike H (1973) Information theory and an extension of the maximum likelihood principle. In: Petrov B, Csake F (eds) International symposium on information theory. Akademiai Kiado, Budapest Ansari M, Gordon D, Akuamoah C (1997) Keynes versus Wagner: public expenditure and national income for three African countries. Appl Econ 29:543550 Bank Negara Malaysia (19702004) BNM annual report. http://www.bnm.gov.my Bardsen G (1989) Estimation of long-run coefcients in error correction models. Oxf Bull Econ Stat 51:345350 Barth JR, Keleher RE, Russek FS (1990) The scale of government and economic activity. South Econ J 13:142183 Bharat K, Panik M, Wahab M (2000) Government expenditure and economic growth: evidence from G7 countries. Appl Econ 32(8):10591068

123

712

M. Samudram et al.

Chang T, Liu W, Caudill S (2004) A re-examination of Wagners law for ten countries based on cointegration and error-correction modeling technique. Appl Financ Econ 14:577589 Department of Statistics Malaysia (19702004) Yearbook of statistics Malaysia. http://www.statistics.gov. my Dritsakis N, Adamopoulos A (2004) A causal relationship between government spending and economic development: an empirical examination of the Greek economy. Appl Econ 36:457464 Engle R, Granger C (1987) Cointegration and error correction: representation, estimation and testing. Econometrica 55:251276 Gandhi V (1971) Wagners law of public expenditure: do recent cross section studies conrm it? Public Financ 26:4455 Ghali K (1999) Government size and economic growth: evidence from a multivariate cointegration analysis. Appl Econ 31:975987 Granger C (1969) Investigating causal relations by econometric models and cross spectral methods. Econometrica 37:424438 Gregory AW, Hansen BE (1996a) Residual-based tests for cointegration in models with regime shifts. J Econom 70:99126 Gregory AW, Hansen BE (1996b) Tests for cointegration in models with regime and trend shifts. Oxf Bull Econ Stat 58:555559 Gupta S (1967) Public expenditure and economic growth: a time-series analysis. Public Financ 22:423426 Holmes JM, Hutton PA (1990) On the causal relationship between government expenditures and national income. Rev Econ Stat 72(1):8795 Hondroyiannis G, Papapetrou E (1995) An examination of Wagners law for Greece: a cointegration analysis. Public Financ 50(1):6779 Johansen S (1988) Statistical analysis of cointegration vectors. J Econ Dyn Control 12:176181 Johansen S, Juselius K (1990) Maximum likelihood estimation and inference on cointegration with applications for the demand for money. Oxf Bull Econ Stat 52:169210 Khan A (1990) Wagners law and the developing economy: a time series evidence from Pakistan. Indian Econ J 38(1):115123 Landau D (1983) Government expenditure and economic growth: a cross-country study. South Econ J 49:783792 Mah J (2000) An empirical examination of the disaggregated import demand of Koreathe case of information technology products. J Asian Econ 11:237244 Mann A (1980) Wagners law: an econometric test for Mexico 19251976. Natl Tax J 33(2):189201 Ministry of Finance (19702004) Economic report. http://www.treasury.gov.my Murthy N (1993) Further evidence of Wagners law for Mexico: an application of cointegration analysis. Public Financ 48(1):9296 Musgrave RA (1969) Fiscal systems. Yale University Press, New Haven Musgrave R, Musgrave P (1988) Public nance in theory and practice. 5th edn. MacGraw-Hill, New York Pesaran H, Shin Y, Smith R (2001) Bounds testing approaches to the analysis of level relationships. J Appl Econom 16:289326 Phillips P, Perron P (1988) Testing for a unit root in time series regression. Biometrika 75:335346 Quah D (1991) Empirical cross-section dynamics in economic growth. Eur Econ Rev 37:426434 Ram R (1987) Wagners hypothesis in time series and cross section perspectives: evidence from real data for 115 countries. Rev Econ Stat 69(2):359393 Rubinson R (1977) Dependency, government revenue and economic growth, 195570. Stud Comp Econ Dev 12:328

123

You might also like