Trends in tax revenues of transition economies: an empirical approach

Abstract

Economic theory postulates that taxes and alternative tax policies have different impacts on economic growth. Some scholars have argued that economic growth can result in an increase in total tax revenues and its ratio to GDP. Our study investigates the impact of economic growth on total tax revenues and different tax share ratios in transitional economies. We find strong support for the hypothesis that economic growth leads to an increase in total tax revenue-to-GDP ratio. We also find that economic growth has led to, and facilitated, changes in the relative shares of different tax sources in transition countries over time differently thus allowing these countries to streamline their taxes and properly plan for and alter their tax system accordingly.

Appendices

Appendix 1: Descriptive statistics and data definition

See Tables 2, 3, 4, 5 .

Table 2 Summary statistics of tax structure variables

Table 3 Summary statistics of control variables

Table 4 Data definition and sources

Table 5 Tax Structure Variables and Columns Abbreviations

Appendix 2: Results of diagnostic tests to ensure robustness of the empirical model

Please refer to footnotes 1 and 2 in the text for additional background on these results (Tables 6, 7).

Table 6 Estimation results: difference GMM

Table 7 Estimation results: system GMM

Appendix 3: The stationarity of tax variables and adf-unit root test results

To check for stationarity (or unit roots) of tax variables and GDP per capita growth in our panel dataset, the Fisher type panel-data unit-root tests are performed. Fisher-type unit-root tests for all tax and GDP variables are based on (a) Augmented Dickey-Fuller (ADF) test and (b) Phillips-Perron (PP) test. The null hypothesis of both the Fisher-ADF and the Fisher-PP unit root tests is that all panels contain a unit root or stationarity. The alternative hypothesis of the tests is that at least one panel is stationary. Panel-specific means (fixed effects) in the model are included, but time trend is excluded.

Lags (5) of the series are used in the Fisher-ADF and Fisher-PP regressions. The lags (5) are used to remove higher-order autoregressive components of the series. Thus, the assumption is that the data are generated by AR (5) process for Fisher tests:

$$\Delta \mathit{ln}{\tau }_{t}=\Delta ln T{R}_{t}-\Delta ln T{R}_{t-1}$$

(5)

The results of this test for all the tax variables and GDP per capita growth are given in Table 8. The first noticeable thing in this table is that almost all the estimated coefficients are individually highly significant, for the p-values are low. This gives us strong evidence against the null hypothesis in favor of the alternative (as p-values are mostly within the range of 0.001 < p < 0.01). These results confirm that the GDP per capita growth and tax variables are all stationary in their change or difference form.

Table 8 Tests for difference stationarity (Fisher-type unit-root test)

Furthermore, the test results indicate that all our dependent and explanatory variables in panel are stationary. Almost all variables in first differences pass the stationarity test at 1% significance level.

Appendix 4: Omitted variables test (model with political budget cycle and year of accession to the eu dummy variables)

To ensure that missing variable phenomenon has not biased the empirical results of our main regression models, we have deployed four dummies as proxies. For political budget cycle we use: (1) one year before election year⁴; (2) one year after election year; proxy for specific tax policies adopted by countries at different point in time. For membership in the European Union (EU) we use two dummies, namely—(3) year of EU accession; and (4) one year after EU accession, controlling for country fixed effects and time dummies.

When running regression with these dummies, we used one fewer dummy than the number of categories for any categorical variable to avoid collinearity using the following model:

$${\tau }_{it}={\beta }_{0}+{\beta }_{1}{\text{D}}_{1i}+{\beta }_{2}{D}_{2i}+ {\beta }_{3}{D}_{3i}+{{\beta }_{4}{D}_{4i}+ \beta {X}_{it}+\epsilon }_{it}$$

(6)

Then, we leave the intercept in (5) to make the model more robust and to obtain unbiased standard errors:

$${\tau }_{it}={\beta }_{1}{\text{D}}_{1i}+{\beta }_{2}{D}_{2i}+ {\beta }_{3}{D}_{3i}+{{\beta }_{4}{D}_{4i}+ \beta {X}_{it}+\epsilon }_{it}$$

(7)

where ${D}_{1i}=\text{PB }(\text{one year before election year dummy})$, ${D}_{2i}=PB+1 (\text{one year after election year dummy})$, ${D}_{3i}=\text{EU }(\text{EU accession year dummy})$, ${D}_{4i}=EU+1 (\text{one year after EU accession year dummy})$.

Equation (7) with added dummy variables is integrated with the tax structure grand regression (the main regression in columns 1 of Table 9 when re-running the model below:

$$\Delta ln {\tau }_{it}=\alpha +{\beta }_{1}\Delta \mathit{ln} {Y}_{it}+{\beta }_{2}\Delta ln {Z}_{it}\eta +{f}_{i}{D}_{i}+{\varphi }_{t}+{\varepsilon }_{it}$$

(8)

Table 9 Impact of control variables on overall tax collection. Period of 1991–2015

The regression results for Eq. (8) are presented in Table 9 below.

First and foremost, when comparing the results in Table 9 with those in column 1 of Table 5, the signs and significance of estimated coefficients for all the original explanatory variables are the same. These results confirm that our original model is robust in explaining the trends in overall tax revenues of the transition economies and new variables do not alter the conclusions arrived at based on that model. Nonetheless, introduction of both political budget cycle and year of membership in the EU dummies produce valuable information.

Specifically, the results in columns 1–4 of Table 9 suggests that the political budget cycle instruments represented in the form of dummy variables ${D}_{1i} \text{one year before election year}$ has a negative impact on tax revenue collection overall when a one-year pre-election year is plagued by political manipulation due to political budget cycle theory (Ferraressi 2021; Golden and Min, 2013). Consequently, fiscal tools such as overall tax revenues and various types of taxes are negatively affected. In contrast to that, ${D}_{2i}$ one year after election variable, has a positive and highly significant impact on the tax revenue collection, suggesting that after election to their respective office, political leaders tend to deviate from their tax reduction promises during the election and raise taxes.

The EU membership dummy variable ${D}_{3i}$ year of EU accession has a negative effect on collection of tax revenues in our sample, suggesting slightly lower revenues compared to the pre-accession period for new EU members. Such a decrease in tax revenues may suggest elimination of tax sources that were previously collected through the local channels that could not continue after EU accession. However, ${D}_{4i}$ one year after EU accession, shows a positive impact on overall tax collection for new EU member countries. This may be due to introducing new tax policies and tax sources to comply with the EU common fiscal regime.