Chapter

14 Tax Revenue in Sub-Saharan Africa: Effects of Economic Policies and Corruption

Author(s):
Sanjeev Gupta, and George Abed
Published Date:
September 2002
Share
  • ShareShare
Show Summary Details
Author(s)
Dhaneshwar Ghura

I. Introduction

Large fiscal deficits have been a daunting problem for a number of countries in sub-Saharan Africa over the past several years. Rapid expansions in expenditure and declining or low revenue levels have been the main cause of fiscal imbalances.1 Recent endogenous growth models have demonstrated that growth can be enhanced by, inter alia, reducing fiscal imbalances, which, in turn, can be achieved by either lowering expenditure or raising revenue.2 However, many countries in the region have reduced expenditure to minimum sustainable levels, especially in health, education, and infrastructure. Thus, raising tax revenue to achieve fiscal sustainability would be a feasible alternative. Also, in order to improve the environment for private sector development and sustained economic growth, governments need to play a supportive role by investing in physical and human capital, and institutional infrastructure. Tax revenue is needed for such expenditure if inflationary financing and the crowding out of the private sector are to be avoided (Hamada, 1994).

The mobilization of tax revenue is an important policy objective. While governments can do little in the short run to change the structural determinants of the tax revenue (such as the composition of value added), they can alter other factors that influence tax revenue, such as the economic policies, the level of corruption, and the quality of tax administration. As documented by Nashashibi and Bazzoni (1994), the wide divergences between the effective and statutory tax rates in many countries in the region indicate that there is scope for raising tax revenue without increasing tax rates by reinforcing tax and customs administrations, reducing tax exemptions, and fighting fraud and corruption. Nevertheless, Heller (1997, p. 41) cautions that “one must be realistic in terms of improvement in revenue ratios that can be reasonably expected to be achieved in many African countries, given the low level of development and the heavily agricultural and informal sector character of their economies.” In addition, Tanzi (Chapter 2 in this volume) cautions that the fight against corruption takes time, needs to be undertaken on several fronts, and can be costly. Furthermore, tax mobilization and reform can be achieved only when there is strong political will and leadership to adopt the necessary measures (Hamada, 1994).

A number of empirical studies have investigated the determinants of tax revenue in developing countries.3 Few of them, however, have focused on the effects of economic policies and corruption on tax revenue, even though these variables have been shown to influence other aspects of economic performance.4Tanzi (1989) argues that the wide fluctuations in tax ratios observed in several countries over short time periods cannot be satisfactorily explained by variations in the traditional determinant of tax revenue—the tax base; rather, changes in the macroeconomic policy environment have played an important role. Also, chand and Moene (Chapter 4 in this volume) note that fiscal corruption has been a key factor behind the poor revenue performance in a number of developing countries.

This paper contributes to the empirical literature by focusing on the impact of economic policies and corruption on tax revenue, using data for 39 sub-Saharan African countries during 1985–96. Among the economic policy-related variables considered in this study are the rate of inflation, the percentage change in the real effective exchange rate, the implementation of structural reforms, and the provision of public services by the government. The effect of corruption, which is typically defined as the abuse of public power for private benefit, is captured by an index that measures the extent to which bribes are generally expected by government officials in relation to, inter alia, tax assessments, trade licenses, and exchange controls.

The tax performance of sub-Saharan African countries varied widely during 1985–96. For example, while the average total tax revenue–GDP ratio for these countries was about 17 percent during this period, 9 countries had ratios below 10 percent and 10 countries had ratios above 20 percent (Table 1). The majority of countries had average tax revenue-GDP ratios below 15 percent during 1985–96. On average, the total tax revenue-GDP ratio declined over time from 18.4 percent in 1985 to 16.3 percent in 1996. The largest decline was experienced by the oil producers—Cameroon, the Republic of Congo, Gabon, and Nigeria—whose average tax ratio fell from 25½ percent in 1985 to 18½ percent in 1996, largely reflecting the decline in oil prices.5 For the non-oil producers, the ratio declined from 17.3 percent to 16 percent.

Table 1.Sub-Saharan Africa: Tax Revenue in Selected Countries1(In percent of GDP)
Average
1985–901991–961985–96
Benin29.610.910.2
Botswana40.934.137.5
Burkina Faso29.410.19.7
Burundi13.614.614.2
Cameroon2,316.212.514.3
Central African Republic29.77.68.6
Chad27.17.57.3
Comoros210.911.811.4
Congo, Republic of225.224.224.7
Côte d’Ivoire218.716.917.5
Equatorial Guinea214.711.713.2
Ethiopia12.28.510.4
Gabon2,324.923.724.3
Gambia, The19.119.419.2
Ghana11.614.212.9
Guinea13.710.911.6
Guinea-Bissau6.56.06.2
Kenya19.922.721.3
Lesotho34.440.037.2
Madagascar9.88.08.9
Malawi18.617.017.8
Mali29.611.210.4
Mauritius20.419.019.7
Mozambique16.217.416.8
Namibia25.633.329.4
Niger28.56.57.5
Nigeria313.712.513.1
Rwanda9.97.68.8
São Tomé and Príncipe11.810.911.3
Senegal214.313.714.0
Seychelles34.732.433.5
Sierra Leone5.49.78.6
South Africa24.425.024.7
Swaziland26.330.328.3
Tanzania14.312.013.2
Togo220.112.616.3
Uganda5.87.86.8
Zambia18.116.217.0
Zimbabwe32.430.931.6
Unweighted averages
Sub-Saharan Africa17.316.416.9
Oil-producing countries20.018.219.1
Non-oil-producing countries16.916.216.6
CFA franc countries14.112.913.5
Oil-producing countries22.120.121.1
Non-oil-producing countries11.810.911.3
Non-CFA franc countries19.118.418.8
Oil-producing countries413.712.513.1
Non-oil-producing countries19.418.719.0

See the Appendix for the sources and definitions of the variables.

CFA franc countries.

Oil-producing countries.

Nigeria.

See the Appendix for the sources and definitions of the variables.

CFA franc countries.

Oil-producing countries.

Nigeria.

While controlling for the elements of the tax base, this paper investigates whether economic policies and corruption can account for part of the variation observed in tax revenue performance in sub-Saharan African countries. The results indicate that, in addition to variables related to income and the structure of the tax base, a number of other factors influence tax revenue, including macroeconomic and structural policies, the provision of public services by the government, and the level of corruption. The rest of the paper is organized as follows. The next section presents the theoretical model and discusses the hypotheses. Section III presents the empirical results, and the last section draws conclusions and suggests possible policy implications.

II. Theoretical Considerations

In order to account for the effects of economic policies and corruption, along with the impact of the elements of the tax base, this paper extends the tax model developed by Heller (1975).

The public decision maker’s utility function is given by

where Y–T (equal to GDP, Y, minus tax revenue, T) is the private sector’s disposable income; D is net domestic government borrowing,6G is total government expenditure, and F + L is net foreign financing, consisting of grants (F) and loans (L), including external arrears accumulation or decumulation (net of amortization). The variables D and (F + L) can be either positive or negative, and thus the first derivatives of U with respect to D and (F + L) are either negative (D and F+L> 0) or positive (D and F + L < 0). All variables in the model are in real per capita terms. The budget constraint faced by the decision maker is given by

Expanding on Leuthold’s (1991) applied tax model, it is assumed that the actual tax revenue-GDP ratio (T/Y) is a function of the desired tax revenue-GDP ratio (T/Y)* and the availability of certain tax bases (B), as well as the status of economic policies (E) and the level of corruption (C). That is,

Desired tax revenue is determined by maximizing (1) subject to (2). Following Heller (1975), it is assumed that the utility function takes a quadratic form as follows:

where the α’s are positive constants, and Ys and Gs are subsistence levels of income and government expenditure, respectively. Empirically, a quadratic utility function is preferable to a log-linear one because the terms D and F + L can be either positive or negative. Since Ys and Gs are not observable, following Leuthold (1991), it is assumed that they are simple linear functions of income, as follows:

and

Maximizing equation (1) with respect to T, G, and D, subject to constraint equation (2), yields the following reduced form for the desired equation for the tax revenue–GDP ratio:

where α = (−α1, + α3 − α1α46 + α4α56) and ß = α24 + α6)/α6. Combining equations (3) and (6) yields

Since β is positive and a could be either positive or negative, (T/Y)* is a negative function of (F + L)/Y and an ambiguous function of the inverse of per capita income (1/F).

The literature on the determinants of tax revenue provides a set of testable hypotheses. This paper focuses on those hypotheses—on income, the tax base, economic policies, corruption, and the external environment—that can be tested using available data for sub-Saharan African countries. The rest of this section briefly discusses the hypotheses relating to the actual variables used; detailed discussions are provided by Tanzi (1989), Farhadian-Lorie and Katz (1989), and Nashashibi and Bazzoni (1994). The Appendix gives the definitions and sources of the variables used in this study.

The theoretical model predicts an ambiguous effect of increases in per capita income on tax revenue. This effect stems from the differential impact of an increase in income on different categories of tax revenues. While a higher level of economic development would be expected to raise the ratio of indirect tax revenue to GDP, it would be expected to lower the trade tax revenue–GDP ratio; thus, the effect on aggregate tax revenue is ambiguous. Farhadian-Lorie and Katz (1989) have noted that trade taxes have historically been a major source of government revenue during the early stages of economic development because they are easier to collect than domestic income or consumption taxes, owing to the rudimentary status of tax administration, as well as the limited availability of “tax handles.” During the later stages of development, however, collection costs are expected to fall, dependence on trade taxes to decline, and dependence on indirect taxes to rise.

Elements of a country’s tax base—better known as tax handles—considered in this study are the share of agriculture in GDP (AGS), the share of oil and non-oil mining activities in GDP (OIL and MINE), and the ratio of the sum of exports and imports to GDP (OPEN).7 The sectoral composition of value added constitutes a key element of the tax base. In many sub-Saharan African economies, a large share of GDP results from agricultural activities.8 However, the agricultural sector is difficult to tax owing to the prevalence of subsistence activities, which are largely informal. The administrative costs for the tax department of organizing and monitoring subsistence activities can be prohibitively high in relation to potential revenue yield. In general, therefore, a negative relationship would be expected between the tax revenue–GDP ratio and the share of agriculture in GDP. Mining activities, on the other hand, are organized, and thus easy to monitor and tax.9 A positive correlation would be expected between the variable OPEN and the tax revenue–GDP ratio: as the international trade sector is a well-organized and monetized sector, administrative costs of the tax system related to this sector should be lower than others.

Tanzi (1989) argues that one has to look beyond the traditional determinants of tax revenue—elements of the tax base—to obtain a satisfactory explanation of the wide fluctuations in tax ratios observed in several countries over short time periods; macroeconomic policy plays an important role. The effect of macroeconomic policy is captured by the inflation rate (INF) and the percentage change in the real effective exchange rate (RERG).10 The effect of inflation on tax revenue can be registered through three main channels. First, according to the Tanzi-Olivera effect, in an inflationary environment, when actual tax payments lag the transactions to be taxed, tax obligations are lower in real terms at the time of tax payments (Tanzi, 1977). Second, excise duties on a number of products (e.g., tobacco, alcohol, and gasoline) may be levied at specific rates that may not necessarily be adjusted in line with inflation (Tanzi, 1989). Finally, high inflation rates reduce the tax base because in order to protect the real value of their wealth, economic agents make portfolio adjustments in favor of assets that typically escape the domestic tax net (such as land, livestock, jewels, and foreign capital). An appreciation of the real effective exchange rate is expected to raise imports and lower exports. The overall effect of a real effective exchange rate appreciation on tax revenue could be positive, given the greater dependence of tax receipts on import rather than export taxes. Nevertheless, an overvaluation of the real effective exchange rate—typically brought about by expansionary financial policies—would be expected to adversely affect overall economic activity, and thus to lower tax revenue.

Tax revenue can also be influenced by the implementation of structural reforms (STRUC). Such reforms can raise tax revenue by improving economic efficiency and resource allocation, enhancing external competitiveness, expanding the productive capacity of the economy, and broadening the tax base.11 In recent years, a number of sub-Saharan African countries have made progress in the implementation of structural reforms. These have included (1) public enterprise restructuring and privatization; (2) retail and producer price decontrol; (3) exchange and trade liberalization; (4) financial sector reform; (5) tax reform; (6) civil service reform; and (7) legal reform. A number of countries that have made progress in structural reforms have benefited from technical assistance aimed at increasing voluntary compliance and self-assessment, expanding the use of final withholding, improving collection procedures, developing audit plans and procedures, and reorganizing tax administration along functional lines (Abed and others, 1998).

Finally, in the area of economic policies, it is hypothesized that, when taxpayers see the benefit of their tax payments in terms of government’s provision of public services, proxied by improvements in an index of human capital development (HCI), their willingness to pay taxes would be expected to increase.12 In a number of developing economies, owing especially to weaknesses in the expenditure management process and the existence of corruption, part of the budgeted outlays do not reach their intended final destinations. Indeed, for the set of countries used in this study, there is a large positive correlation between declining corruption and rising human capital (a correlation coefficient of 0.48; see AppendixTable A1).13 The index of human capital is intended to measure the visible impact of government expenditure on actual priority outlays.

It is hypothesized that corruption (CORRUPT) lowers the tax revenue–GDP ratio. Klitgaard (1998) notes that acts of corruption include (but are not limited to) bribery, extortion, influence peddling, nepotism, fraud, and embezzlement. Tanzi (Chapter 2 in this volume) provides a set of factors that encourage fiscal corruption, including complicated tax laws, excessive discretionary power vested in tax administrators, the necessity for frequent contacts between taxpayers and tax officials, weak legal and judicial systems, lack of accountability and transparency in the tax administration, and low salaries in the public sector. Corrupt tax and custom officials allocate a proportion of their working hours to (1) collecting bribes in exchange for alleviating tax burdens of taxpayers offering these bribes; and (2) complicating procedures for taxpayers who refuse to participate in the bribery scheme, thus forcing them out of business or into the informal sector. These activities lower tax revenue for the public treasury.14 Pervasive corruption in an economy is expected to lower investment and economic growth, and thus weaken the tax base.15 The index of corruption used in this study is taken from the International Country Risk Guide, published by the Political Risk Services Group in Syracuse, New York. This index, which takes integer values ranging from 0 (high corruption) to 6 (low corruption), measures the extent to which bribes are generally expected by government officials in relation to, inter alia, tax assessments, trade licenses, and exchange controls.16

With respect to the external environment, the theoretical model predicts a negative effect of increases in external financing on the domestic tax revenue–GDP ratio. Two variables are used to capture this effect—the ratio of external grants to GDP (GRANTY) and the change in the debt stock-GDP ratio (CHDETY).17 The effects of changes in the terms of trade on the tax revenue–GDP ratio are ambiguous. If a large proportion of a country’s imports is price inelastic, a deterioration in the terms of trade owing to an increase in import prices could improve the tax base. However, if the deterioration in the terms of trade is due to a decline in export prices and the country depends on revenue from export taxes, the tax base would be expected to shrink. In addition, the decline in income associated with a decline in the terms of trade would be expected to lower the tax base.

III. Empirical Framework and Results

Empirical Framework

An empirical counterpart of equation (7) for the ith sub-Saharan African country at time t is written as follows:

where TRY is the tax revenue–GDP ratio; PCI is per capita income;18AGS is the share of agriculture in GDP; OPEN is the ratio to GDP of the sum of exports and imports; INF is the rate of inflation; OIL is a dummy variable that takes a value of 1 if the ith country is an oil producer; MINE is a dummy variable that takes a value of 1 if the ith country is not an oil producer but whose mining share is at least 5 percent of GDP; RERG is the percentage change in the real effective exchange rate; STRUC is a dummy variable capturing the implementation of structural reforms by the ith country; HCI is an index of human capital; CORRUPT is an index of corruption that varies from 0 (high corruption) to 6 (low corruption); GRANTY is the external grants-GDP ratio; CHDETY is the change in the stock of external debl-GDP ratio; and TTG is the percentage change in the terms of trade. The subscript s for HCI denotes that this variable is time-invariant over two subperiods (1985–90 and 1991–96). The coefficients t1t5 broadly capture the effects of variables related to income and the tax base that are typically used in the literature. The coefficients n1n8 are intended to capture the effects of variables related to economic policies and corruption, typically ignored in the empirical literature. The Appendix provides the definitions and sources of the variables; AppendixTable A1 provides the matrix of correlation coefficients for these variables; and AppendixTable A2 provides the period averages of the data.

A regression framework is used to estimate the tax equation, with an unbalanced panel data set for 39 countries covering the period 1985–96.19 As the data are in panel form, the error term for equation (8) accordingly has three components: ui and vt, which capture country- and time-specific effects, respectively, and eit, which is an error term common to all observations. To deal with time effects, the data are processed to remove the time means from the series, and the resulting model is estimated without an intercept. Country heterogeneity is captured by the inclusion of country-specific information in the indicators for the level of human capital development, the stance of economic policies, changes in the terms of trade, the levels of external indebtedness and grants, and the level of corruption. In addition, dummy variables for subgroups of countries (CFA, STRUC, OIL, MINE) are used to account for the possibility of fixed effects stemming from a priori information regarding country characteristics and institutional arrangements.

In order to correct for possible simultaneity bias stemming from the variables PCI, OPEN, INF, STRUC, and CORRUPT, an instrumental variables technique is used.20 Since the variable STRUC is binary (0,1), a logistic model of the following type is used obtain its predicted values: loge(p/(1 − p)) = a + bX, where p = Pr(Y = 1 | X) is the response probability to be modeled, Y is the binary (0,1) response, X is the vector of instruments, and b is the vector of coefficients. To deal with the problem of heteroscedasticity, a feasible instrumental variables generalized least squares (IV-GLS) procedure is used.21

As noted above, data on the level of corruption are available for only 27 of the 39 countries. To avoid losing observations in the regression analysis, the instrumental variables mentioned above are used to estimate the missing data for this variable. As shown in AppendixTable A2, this set of instruments reproduces the available data fairly accurately, thereby providing a reasonable degree of confidence in the data that are generated for the countries with no available data. The correlation coefficient between the generated corruption (CORRUPTP) and actual corruption (CORRUPT) variables is 0.74 (AppendixTable A1).

Econometrics Results

The regression results are given in Table 2. Following the traditional empirical literature, regression (1) includes only variables related to income and the tax base; this is taken to be the base regression. Regressions (2)-(4) also use variables related to macroeconomic and structural policies, the extent of public services provided by the government, corruption, and the external environment. The last column of Table 2 provides an indication of the relative importance of the explanatory variables in explaining the tax revenue–GDP ratio, as captured by the beta coefficients using the results of regression (4). The following observations can be made based on the results.

  • The results of the base regression are broadly consistent with those available in the empirical literature. They indicate that the tax revenue–GDP ratio grows with an increase in income, a decline in the share of agriculture in GDP, greater openness of the economy, and the existence of oil and non-oil mining sectors.22 As indicated by the beta coefficients, among the variables capturing the effects of income and the tax base, the degree of openness (OPEN) exerts the largest impact on the tax ratio, followed by the income variable (l/PCI), the existence of an oil sector (OIL), the agricultural share (AGS), and the existence of a non-oil mining sector (MINE).
  • When the base regression is augmented to include other variables, the main results relating to income and the elements of the tax base do not change by much. Nevertheless, owing to the high degrees of correlations between human capital (HCI), on the one hand, and the inverse of per capita income (PCI) and the agriculture share (AGS), on the other—correlation coefficients of −0.72 and −0.66, respectively (see AppendixTable A1)—the magnitude and statistical significance of the impact of loge(1\PCI) and AGS on the tax ratio fall.
  • The results support the theoretical view provided by Tanzi (1989) and some existing empirical evidence that the macroeconomic policy environment matters for tax revenue performance.23 An increase in inflation (a proxy for expansionary financial policies) lowers the tax ratio. An appreciation of the real effective exchange rate has a positive although statistically insignificant effect on the tax ratio.
  • Structural reforms (STRUC) have positive and significant effects on the tax revenue–GDP ratio. This result indicates that, on average, countries that made progress in the implementation of structural reforms were able to raise their average tax revenue–GDP ratios higher than countries that did not.
  • The effect of human capital development (HCI), another economic policy-related variable used as a proxy for the provision of public services by the government, is positive and significant. It could be inferred from this result that, when taxpayers see the benefits of their tax contributions, their willingness to voluntarily comply with their tax obligations increases. It should be noted that HCI is measured in such a way as to avoid the problem of causality and simultaneity bias (see the Appendix).
  • As shown by the beta coefficients, of the economic policy-related variables, inflation exerts the largest impact on the tax revenue–GDP ratio, followed by the implementation of structural reforms. Thus, economic policies that emphasize a prudent financial stance and the implementation of structural reforms can be expected to raise tax revenue. The relative impact of the provision of public services by the government is small.
  • There is strong evidence that an increase in the level of corruption (as captured by a decline in CORRUPTP) lowers the tax revenue–GDP ratio. The important role played by corruption in influencing tax revenue is confirmed by its relatively high beta coefficient. Thus, efforts to lower corruption would be expected to increase tax revenue significantly.
  • As the levels of corruption and human capital development are highly correlated (see AppendixTable A1), when these two variables are included in the same regression the magnitude and statistical significance of their impact fall.
  • Increases in external grants lower tax revenue. While this result could be indicative of substitution between domestic tax revenue mobilization and the availability of external grants, it could also reflect a reverse causality problem, whereby countries with lower tax revenue–GDP ratios have been recipients of larger amounts of grants. Several factors guide the flow of grants, including the level of development, the status of implementation of macroeconomic and structural policies, and the level of corruption. The regression analysis controls for these factors and still finds a significant independent effect of external grants on tax revenue, indicating that, on average, grants tend to substitute for domestic tax revenue mobilization.
Table 2.Estimates of the Tax Equation and Beta Coefficients1
Regression Number
Explanatory Variables2(1)(2)(3)(4)Beta Coefficients3
Income loge (1/PCI)−3.223***−1.248**−2.407***−1.696***−0.190
(8.83)(2.51)(6.72)(3.38)
Tax base
Share of agriculture in GDP ratio−0.078***−0.105***−0.045**−0.065***−0.121
(AGS)(4.41)(5.59)(2.22)(2.89)
Openness (OPEN)0.127***0.106***0.122***0.118***0.412
(17.51)(12.59)(13.10)(12.57)
Dummy variable for oil-producing2.384***3.031***4.456***4.044***0.144
countries (OIL)(3.96)(4.76)(6.79)(5.91)
Dummy variable for non-oil mining1.643***2.184***1.501***1.722***0.088
countries (MINE)(5.38)(5.31)(3.58)(4.01)
Economic policies
Inflation (INF)−0.096***−0.082***−0.084***−0.284
(7.24)(5.69)(5.87)
Percentage change in real effective0.0100.0150.0150.025
exchange rate (RERG)(0.92)(1.42)(1.35)
Structural reforms (STRUC)1.292***1.443***1.132***0.066
(3.36)(3.84)(2.83)
Provision of public services, proxied by0.149***0 080**0.005
human capital index (HCI)(4.44)(2.08)
Corruption (CORRUPTP)1.686***1.242***0.143
(5.24)(3.28)
External environment
External grants-GDP ratio (GRANTY)−0.092**−0.080**−0.030**−0.002
(2.51)(2.44)(2.54)
change in external debt-GDP ratio0.0200.0210.0190.027
(CHDETY)(1.07)(1.09)(0.99)
Percentage change in terms of trade (TTG)0.0010.0070.0050.009
(0.14)(0.76)(0.58)
Dummy variable for CFA franc countries−3.706***−5.149***−4.815***−4.659***−0.263
CFA(16.13)(10.69)(9.16)(8.78)
Mean square error0.9470.9470.9750.964
F-value4486.84***212.79***228.96***216.33***
Number of observations415415415415

An instrumental variables generalized least squares (IV-GLS) procedure is used for estimation. The numbers in parentheses below the estimated coefficients are the absolute values of the t-ratios ***, **, and * denote statistical significance at the 0.1, 0.05, and 0.10 level’s- respectively.

See the Appendix for definitions and sources of variables used.

Beta coefficients using the estimated coefficients reported in regression (4). The beta coefficient of an explanatory variable X, for example, is the ratio of the product of its estimated coefficient and its standard error to the standard error of the depentiens variable.

Test statistic for the test of the null hypothesis that the joint effect of all variables included on the right-hand side of the estimated equations is zero.

An instrumental variables generalized least squares (IV-GLS) procedure is used for estimation. The numbers in parentheses below the estimated coefficients are the absolute values of the t-ratios ***, **, and * denote statistical significance at the 0.1, 0.05, and 0.10 level’s- respectively.

See the Appendix for definitions and sources of variables used.

Beta coefficients using the estimated coefficients reported in regression (4). The beta coefficient of an explanatory variable X, for example, is the ratio of the product of its estimated coefficient and its standard error to the standard error of the depentiens variable.

Test statistic for the test of the null hypothesis that the joint effect of all variables included on the right-hand side of the estimated equations is zero.

IV. Conclusions and Policy Implications

This paper analyzes tax revenue performance in sub-Saharan Africa, using panel data for 39 countries in the region during 1985–96. A relatively large set of factors that can potentially influence tax revenue performance—income, the structure of the economy, macroeconomic and structural policies, the extent of provision of public goods by the government, the level of corruption, and the external environment—is considered in the econometrics analysis. The effect of corruption, which is typically defined as the abuse of public power for private benefit, is captured by an index that measures the extent to which bribes are generally expected by government officials in relation to, inter alia, tax assessments, trade licenses, and exchange controls.

The analysis confirms the important role played by income and the elements of the tax base in influencing the tax revenue–GDP ratio. The latter rises with income and the level of openness of the economy, as well as with reductions in the share of agriculture in GDP. The results also indicate that a number of other factors typically not considered in the empirical literature significantly influence the tax ratio. In particular, the economic policy environment and the level of corruption matter for the tax revenue–GDP ratio: the latter declines with rising inflation—a proxy for expansionary financial policies—and corruption. Also, there is evidence that countries that have implemented structural reforms on a sustained basis have raised their tax revenue higher than countries that have not. Furthermore, an increase in the level of human capital—a proxy for the extent of public service provided by the government—is associated with an increase in tax revenue. In addition, increases in the level of external grants are associated with lower tax ratios.

An analysis of beta coefficients indicates that, among the economic policy–related variables, inflation has the largest impact on the tax revenue–GDP ratio, followed by the implementation of structural policies. Thus, for a given tax regime and rate, economic policies that promote a noninflationary environment (through a prudent financial stance) and the implementation of structural reforms can be expected to raise tax revenue. Also, the evidence strongly suggests that measures taken to reduce corruption would be expected to enhance tax revenue significantly. Among the variables capturing the effects of income and the tax base, the degree of openness of the economy exerts the largest impact on the tax ratio, followed by income and the agricultural share.

Although measures taken to promote economic reforms and reduce corruption would be expected to enhance tax revenue, a number of caveats are in order. The fight against corruption takes time, needs to be undertaken on several fronts, and can be costly (Tanzi, Chapter 2 in this volume). In addition, the implementation of policies to lower inflation and promote structural reforms may encounter resistance both from the government (which stands to lose seigniorage revenue) and special interest groups (which stand to lose certain privileged positions, such as monopoly power). Thus, projections of large tax revenue gains over a short time period through economic policy reforms and measures to reduce corruption (including through the reform of tax administration) may not be realistic. Finally, in view of the generally low levels of development of sub-Saharan African economies, as well as of the agricultural and informal character of these economies, caution must be exercised in projecting revenue improvements that can reasonably be expected in many of these countries (Heller, 1997).

Appendix
Definitions and Sources of Variables24
Tax revenue
TRYTotal tax revenue–GDP ratio (in percent, in real terms).
Income
PCIPer capita income, calculated as per capita real GDP, converted into U.S. dollars using 1990 nominal exchange rate.
Tax base
AGSShare of agriculture in GDP (in percent). Source: World Bank, World Development Indicators database.
OPENThe ratio to GDP of the sum of exports and imports (in percent).
OILA dummy variable that takes a value of one for oil-producing countries—Cameroon, the Republic of Congo, Gabon, and Nigeria—and zero otherwise.
MINEA dummy variable that takes a value of one for non-oil mining countries—Botswana, Equatorial Guinea, Guinea, Namibia, Niger, Sierra Leone, Togo, Zambia, and Zimbabwe—and zero otherwise.
Macroeconomic and structural policies
INFRate of change of the consumer price index (in percent).
RERGPercentage change in the real effective exchange rate (RER). A positive value for RERG denotes an appreciation of the RER. Owing to data limitations for Comoros and Säo Tomé and Príncipe, the following proxy is used: CPI/(ERI*WPIUS), where CPI is the domestic consumer price index, ERI is an exchange rate index, and WPIUS is the U.S. wholesale price index. Source for RER: IMF, Information Notice System.
STRUCA dummy variable for capturing the effects of structural reforms. It takes a value of one for sustained adjusters and zero otherwise. Two sets of countries are included. First, there are 5 countries with small macroeconomic imbalances during 1985–96—Botswana, Mauritius, Namibia, Seychelles, and Swaziland—that implemented structural reforms without IMF-supported programs. Second, there are 19 countries that successfully implemented Structural Adjustment Facility (SAF) or Enhanced Structural Adjustment Facility (ESAF)-supported programs on a sustained basis. This country group includes countries that have completed three years of SAF/ESAF-supported programs and excludes countries with large undrawn balances at the expiration or cancellation of the programs. The dummy variable takes a value of one starting in the first year of the IMF-supported program to the end of the period. The sustained adjusters and their first program years are as follows: Benin (1989), Burundi (1987), Burkina Faso (1993), Côte d’Ivoire (1994), Ethiopia (1993), The Gambia (1987), Ghana(1988), Guinea (1992), Kenya (1988), Lesotho (1988), Malawi (1988), Mali (1989), Mozambique (1987), Niger (1987), Senegal (1987), Tanzania (1988), Togo (1987), Uganda (1987), and Zimbabwe (1993). Burundi and The Gambia are taken to be sustained adjusters only through 1993, owing to political difficulties during 1994–96. Other countries that had SAF- or ESAF-supported programs during the period 1985–97 but which are not classified as sustained adjusters are Cameroon, the Central African Republic, Comoros, Equatorial Guinea, Madagascar, and Sierra Leone.
Provision of public services by government
HCIIndex of human capital development. Four variables are used to construct this index (secondary school enrollment ratio; literacy rate; life expectancy at birth; and 1,000 minus the infant mortality rate).25 These variables are available only at irregular intervals for most countries and fluctuate substantially over time. For each variable, averages are computed over two subperiods: 1985–90 and 1991–96. These averages are transformed such that their mean values are equal to 100 and their standard deviations are equal to ten. The average of these four transformed variables is used for HCI.26 Source: World Bank, World Development Indicators database, 1997.
Corruption
CORRUPTAn index of corruption, which takes integer values ranging from zero (high corruption) to six (low corruption). Source: International Country Risk Guide, published by Political Risk Services Group in Syracuse, New York.
CORRUPTPData on the level of corruption are available for only 27 of the 39 countries included in the study. In order to avoid the loss of valuable data in the regression analysis, a methodology is used to estimate the data for the countries for which data are missing (see text).
External environment
GRANTYExternal grants-GDP ratio (in percent, in real terms).
CHDETYNet external indebtedness (including arrears accumulation or decumulation), as measured by the change in the external debt-to-GDP ratio (in percent). In order to exclude the impact of revaluation stemming from changes in exchange rates, dollar values are used in the estimation of this variable. Source: World Bank, World Development Indicators database.
TTGPercentage change in the terms of trade.
Dummy variable
CFAA dummy variable that takes a value of one for the CFA franc countries—Benin, Burkina Faso, Cameroon, the Central African Republic, chad, Comoros, the Republic of Congo, Côte d’lvoire, Equatorial Guinea, Gabon, Mali, Niger, Senegal, and Togo—and zero otherwise.
Table A1.Matrix of Correlation Coefficients
ATRYloge(1/PCI)AGSOPENINFRERGGRANTYCHDETYTTGHCICORRUPTCORRUPTP
loge(1/PCI)−0.63***1.00
AGS−0.74***−0.81***1.00
OPEN0.67***−0.36***−0.52***1.00
INF−0.16***0.29***0.16***−0.051.00
RERG0.08−0.12**−0.13**0.04−0.10**1.00
GRANTY−0.26***0.36***0.22***−0.060.19***−0.081.00
CHDETY−0.10**0.14**0.080.030.21***−0.16***0.37***1.00
TTG0.15**−0.14**−0.15**0.08−0.040.10**−0.030.001.00
HCI0.64***−0.72***−0.66***0.39***−0.14***0.08−0.21***−0.030.14**1.00
CORRUPT0.27***−0.11−0.24***0.16**−0.110.050.06−0.12**0.020.37**1.00
CORRUPTP0.43***−0.28***−0.44"0.20***−0.050.05−0.07−0.19***−0.010.48**0.74***1.00
CFA−0.27***−0.10**0.10**−0.17***−0.37***0.03−0.01−0.06−0.03−0.26***−0.25***−0.39***
STRUC0.27***−0.05−0.090.30***−0.090.00−0.07−0.21***0.030.18**0.20***0.37***
OIL0.09−0.31***−0.23***−0.01−0.070.01−0.25***0.010.000.08−0.33***−0.25***
MINE0.22***−0.19***−0.23***0.14**0.10**0.04−0.20***−0.10**0.030.020.180.08
1See Appendix text for definitions and sources of variables. ****, **, and * denote statistical significance at the 0.01, 0.05, and 0.10 level’s, respectively.
1See Appendix text for definitions and sources of variables. ****, **, and * denote statistical significance at the 0.01, 0.05, and 0.10 level’s, respectively.
Table A2.Period Average of Data by Variable and Country, 1986–961
CountryTRYPCIAGSOPENINFRERGGRANTYCHDETYTTGHCICORRUPTCORRUPTPSUS2CFAOILMINE
Benin10.240835.845.16.4−1.34.64.9−1.8952.60.7100
Botswana37.42,8445.489.511.30.31.31.16.01103.73.81.0001
Burkina Faso9.930034.629.13.4−4.05.72.93.3913.63.80.4100
Burundi14.220054.726.09.2−4.55.25.80.4943.30.6000
Cameroon14.095928.327.05.4−1.30.27.5−5.71052.52.00.0110
Central African Republic8.547345.726.43.1−2.96.14.2−3.1972.50.0100
Chad7.521440.939.54.6−5.012.46.35.0932.60.0100
Comoros11.452538.828.35.22.014.02.9−3.11012.80.0100
Congo, Republic of23.91,24311.872.06.7−0.80.311.95.31043.03.10.0110
Côte d’Ivoire17.589931.455.36.50.50.38.8−3.71013.53.00.3101
Equatorial Guinea13.043056.187.13.6−5.01.912.6−0.6952.10.0101
Ethiopia10.316755.717.77.5−8.12.93.6−3.8932.62.90.4000
Gabon23.04.9908.959.54.7−3.00.36.8−l.l1011.61.90.0110
Gambia. The19.430829.387.414.0−2.55.46.2−4.6923.13.00.6000
Ghana13.242148.140.231.3−8.52.95.7−3.110633.20.8000
Guinea11.648923.847.917.00.33.56.62.8893.62.80.5001
Guinea-Bissau6.222545.134.455.2−7.517.719.5−4.9902.02.10.0000
Kenya21.532430.340.816.3−2.01.53.53.51073.03.80.8000
Lesotho37.035815.8133.013.0−0.66.77.51.51063.70.8000
Madagascar8.825334.527.219.8−3.82.85.8−2.01034.03.40.0000
Malawi17.721343.041.727.9−1.10.37.8−2.9933.63.20.8000
Mali10.528846.042.33.0−4.61.77.4−1.6921.92.10.7100
Mauritius19.72,30211.493.37.2−1.60.44.14.21193.91.0000
Mozambique17.210139.360.157.4−9.416.619.71.0924.03.60.9000
Namibia30.01,81612.393.212.1−1.03.40.23.11104.44.31.0001
Niger7.431236.829.93.2−6.04.81.8−1.6903.52.90.9101
Nigeria13.136933.446.434.5−5.504.1−3.81012.02.20.0010
Rwanda8.733738.019.913.4−2.84.13.2−1.1933.40.0000
São Tomé and
Príncipe11.150927.157.038.9−3.615.246.1−6.01134.00.0000
Senegal13.874721.038.13.8−3.71.82.9−2.1982.93.00.9100
Seychelles33.85,1784.350.91.6−1.61.52.63.41266.01.0000
Sierra Leone8.619644.630.865.7−5.606.63.5841.62.00.0001
South Africa24.72,7585.240.612.8−0.600.1−0.81165.24.70.0001
Swaziland28.41,08012.9140.411.4−0.90.80.50.61093.91.0000
Tanzania12.815656.844.329.8−9.93.16.1−3.610003.62.90.8000
Togo15.842336.456.360−2.11.84.4−0.71002.02.70.9101
Uganda6.720753.521.762.0−3.22.94.8−6.3962.83.20.9000
Zambia17.044219.661.088.51.84.213.80.91002.73.00.0001
Zimbabwe31.867114.259.220.4−4.41.73.52.21113.53.80.4001

See Appendix text for definitions and sources of variables.

Proportion of time during the 11 -year sample period characterized by sustained adjustment.

See Appendix text for definitions and sources of variables.

Proportion of time during the 11 -year sample period characterized by sustained adjustment.

References

    AbedGeorge and others1998Fiscal Reforms in Low-Income Countries: Experience Under IMF-Supported ProgramsIMF Occasional Paper No. 160 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    AguirreCarlos A.Peter S.Griffith andM. ZühtuYücelik1981“Tax Policy and Administration in Sub-Saharan Africa,”Part I, inTaxation in Sub-Saharan Africa IMF Occasional Paper No. 8 by the Fiscal Affairs Department (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    BardhanPranab1997“Corruption and Development: A Review of Issues,”Journal of Economic LiteratureVol. 35 (September) pp. 132046.

    • Search Google Scholar
    • Export Citation

    Farhadian-LorieZiba andMenachemKatz1989“Fiscal Dimensions of Trade Policy,” in Fiscal Policy Stabilization and Growth in Developing Countriesed. by Mario I.Blejer andKe-youngChu (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    GhuraDhaneshwar andMichael T.Hadjimichael1996“Growth in Sub-Saharan Africa,”Staff PapersInternational Monetary FundVol. 43 (September) pp. 60534.

    • Search Google Scholar
    • Export Citation

    HamadaKoichi1994“Broadening the Tax Base: The Economics Behind It,”Asian Development ReviewVol. 12 (No. 2) pp. 5184.

    HellerPeter S.1975“A Model of Public Fiscal Behavior in Developing Countries: Aid, Investment, and Taxation,”American Economic ReviewVol. 65 (June) pp. 42945.

    • Search Google Scholar
    • Export Citation

    HellerPeter S.1997“Strengthening Revenue Mobilization Efforts in Sub-Saharan Africa,” in Deepening Structural Reform in Africa: Lessons from East Asiaed. by LauraWallace (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    KhanMohsin S.1987“Macroeconomic Adjustment in Developing Countries: A Policy Perspective,”World Bank Research ObserverVol. 2 (January) pp. 2342.

    • Search Google Scholar
    • Export Citation

    KlitgaardRobert1998“International Cooperation Against Corruption,”Finance & Development (March) pp. 36.

    LeutholdJane H.1991“Tax Shares in Developing Economies: A Panel Study,”Journal of Development EconomicsVol. 35 (January) pp. 17385.

    • Search Google Scholar
    • Export Citation

    MusgraveRichard1987“Tax Reform in Developing Countries,” in The Theory of Taxation for Developing Countriesed. by DavidNewbery andNicholasStern (New York: Oxford University Press).

    • Search Google Scholar
    • Export Citation

    NashahibiKarim andStefaniaBazzoni1994“Exchange Rate Strategies and Fiscal Performance in Sub-Saharan Africa,”Staff PapersInternational Monetary FundVol. 41 (March) pp. 76122.

    • Search Google Scholar
    • Export Citation

    StotskyJanet G. andAsegedechWoldeMariam1997“Tax Effort in Sub-Saharan Africa,”IMF Working Paper 97/107 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    TanziVito1977“Inflation, Lags in Collection, and the Real Value of Tax Revenue,”Staff PapersInternational Monetary FundVol. 24 (March) pp. 15467.

    • Search Google Scholar
    • Export Citation

    TanziVito1981“A Statistical Evaluation of Taxation in Sub-Saharan Africa,”Part II inTaxation in Sub-Saharan Africa IMF Occasional Paper No. 8 by the Fiscal Affairs Department (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    TanziVito1987“Quantitative Characteristics of the Tax Systems of Developing Countries,” in The Theory of Taxation for Developing Countriesed. by DavidNewbery andNicholasStern (New York: Oxford University Press).

    • Search Google Scholar
    • Export Citation

    TanziVito1989“The Impact of Macroeconomic Policies on the Level of Taxation and the Fiscal Balance in Developing Countries,”Staff PapersInternational Monetary FundVol. 36 (September) pp. 63356.

    • Search Google Scholar
    • Export Citation

    TanziVito1992“Structural Factors and Tax Revenue in Developing Countries: A Decade of Evidence,” in Open Economies: Structural Adjustment and Agricultureed. by IanGoldin andL. AlanWinters (Cambridge, England; New York: Cambridge University Press).

    • Search Google Scholar
    • Export Citation

    TanziVito andHowellH. Zee1997“Fiscal Policy and Long-Run Growth,”Staff PapersInternational Monetary FundVol. 44 (June) pp. 179209.

    • Search Google Scholar
    • Export Citation
I would like to thank Marcel Fafchamps, Menachem Katz, Carlos Leite, Ian Lienert, Joseph Ntamatungiro, Dominique Simard, and Janet Stotsky for useful comments. Yasuyuki Todo contributed to an earlier version of this paper as a summer intern in the IMF’s African Department in 1997.
1Nashashibi and Bazzoni (1994) provide an analysis of the trends in revenue and expenditure, as well as economic performance, in the region during 1980–91.
2See Tanzi and Zee (1997) for a survey of the literature on the effects of fiscal policies on growth.
3See, for example, Heller (1975), Tanzi (1981, 1987, and 1992), Farhadian-Lorie and Katz (1989), Leuthold (1991), Nashashibi and Bazzoni (1994), and Stotsky and WoldeMariam (1997). Stotsky and WoldeMariam (1997) provide a survey of previous empirical work.
4Bardhan (1997) reviews the literature on corruption and development. Mauro (chapter 9 in this volume) finds evidence of adverse effects of corruption on economic growth. Ghura and Hadjimichael (1996) find evidence to support the positive effects of macroeconomic stability on economic growth in sub-Saharan Africa.
5For the oil producers in the sample, tax revenue includes oil revenue from all sources—that is, from oil production shared between private oil companies and the government and oil company profits or income taxes.
6The variable D includes net bank and nonbank financing, net domestic arrears, and, for simplicity, nontax revenue.
7The concept of tax handles is explained by Musgrave (1987, p. 244). Leuthold (1991, p. 175) summarizes tax handles as “tax bases that lend themselves to taxation.”
9Data on mining shares are incomplete for the set of countries included in this study. To circumvent this problem, two dummy variables are used to represent oil-producing countries (OIL), and non-oil producers whose average share of mining value added in GDP during 1985–96 was greater than or equal to 5 percent of GDP (MINE). The oil producers are Cameroon, the Republic of Congo, Gabon, and Nigeria. The other mining countries are Botswana, Equatorial Guinea, Guinea, Namibia, Niger, Sierra Leone, Togo, Zambia, and Zimbabwe.
10In the empirical literature, the impact of macroeconomic policies on tax revenue has received little attention; the papers by Farhadian-Lorie and Katz (1989) and Nashashibi and Bazzoni (1994) are among the few exceptions.
11See Khan (1987) for a general analysis of structural reforms. The effects of these reforms are captured by the use of a dummy variable for countries classified as sustained adjusters, which are considered to have made relatively good progress in implementing structural reforms (see the Appendix).
12Four social indicators are used to construct this variable—secondary school enrollment ratio, literacy rate, life expectancy at birth, and the infant survival rate. See the Appendix for a description of the procedure used for the aggregation.
13Tanzi (Chapter 2 in this volume) discusses how spending decisions are affected by corruption.
14See chand and Moene (Chapter 4 in this volume) for an analysis of this phenomenon.
15Empirical evidence provided by Mauro (Chapter 9 in this volume) suggests that large economic payoffs can be achieved by reducing corruption.
16Data on the level of corruption are available for only 27 of the 39 countries included in the study. See the next section for the methodology used to estimate the data for the countries for which data are missing.
17A large stock of external public debt can lower the tax base; see Tanzi (Chapter 2 in this volume).
18The best equation fit was obtained when the natural logarithm of (1/PCI) was used in the estimation of equation (8).
19With a one-period lag used for instruments, one observation is lost per country. Thirty-one countries have data for the full period (1986–96); out of the eight remaining countries, four have data for the period 1987–96, two have data for 1988–96, and two have data for 1989–96. A total of 415 observations are available for the regression estimation.
20The instruments used are the contemporaneous, squared, and lagged values of population, population growth, urbanization rate, growth in the terms of trade, agriculture share, the external grants–GDP ratio, the change in external debt—GDP ratio, the external debt-GDP ratio, and growth in the real effective exchange rate. In addition, an index of human capital (HCI), HCI squared, the lagged broad money-GDP ratio, CFA, OIL, and MINE are used.
21This procedure is implemented in two steps. First, an instrumental variables technique is used to estimate the regression equation with pooled data. Second, the residuals from this step are used to calculate the standard deviation for each country; the inverse of the country-specific standard deviations are then used to weigh all the included variables (including predicted ones), and the equation is reestimated with the pooled transformed data.
22See, for example, Tanzi (1981, 1987, and 1992); Leuthold (1991); and Stotsky and WoldeMariam (1997). See also Farhadian-Lorie and Katz (1989) for an analysis with respect to trade taxes.
24Unless otherwise indicated, data are from the IMF, World Economic Outlook database. The data on tax revenue were obtained from the IMF’s African Department database. See Table 1 for a list of countries included in this study. Angola, Cape Verde, the Democratic Republic of Congo (formerly Zaïre), Eritrea, and Liberia are excluded from the study owing to data limitations.
25Infant mortality rate is the number of infants per thousand live births who die before reaching one year of age.
26For Namibia and Säo Tomé and Príncipe, only three out of the four human capital indicators were available. The average of the three transformed indicators is defined as HCI for these two countries. For a few countries, no data were available for one of the two sub-periods. In such a case, the value of HCI for the subperiod for which the data were available is used for the other subperiod.

    Other Resources Citing This Publication