Determinants of Tax Revenue Efforts in Developing Countries
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Author’s E-Mail Address: abhijit@ucsc.edu

This paper contributes to the existing empirical literature on the principal determinants of tax revenue performance across developing countries by using a broad dataset and accounting for some econometric issues that were previously ignored. The results confirm that structural factors such as per capita GDP, agriculture share in GDP, trade openness and foreign aid significantly affect revenue performance of an economy. Other factors include corruption, political stability, share of direct and indirect taxes etc. The paper also makes use of a revenue performance index, and finds that while several Sub Saharan African countries are performing well above their potential, some Latin American economies fall short of their revenue potential.

Abstract

This paper contributes to the existing empirical literature on the principal determinants of tax revenue performance across developing countries by using a broad dataset and accounting for some econometric issues that were previously ignored. The results confirm that structural factors such as per capita GDP, agriculture share in GDP, trade openness and foreign aid significantly affect revenue performance of an economy. Other factors include corruption, political stability, share of direct and indirect taxes etc. The paper also makes use of a revenue performance index, and finds that while several Sub Saharan African countries are performing well above their potential, some Latin American economies fall short of their revenue potential.

I. Introduction

Reaching the Millennium Development Goals (MDGs) will require a concerted effort from both developed and developing countries. Aid from developed countries will have to rise significantly to achieve the MDGs. Although the donors have pledged to increase development aid by US$18.5 billion (from a 2002 level of US$58 billion), the World Bank (2004) estimates that developing countries could effectively use at least US$30 billion initially. The developed countries also need to aim for improved market access for developing countries’ exports by eliminating tariffs and domestic subsidies.

However, because excessive reliance on foreign financing may in the long run lead to problems of debt sustainability, developing countries will need to rely substantially on domestic revenue mobilization. The experience with domestic resource mobilization of developing countries over the last 25 years has been mixed. In countries such as Botswana, Israel, Kuwait and Seychelles, the central government revenue’s share in GDP has been more than 40 percent on average. On the other hand, countries such as Argentina, Niger, Guatemala and Burkina Faso have struggled to raise their revenue above 11 percent.

In this paper we investigate the main factors that may explain the variation in resource mobilization of developing countries. More specifically, we look at the main determinants of revenues (excluding grants) of the central government, and analyze the extent to which factors such as government policies, the structure of the economy, institutions and the stage of development explain their variation. While a number of studies have analyzed the principal determinants of tax revenue, in this paper we extend the literature by using a broader dataset and correcting for some of the econometric issues that were previously ignored. The dataset is extended by using a larger number of countries over a sufficiently long time horizon. Moreover, we incorporate new variables such as specific sources of tax revenue, political stability, economic stability, law and order etc. as potential determinants of revenue performance. We address some potential econometric problems by employing econometric specifications that take into account, among other things, the persistence of revenue performance and the possibility of some of the explanatory variables being influenced by revenue performance.

Our principal findings are that structural factors such as per capita GDP, share of agriculture in GDP, and trade openness are strong determinants of revenue performance. We also find that although foreign aid improves revenue performance, foreign debt does not have a significant effect. Among the institutional factors, we find that corruption is a significant determinant of a country’s revenue performance. Political and economic stability matters as well, but this finding is not robust across specifications. Finally, we find that those countries that depend on taxing goods and services as their primary source of tax revenue, have relatively poor revenue performance. On the other hand, countries that rely more on income taxes, profit taxes, and capital gains taxes, perform much better.

We also construct a revenue performance index that allows us to compare actual revenue performance with predicted revenue performance. We find that several African countries, including a number of countries from Sub-Saharan Africa, perform significantly better than predicted. However, several countries from Latin America and Eastern Europe perform below their predicted revenue performance.

After reviewing the literature, we briefly describe the data. Then we introduce the empirical model and discuss the main econometric results. Next, we develop the revenue performance index and use this index to rank countries. To end, we conclude and make some policy recommendations.

II. Recent Research Findings

What affects revenues (measured as the ratio tax revenues to GDP) has been the subject of a long debate. Before turning to the evidence, we discuss factors that are typically included in the specifications. Researchers have included several variables such as per capita GDP, the sectoral composition of output, the degree of trade and financial openness, the ratio of foreign aid to GDP, the ratio of overall debt to GDP, a measure for the informal economy, and institutional factors such as the degree of political stability and corruption as potential determinants of revenue performance.

Per capita income is a proxy for the overall development of the economy and is expected to be positively correlated with tax share as it is expected to be a good indicator of the overall level of economic development and sophistication of the economic structure. Moreover, according to Wagner’s law, the demand for government services is income–elastic, so the share of goods and services provided by the government is expected to rise with income. The sectoral composition of output also matters because certain sectors of the economy are easier to tax than others. For example, the agriculture sector may be difficult to tax, especially if it is dominated by a large number of subsistence farmers. On the other hand, a vibrant mining sector dominated by a few large firms can generate large taxable surpluses.

The degree of international trade—measured by the share of exports and imports—should also matter for revenue performance. Imports and exports are amenable to tax as they take place at specified locations. Furthermore, most developing countries shifted away from trade taxes in the 1990s, which was largely due to the widespread liberalization of trade undertaken under the Uruguay Round. The effect of trade liberalization on revenue mobilization may be ambiguous. If this liberalization occurs primarily through reduction in tariffs then one expects losses in tariff revenue. On the other hand, Keen and Simone (2004) argue revenue may increase provided trade liberalization occurs through tariffication of quotas, eliminations of exemptions, reduction in tariff peaks and improvement in customs procedure. Rodrik (1998) also points out that there is a strong positive correlation between trade openness and the size of the government, as societies seem to demand (and receive) an expanded role for the government in providing social insurance in more open economies subject to external risks.

The degree of external indebtedness of a country may affect revenue performance as well. To generate the necessary foreign exchange to service the debt, a country may choose to reduce imports. In such a scenario, import taxes will be lower. Alternatively, the country may choose to increase import tariffs or other taxes with a view to generate a primary budget surplus to service the debt.

Foreign aid has also been identified as a factor that may affect revenue performance. A key distinction appears to be whether the aid is used productively or simply to finance current consumption expenditures. Moreover, the composition of aid has an important effect on revenue performance. For example, Gupta et al. (2004) find that concessional loans are associated with higher domestic revenue mobilization, while grants have the opposite affect.

The empirical findings have been mixed because of their sensitivity to the set of countries and the period of analysis. 2 The majority of these studies employ cross section empirical methods and hence ignore on the variation over time. Lotz and Morss (1967) find that per capita income and trade share are determinants of the tax share, and this finding has been replicated since (e.g., see Piancastelli (2001)). Chelliah (1971) relates the tax share to explanatory variables such as mining share, non-mineral export ratio and agriculture share. Several studies, including Chelliah, Baas and Kelly (1975) and Tait, Grätz and Eichengreen (1979), update Chelliah (1971) and obtain similar results. In a related study covering developing countries, Tanzi (1992) finds that half of the variation in the tax ratio is explained by per capita income, import share, agriculture share and foreign debt share. Recently, some studies have looked at the importance of institutional factors in determining revenue performance. For example, Bird, Martinez-Vasquez and Torgler (2004) find factors such as corruption, rule of law, entry regulations play key roles.

Several regional studies have looked into determinants of resource mobilization. For sub-Saharan African countries, Tanzi (1981) finds that mining and non-mineral export share positively affect the tax ratio. Focusing on the same region, Leuthold (1991) uses panel data to find a positive impact from trade share, but a negative one from the share of agriculture. In a similar study, Stotsky and WoldeMariam (1997) find that both agriculture and mining share are negatively related to the tax ratio, while export share and per capita income have a positive effect. They also find a positive but weak link between IMF programs and tax share. Ghura (1998) concludes that the tax ratio rises with income and degree of openness, and falls with the share of agriculture in GDP. He also finds that other factors like corruption, structural reforms and human capital development affect the tax ratio. While a rise in corruption is linked with a decline in tax ratio, structural reforms and an increase in the level of human capital is associated with an increase in tax ratio. In a study of Arab countries, Eltony (2002) observes that mining share has a negative impact on the tax ratio for oil exporting countries, but a positive impact for non-oil exporting countries.

To summarize, most studies find that per capita GDP and degree of openness is positively related to revenue performance, but a higher agriculture share lowers it. The effect of mining share and revenue performance is ambiguous. Studies such as Tanzi (1991) and Eltony (2002) found that foreign debt is positively related to resource mobilization.

III. Data Description

We use a panel dataset that covers 105 developing countries over 25 years. The choice of countries and years is primarily motivated by the desire to use consistently measured variables. Table 1 gives summary statistics of the key variables. The variable of interest is central government revenue (excluding grants) as a percentage of GDP, and is taken from Government Financial Statistics (GFS) and WEO Economic Trends in Africa (WETA). Among the explanatory variables, we include structural variables such as per capita GDP. share of agriculture in GDP, share of manufacturing in GDP, share of imports in GDP, ratio of debt and aid to GDP. Their sources are primarily the International Financial Statistics (IFS) and World Development Indicators (WDI). Information on the proportion of tax revenue collected from goods and services, income profit and capital gains, and trade comes from GFS, and information on the highest marginal tax rate (for corporate and individual tax rates) is from the WDI. We include the Trade Restrictiveness Index, which has a measure for average tariffs and which ranks countries based on non-tariff barriers and tariff rates. Finally, we use variables that capture institutional factors such as political stability, economic stability, corruption, law and order and government stability. These are obtained from the Intra Country Risk Guide (ICRG) dataset. We define those measures such that a higher number implies a better state of the world.

Table 1:

Summary of the Variables

article image

IV. Empirical Analysis

A. Graphical Analysis

Before we turn to the regression results, we briefly show the observed relationship between revenue performance and some explanatory variables (see Figures 1-6). A first observation is that agriculture share appears to have a strong negative relationship with revenue performance. There is no apparent correlation between manufacturing share and revenue performance. It also appears that per capita GDP and import share have a strong positive relationship with revenue performance. Similarly, political and economic stability appear strongly related to revenue performance.

Figure 1:
Figure 1:

Central Government Revenue and Agriculture

(In percent of GDP)

Citation: IMF Working Papers 2007, 184; 10.5089/9781451867480.001.A001

Figure 2:
Figure 2:

Central Government Revenue and Manufacturing

(In percent of GDP)

Citation: IMF Working Papers 2007, 184; 10.5089/9781451867480.001.A001

Figure 3:
Figure 3:

Central Government Revenue (% of GDP) and Log of Per Capita GDP

Citation: IMF Working Papers 2007, 184; 10.5089/9781451867480.001.A001

Figure 4:
Figure 4:

Central Government Revenue and Imports

(In percent of GDP)

Citation: IMF Working Papers 2007, 184; 10.5089/9781451867480.001.A001

Figure 5:
Figure 5:

Central Government Revenue (% of GDP) and Political Stability

Citation: IMF Working Papers 2007, 184; 10.5089/9781451867480.001.A001

Figure 6:
Figure 6:

Central Government Revenue (% of GDP) and Economic Stability

Citation: IMF Working Papers 2007, 184; 10.5089/9781451867480.001.A001

B. Baseline Regression Analysis

In our baseline panel regressions we use fixed and random effects specifications. The fixed effect assumes that certain country-specific characteristics are not captured by the explanatory variables, and that these are uncorrelated with the error term. The fixed effect specification is

yit=αi+β.Xit+γ.Yit+δ.Zit+εit,

where yit is a the ratio of central government revenue (excluding grants) to GDP in country i during period t, αi is the country fixed effect, Xit is set of structural variables, and the vectors Yit and Zit include institutional and policy variables. Alternatively, the random effects specification is

yit=α+β.Xit+γ.Yit+δ.Zit+ui+εit,

with ui the random effect.

The structural variables include the log of per capita GDP, the share of agriculture in GDP, the ratio of imports to GDP, share of aid and debt in GDP. The institutional variables include corruption, law and order, government stability, political stability and economic stability.

Finally, the policy variables include the various sources of tax revenue as a percentage of revenue, the highest corporate and income tax rate, and average tariffs.

The results of the baseline regressions, using the fixed- and random-effects specifications, are summarized in Tables 2 and 3. Wherever necessary, the regressions also include dummies for landlocked and resource-rich countries.3 The standard errors are adjusted for intra-group correlations. Because of the high degree of collinearity between the agriculture share and the log of GDP per capita (R2 = 0.81), we use those variables in separate specifications.

Table 2:

Determinants of Revenue Performance (Fixed Effects Estimation)

article image
Note: Robust z statistics in brackets.+ significant at 10%; **significant at 5%; *significant at 1%.
Table 3:

Determinants of Revenue Performance (Random Effects Estimation)

article image
Note: Robust z statistics in brackets.+ significant at 10%; **significant at 5%; *significant at 1%.

A first finding is that coefficient on log of per capita GDP is significantly positive in all the random-effects regressions and in most fixed–effects specifications. This is in line with other studies that found that the capacity to collect and pay taxes increases with the level of development (see for example Chelliah, 1971).

Our results also suggest a strong negative and significant relationship between agriculture share and revenue performance. For example, a one percent increase in the share of agriculture sector could reduce revenue performance by as much as 0.4 percent. This relationship could work through both the supply and the demand side. On the supply side, if a large part of the agriculture sector is subsistence, then this sector may be hard to tax. Moreover, it may be politically infeasible to tax the agriculture sector. On the other hand, a large agriculture sector may reduce the need to spend on public goods and services, which tend to be relatively urban-based.

Next, in most specifications we find a strong positive relationship between openness and revenue performance. For example, an increase in the ratio of imports to GDP of one percent may increase revenue performance by up to 0.15 percent. One explanation for this finding is that trade-related taxes are easier to impose because the goods enter or leave the country at specified locations.

We also find that foreign aid has a positive effect on revenue performance, but the relationship appears weaker than that for some other variables. Gupta et al. (2004) had already pointed out that if foreign aid comes primarily in the form of loans, then the burden of future loan repayments may induce policymakers to mobilize higher revenues. However, aid in the form of grants may created a moral hazard problem if it decreases incentives to increase the tax base. We found that debt is negatively related with revenue performance, although the relationship is not very strong.

Our results for the institutional factors are mixed. We do not find a significant effect from the variables that capture government stability, corruption, and law and order. However, across some specification, the impact is significant when institutions are measured by political and economic stability.

We also investigate how the various sources of tax revenues affect the share of central government revenue in GDP. We find that countries that rely more on taxes on goods and services as a source of revenue have lower revenue performance. Since most of the taxes on goods and services are indirect taxes, they tend to be regressive in nature. As a result, they may exacerbate the inequality in income distribution and reduce the tax base, which in some cases may result in a reduction in the share of revenue in GDP. In contrast, greater reliance on taxation of income, profits and capital gains appears to improve revenue performance. To the extent that these taxes are progressive, they reduce income dispersion and generate higher revenue. We also find that the share of tax revenue from trade does not affect revenue performance significantly.

Finally, revenue performance does not appear to be determined significantly by corporate and individual tax rates, or by average tariffs, once we have taken into account the structural variables, institutional variables and various sources of tax revenue. As a result, we drop these variables from subsequent analysis.4

C. Panel-Corrected Standard Error Estimation

Most of the previous empirical analyses did not consider that revenue performance tends to be highly persistent over time (Leuthold (1991) is an exception). This persistence is illustrated in Figure 7 for a subset of the countries in our dataset.

Figure 7:
Figure 7:

Variation in Revenue Performance

Citation: IMF Working Papers 2007, 184; 10.5089/9781451867480.001.A001

In the presence of serial correlation, the empirical model becomes

yit=α+β.Xit+γ.Yit+δ.Zit+ui+εit,

where

εit=ρi.εit1+vit.

After testing for first-order serial correlation in the residuals with a Wooldridge test, we estimate the model using panel-corrected standard error estimates (PCSE). 5 The PCSE uses Prais-Winsten regression, and assumes that the disturbances are heteroskedastic and contemporaneously correlated across panels. It can be used in the presence of an AR(1) with a common coefficient across all the panels (ρi = ρ,i), and also with specific coefficient for each panel (ρiρj, ∀ij). When autocorrelation with a common coefficient of correlation is specified, the common correlation coefficient is computed as

ρ=ρ1+ρ2+ρ3+.......+ρmm,

In this expression, ρi is the estimated autocorrelation coefficient for panel i and m is the number of panels.

Although the PCSE estimates yields larger standard errors, the results are similar to the baseline results (see Tables 4 and 5). As before, revenue performance increases with per capita GDP and import share, and declines with agriculture share in GDP. The impact of foreign aid is now stronger, especially when the autocorrelation process is different for each panel. In this context, an anticipated increase in aid from around US$80 billion in 2004 to US$130 billion in 2010 would increase revenue performance by as much as 0.6 percent. Among the institutional factors, corruption has a significantly adverse effect on revenue performance (confirming the result by Ghura (1999)). Political and economic stability are significant only for some specifications, just like in the baseline estimations. We also confirm our earlier findings that revenue performance in countries with heavy reliance on taxes from goods and services is weaker, it is better for those countries that rely more on taxes from income, profits and capital gains. Finally, relatively high reliance on tax revenue from trade remains associated with poor revenue performance, but this finding is not robust across specifications.

Table 4:

Determinants of Revenue Performance (Common Correlation Coefficient)

article image
Note: Robust z statistics in brackets.+significant at 10%; **significant at 5%; *significant at 1%.
Table 5:

Determinants of Revenue Performance (Panel Specific Correlation Coefficient)

article image
Note: Robust z statistics in brackets.significant at 10%; *significant at 5%; *significant at 1%.

D. Sensitivity Analysis

Testing for Endogeneity

Countries that find it difficult to mobilize revenue from domestic sources would be expected to rely more heavily on foreign aid and debt as a source of revenue. Therefore, there can be an endogeneity problem among foreign aid, debt and revenue performance.

To allow for this endogeneity, we re-estimate the specifications presented in columns III-VI and IX-XII of Tables 4 and 5 with lagged values of aid share and debt share, instead of contemporaneous values. The results are given in Table 6.

Table 6:

Determinants of Revenue Performance (Lagged Values of Foreign Aid and Debt)

article image
Note: Robust z statistics in brackets.significant at 10%; **significant at 5%; *significant at 1%.

It appears that endogeneity is not a severe problem, because the findings in Table 6 remain similar to the earlier results. While debt continues to be weakly related to revenue performance, foreign aid has a positive and significant impact on revenue performance (particularly for the case where countries have different degrees of persistence in revenue performance). We also see that the sources of tax revenue are strong determinants of revenue performance, since the coefficient on the share of taxes from goods and services, as well as that from income, profits and capital gains are significant across all specifications.

Dynamic Panel Data

Instead of allowing for serial correlation in the error term, the econometric specification could also capture the persistence in revenue performance (described in Section IV.C) by including the lagged value of the dependent variable. Because the lagged dependent variable is correlated with the error term, it is well known that this creates some estimation problems. To overcome these problems, Arellano and Bond (1991) proposed a generalized method-of-moments estimator using lagged levels of the dependent variable and the predetermined variables and differences of strictly exogenous variables. This method is referred to as difference-GMM. A problem with the original Arellano-Bond estimator is that lagged levels of variables may be poor instruments if those variables are highly persistent. In such cases, Arellano and Bover (1995) and Blundell and Bond (1998) describe how additional moment conditions can increase efficiency. This procedure is referred to as system-GMM.

Table 7 reports the results from the dynamic panel methods. 6 Our results confirm that lagged revenue share is a strong and significant predictor of current revenue performance, across both difference- and system-GMM. Overall, the results from the difference-GMM are quite weak, and only agriculture share, aid share and debt share are significant predictors of revenue performance. However, once we use system-GMM to take into account the near random walk of revenue performance, the results are broadly similar to the baseline results.

Table 7:

Determinants of Revenue Performance (Dynamic Panel Specification)

article image
Note: Robust z statistics in brackets.significant at 10%; ** significant at 5%; *significant at 1%.All variables are in difference.Second-order autocorrelations of residual are always rejected.Aid share and Debt share are treated as endogenous variables because they can be influenced by revenue performance.

Looking at columns (V) to (VIII) in Table 7 we find that per capita GDP, agriculture share and import share are significant predictor of revenue performance. However, the impact of per capita GDP is substantially smaller in the dynamic specification. The impact of agriculture share and import share are also marginally smaller in the dynamic specification. Both foreign aid share and debt share significantly affect the revenue performance. While aid share has a positive impact, a higher debt share is associated with a lower revenue performance. Finally, as in the baseline specification, share of revenue from taxing goods and services is negatively related to revenue performance, while share of revenue from income, profit and capital gains has a positive impact.

Sub Sample Analysis

Next, we look closer at the revenue performance of countries that belong to similar income groups. To proceed, we split the sample according to the World Bank’s classification of countries according to income group (see Appendix B for the list of countries by income group). The estimation results are given in Tables 8-10. Several interesting findings emerge.

Table 8:

Determinants of Revenue Performance (Low-Income Countries)

article image
Note: Robust z statistics in brackets.significant at 10%; **significant at 5%; *significant at 1%.
Table 9:

Determinants of Revenue Performance (Middle-Income Countries)

article image
Note: Robust z statistics in brackets.+ significant at 10%; **significant at 5%; *significant at 1%.
Table 10:

Determinants of Revenue Performance (High-Income Countries)

article image
Note: Robust z statistics in brackets.+ significant at 10%; ** significant at 5%; *significant at 1%.

We find that the share of agriculture in GDP is a significant determinant of revenue performance across all income ranges. On the other hand, while per capita GDP has a strong impact on revenue mobilization in high-income countries, its effect is somewhat weaker in low-income and middle-income countries. For the low- and middle-income countries we also find a strong and positive relationship impact from openness to trade; this relationship is not always significant for high-income countries.

For low-income countries, foreign aid has a significant positive effect on revenue performance across most specifications. For these countries, an increase in foreign aid by 1 percent can improve revenue performance by as much as 0.11 percent. This relationship is not statistically significant for middle-income and high-income countries. There is no significant relationship between foreign debt and revenue performance in any of the groups.

Among institutional factors, the coefficient on corruption is significant for low-income and middle-income countries. Indeed, for these countries, a reduction in corruption (implying an increase in the corruption index) would substantially increase revenue. For example, in low-income countries, an increase in the corruption index of one unit would improve revenue performance by about 1.5 percent; and in middle-income countries, the effect is slightly greater than 0.5 percent. On the other hand, the coefficients on government stability and law and order are not statistically significant in any of the groups.

Next, the results suggest that political stability is weakly related to revenue performance for low- and middle income countries but not for high-income countries. For low-income countries, an increase in the political stability index of one unit can increase revenue performance by 0.08 percent; for middle-income countries the effect would be 0.07 percent. However, political stability has a weak negative relationship in high-income countries. Also, economic stability has a weak impact on revenue performance, and only in low-income countries.

Finally, we find that in low-income and high-income countries, but not in the middle-income group, greater reliance on taxing goods and services as a source of revenue is associated with poor revenue performance. Furthermore, greater reliance on taxing income, profits and capital gains is associated with improved revenue performance across all income groups.

Using various forms of panel data estimations, and correcting for the observed persistence in revenue performance, our results confirm that the principal determinants of revenue performance include factors like per capita GDP, agriculture’s share in GDP, trade openness foreign aid, corruption, political stability and specific sources of tax revenue. Although the results are broadly similar across most specifications, we prefer the results from the panel-corrected standard error estimates with panel specific correlation coefficient and system-GMM estimates.

V. Assessment of Revenue Performance

So far our analysis has focused on finding the main factors that affect revenue performance in a sample of developing countries. However, as pointed out by Chelliah (1971) and Chelliah et. al. (1975), this does not tell us whether a country could not, if it wanted, attain higher revenue performance. Countries inherently have different capacities to raise revenues, and this must be taken into consideration while making cross-country revenue comparisons. We follow these studies and compute the revenue effort for the countries in our sample.

Our starting point is to take the estimated coefficients of the regressions in the previous section to compute the ‘predicted’ revenue performance of the countries in the sample. Next, we use this predicted revenue performance to construct an index of revenue effort by taking the ratio of the actual revenue performance and the predicted values. Thus, a country that lies on the regression line will have a revenue performance index equal to 1, and countries that have actual revenue performance above (below) predicted revenue performance have a revenue effort index bigger than (smaller than) one.

Of course this approach has a number of limitations. First, there might be some unobserved variables that affect revenue performance. Second, while calculating the tax potential we must focus only on factors, which are ‘given’ i.e., beyond the control of the government. Finally, the revenue effort index will not be robust to the regression specification. Therefore, in deciding which equation to use, one needs to consider the statistical fit as well as the economic rationale. Aware of these qualifications, we proceed and we present the revenue effort indices in Table 11.7

Table 11:

Revenue Effort Indices for Developing Countries (1980–2004)

article image
Author’s Calculations.