The previous data story illustrated the tax systems in Africa. Two data sources are available, namely data sets provided by the OECD and by ATAF. They show substantial differences that are hard to reconcile. In this analysis, we focus on the OECD data and aim to describe in how far socio-economic circumstances affect the prevalence of taxation in African countries. Methodological, panel data regressions are executed to link socio-economic variables to the tax quotas (tax revenues in relation to GDP). Overall, in the recent decade results show limited effects of a variation in the considered variables. Path dependency seems to be more relevant.
However, it cannot be ruled out that the more or less binding decisions from the past were affected by the socio-economic circumstances at that time. The limited data coverage for former decades hinders the ability to conduct a more in-depth analysis. Overall, it can be stated that a higher level of GDP per capita affects tax quotas positively. Also an older population seems to be linked to higher tax revenues, especially via social contributions and valued-added tax (VAT). The positive relation to GDP per capita is unlikely to be driven by business cycle fluctuations, instead it likely reflects the achieved level of economic development . The relation to GDP per capita is especially strong for income taxation (direct taxation). Indirect taxes do not benefit from higher GDP levels. Here, the quality of governance is related to a stronger reliance on this source of tax revenues. It holds for VAT as well as other indirect taxes.
The Africa Monitor covers 55countries but only 33 countries reported tax data processed by the OECD. It is likely that non-reporting is not random and correlates to the same socio-economic variables as the tax quotas. This has already been discussed in the Data Story Taxation in Africa. For the year 2021, the GDP per capita among reporting countries was higher, as well as the quality of the governance. Reporting countries had a higher share of old and a lower share of young people in the mean, but no differences in the median. Non-reporting countries were more highlyindebted. Bringing these factors together, the only statistically significant regressors in a probit regression of reporting versus non-reporting are the quality of governance and the share of old people. With respect to the marginal effects, the quality of governance has some relevance of reporting to the OECD (Table 1).
| Table 1: Coverage in the OECD data set Probit regression (population averages) | ||||||
|---|---|---|---|---|---|---|
| Coefficient | p-value (robust) | Marginal Effect | ||||
| GDP per Capita | 0.000 | 0.211 | 0.000 | |||
| Governance | 0.008 | 0.051 | 0.003 | |||
| Young pop. | -0.001 | 0.436 | 0.000 | |||
| Old pop. | 0.003 | 0.063 | 0.001 | |||
| Debt | 0.000 | 0.536 | 0.000 | |||
| Constant | 0.267 | 0.247 | ||||
| p value all coefficients | 0.770 | |||||
| Number of obs. | 52 | |||||
Source: OECD, own calculations.
To analyze the socio-economic circumstances of the prevalence of tax revenues relative to GDP, we apply three different regression approaches, with each having specific advantages and disadvantages. The fixed effects approach controls for country-specific characteristics that do not change over time. One disadvantage of this approach is that very long run effects between the regressors and the tax quota may remain undetected, since the variation between the countries is neglected. The random effects specification includes this kind of information. However, the estimation of the coefficients is more likely to be affected by omitted variables. In contrast omitted variables that are stable over time are accounted for in a fixed effects regression. Finally, a dynamic regression, which includes a lagged dependent variable, provides an indication of how the contemporaneous outcome would change, in response to a change in the explanatory variables.
An interesting result is that GDP per capita has a statistically significant impact in the fixed effects specification but not in the dynamic one (Table 2). The long run coefficient in the dynamic regression is substantially smaller, too. This corresponds to the hypothesis that the impact of GDP per capita is not a business cycle phenomenon. It seems that it is rather about the overall level of economic development. The coefficient for the quality of governance is quite substantial, but it is not significant. A high coefficient also shows up for the share of old people. The variation of the estimator is a bit more favorable in the dynamic regression, where it results significant. A more elderly society seems to rely more on taxes. This may depend on the necessity to spend more on social issues. Interestingly, the reliance on Grants seems to diminish the need for an extended taxation system. The coefficient for the quality of governance is quite substantial but not statitically significant.
| Table 2: Panel regression for the overall tax revenues rel. to GDP | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Fixed Effects | Random Effects | Dynamic (FE) | |||||||||
| Coeff. | t-value | Coeff. | t-value | Coeff. | t-value | ||||||
| GDP per capita | 0.0435 | 2.46 | 0.0185 | 1.73 | 0.0065 | 0.42 | |||||
| Current Account | -0.0136 | -0.58 | -0.0169 | -0.72 | -0.0111 | -0.59 | |||||
| Quality of Gov. | 0.2713 | 0.37 | 0.4893 | 0.82 | 0.0475 | 0.12 | |||||
| Young Population | 0.0545 | 0.2 | -0.1151 | -0.64 | -0.0028 | -0.02 | |||||
| Old Population | 0.5131 | 1.14 | 0.5807 | 1.12 | 0.3252 | 1.7 | |||||
| Debt | 0.0256 | 1.7 | 0.0196 | 1.54 | -0.0009 | -0.1 | |||||
| IMF | -0.0023 | -1.06 | -0.0017 | -0.8 | 0.0003 | 0.21 | |||||
| Grants | -0.0854 | -1.31 | -0.0853 | -1.57 | -0.0165 | -0.45 | |||||
| Constant | -16.8362 | -0.74 | 7.0919 | 0.45 | -0.5746 | -0.04 | |||||
| Lagged Dep. | 0.6277 | 16.69 | |||||||||
| Explained Variation | |||||||||||
| within | 0.1638 | 0.1516 | 0.4668 | ||||||||
| between | 0.328 | 0.4813 | 0.954 | ||||||||
| overall | 0.3135 | 0.4554 | 0.9158 | ||||||||
Source: OECD, own calculations.
The difference in the explained between variation between fixed effects and random effects corresponds to the “very long-run” impact of the considered variables. The remaining variation is due to other factors not captured here. It might be time invariant, like colonial history or topographic aspects. Overall, the explaining power of the random effects model is quite remarkable, but the larger share of the variation remains unexplained.
The explanatory power of the model for the income tax quota (income tax revenues relative to GDP) is much smaller than for all tax revenues together. Hardly any of the within variation can be explained by the model. Thus, it is more likely that the explanatory variables are connected to long run trends.
Results for GDP per capita and the share of older people exhibit a smaller but still significant coefficient. The relevance of Grants cannot be confirmed for income taxes. Interestingly, the appearance of the IMF is negatively linked to the tax quota (Table 3). However, this could be due to an endogeneity issue. It is unlikely that this relates to IMF policies (less direct taxes, more indirect) since positive effects cannot be found for indirect taxes (see Table 5, 6). The estimated coefficient for the quality of governance is negative but results insignificant.
| Table 3: Panel regression for the income tax revenues rel. to GDP | ||||||
|---|---|---|---|---|---|---|
| Fixed Effects | Random Effects | Dynamic (FE) | ||||
| Coeff. | t-value | Coeff. | t-value | Coeff. | t-value | |
| Current Account | -0.0032 | -0.27 | -0.0052 | -0.41 | 0.0006 | 0.05 |
| Quality of Gov. | -0.2303 | -0.71 | -0.0544 | -0.18 | -0.0688 | -0.33 |
| Young Population | -0.0130 | -0.1 | -0.0478 | -0.51 | -0.0232 | -0.32 |
| Old Population | 0.3062 | 1.49 | 0.2326 | 1.19 | 0.2258 | 2.70 |
| Debt | 0.0077 | 0.96 | 0.0062 | 0.96 | -0.0029 | -0.56 |
| IMF | -0.0039 | -1.98 | -0.0030 | -1.59 | -0.0006 | -1.02 |
| Grants | -0.0240 | -0.56 | -0.0027 | -0.07 | -0.0067 | -0.23 |
| Constant | -7.8976 | -0.79 | 2.3162 | 0.28 | 0.8756 | 0.11 |
| Lagged Dep. | 0.6089 | 8.64 | ||||
| Explained Variation | ||||||
| within | 0.0867 | 0.0811 | 0.4148 | |||
| between | 0.206 | 0.2461 | 0.9024 | |||
| overall | 0.1886 | 0.2246 | 0.8398 | |||
The prevalence of the value added tax (VAT) is hardly connected to the variables considered, according to the fixed effects estimation. However, in the random effects model the between variation is explained by a substantial degree. Thus, the decision to introduce or to extend the VAT is rather a long-run decision and the tax is not adjusted frequently. Accordingly, the prevalence of the VAR is path dependent. The quality of governance (positive) and the share of young people (negative) seem to have an impact on this kind of decision.
In the short run the current account balance has a negative impact on the VAT quota. This is plausible since the VAT tax also applies to imports and not just domestic value added. In contrast exports are typically not part of the tax base. If imports increase faster than exports (deterioration of the current account) the tax base for the VAT relative to GDP increases. That the coefficient for the current account is even higher in the dynamic model gives a hint that the before described endogenous reaction is more plausible and less any strategic consideration.
| Table 5: Panel regression for value added tax revenues rel. to GDP | ||||||
|---|---|---|---|---|---|---|
| Fixed Effects | Random Effects | Dynamic (FE) | ||||
| Coeff. | t-value | Coeff. | t-value | Coeff. | t-value | |
| Current Account | -0.0062 | -0.67 | -0.0070 | -0.7 | -0.0096 | -1.61 |
| Quality of Gov. | 0.1687 | 0.75 | 0.3491 | 1.85 | 0.1981 | 1.04 |
| Young Population | -0.0568 | -0.71 | -0.1134 | -1.84 | -0.0194 | -0.39 |
| Old Population | 0.0224 | 0.13 | 0.0516 | 0.31 | 0.0353 | 0.37 |
| Debt | -0.0001 | -0.02 | 0.0001 | 0.01 | -0.0022 | -0.54 |
| IMF | -0.0002 | -0.17 | -0.0004 | -0.49 | 0.0002 | 0.41 |
| Grants | -0.0159 | -0.4 | -0.0093 | -0.28 | 0.0020 | 0.08 |
| Constant | -0.2462 | -0.04 | 11.7545 | 2.09 | 2.4223 | 0.48 |
| Lagged Dep. | 0.4385 | 5.8 | ||||
| Explained Variation | ||||||
| within | 0.0598 | 0.0372 | 0.2583 | |||
| between | 0.1912 | 0.5268 | 0.8916 | |||
| overall | 0.1768 | 0.4777 | 0.8334 | |||
The explanatory power of the panel models for prevalence of other indirect taxes is similar to the results for the VAT. However, the between variation is already well explained in the fixed effects model. Short-run and long-run impacts go into the same direction. Particularly relevant are the quality of governance and the indebtedness. Both have a positive impact with similar magnitude in the random as well as the fixed effects regression.
The level of grants has a negative impact. It seems that grants have a tendency to crowd out this kind of taxation.
| Table 6: Panel regression for other indirect tax revenues rel. to GDP | ||||||
|---|---|---|---|---|---|---|
| Fixed Effects | Random Effects | Dynamic (FE) | ||||
| Coeff. | t-value | Coeff. | t-value | Coeff. | t-value | |
| Current Account | -0.0002 | -0.03 | 0.0003 | 0.05 | 0.0010 | 0.19 |
| Quality of Gov. | 0.3792 | 1.76 | 0.3377 | 2.37 | 0.0566 | 0.48 |
| Young Population | -0.0175 | -0.21 | -0.0257 | -0.54 | -0.0243 | -0.6 |
| Old Population | -0.0477 | -0.54 | -0.0071 | -0.08 | -0.0312 | -0.5 |
| Debt | 0.0094 | 1.86 | 0.0090 | 2.05 | 0.0026 | 0.99 |
| IMF | 0.0007 | 0.89 | 0.0006 | 0.8 | 0.0003 | 0.7 |
| Grants | -0.0521 | -1.94 | -0.0658 | -2.78 | -0.0296 | -1.84 |
| Constant | 6.6833 | 0.68 | 6.6929 | 1.41 | 1.1304 | 0.27 |
| Lagged Dep. | 0.5679 | 10.18 | ||||
| Explained Variation | ||||||
| within | 0.1548 | 0.152 | 0.495 | |||
| between | 0.4128 | 0.4874 | 0.9034 | |||
| overall | 0.3733 | 0.4359 | 0.8537 | |||
The within variation remains almost unexplained, where higher debt has a tendency to lower the relative importance of the VAT. When between variation affects the estimator, the quality of governance seems to have a positive impact on the relevance of the VAT, while GDP per capita a negative. For the dynamic regression, the positive impact of the current account corresponds to the finding that was already present for the VAT alone.
| Table 7: Panel regression for the relation between value added tax and total tax | ||||||
|---|---|---|---|---|---|---|
| Fixed Effects | Random Effects | Dynamic (FE) | ||||
| Coeff. | t-value | Coeff. | t-value | Coeff. | t-value | |
| Current Account | -0.0001 | -0.23 | -0.0001 | -0.14 | -0.0005 | -1.67 |
| Quality of Gov. | 0.0071 | 0.53 | 0.0180 | 1.96 | 0.0126 | 1.23 |
| Young Population | -0.0022 | -0.57 | -0.0034 | -1.25 | 0.0000 | 0.01 |
| Old Population | -0.0013 | -0.20 | -0.0008 | -0.15 | -0.0010 | -0.30 |
| Debt | -0.0006 | -1.98 | -0.0004 | -1.55 | -0.0003 | -1.34 |
| IMF | 0.0000 | 0.63 | 0.0000 | -0.08 | 0.0000 | 0.92 |
| Grants | 0.0009 | 0.36 | 0.0012 | 0.67 | 0.0012 | 0.76 |
| Constant | 0.2789 | 0.56 | 0.8251 | 2.78 | 0.2785 | 0.89 |
| Lagged Dep. | 0.3852 | 7.55 | ||||
| Explained Variation | ||||||
| within | 0.0363 | 0.0235 | 0.2241 | |||
| between | 0.0035 | 0.3759 | 0.8073 | |||
| overall | 0.0006 | 0.2992 | 0.6909 | |||
The rather path dependency and the low explanatory power of the panel models hinder the potential to forecast how the tax systems in Africa will evolve in the future even if several socio-economic trends would be foreseeable. However, a tendency to more taxation for richer and older-growing societies is plausible and backed by the data.
Social security contributions
Results for social security contributions differ substantially from the results for the income taxation. Overall, the between variation stays almost unexplained. The variables seem to be related to changes within the countries and provide little explanations why a country decides in favor of a larger social security system. The coefficient for the share of elderly seems to be comparable to the results of the income taxation. In a pay-as-you-go pensions system the link is plausible. Higher contribution rates are needed to meet the financing needs.
Remarkably, the share of the young population positively affects the social security contributions. However, the effect is smaller than for the share of the elderly population.
GDP per capita has a negligible impact. Rather relevant is the level of debt. More indebted countries seem to lean toward higher social security contributions. It would be interesting to check how the respective social security systems are organized. If the social insurances are purchasing government bonds a potential link to the debt level may be plausible.