Tax revenue and economic growth in developing country: an autoregressive distribution lags approach

In the literature, there are various theories and frameworks that are used and are being discussed by various academicians and researchers to investigate the relationship between taxation and economic growth. Three main theories, namely, optimal taxation theory, exogenous growth theory and endogenous growth theory, are briefly discussed below.

Optimal taxation theory involves designing and undertaking a tax that lowers distortion and inefficiency in the market equilibrium under certain economic conditions. The fundamental theory of optimal taxation comprises choosing the tax that will maximise welfare function of the society considering a given set of constraints. In addition, if the first best outcome is not feasible and you have to seek for the second best then the design and implementation of the optimal tax requires knowing how to increase the number of outcome from a heterogeneous population using socially optimal way (Mankiw, Weinzierl, & Yagan, 2009).

In the literature, exogenous growth theory is also known as neoclassical theory to evaluate the correlation between taxation and growth. This theory is the mirror image of the endogenous growth theory to be discussed shortly, also known as the new growth theory. Solow model is one of the examples of exogenous theory and the pioneer of this theory is Robert Solow (1956). According to this theory, fiscal policy action by the government will not have any effect on the long-term growth of the economy, but it implies that any changes in economic growth will be caused by the key factors of production such as labour, capital and technological progress that are being determined outside of the model (Solow, 1956). However, in the neoclassical framework, taxes imposed by the government can have an impact on growth in the transition to a new steady state if they affect the rate of savings and consequently investment level.

Endogenous growth models (EGMs) consist of a diverse body of theories that model economic growth through the medium of technological discoveries and progress that emerged during the 1980s. In the neoclassical growth model as it has been described, economic growth is determined by the rates of savings and capital accumulation. In this growth model, technological progress is taken as exogenous, meaning taken as given and being determined outside the model. In the EGM, technology progress is seen as the key determinant of long-term economic growth which the neoclassical growth model could not account for. Hence, technological progress becomes endogenous in endogenous growth models. As the matter of fact, it is the effect of diminishing returns in the neoclassical growth model that limits the expansion of output and economic growth. To overcome this limitation to economic growth, EGM inculcates increasing returns to scale (Barro, 1990). The classical Cobb Douglas production function exhibits constant returns to scale to the factor inputs say labor and capital. This leaves no reward or incentive for individuals to engage in economic activities that encourage technological innovations. Therefore, any theory that endogenises technological progress cannot be based on a competitive equilibrium where factors of production are being rewarded according to their marginal products. As far as the role of fiscal policy is concerned, this theory speculated that the state spending and taxation should have both a temporary and permanent effect on the growth of the economy. Tax will cause inefficiency and distortion in the product and labour markets, and the productive expenditure will affect the growth rate in the long-term (Barro, 1990).

In developing economies, the government has to play an active role in achieving sustainable economic growth. Therefore, fiscal policy is a very crucial instrument of government in promoting economic growth (Babatunde et al., 2017). An important part of the fiscal policy is taxation apart from government expenditure. Many economists believe that tax revenue is one of the most significant factors that contribute to a country's growth. For years, it has provided developing countries with a stable and predictable fiscal environment to promote growth and to finance the needed social and physical infrastructures. In Tanzania, for instance, the economy is highly dependent on tax revenue as a source of government expenditure for developmental purposes (Kweka & Morrissey, 2000). Combined with economic growth, tax revenue reduces long-term reliance on aid and ensures good governance by promoting openness and accountability of governments to their citizens (Romer and Romer, 2010).

In Tanzania, taxes on domestic goods and services (mainly VAT and excises) were the largest tax base from 1999 to 2008. From 2001, however, revenues from direct taxes (personal and corporate) in percent of GDP have gradually increased even though they are still below indirect taxes. From 2005, onwards, the same applies for trade taxes (tariff). Therefore, by 2008, indirect taxes (4.6%), direct taxes (4.5%) and trade taxes (4.1%) contributed with almost the same tax share of GDP. Trade tax revenues declined from 4% of GDP in 1996 to 2.6% in 2005, before increasing to 4.1% in 2008 (Fjeldstad & Heggstad, 2011). Non-tax revenues are mainly user fees, charges from various ministries and dividend from government institutions and have been kept relatively stable as a share of GDP throughout the period under study.

This study on the effect of taxation on economic growth employs secondary data obtained from published sources, that is, second-hand type of information obtained from secondary sources from 1996 to 2019. The variables of interest selected in this study based on the availability and reliability of data in which five variables (income tax, taxes on domestic goods and services, FDI, domestic investment (DI), inflation and real GDP) are being included. The data were obtained from Tanzania National Bureau of Statistics (NBS) and Tanzania Revenue Authority (TRA).

The econometric literature makes clear that the estimates of ordinary least squares (OLS) are biased in nature and that they are inconsistent especially when using time series data. The reason is that the OLS estimates are affected by the presence of endogeneity, for that reason, they are unable to promise stable and robust forecasting (Messaoud & Teheni, 2014). This study applies autoregressive distribution lags (ARDL) approach to deal with these shortcomings.

The inclusion of inflation in the empirical model of this study is informed by the advice of Taylor (1994) that to be able to rise GDP growth, cost-oriented anti-inflation programmes have to be supported by high transfers from outside the country. There exist recursive relationship between tax revenue and economic growth accruable to the government. The relevance of foreign direct investment (FDI) apart from DI in the model is supported by the two-gap model of Chenery and Strout (1966) which showed that developing countries are constrained with low level of foreign exchange earnings and domestic savings. The expectation is that, in combining with other variables, foreign direct investment provides an optimal avenue to break the circle of poverty, solve the two gaps simultaneously and increase growth of the economy. The functional relationship is expressed as: (1)GDP=f(YTAX, TGS, FDI, DI, INF){\rm{GDP}} = {\rm{f}}({\rm{YTAX}},\,{\rm{TGS}},\,{\rm{FDI}},\,{\rm{DI}},\,{\rm{INF}}) where GDP is the growth rate of real gross domestic product per capita; YTAX is the income tax of households and firms, TGS is taxes on domestic goods and services, FDI is the foreign direct investment; DI is the domestic investment; and INF is the rate of inflation.

The generalised ARDL (ρ, q) model is specified as follows: (2)Yt=α0j+∑i=1ρδjYt−1+∑i=0qβj'Xt−1+εjt{Y_t} = {\alpha _{0j}} + \sum\limits_{i = 1}^\rho {{\delta _j}{Y_{t - 1}}} + \sum\limits_{i = 0}^q {\beta {'_j}{X_{t - 1}}} + {\varepsilon _{jt}} where Y′_t is a vector and the variables in (X′_t)′ are allowed to purely I(0) or I(1) or cointegrated. β and δ are coefficients, j = 1,....., k; α is a constant. ρ and q are optimal lag orders; ε_jt is a vector of the error term, unobservable zero mean white noise vector process (serially uncorrelated or independent). The dependent variable is a function of its lagged values, the current and lagged values of other exogenous variables in the model.

Since time series data are being used, it is imperative to test if they are stationary to avoid spurious regression results. This is also important to ensure that no variable of interest is integrated of order two or higher. The Dickey-Fuller (ADF) and Phillips-Perron (PP) unit root test results show that all the variables are integrated of either order 0 or 1 (see Tables 1 and 2). The result presented in Table 2 shows that all the series were stationary at first difference since their respective probability values were less than the 5% significance level.

The ARDL bound test for cointegration and the result for the series model are being presented in Table 3. When GDP is the dependent variable, the test results indicated that F_GDP(YTAX, TGS, FDI, DI, INF) = 43.944; for YTAX model, F_YTAX(GDP, TGS, FDI, DI, INF) = 11.949; for TGS model F_TGS (GDP, YTAX, FDI, DI, INF) = 13.629; for FDI model, F_FDI(GDP, YTAX, TGS, DI, INF) = 4.496; for DI model F_DI (GDP, YTAX, TGS, FDI, INF) = 3.016 and for INF model, F_INF(GDP, YTAX, TGS, FDI, DI) = 3.544. The results indicated that there is a long-run relationship among the variables when GDP, YTAX, TGS, FDI, and INF are dependent variables, while for the DI, the null hypothesis of no cointegration was not rejected. For models that were found to have long-run relationship, the F-statistics were found to be higher than the upper bound critical value at 1% significance level. These obtained results suggested that the variables share a long-run relationship; therefore, we can proceed with the estimation of a long-run and short-run ARDL model, while for the model with DI as a dependent variable only short-run relationship is being estimated.

The long-run coefficients for the cointegrated models estimated are presented in Table 4. With GDP as the dependent variable, the results illustrated that income tax (YTAX) has a negative significant relationship with economic growth at 5% level. This implies that a 1% increase in income tax will result into about 5875 times decrease in GDP. Taxes on domestic goods and services (TGS) and DI were found to be positively related with GDP and statistically significant at 1% level. More specifically, 1% increase in DI will result into about 1.09% increase in GDP. With income tax as a dependent variable, GDP and TGS were found to be statistically significant at 5% and 1%, respectively.

Regarding TGS model, GDP and income tax were found to be positively related with TGS and significant at 1% level. More specifically, 1% increase in GDP will result into about 0.5% increase in TGS. On the other hand, for the FDI and inflation model, all the variables were found not be statistically significant at any level of significance, hence accepting null hypothesis for all regression coefficients in these two last two models.

Table 5 provides the ARDL short-run coefficients among the estimated parameters. The results indicated that the lag value of GDP and the current value of YTAX have a negative effect on economic growth in the short-run at 1% of statistical significant. For the case of income tax, lagged value of income tax, GDP and FDI were found to be statically significant. GDP and inflation were found to positively cause changes on TGS and to be statistically significant at 1 % and 10%, respectively. Moreover, just like in the long-run estimation when the variable INF is dependent variable, all other variables were found not to be statistically significant.

The ECM_t−1 for the five cointegrated models found to be negative and statistically significant at different levels of significance. This adjustment coefficient shows how quickly short-term disturbances return to the long-run equilibrium. For GDP model, the value of −0.53 indicated that the short-run deviations for the long run were corrected about 53 % each year. While for FDI, the ECM_t−1 value of −0.51 indicated that the short-run deviations for the long-run were corrected about 51%. Regarding the YTAX model, the value of −0.57 indicated that the short-run deviations for the long run were corrected about 57% each year. With the same implication for TGS, the ECM_t−1 value of −0.87 indicated that the short-run deviations for the long run were corrected about 87%. These models pass the diagnostic tests for heteroscedasticity and serial correlation, Durbin-Watson, Breusch-Godfrey and White tests were applied and results are presented in Table 5.

Following the cointegration analysis and the fact that some of the models were found to be cointegrated give raise to determine pair-wise Granger causality to see the direction of the causality between the variables. In any case, if cointegration exists, it has to be either bidirectional, unidirectional or neutral causality between the variables. The results are obtained by using the pair-wise with Granger causality and Wald test. The Wald chi-square statistics of the variables indicated the short-run causality effects. The obtained results indicate that there is bidirectional Granger causality between GDP and taxes on domestic goods (TGS) at 1% significance level. The same results were obtained between GDP and DI, that is, they Granger cause each other. On the other hand, the chi-square statistics demonstrated that at 1% level, there is unidirectional Granger causality between GDP and inflation. The results also indicated that FDI Granger causes TGS. The findings also suggest that for case of income tax and inflation, there are no significant causality effects from any of the variable chosen in this study as presented on Table 6.