TAX-AND-SPEND, SPEND-AND-TAX OR FISCAL SYNCHRONIZATION: SOME RESULTS FOR SABAH MUNICIPALS

A.M. DAYANG-AFFIZZAH
Department of Economics
Faculty of Economics and Business
Universiti Malaysia Sarawak

 MUZAFAR SHAH HABIBULLAH
Department of Economics
Faculty of Economics and Management
Universiti Putra Malaysia

INTRODUCTION 

One of the most researched topics in macroeconomics is the empirical testing of the relationship between government expenditures and tax revenues. Establishing the direction of interdependence between the two macroeconomic variables, namely, government expenditures and tax revenues would assist policy makers in identifying the source of any fiscal imbalances that might exist. Consequently this would facilitate efforts to develop a suitable fiscal reform strategy.

The discussion of the causal link between revenues and expenditures has resulted into several competing hypotheses. First, the fiscal synchronization hypothesis implies that taxation and spending decisions are simultaneously made by the fiscal authorities. In the Granger sense, this is known as a bi-directional relationship between tax revenue and government spending. Second, a unidirectional causality that runs from revenue to expenditure supports the so-called tax-and-spend hypothesis. The hypothesis indicates that, since government revenue causes changes in government expenditure, the control of tax revenue should represent a good policy to reduce the size of the government expenditure. On the other hand, the spend-and-tax hypothesis implies that government expenditure leads to changes in tax revenue. In other words, the chain of causality is running from government spending to tax revenue. Lastly, Baghestani and McNown (1994) relates to the institutional separation of the expenditure and taxation decisions of government. This perspective suggests that revenues and expenditures are independent of one another.

In a recent study on Malaysia, Aziz et al. (2000) have pointed out that Malaysia’s government expenditures have almost consistently exceeded government revenues throughout most of the past decades since 1959 except for the 1959-61 and 1993-97 periods. The government’s commitment in pursuing rapid economic development programmes as embodied in the various five year Malaysian development plans largely accounts for the fiscal deficits incurred. Nevertheless, the expanded role of the public sector has resulted in rapid growth of government expenditures. When testing for the above hypotheses, Aziz et al. (2000) found out that Malaysia’s revenues and expenditures data support the fiscal synchronization hypothesis. It implies bi-directional causal relationship in the Granger sense between government revenue and spending.

The purpose of the present study is to investigate the causal relationship between revenues and spending at the municipals level in the state of Sabah for the period of 1965 to 2003. Using the concept of cointegration and vector error correction models, inference can be made concerning the respective hypothesis set forth. Section 2 will provide a brief overview of the hypotheses along with a review of related literature on tax-spend debate. Section 3 discusses the methodology and data used in the analysis. Section 4 provides the empirical results while section 5 contains our conclusion. 

REVIEW OF RELATED LITERATURE 

The relationship between government revenue and expenditure can be categorized into four main competing hypotheses. First, the fiscal synchronization hypothesis implies that taxation and spending decisions are simultaneously made by the fiscal authorities. In the Granger sense, this is known as a bi-directional relationship between tax revenue and government spending. According to Musgrave (1966), voters compare the marginal benefits and marginal costs of government services when formulating a decision in terms of the appropriate levels of government revenues and expenditures. Thus, revenue and expenditure decisions are jointly determined under fiscal synchronization hypothesis. Second, a unidirectional causality that runs from revenue to expenditure supports the so-called tax-and-spend hypothesis. The hypothesis indicates that, since government revenue causes changes in government expenditure, the control of tax revenue should represent a good policy to reduce the size of the government expenditure. The tax-and-spend hypothesis advocates by Friedman (1978) argues that changes in government revenues lead to changes in government expenditures. Friedman suggests that tax increases will only lead to expenditure increases resulting in the ability to reduce budget deficits. On the other hand, the spend-and-tax hypothesis implies that government expenditure leads to changes in tax revenue. In other words, the chain of causality is running from government spending to tax revenue. Peacock and Wiseman (1979) argue that temporary increases in government expenditures as a result of crises can lead to permanent increases in government revenues. Lastly, when there is no causality between revenue and expenditure, then the act of collecting tax revenue and spending are independent from each other as forwarded by Baghestani and McNown (1994).

Empirical evidences with respect to the above competing hypotheses are at best mixed. Musgrave (1966), Meltzer and Richard (1981), Miller and Russek (1990), Bohn (1991) and Bhat et al. (1993) supported the fiscal synchronization hypothesis. Studies by Friedman (1972, 1978), Buchanan and Wagner (1977), 1978), Darrat (1998), Blackley (1986), Marlow and Manage (1987) and Joulfaian and Mookerjee (1990); all indicated that government tax revenues lead expenditures. On the other hand, the spend-and-tax hypothesis is supported by studies done by peacock and Wiseman (1961, 1979), Jones and Joulfaian (1991), Anderson et al. (1986), von Furstenberg et al. (1986) and Provopoulos and Zambaras (1991). Conversely, apart from those studies, Baghestani and McNown (1994) found that government expenditure and revenue are not interdependent.

Other studies on the government revenue-expenditure nexus found a diversity of results, depending on the time period used, lag length and also between different levels of government. For instance, Manage and Marlow (1986) found out that using different lag lengths gave different sets of results. Varying the lag length between two and five, the result indicates that in all cases the lower and most upper lag lengths suggest unidirectional relationship; a causal relation that runs from expenditures to revenues. On the other hand, the intermediate lag length provides support for bi-directional causal relationship between the two variables. Ram (1988) used both annual and quarterly data to examine the expenditures-revenues nexus for both Federal government and State and Local government. The study arrived at conflicting results too. For example, using annual data, the results support the fiscal synchronization hypothesis at the federal government level. However, when quarterly data was used, the results suggest that causality runs from revenues to expenditures, thus, supporting the tax-and-spend hypothesis. But, at the state and local level, both annual and quarterly data indicate results that support the spend-and-tax hypothesis. In another study, Owoye (1995) studied the causal relationship between taxes and expenditures in the G7 countries. Despite the similarities in terms of the economic settings of the sample countries, Owoye found that the results of the causality relationships are not similar. The empirical results obtained from the error-correction models support the fiscal synchronization hypothesis for the U.S., Germany, U.K., France and Canada. This implies that the fiscal authorities in these countries make tax and spending decisions jointly. On the other hand, causality runs from revenues to expenditures in Japan and Italy, thus, supporting the tax-and-spend hypothesis.

A study on selected EU economies by Kollias and Makrydakis (2000) found out that the fiscal decision-making process in Greece, Ireland, Spain and Portugal is diverse. In the case of Greece and Ireland, the evidence supports overwhelmingly the principal of synchronization in the making of spending and tax decisions by the fiscal authority. In the case of Spain, causality is found to run from revenues to expenditures implying that the Spanish fiscal authorities decide first on the amount of tax collections and then decide on how much to spend. In the case of Portugal, decisions on government spending and tax levying turn out to be independent actions. On the other hand, study by Chang et al. (2002) on three newly industrialized countries of Asia (South Korea, Taiwan and Thailand) and seven industrialized countries (Australia, Canada, Japan, New Zealand, South Africa, UK and the USA) found out that data for Japan, South Korea, Taiwan, UK and the USA support the tax-and-spend hypothesis, while the spend-and-tax hypothesis holds only for Australia and South Africa. In the case of Canada, it supports the fiscal synchronization hypothesis. Further, Chang et al. (2002) found out that revenues and expenditures in New Zealand and Thailand are not interrelated.

The studies on tax-spend nexus for the developing economies are quite numerous and among others include Chang and Ho (2002) for China; Fuess et al. (2003) and Chang and Ho (2002) for Taiwan; AbuAl-Foul and Baghestani (2004) on Egypt and Jordan; Ewing and Payne (1998) for Latin America; and Carneiro et al. (2004) for Guinea-Bissau.

Using annual time series data for China over the period 1977 to 1999, and employing the multivariate error correction models in their study, Chang and Ho (2002) found out that bi-directional Granger causality exist between government revenues and government expenditures, thus, supporting the fiscal synchronization hypothesis for China. As for Taiwan, both studies by Chang and Ho (2002) and Fuess et al. (2003) support the tax-and-spend hypothesis that one-way causality from government receipts to expenditures. For the Latin American countries, Ewing and Payne (1998) found out that Chile and Paraguay support the fiscal synchronization hypothesis. For Colombia, Ecuador and Guatemala, there is evidence of causality from revenues to expenditures thus supporting the tax-and-spend hypothesis. The study by AbuAl-Foul and Baghestani (2004) on Egypt and Jordan found out that data for Egypt indicate a unidirectional causation from revenue to spending, with higher revenue leading to higher spending. Results for Jordan indicate bi-directional causation between revenue and spending, thus support the fiscal synchronization hypothesis. For Guinea-Bissau, Carneiro et al. (2004) found out that there exists a stable long-run relationship between government expenditures and revenues. The spend-and-tax hypothesis means that the government seems to spend first and then raise tax revenues and/or request/receive grants to finance its expenditures, rather than adopting the approach of raising funds to finance spending later.

METHODOLOGY 

Testing for Long-Run Relationship between Economic Variables

To determine the long-run relationship between revenue and expenditure, we employ the Johansen (1988) and Johansen and Juselius (1990) multivariate maximum likelihood estimation procedure. Detailed exposition on the Johansen-Juselius technique has been provided in Dickey et al. (1991), Cuthbertson et al. (1992) and Charemza and Deadman (1992). However, a brief discussion on the Johansen-Juselius technique is provided below. We begin with by defining a k-lag vector autoregressive (VAR) representation

            X_t = a + P₁X_t-1 + P₂X_t-2 + ... + P_kX_t-k + n_t     (t=1, 2,...,T)                                                  (1)

where X_t is a px1 vector of non-stationary I(1) variables, a is a px1 vector of constant terms, P₁, P₂...P_k are pxq coefficient matrices and n_t is a px1 vector of white Gaussian noises with mean zero and finite variance. Equation (1) can be reparameterised as

            DX_t = a + G₁DX_t-1 + G₂DX_t-2 + ... + G_k-1DX_t-k+1 + P_kX_t-k + n_t                                                            (2)

where G_i = -I + P₁ + P₂ + ... + P_i         (i=1, 2,...k-1) and P is defined as

            P = -I + P₁ + P₂ +...+ P_k.                                                                                       (3)

Johansen (1988) shows that the coefficient matrix P_k contains the essential information about the cointegrating or equilibrium relationships between the variables in the data set. Specifically, the rank of the matrix P_k indicates the number of cointegrating relationships existing between the variables in X_t. In this study, for a two case variables, X_t = (revenue and expenditure) and so p=2. Therefore, then the hypothesis of cointegration between revenue and expenditure is equivalent to the hypothesis that the rank of P_k = 1. In other words, the rank r must be at most equal to p-1, so that r£ p-1, and there are p-r common stochastic trends. If the r=0, then there are no cointegrating vectors and there are p stochastic trends. 

The Johansen-Juselius procedure begins with the following least square estimating regressions

            DX_t = a₁ + G_iDX_t-i + w_1t                                                                                              (4)

            X_t-p = a₂ + G_iDX_t-i + w_2t                                                                                              (5)

Define the product moment matrices of the residuals as S_ij = T ^-1v_itv_jt (for i,j=1,2), Johansen (1988) shows that the likelihood ratio test statistic for the hypothesis of at most r equilibrium relationships is given by

            -2lnQ_r = -Tln(1-l_i)                                                                                         (6)

where l₁ > l₂ >...l_p are the eigenvalues that solve the following equation

            ôlS₂₂-S₂₁S₁₁’S₁₂ô=0.                                                                                                  (7)

The eigenvalue are also called the squared canonical correlations of w_2t with respect to w_1t. The limiting distribution of the -2lnQ_r statistic is given in terms of a p-r dimensional Brownian motion process, and the quantiles of the distribution are tabulated in Johansen and Juselius (1990) for p-r=1,...,5 and in Osterwald-Lenum (1992) for p-r=1,...10.

Equation (6) is usually referred to as the trace test statistic which is rewritten as follows

            L_trace = -Tln(1-l_i)                                                                                                        (8)

 where l_r+1,...l_p are the p-r smallest squared canonical correlation or eigenvalue. The null hypothesis is at most r cointegrating vectors. The other test for cointegration is the L-maximal eigenvalue test based on the following statistic

            L_max = -T.ln(1-l_r+1)                                                                                                      (9)

 where l_r+1 is the (r+t)^th largest squared canonical correlation or eigenvalue. The null hypothesis is r cointegrating vectors, against the alternative of r+1 cointegrating vectors. Comparing the two tests, Johansen and Juselius (1990) indicate that the trace test may lack power relative to the maximal eigenvalue test which will produce clearer results.

 Sources of Data

Data on total revenues and total expenditures for eighteen municipals in the state of Sabah for the period 1965 to 2003 are collected from the various issues of the Statistical Yearbook Sabah published by the Department of Statistics Malaysia, Sabah. The 18 municipalities include Beaufort, Keningau, Kota Belud, Kota Kinabalu, Kuala Penyu, Kudat, Labuk and Sugut, Lahad Datu, Papar, Ranau, Sandakan, Sipitang, Tambunan, Tawau, Tenom and Tuaran. Other municipals such as Labuan, Kota Kinabalu Rural, Panampang, Pensiangan, Sandakan Rural, and Tawau Rural are not included in this study due to unavailability of consistent data. All time series variables were transformed into natural logarithms. 

THE EMPIRICAL RESULTS 

Before testing for cointegration by using the Johansen-Juselius procedure, we test for the order of integration of all variables for all municipalities. Table 1 show the results of the unit root test for the test of the order of integration of revenue and expenditure for the eighteen municipals in Sabah under investigation. Clearly, in all cases, the augmented Dickey-Fuller test (Dickey and Fuller, 1981) statistics indicate that the two series in Sabah are difference stationary, in other words, they are I(1) in levels, except for Kinabatangan and Semporna.

Having noted that both series are of the same order of integration, we run the cointegration test following the procedure provide by Johansen and Juselius (1990). These results are tabulated in Table 2. The null hypothesis of no cointegration cannot be rejected in all cases of the tax-spend nexus using both the trace and L-max statistics at the one percent significance level. However, in the cases of Kudat, Tambunan and Tawau, the null hypothesis of no more than zero and no more than one cointegrating relation is soundly rejected at the one percent significance level using the trace statistics, while in the cases of Tambunan and Tawau using the L-max statistics. Since according to Cheung and Lai (1993), the trace test shows more robustness to both skewness and excess kurtosis in the residuals than does the L-max test, therefore, we will only emphasis on the used of trace statistics to make inferences for non-cointegration between revenue and expenditure in this study.

Furthermore, knowing that the Johansen-Juselius cointegration procedures are distorted in small sample, we proceed our analysis employing the vector error correction model to infer cointegration among the series. According to the ‘Granger Representation Theorem’ not only does cointegration imply the existence of an error correction model but also the converse applies, that is, the existence of an error correction model implies cointegration of the variables. Recent developments in cointegration and error correction model as pointed by Pesavento (2004) suggest that the Johansen’s test for cointegration has low power in both large and small sample compared to the error correction model. In fact, Kremers et al. (1992) have argued that the standard t-ratio for the coefficient on the error-correction term in the dynamic equation is a more powerful test for cointegration. Banerjee et al. (1986) and Kremers et al. (1992) show that standard asymptotic theory can be used when conducting the test in the context of an error correction model; specifically, the t-statistics on the error correction term coefficients have the usual distribution.

Since our task is to determine the long-run relationship and the causal direction between the two variables in question, we estimate the following vector error correction model. We specify the following two-variable vector error correction models (VECM) as

                                                  (10)

                                                (11)

where ecm_t-1 is the lagged residual from the cointegration between y_t (say, expenditure) and x_t (revenue) in level. Granger (1988) points out that based on equation (10), the null hypothesis that x_t does not Granger cause y_t is rejected not only if the coefficients on the x_t-j, are jointly significantly different from zero, but also if the coefficient on ecm_t-1 is significant. The VECM also provides for the finding that x_t-j Granger cause y_t, if ecm_t-1 is significant even though the coefficients on x_t-j are not jointly significantly different from zero. Furthermore, the importance of a‘s and b‘s and represent the short-run causal impact, while g’s gives the long-run impact. In determining whether y_t Granger cause x_t, the same principle applies with respect to equation (11). Above all, the significance of the error correction term indicates cointegration, and the negative value for g’s suggest that the model is stable and any deviation from equilibrium will be corrected in the long-run.

The results of estimating equations (10) and (11) are presented in Table 3. In our study, we attempt to determine whether revenues and expenditures among the municipalities in the state of Sabah are related and when these variables are related or exhibit long-run relationship, we would expect the estimated parameters of the error correction terms in equations (10) and (11) are significant. From the VECM results in Table 3, we presented the t-statistics of the error correction term, ecm_t-1, where we can infer the long-run causality between the variables. The significance (at least one) of the error correction term implies cointegration or exhibit long-run relationship between revenue and expenditure. Generally, results in Table 3 indicate that expenditure and revenue are cointegrated in 5 out of 16 municipalities. These municipalities are Keningau, Kudat, Labuk and Sugut, Lahad Datu and Ranau. In all five cases, at least one error correction term is statistically significant at the 5 percent level in the two variable VAR systems. In other words, both these two variables are bound together by the long-run relationships. Further, the significance of the error correction term (ecm) in the revenue equation support the spend-and-tax hypothesis for Keningau, Kudat, Labuk and Sugut, and Ranau. On the other hand, Lahad Datu Municipal supports the tax-and-spend hypothesis.

Further Analysis with GDP as a Third Variable

Studies on tax-spend nexus have indicated that gross domestic product as a third variable in the analysis can reduced misspecification problem. Dalagamas (2000) has pointed out that omission of certain key macroeconomic variables will result in spurious regression problems. Payne (1997) has pointed out that the inclusion of other exogenous variables will give a better reflection of the effect of the overall conditions in the economy on the variables being analyzed. In his study on tax-spend nexus for Canada, by not taking into account gross domestic product (GDP) will lead to misspecification problem. Fuess et al. (2003) put forward that the interrelationship between government revenues and expenditures is likely to differ according to the inclusion of a third variable as a control variable. The rational for including GDP as a third variable is that revenue and expenditure growth should be related to the overall conditions in the economy (see also Baghestani and McNown, 1994; Koren and Stiassny, 1998; Chang et al., 2002). In this study, we used Sabah’s gross domestic product to proxy for the state of the economy.

Results of tri-variate cointegration test using the Johansen-Juselius approach are presented in Table 4. The results clearly indicate that the null hypothesis of no cointegration between revenue and expenditure with the present of a third variable can be rejected in 8 out of 16 cases. Municipalities that exhibit long-run relationship between revenue and expenditure, inferring from the trace test include Beaufort, Keningau, Kota Kinabalu, Kuala Penyu, Labut and Sugut (with two cointegrating vectors), Ranau, Tambunan and Tuaran. This is an improvement from the bi-variate case in Table 2.

On the other hand, Table 5 shows the results of estimating the vector error correction model for the tri-variate case. Interestingly, the results suggest that 13 out of 16 cases exhibit long-run relationship between revenue and expenditure in the state of Sabah. The VECM model indicates that at least one of the error correction terms is statistically significant at the 5 percent level. The only municipals that do not show any long-run relationship between revenue and expenditure are Kota Belud, Kudat and Sipitang; which is consistent with earlier results as presented in Table 4.

As suggested by the results in Table 5, there are four competing tax-spend hypotheses derived by the models. The tax-and-spend hypothesis is supported by Papar, Sandakan, Tambunan, Tawau, Tenom and Tuaran. The spend-and-tax hypothesis is supported by Keningau, Lubuk and Sugut, Lahad Datu and Ranau. Fiscal synchronization is supported only by Kota Kinabalu. Lastly, the hypothesis that revenue and expenditure is independent of each other is supported by Beaufort and Kuala Penyu. These results are summarized in Table 6.

CONCLUSION 

The objective of the present study is to extend the literature on the tax-spend debate to a sample of municipalities in Sabah, one of the fourteen states in Malaysia. To test several hypotheses concerning the temporal relationship between revenues and expenditures, we utilize both the multivariate cointegration approach proposed by Johansen and Juselius (1990), and the error correction model approach proposed by Kremers et al. (1992) and Pesavento (2004).  

Our more robust model by including GDP as a third variable suggests that the results are mixed. On one hand, municipals in Papar, Sandakan, Tambunan, Tawau, Tenom and Tuaran support the tax-and-spend hypothesis. This implies that these municipals decide first on the amount of tax collections and then decide on how much to spend. Under this scenario, these municipals should focus attention on adjusting revenues in order to control spending and the size of the budget deficits.

On the other hand, municipals in Keningau, Lubuk and Sugut, Lahad Datu and Ranau support the spend-and-tax hypothesis. This would imply that the fiscal authorities of these municipals decide to spend first and then increase tax collection to cover expenses. Under this scenario, the fiscal authorities of these municipalities should try to control spending in order to restore fiscal discipline and control the size of its fiscal deficits. Only Kota Kinabalu supports the fiscal synchronization hypothesis and implies that under this scenario, the fiscal authority of Kota Kinabalu should try to raise revenues and cut spending simultaneously in order to control budget deficits. On the other hand, municipals in Beaufort and Kuala Penyu suggest that revenues and expenditures are not interrelated in the long-run. This implies that the independent determination of revenue and spending suggest the absence of coordination between expenditure and revenue decisions in the respective municipalities.