Is the Thai government revenue-spending nexus asymmetric?

Abstract

Recent evidence suggests that there might be an asymmetric process of adjustment towards the long-run equilibrium in the government revenue-spending nexus. In this paper, the long-run relationship between government revenue and spending is examined in the case of Thailand using annual data over the period 1991–2019. Specifically, this is an attempt to determine whether this long-run relationship is linear or nonlinear. In doing so, both linear and nonlinear cointegration tests are employed. The empirical results suggest that the positive long-run relationship between revenue and spending is linear and stable when revenue is the dependent variable. By estimating the TAR and MTAR models, evidence of a nonlinear revenue-spending relationship is not found. Therefore, there does not seem to be asymmetric adjustment toward the long-run equilibrium. The results of causality tests based on the estimated ECMs of linear cointegrating equations show no causality between revenue and spending in the short run, supporting the fiscal institutional separation hypothesis. In the long run, there is unidirectional causality running from government spending to revenue, which supports the fiscal spend-and-tax hypothesis.

Keywords:

• Government revenue
• government expenditures
• unit root
• cointegration
• structural break
• symmetric or asymmetric adjustment
• causality

JEL Codes:

• C32
• E62

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The estimation techniques of the present paper are similar to the techniques used by Paleologou (2013) in that the residual-based cointegration test is estimated, and followed by nonlinear cointegration tests. However, both direct and reverse relations between the two variables are estimated in the residual-based tests.

2 In testing for cointegration, Engle and Granger (1987) suggest that a reverse relationship between variables should also be performed because a long-run relationship can be found in one way or the other, or both. Furthermore, most previous studies employ the standard Granger causality tests, which are the short-run analysis only. Granger (1988) suggests that both short- and long-run causality tests can be performed on the estimated error correction models of the cointegrating equations. Narayan (2005) and Nyamango et al. (2007) examine two long-run equations by letting government expenditure as a dependent variable in the first equation and government revenue as a dependent variable in the second equation. They find that there are bidirectional long-run causal linkages between government revenue and expenditure in South Africa. In addition, Parida (2012) estimates the two long-run equations in the case of India. Many previous studies ignore this issue.

3 The t-statistic is the augmented Dickey-Fuller (ADF*) statistic, which is different from the ADF statistic of Engle and Granger (1987). In addition, Phillips’ (1987) procedure is also calculated as the Z*_t statistic.

4 The speed of adjustment is ${\lambda }_{i1}=I_t{\widehat{\rho }}_1$ in the first regime and ${\lambda }_{2i}=(1-I_t){\widehat{\rho }}_2$ in the second regime while I_t is expressed in Eq. (9) for the TAR model and in Eq. (10) for the MTAR model.

5 Even though the estimated Eq. (4) is not stable and any inference cannot be made in a linear cointegration test, the long-run relationship might be nonlinear with asymmetric adjustment toward the long-run equilibrium.

6 According to Hansen and Seo (2002), the F-test (Ф) for the TAR and MTAR models has a non-standard distribution due to the presence of nuisance parameters that are only identified by the alternative hypothesis. Therefore, the test critical values must be computed.

7 The results also indicate that the convergence condition is met, i. e., ρ₁ < 0. ρ₂ < 0 and (1 + ρ₁)(1 + ρ₂). According to Pettrucelli and Woolford (1984), this convergence condition is the condition for the stationarity of the residual series. Even though the t-Max statistic has lower power of testing than the Ф-statistic, both statistics accept the null hypothesis of no nonlinear cointegration.

8 It should be noted that the Wald F-statistic is not the same as the F-statistic for the estimated ECMs.

Additional information

Notes on contributors

Komain Jiranyakul

Komain Jiranyakul is currently an associate professor of economics at School of Development Economics, National Institute of Development Administration, Thailand. His research fields include trade and development, financial economics and applied econometrics. He recently published papers in Journal of Advanced Studies in Finance Journal of Financial Economic Policy, Economics Bulletin, Asian Economic Journal, and Journal of Asian Economics.