Fiscal Politics
Chapter

Chapter 19. IMF Conditionality and Revenue Performance

Author(s):
Vitor Gaspar, Sanjeev Gupta, and Carlos Mulas-Granados
Published Date:
April 2017
Share
  • ShareShare
Show Summary Details
Author(s)
Ernesto Crivelli and Sanjeev Gupta 

Introduction

Fiscal adjustment has been an important element of IMF programs.2 Experience shows that expenditure reductions have generally been achieved while increases in revenues have fallen short in relation to program targets (IEO, 2003). This is despite the fact that reform of the tax system—including both tax policy and revenue administration measures—has been frequently subjected to conditionality, to support the implementation of needed structural tax measures.

Conditionality typically covers both the design of IMF-supported programs—that is, the macroeconomic and structural policies—and the specific tools used to monitor progress toward the program goals outlined by the country in cooperation with the IMF. The program’s objectives and policies depend on country circumstances, but the overarching goal is to restore or maintain balance of payments viability and macroeconomic stability, while in low-income countries, reducing poverty. Until the early 1980s, IMF conditionality largely focused on macroeconomic policies. Subsequently, the complexity and scope of structural conditions increased, reflecting the IMF’s growing involvement in low-income and transition economies, where severe structural problems hampered economic stability and growth. Over the years, program conditionality has become better tailored to individual country needs, more streamlined, and focused on core areas of IMF expertise (IMF, 2012).

Conditionality can take different forms, including prior actions (PA), quantitative performance criteria (QPC), indicative targets (IT), or structural benchmarks (SB). Prior actions are measures that a country agrees to take before the IMF’s Executive Board approves financing or completes a review. Quantitative performance criteria are specific and measurable conditions that have to be met to complete a review. Indicative targets are used to supplement QPCs for assessing programs. Structural benchmarks are reform measures that are important to achieve program goals and are intended to assess program implementation during a review.

Over the last 20 years, some form of revenue conditionality has been included in the 441 approved IMF programs, thus supporting the implementation of structural tax measures.3 In recent years, the use of revenue conditionality has increased, partly reflecting greater reliance on the IMF’s technical assistance and the desire of countries to implement this technical advice.4 A quick glance at the data suggests that revenue conditionality in IMF-supported programs appears to have been associated with higher revenue collection in low- and middle-income countries. Figure 19.1 displays tax-to-GDP ratios in countries where tax reform was supported by a period of at least two consecutive years of revenue conditionality. In more than 75 percent of such cases, the tax-to-GDP ratio increased as compared to the year prior to the inclusion of the revenue conditionality.

Figure 19.1.Tax Revenue and IMF Revenue Conditionality

(Before and after a period of consecutive conditionality, 1993–2013)

Sources: IMF, Government Finance Statistics and MONA database; and authors’ calculations.

One would expect an IMF program to contribute to improving revenue collection, regardless of revenue conditionality. This is because the government should be strengthening revenue collection as part of the agreed fiscal adjustment in the context of the program, so as to give a positive signal to creditors and investors (Przeworsky and Vreeland, 2000). Even in the absence of a Fund program, higher revenue collection would be needed to help deal with a severe fiscal crisis. A first key question then is whether there is a role for revenue conditionality in strengthening revenue collection.

Table 19.1 shows average annual changes in tax revenue for 1994–2013 for low- and middle-income countries. Revenue collection appears to have grown faster in countries with IMF programs that included revenue conditionality. Tax revenue increased faster in this group of countries as compared to the sample as a whole, and in particular faster than in countries without revenue conditionality either with or without an IMF program. This result is particularly strong for low-income countries in which average annual revenue growth in IMF program countries that included revenue conditionality is more than twice the observed revenue growth for the sample as a whole as well as for countries with no revenue conditionality.

Table 19.1.Tax Revenue Performance and IMF Revenue Conditionality(Average annual changes, in percent of GDP, 1994–2013)
1994–20131994–981999–20032004–082009–13
All Countries0.130.030.290.34−0.36
IMF Program with Revenue Conditionality0.300.160.460.240.20
IMF Program without Revenue Conditionality0.02−0.220.400.21−0.52
No IMF Program0.120.110.150.41−0.45
Middle-Income Countries
IMF Program with Revenue Conditionality0.33−0.110.650.160.38
IMF Program without Revenue Conditionality−0.09−0.160.380.28−1.66
No IMF Program0.100.140.140.41−0.57
Low-Income Countries
IMF Program with Revenue Conditionality0.360.510.180.330.12
IMF Program without Revenue Conditionality0.14−0.360.290.140.31
No IMF Program0.01−0.020.150.050.43
Sources: IMF, Government Finance Statistics and MONA database; and authors’ calculations.
Sources: IMF, Government Finance Statistics and MONA database; and authors’ calculations.

The second key question relates to the design of revenue conditionality in IMF programs, and in particular the extent to which revenue conditionality has focused more on broad-based consumption taxes—such as the VAT—or income taxes, and their overall contribution to revenue. The recent work (Arnold et al., 2011; OECD, 2010) suggests a growth-hierarchy amongst taxes that favors broad-based consumption taxes for not discouraging savings and investment. Income taxes are believed to have the most adverse effects on growth as they interfere directly with economic decisions—in particular, labor force participation. Thus, an analysis of revenue conditionality in IMF programs can help better understand the contribution of IMF programs to economic growth (Dicks-Mireaux et al., 2000; Przeworsky and Vreeland, 2000). The final issue is whether the design of revenue conditionality—focusing on tax policy or tax administration; specific or more general in nature—makes a difference to revenue collection.

There are limited studies that have analyzed the impact of IMF programs on tax revenue collection.5 Most prominently, Bulir and Moon (2003) studied fiscal developments in 112 countries during the 1990s and Cho (2009) in 93 developing countries during 1951–2000 and found that IMF programs had no effect on revenue collections. By contrast, Brun, Chambas and Laporte (2010) concluded that IMF-supported programs had a positive impact on total revenues in sub-Saharan Africa during 1984–2007. There is more extensive literature on the impact of IMF programs on the overall fiscal balance. Using alternative empirical methodologies these studies mostly conclude that participation in IMF programs improves fiscal outcomes (Conway, 1994; Evrensel, 2002; Dreher and Vaubel, 2004; Easterly, 2005; Atoian and Conway, 2006; Nsouli et al., 2006; Mumssen et al., 2013). However, there is no recent econometric assessment of the extent to which, conditional on other revenue-relevant developments, revenue conditionality contained in IMF programs has affected tax revenue collection—including its main components—nor of the underlying design factors of conditionality that may contribute to higher revenues.

This paper analyzes the impact of revenue conditionality in IMF programs on tax revenue collection in 126 low- and middle-income countries over the period 1993—2013. In doing so, it specifically addresses the questions raised above by using a newly assembled and broad (unbalanced) panel dataset on tax revenue—including all main tax components—and takes advantage of a database on IMF programs that includes detailed information on revenue conditionality. The essence of the empirical strategy is to examine the relationship between IMF programs—with and without revenue conditionality—and tax revenue performance as compared to countries with no IMF program, by looking at potentially differential effects on various types of taxes. Robustness tests are performed to account for differential characteristics in the design of revenue conditionality, to better understand potential differences related to the country’s income level, or initial conditions, as well as the strength of institutions. Finally, cyclically adjusted revenues are considered to account for the effect of the economic cycle on revenues.

The paper is organized as follows. The next section describes the data set and presents the empirical specification and estimation strategy. The main results are presented in the third section, with further robustness analysis in the fourth section. A summary of the results and policy implications are presented in the last section.

Data and Methodology

Data

The dataset comprises an unbalanced panel of 126 low- and middle-income countries over the period 1993–2013. Data on tax revenues are drawn from three sources: the IMF’s Government Finance Statistics (GFS), the IMF’s World Economic Outlook (WEO), and the Organisation for Economic Co-operation and Development (OECD)’s Revenue Statistics in Latin America. These data comprise besides total tax revenue, taxes on goods and services6, VAT, taxes on corporate profits (CIT), the personal income tax (PIT), and taxes on international transactions (Trade), all expressed relative to GDP. Full details of the dataset and summary statistics are provided in Annex 19.1. Figure 19.2 illustrates average tax revenue performance for all countries in the sample, showing an average increase in tax revenue collection by about 2 percentage points of GDP, over the sample period. Until 2008, revenue collection in middle-income countries increased by around 3 percentage points of GDP, on average, about 1 percentage point of GDP more than in low-income countries. After 2008, however, low-income countries were able to strengthen revenue collection further, whereas in middle-income countries, the effects of the global financial crisis resulted in lower tax-to-GDP ratios. This translated into almost a convergence between the two groups of countries in observed tax-to-GDP ratios.

Figure 19.2.Average Tax Revenue in Low- and Middle-Income Countries

(Percent of GDP, 1993–2013)

Sources: Country documents; IMF, Government Finance Statistics database; and authors’ calculations.

Data on IMF programs as well as on revenue conditionality included in these programs are taken from the IMF’s Monitoring of Fund Arrangements (MONA) database, as explained in Annex 19.1. Revenue conditionality may be either quantitative (e.g., increasing the VAT rate to 18 percent) or structural (e.g., submitting legislation to parliament for introducing a VAT). At the same time, revenue conditionality can be related to tax policy or tax administration reform. Finally, revenue conditionality can be specific or general. Specific revenue conditionality can be identified with a tax type and is associated with a specific revenue target (e.g., increase the tax-free threshold under the personal income tax by a certain amount).7 General conditionality, in contrast, cannot be linked to a specific tax type and its main objective is usually either to support the initial steps in a wide-ranging tax reform (such as submission to cabinet of a tax reform proposal) or to strengthen aspects of the revenue administration (e.g., adopt a new IT system in the revenue agency).

The incidence of revenue conditionality in IMF programs is represented by binary variables (including for total tax, and for each of the main taxes) that equal one if a country in a given year had an IMF program with met8 revenue conditionality and zero otherwise. In cases in which revenue conditionality cannot be linked to a specific tax (general conditionality), it is assumed that the revenue conditionality applies to all taxes in that specific year. In most cases, the first lag of the revenue conditionality dummy is considered, to account for delayed reaction of tax revenue to the tax measure implied in the conditionality. This is particularly relevant for conditionality added during a program review taking place late in the year.

A large number of developing countries had IMF programs in the past twenty years. Since 1993, 96 of the 126 countries in the sample had such a program for at least 1 year. The number of years a given country had a program varied substantially. Over the entire sample, about 43 percent of the time, countries had IMF programs (Annex Table 19.1.1). IMF programs were more frequent in low-income countries (about 63 percent). Revenue conditionality has been an important component of IMF-supported programs. Since 1993, about 1,600 revenue conditions were included in the 441 newly approved IMF programs (about 20 percent of the total number of conditions). Figure 19.3 shows the number of years in which a Fund program included revenue conditionality. On average, countries had 5 years with revenue conditionality, which means those IMF programs included revenue conditionality in at least 5 occasions over the sample period (there might be more than one revenue condition in a given year, for example, applying to different taxes). In addition, countries had, on average, 3 years of consecutive revenue conditionality over the sample period. Finally, revenue conditionality has taken mostly the form of structural benchmarks (80 percent), with the remainder conditionality taking the form of prior actions.

Figure 19.3.Years with Revenue Conditionality

(By country, 1993–2013)

Sources: IMF, MONA database; and authors’ calculations.

The number of countries that included revenue conditionality in IMF-supported programs has varied over time. It increased during the 1990s—reflecting the structural nature of IMF programs in the former transition economies (Figure 19.4). As a result, more than 40 countries with an IMF program included at least one revenue condition by 2000. Subsequently, revenue conditions fell in the early 2000s with streamlining of conditionality in IMF programs (IMF, 2005). However, there was a resurgence of revenue conditionality after 2008, presumably reflecting challenges in implementing tax reforms and the need to shore up revenues in the aftermath of the global financial crisis. While during the 1990s, middle-income countries made up the bulk of the IMF programs with revenue conditionality (about 60 percent), more recently, low-income countries have increasingly included revenue conditionality (about 50 percent since 2006). Noteworthy, the bulk of the revenue conditionality in IMF programs has focused on taxes on goods and services (56 percent), followed by conditionality on taxes on income (32 percent), and on international transactions (12 percent).

Figure 19.4.Countries with Revenue Conditionality in IMF Programs, 1993–2013

(By year and income level)

Sources: IMF, MONA database; and authors’ calculations.

Finally, a closer look at compliance with revenue conditionality in IMF programs suggests that overall compliance has been close to 76 percent, with the strongest compliance observed for conditionality on taxes on goods and services (about 80 percent).9 A low compliance record (including countries complying with 50 percent or less of the total number of revenue conditions) is mostly explained by noncompliance with conditionality on income taxes. Within this group, only 5 countries (accounting for about 2 percent of the total number of conditions in the sample) had a compliance record below 30 percent. In addition, countries with a low revenue conditionality compliance record were ranked very low in terms of quality of institutions (with a score below 2 in the ICRG ranking).

Empirical Specification and Estimation

The impact of revenue conditionality in IMF programs on tax revenues is explored by estimating equations of the form:

where T denotes tax revenues in country i = 1,. . .,N at time t = 1,. . .,L, expressed relative to GDP in logs,10PC is a dummy variable for IMF programs including met revenue conditionality (equal to 1 if an IMF program with country i includes met revenue conditionality in year t-1 and 0 otherwise), and PNOC is a dummy variable for country i having an IMF program in t-1 without revenue conditionality. X is a vector of controls, and country and time-specific effects are also included. The lagged dependent variable allows for sluggish response in the tax-to-GDP ratio. Eq. (19.1) is estimated separately for total tax revenue (Total Tax), as well as revenues from taxes on goods and services (G&S), the value-added tax (VAT), taxes on income (Income), taxes on corporate profits (CIT), the personal income tax (PIT), and tax on international transactions (Trade).

The control variables in X are drawn from previous studies on the determinants of tax-to-GDP ratios (Ghura, 1998) and tax effort (see, for example, Sen Gupta, 2007; Baunsgaard and Keen, 2010; Pessino and Fenochietto, 2010). In particular, the overall development of the economy, measured by GDP per capita, is expected to show a positive correlation with revenue reflecting a growing demand for public services with rising income per capita, and because of a higher degree of economic and institutional sophistication. A higher share of agriculture in value-added is expected to be negatively associated with revenue because agriculture is harder to tax. The degree of trade openness, measured as the sum of the shares of imports and exports in GDP, can present either sign. Rodrik (1998) argues that more open countries are vulnerable to risks and, given the need for social insurance, therefore tend to have bigger governments. Moreover, since trade taxes are easier to collect, especially in developing countries, a positive relationship between trade openness and revenues can be expected. However, higher trade openness could be the result of trade liberalization through tariff reductions. This would be consistent with a negative relationship between trade openness and revenue. Other control variables include inflation, which may have revenue effects through both unindexed tax systems and the generation of seigniorage; the level of external indebtedness, which reflects the need to generate revenue to service debt; and the quality of institutions as proxied by the Transparency International’s corruption perception index, which takes values from 0 (high corruption) to 10 (low corruption).

Eq. (19.1) is estimated using a system-Generalized Method of Moments (GMM) model,11 allowing for an unbiased estimate of all variables, including the coefficient on the lagged dependent variable. The system-GMM takes Eq. (19.1) in differences and levels as a system, using lagged changes as instruments in the latter, and lagged levels as instrument for changes in the former. This estimator is best suited for situations with “small T, large N” panels as is the case in this paper with T=21 years and N=126 countries. As an alternative to the GMM results, we present main results using the Anderson and Hsiao (1981) instrumental variables approach.

Of major concern in this literature is the treatment of endogeneity of IMF revenue conditionality as IMF loans tend to be extended in response to economic imbalances (Conway, 2003). As such, countries with a low tax-to-GDP ratio—reflecting the underlying macroeconomic and structural weaknesses—may need to request IMF support to strengthen their fiscal position, thereby creating a potential problem of reverse causality. System-GMM models are well-suited to address cases in which independent variables are not strictly exogenous, meaning they are correlated with past and possibly current realizations of the error (Roodman, 2009a). Using second and deeper lags of the potentially endogenous variables (and their differences) makes them predetermined, meaning not correlated with the error term.

Also of concern is the possibility of sample selection bias associated with participation in an IMF program. With regard to tax revenues, countries that have an IMF program may not be directly comparable to those without one because the former must address macroeconomic imbalances that will influence fiscal policy and the ability of the government to collect taxes. This creates a potential selection bias problem. We address this issue by following the literature on the determinants of IMF programs (Moser and Sturm, 2011; Barro and Lee, 2005) that relies on identifying suitable instruments to isolate the effects of IMF programs on tax revenue. We instrument the IMF program variable in the system-GMM equations with economic as well as political variables that have been found to be well correlated with IMF program involvement.12 For our main results, we use three variables: the lag of a five-year moving average of a dummy indicating whether or not a country was under an IMF arrangement, as in Przeworski and Vreeland (2000), which is a core political variable used extensively in the literature, indicating persistence in IMF involvement; the level of international reserves in months of imports, which in low level may indicate a higher likelihood of balance of payments difficulties and vulnerability to speculative attacks, thus making the country more likely to request and receive IMF assistance; and the Swiss Economic Institute’s (KOF) index of political globalization, as a measure of a country’s integration in the world politics, that may facilitate access to multilateral financing.

For robustness, we include additional political variables such as the index of political plurality and the chief executive years in office, from the World Bank’s Database of Political Institutions; and the Freedom House polity indexes on political rights and civil liberties. Finally, Kuncic’s (2014) indicator on legal institutional quality was considered. Full details of the dataset and summary statistics are provided in Annex 19.1.

These instruments, while likely correlated with the IMF program variable, are less likely to be directly correlated with tax revenue.13 While our identification strategy is commonly used in the literature to address potential endogeneity and selection bias in the IMF program variable, such methodology implies a number of exclusionary restrictions for estimation of an IMF participation equation that clearly depends on the appropriate choice of instruments. These instruments may be correlated with other variables (besides tax revenue) that could ultimately be correlated with the dependent variable, other than via their impact on IMF conditions. As a result, the exclusion restriction may be violated, and the estimation strategy potentially biased.

To test the validity of the instruments in system-GMM we present not only the Hansen statistic14 (Roodman, 2009b) but also the difference-in-Hansen test of exogeneity of the instruments (Bond, Hoeffler, and Temple, 2001). In addition, we test for under identification (Kleibergen and Paap (2006) LM test) and weak instruments (Cragg and Donald (1993) and Kleibergen and Paap (2006) Wald tests) following Bazzi and Clemens (2013) for both the levels equation as well as for the difference equation. Finally, due to the presence of heteroskedasticity in the data, robust standard errors are presented. Generally, the diagnostics performed on the estimations below are satisfactory, with a tolerable value for the Hansen test, and with the Arellano and Bond (1991) test for first and second order serial correlation (M1 and M2) suggesting that the former is present but the latter is not, which is consistent with the underlying assumptions. The difference-in-Hansen p-values imply that we cannot reject the hypothesis that the subset of instruments used is indeed exogenous. In addition, the Kleibergen and Paap (2006) reported p-values imply that we can reject the null hypothesis of under identification for the levels and difference equations. Also the diagnostics on weak instruments using the Cragg and Donald (1993) Wald test are generally acceptable, when compared to the reported critical values from Stock and Yogo (2005) for the bias of the Instrumental Variable estimates greater than 10 or 30 percent of the OLS bias. Finally, we have tested for unit roots (due to the coefficient of the lagged dependent variable being close to one in some cases). For this we use Maddala and Wu’s (1999) Fisher-type test15 (suitable for unbalanced panels), using several alternative specifications (with/without time trend, demean) as proposed in Choi (2001). In all cases, the tests’ p-values strongly reject the null hypothesis that the panels contain a unit root.

Main Results

This section analyses the impact of IMF programs with and without revenue conditionality on total tax revenue and its main components. The question here is whether an IMF program with revenue conditionality has a differential positive impact on revenue collection as compared to countries with an IMF program that does not include revenue conditionality or to countries with no IMF program.

Table 19.2 reports the results for total tax revenue (Column 1), as well as four of its components: taxes on goods and services (Column 2), of which VAT (Column 3), taxes on income16 (Column 4), and tax on international trade (Column 5).17 In general, we do find support for the underlying hypothesis that revenue conditionality contained in IMF-supported programs has a positive impact on tax revenue. The effect of revenue conditionality on tax revenue is found to be positive and significant for the total as well as for taxes on goods and services, which includes VAT. In contrast, an IMF program without revenue conditionality has no significant impact on revenue collection. In addition, we formally test whether the coefficients of the two dummies (Program with and without conditionality) are significantly different from each other. The reported p-values for β1 = β2 suggest that both coefficients are statistically different from each other for total tax revenue, taxes on goods and services, and the VAT.

The estimated coefficient on total tax revenue implies that IMF programs with revenue conditionality could raise tax revenue by about 1 percentage point of GDP in a given year, with half of this revenue gain explained by the positive impact on taxes on goods and services. The lagged dependent variable captures the tax revenue gain over time for years with consecutive revenue conditionality. Taking this into account and recalling that countries had on average 3 consecutive years of IMF programs containing revenue conditionality over the sample period, it implies a revenue gain of about percentage points of GDP by the third year of consecutive revenue conditionality, with half of the gain explained by taxes on goods and services.18

The highly significant impact of IMF revenue conditionality on taxes on goods and services—in particular on VAT—could be explained by the large share of revenue conditionality attached to these taxes as discussed in the Data and Methodology section. Besides their large contribution to tax revenue, the superiority of broad-base consumption taxes has been highlighted, not only in terms of efficiency and welfare gains (Keen and Ligthart, 2001) but also in terms of helping strengthen the tax administration, thus improving tax collection in the aggregate. The result on the VAT, in particular, also confirms previous empirical results on the positive relationship between the adoption of a VAT—which has been found to be positively correlated with having an IMF program—and improvements in tax revenue collection (Keen and Lockwood, 2010).

All in all, this result suggests that revenue conditionality has supported the development of growth-enhancing tax instruments (Arnold et al., 2011; Acosta-Ormaechea and Yoo, 2012). A proportional tax—such as the value-added tax—on all consumption, however, can have negative distributional impact. This effect is usually mitigated by exempting a few sensitive food and other items under the VAT, and adopting a turnover threshold that confers a competitive advantage to smaller and presumably less well-off traders who serve relatively poor customers; this is tantamount to a de facto exemption (Jenkins, Jenkins, and Kuo, 2006). Moreover, if revenues from the VAT finance increased social expenditures, then the net distributional outcome can be progressive (Muñoz and Cho, 2004). Empirical evidence for 140 countries shows that IMF programs have a positive effect on raising social spending (on health and education) in low-income countries (Clements, Gupta, and Nozaki, 2013).

The effect of IMF programs (with or without revenue conditionality) on taxes on income and on international trade is not statistically significant. In contrast to taxes on goods and services, the focus of conditionality on taxes on income has been less frequent due to their relatively low contribution to tax revenue.19 The result can also be explained by the proliferation of tax incentives (including excessive allowances on the personal income tax or corporate income tax holidays, etc.) (Zee, Stotsky, and Ley, 2002). As for taxes on international transactions (trade taxes), the result is expected as trade liberalization has been generally supported in IMF programs to replace harmful trade taxes with broad-base consumption taxes (Baunsgaard and Keen, 2010), and as such, no impact—or even a negative impact—on trade tax should be expected.

Attention focuses now on identifying the differential impact on tax revenue from conditionality related to tax policy as opposed to tax administration measures; as well as specific as opposed to more general revenue conditionality in IMF programs, as defined in the Data and Methodology section. Table 19.3 presents the results for conditionality on tax policy and tax administration, whereas Table 19.4 presents the results for specific versus general conditionality.

Table 19.2.IMF Revenue Conditionality on Tax Revenues1,2
(1)(2)(3)(4)(5)
Total TaxG&SVATIncomeTrade
Tax, Lagged0.9573***0.9101***0.9206***0.7568***0.9080***
(0.0406)(0.0512)(0.0415)(0.0807)(0.1031)
IMF Program No Conditionality,
Lagged0.0054−0.0656−0.0184−0.04330.0022
(0.0282)(0.0411)(0.0347)(0.0611)(0.0600)
IMF Program with Conditionality,
Lagged0.0760***0.1089**0.1013**0.0199−0.0920
(0.0349)(0.0498)(0.0509)(0.0921)(0.0835)
Trade Openness0.0009**0.00060.00070.00130.0010
(0.0004)(0.0007)(0.0006)(0.0013)(0.0014)
Inflation−0.00280.06140.4281*0.1368−0.1099
(0.0021)(0.1299)(0.2604)(0.1534)(0.1280)
GDP Per Capita (log)0.00540.01720.00890.0500***0.0019
(0.0120)(0.0132)(0.0100)(0.0199)(0.0118)
Agriculture Share in Value Added0.00150.00060.00090.00010.0034
(0.0013)(0.0019)(0.0016)(0.0028)(0.0040)
External Debt−0.0005−0.0003−0.0001−0.0025*−0.0002
(0.0006)(0.0005)(0.0007)(0.0013)(0.0009)
β1 = β2 (p value)0.0090.0170.0550.5950.446
Ml (p value)0.0140.0000.0000.0010.000
M2 (p value)0.4280.4320.2620.5650.407
Hansen-Overidentification
(p value)0.6460.8250.5130.3890.797
Diff-in-Hansen-test of Exogeneity
(p value)0.3630.6080.3680.3030.510
For Levels Equation
Kleibergen-Paap LM Test
(p value)0.0000.0000.0000.0000.000
Cragg-Donald Wald F Stat333.45186.0982.32137.65103.29
For Difference Equation
Kleibergen-Paap LM Test (p value)0.0780.0990.0860.0560.000
Cragg-Donald Wald F Stat13.258.566.108.255.92
Observations1,8511,7036291,7181,683
Number of Instruments7474737476
Number of Countries12211481114114
Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is total tax revenue, and revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**,*) indicate significance at 1 (5, 10) percent.

One step, robust, with instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is total tax revenue, and revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**,*) indicate significance at 1 (5, 10) percent.

One step, robust, with instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Table 19.3.Tax Policy vs. Tax Administration Revenue Conditionality on Tax Revenues1,2
(1)(2)(3)(4)(5)
Total TaxG&SVATIncomeTrade
Tax, Lagged0.9648***0.8596***0.9253***0.8232***0.8855***
(0.0464)(0.0528)(0.0405)(0.0812)(0.0525)
IMF Program No Conditionality, Lagged−0.0306−0.0024−0.0268−0.02670.0427
(0.0366)(0.0476)(0.0347)(0.0641)(0.0741)
Conditionality on Tax Policy, Lagged0.1414**0.06120.1433*0.06280.0773
(0.0687)(0.0796)(0.0889)(0.1495)(0.1499)
Conditionality on Tax Administration, Lagged−0.04120.1691**0.0534−0.07190.0108
(0.0690)(0.0889)(0.0816)(0.0876)(0.0858)
Tax Policy vs. Tax Administration (p value)0.1480.1550.5510.4690.749
Constrained Coefficient0.0505**0.0602**0.0959***−0.03300.0456
(0.0277)(0.0312)(0.0398)(0.0692)(0.0735)
M1 (p value)0.0130.0000.0000.0000.000
M2 (p value)0.5190.4600.3150.6360.545
Hansen-Overidentification (p value)0.6760.6060.6130.2710.767
Diff-in-Hansen-Test of Exogeneity (p value)0.4340.3680.5180.2540.565
For Levels Equation
Kleibergen-Paap LM Test (p value)0.0000.0000.0000.0000.000
Cragg-Donald Wald F Stat51.8537.3922.1320.0712.22
For Difference Equation
Kleibergen-Paap LM Test (p value)0.0060.0000.0140.0030.000
Cragg-Donald Wald F Stat7.5613.137.104.058.19
Observations1,8501,7036291,7181,702
Number of Instruments7474737273
Number of Countries12211481114115
Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is total tax revenue, and revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is total tax revenue, and revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Table 19.4.IMF Specific vs. General Revenue Conditionality on Tax Revenues1,2
(1)(2)(3)(4)
G&SVATIncomeTrade
Tax, Lagged0.9116***0.9198***0.8155***0.8854***
(0.0536)(0.0421)(0.0565)(0.0609)
IMF Program No Conditionality, Lagged−0.0606−0.0173−0.01410.0823
(0.0401)(0.0379)(0.0423)(0.0608)
Specific Conditionality, Lagged0.2517***0.0928*0.3038***−0.3503
(0.1041)(0.0525)(0.1162)(0.4317)
General Conditionality, Lagged0.04670.1020−0.0383−0.0593
(0.0680)(0.1652)(0.0648)(0.0853)
Specific vs. General Conditionality (p value)0.1540.9560.01470.508
Constrained Coefficient0.1185***0.0933*−0.0704
(0.0458)(0.0514)(0.0837)
M1 (p value)0.0000.0000.0000.000
M2 (p value)0.4840.2460.7850.398
Hansen-Overidentification (p value)0.7300.5280.4470.892
Diff-in-Hansen-Test of Exogeneity (p value)0.5390.3890.5910.711
For Levels Equation
Kleibergen-Paap LM Test (p value)0.0000.0000.0000.000
Cragg-Donald Wald F Stat65.8767.6131.069.44
For Difference Equation
Kleibergen-Paap LM Test (p value)0.0170.0100.0310.084
Cragg-Donald Wald F Stat6.846.217.511.10
Observations1,6846201,6991,683
Number of Instruments74737474
Number of Countries11380113114
Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Concerning the impact on tax revenue from conditionality on tax policy and tax administration, while the estimated coefficients in Table 19.3 show mixed results regarding the effectiveness of different types of conditionality on the different taxes, a formal test on the estimated coefficients suggests that the impact of conditionality on tax policy is not statistically different from the impact of conditionality on tax administration (with p-values well above 10 percent). The reported constrained coefficients also imply a similar revenue impact to that estimated in Table 19.2, with taxes on goods and services explaining about half the total revenue impact. This result also suggests that the impact on revenue collection can indeed be substantial when both types of revenue conditionality are present in a given year, which confirms that tax policy and tax administration complement each other in a successful tax reform.

Concerning the impact of specific versus general conditionality, while the estimated coefficients in Table 19.4 appear to favor specific conditionality, again a formal test on the estimated coefficients suggests that the impact of specific conditionality is not statistically different from the impact of general conditionality except for taxes on income, for which specific conditionality appears to have a positively significant and large effect on revenue collection. Again here, the reported constrained coefficients imply similar revenue impact to those estimated in Table 19.2 for taxes on goods and services and the VAT. For taxes on income, the revenue gain differential with respect to the overall sample—and in particular with respect to more general revenue conditionality—is expected and explained by the clearer link that exists between the revenue target and the specific conditionality added to help attain this target.

The results above having given a sense of robustness across different types of conditionality, the focus now is on alternative instruments within the system-GMM, as well as on alternative estimation methods. Table 19.5 presents the results for the impact of IMF programs (with and without conditionality) on tax revenues, when considering an extended set of alternative political economy instruments in system-GMM as indicated in the Data and Methodology section. As explained there, these instruments, while also correlated with the IMF program variable, are less likely to be directly correlated with tax revenue, with reported tests confirming the relevance of the chosen instruments.

Table 19.5.IMF Revenue Conditionality on Tax Revenues with Alternative Instruments1,2
(1)(2)(3)(4)(5)
Total TaxG&SVATIncomeTrade
Tax, Lagged0.6825***0.5085***0.5672***0.7474***0.9159***
(0.1703)(0.1596)(0.1784)(0.2815)(0.0951)
IMF Program No Conditionality, Lagged−0.0226−0.08380.02920.0328−0.0996
(0.0584)(0.0874)(0.0827)(0.1133)(0.1012)
IMF Program with Conditionality, Lagged0.1566**0.1638**0.1835**−0.01150.1209
(0.0742)(0.0796)(0.0812)(0.1448)(0.1349)
β1 = β2 (p value)0.0680.0920.0570.8090.204
M1 (p value)0.0440.0340.0060.0140.000
M2 (p value)0.6350.6570.3840.7600.317
Hansen-Overidentification (p value)0.1740.7980.4540.7240.879
Diff-in-Hansen-Test of Exogeneity (p value)0.5840.7150.8440.8780.409
For Levels Equation
Kleibergen-Paap LM Test (p value)0.0000.0000.0000.0000.000
Cragg-Donald Wald F Stat81.4644.0825.8536.9528.69
For Difference Equation
Kleibergen-Paap LM Test (p value)0.0000.0180.0120.0000.000
Cragg-Donald Wald F Stat25.5618.5211.2516.3514.13
Observations1,8511,7036291,7181,683
Number of Instruments5473685454
Number of Countries12211481114114
Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is total tax revenue, and revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with (collapsed) instruments based on first lag of differences in past IMF program (five-year moving average, lagged), political globalization index, political plurality index, chief executive years in office, polity index, legal institutional quality, corruption perception index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 4.67.

Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is total tax revenue, and revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with (collapsed) instruments based on first lag of differences in past IMF program (five-year moving average, lagged), political globalization index, political plurality index, chief executive years in office, polity index, legal institutional quality, corruption perception index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 4.67.

The results in Table 19.5 are qualitatively identical to those presented in Table 19.2 above with a significantly positive effect of IMF programs with revenue conditionality on total tax revenue, as well as on taxes on goods and services and the VAT. Similarly, the reported p-values suggest that the estimated coefficients on IMF program with and without revenue conditionality are statistically different from each other for these taxes. After extending the set of instruments for the probability of being in an IMF program, however, the impact of revenue conditionality on tax revenue increases, which is reflected in larger estimated coefficients. In addition, Annex 19.2 (available online only)20 presents the same results using the Anderson and Hsiao (1981) instrumental variables approach with an extended set of political economy instruments. The results are qualitatively similar to that in Table 19.2, though with lower estimated coefficients.21

Further Analysis22

A number of robustness analyses are presented in this section.23 A first robustness check consists of trying to identify any differential effect of IMF programs with revenue conditionality on tax revenue based on the level of development of the country, or the strength and quality of the country’s institutions. Table 19.6 presents the results for low-income countries that are eligible for IMF PRGT concessional financing (Column 1), as well as middle-income countries (Column 2) considered separately.24Table 19.6, Columns 3–4 present the results for countries grouped on the basis of the ICRG ranking of corruption, which is taken as a proxy for the strength of a country’s institutions. For the analysis, countries with strong institutions are those with a score equal to or above 3, whereas countries with weak institutions are those with a score below 3.25

The estimated coefficient on IMF program with revenue conditionality for low-income countries (Column 1) is significantly positively related to tax revenue, implying a revenue gain of about 1½ percentage points of GDP– and with a coefficient that is statistically different than the coefficient for IMF program without conditionality. For middle-income countries (Column 2), however, while the impact of revenue conditionality appears to be also significantly positively related to tax revenue, the estimated coefficient is not statistically different from that for the IMF program without conditionality, and the resulting constrained coefficient is not significantly different from zero. This result shows how revenue conditionality in IMF programs can be instrumental in helping low-income countries address implementation challenges and capacity constraints in the adoption of tax reforms.

Table 19.6.By Income Level and Strength of Institutions1,2
(1)(2)(3)(4)
Low-Income
ConcessionalStrongWeak
FinancingMiddle IncomeInstitutionsInstitutions
Tax, Lagged0.8183***0.8994***0.8903***0.8996***
(0.1364)(0.0117)(0.0094)(0.0351)
IMF Program No Conditionality, Lagged−0.0570−0.0992−0.0113−0.0215
(0.0663)(0.0837)(0.0201)(0.0294)
IMF Program with Conditionality, Lagged0.1574**0.1973*0.0514**0.0230
(0.0709)(0.1076)(0.0250)(0.0765)
β1 = β2(p value)0.0420.1060.0450.637
Constrained Coefficient0.0288−0.0117
(0.0267)(0.0207)
Observations1,0341,4701,106974
Number of Instruments40318280
Number of Countries638911185
Source: Authors’ calculations.

Dependent variable is total tax revenue, relative to GDP. Full set of year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with (collapsed) instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Source: Authors’ calculations.

Dependent variable is total tax revenue, relative to GDP. Full set of year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with (collapsed) instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

As for the analysis on the impact of IMF programs with revenue conditionality when considering the strength of a country’s institutions, Table 19.6 (Columns 4—5) shows a clear indication that revenue conditionality in IMF programs will potentially have the largest impact on countries with the strongest institutions (or lowest corruption). The estimated coefficient on IMF program with revenue conditionality is only significantly positively related to tax revenue in countries with strong institutions. Also here, the reported p-values suggest that IMF program with revenue conditionality are statistically different than IMF program without conditionality for countries with strong institutions. This result confirms earlier results on the importance of institutions for fiscal policy implementation in low-income countries (Lledo and Poplawski- Ribeiro, 2013).

An additional robustness check consists in analyzing the differential impact of IMF programs with revenue conditionality once initial conditions are accounted for. For this purpose, the sample is split to include countries above and below the average tax-to-GDP ratio, as well as countries in the 25th percentile with the lowest and highest tax-to-GDP ratio, respectively, which is equivalent to tax revenue approximately lying below 10 percent of GDP and above 20 percent of GDP, respectively. The underlying hypothesis is that countries with a relatively low tax revenue collection may rely more on revenue measures supported by revenue conditionality to close the potential fiscal gap as opposed to countries where the tax effort is already high.

Table 19.7.Initial Conditions: Measured by Tax-to-GDP Ratio1,2
(1)(2)(3)(4)
Lower 25thHigher 25th
Below AveragePercentileAbove AveragePercentile
Tax, Lagged0.9349***0.8001***0.8896***0.8476***
(0.0504)(0.1348)(0.0520)(0.0503)
IMF Program No Conditionality, Lagged−0.0214−0.11970.0336−0.0669
(0.0355)(0.1153)(0.0379)(0.0566)
IMF Program with Conditionality, Lagged0.0709*0.2063**0.0803**0.0966
(0.0394)(0.0938)(0.0350)(0.0606)
β1 = β2 (p value)0.0660.0790.4190.007
Constrained Coefficient0.0584***
(0.0222)
Observations7185171,133603
Number of Instruments74377372
Number of Countries785910270
Source: Authors’ calculations.

Dependent variable is total tax revenue, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with (collapsed) instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Source: Authors’ calculations.

Dependent variable is total tax revenue, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with (collapsed) instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

The results in Table 19.7 show a relatively small difference (not statistically different) in the revenue gain associated with IMF revenue conditionality for countries with tax-to-GDP ratios below and above the average. The difference is, however, much more pronounced for countries with the lowest and highest tax-to-GDP ratios. While the revenue gain associated with revenue conditionality in the first group of countries is close to 2 percentage points of GDP, there appears to be no significant impact on countries that already face the highest tax revenue ratio (above 20 percent of GDP).

A further robustness test consists of identifying changes in tax-to-GDP ratios that are not related to the current state of the economy when analyzing the impact of revenue conditionality in IMF programs, that is, the component of tax revenue that does not respond systematically to output conditions, but is instead the consequence of exogenous political processes or extraordinary non-economic circumstances. This analysis is particularly relevant in the context of IMF programs that are usually negotiated in the context of large macroeconomic imbalances and lower-than-potential economic growth. Following Fatas and Mihov (2003, 2006), cyclically-adjusted tax revenue (and components) are obtained by estimating for each country equations of the form:

where T is tax revenue (and components), expressed relative to GDP, ΔYis real GDP growth, and Z is a set of controls.26 In order to control for possible endogeneity of tax revenue with respect to GDP, the instrumental variables (IV) estimator is applied, where Δ Y(-1) and Δϒ(-2) are used as instruments.27 The residuals of Eq. (19.2) for each country represent the discretionary component of tax revenue and enter Eq. (19.1) as the dependent variable. The results in Table 19.8 are qualitatively similar to those in the Further Analysis section with a highly significant and positive impact of IMF programs with revenue conditionality on total tax revenue as well as for taxes on goods and services and VAT.

Table 19.8.Cyclically Adjusted Tax Revenue1,2
(1)(2)(3)(4)(5)
Total TaxG&SVATIncomeTrade
Tax, Lagged0.9400***0.9033***0.5026**0.7620***0.8265***
(0.0520)(0.0549)(0.2336)(0.0944)(0.0871)
IMF Program No Conditionality, Lagged−0.0059−0.01760.1531−0.01820.1494*
(0.0343)(0.0311)(0.1240)(0.0710)(0.0850)
IMF Program with Conditionality, Lagged0.1047*0.0689*0.1980**−0.0461−0.0999
(0.0599)(0.0415)(0.0972)(0.1026)(0.0782)
β1 = β2(p value)0.0840.0540.0670.8110.053
Observations1,7121,4925161,5071,491
Number of Instruments5587535555
Number of Countries12211375113114
Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is the cyclically adjusted measure of total tax revenue, and revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1(5, 10) percent.

One step, robust, with (collapsed) instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is the cyclically adjusted measure of total tax revenue, and revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of controls and year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1(5, 10) percent.

One step, robust, with (collapsed) instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

A final test considers the number of revenue conditions in IMF programs as an alternative benchmark to determine the impact of conditionality on tax revenue collection. While it is difficult to objectively measure the relative stringency of conditionality in IMF programs, using the number of conditions is a measure that has been widely accepted in the literature (see, for example, Dreher and Jensen, 2007; Copelovitch, 2010). In the analysis presented below, the IMF program with revenue conditionality variable is measured to account for the share of revenue conditions to the total number of conditions for each country/ year during an IMF program.28 The results in Table 19.9 on the impact of IMF programs with revenue conditionality are robust to this alternative indicator on revenue conditionality. We find a significantly positive impact of IMF programs with revenue conditionality on total tax revenue, as well as on taxes on goods and services, and VAT. Similarly, the reported p-values suggest that the coefficients for IMF program with and without revenue conditionality are statistically different from each other. The main difference with the results reported in the Further Analysis section corresponds to the size of the estimated coefficients, suggesting a somehow larger revenue impact associated with revenue conditionality, particularly with respect to taxes on goods and services.

Table 19.9.Share of Revenue Conditionality in Total1,2
(1)(2)(3)(4)(5)
Total TaxG&SVATIncomeTrade
Tax, Lagged0.9239***0.8801***0.8923***0.7450***0.8937***
(0.0398)(0.0596)(0.0383)(0.0828)(0.0622)
IMF Program No Conditionality, Lagged−0.0024−0.08110.0099−0.04670.0680
(0.0317)(0.0544)(0.0301)(0.0988)(0.0678)
IMF Program with Conditionality, Lagged0.1903**0.4116*0.2148**0.0625−0.0240
(0.0926)(0.2272)(0.1166)(0.3431)(0.2112)
β1 = β2 (p value)0.0840.0570.0910.7970.720
Observations1,8511,7046291,7191,703
Number of Instruments7744617474
Number of Countries12211481114115
Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is total tax revenue, and revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.

Dependent variable is total tax revenue, and revenue from taxes on goods and services, VAT, income, and trade, respectively, relative to GDP. Full set of year dummies in all regressions. Robust standard errors in parentheses; *** (**, *) indicate significance at 1 (5, 10) percent.

One step, robust, with instruments based on first lag of differences in past IMF program (five-year moving average, lagged), international reserves (in months of imports, lagged), political globalization index, and second lags of their levels in the differenced equation. Stock-Yogo 30 percent maximum IV relative bias is 5.39.

Concluding Remarks

In recent years, the number of revenue-related structural benchmarks in IMF programs has increased. This form of conditionality is agreed with the authorities and monitored by IMF staff, but is not a precondition for the continuation of the program. The question is whether this form of revenue conditionality has a positive impact on the revenue performance of a country. The evidence to date has been mixed. This paper revisits the issue by using more up-to-date and detailed data for 126 low- and middle-income countries during 1993—2013. The analysis extends beyond total tax revenues by disaggregating data by tax types. Since much of the conditionality tends to be related to a specific tax, the paper analyzes its impact on different taxes.

The results are revealing. Revenue conditionality indeed matters. It could potentially improve revenue performance by about 1 percentage point of GDP in a given year. It matters more in low-income countries than in middle-income ones, particularly those countries where revenue ratios are below the group average. It has the maximum impact on taxes on goods and services as well as the VAT—a tax which is relatively more friendly towards promoting growth. These results hold even after revenues are adjusted for economic cycle. Once conditionality is targeted to a specific tax, it affects revenue performance, particularly of income taxes. Unfortunately, in countries where corruption is high, revenue conditionality makes no difference to revenue performance.

Finally, notwithstanding the robust empirical evidence presented here on the positive impact of IMF conditionality on revenue collection, it might be that the focus on revenue conditionality in IMF programs is too narrow. A closer look at the effect of specific policy instruments or tax measures—possibly introduced or removed because of IMF conditionality—could provide a broader insight in explaining tax revenue performance. This subject is beyond the scope of the paper and needs to be explored in future research.

Annex 19.1. Data

The countries in the sample are the following:

Low-income countries: Bangladesh, Benin, Burkina Faso, Burundi, Cambodia, Central African Republic, Chad, Comoros, Democratic Republic of the Congo, Eritrea, Ethiopia, The Gambia, Ghana, Guinea, Guinea-Bissau, Haiti, Kenya, Kyrgyz Republic, Lao P.D.R., Liberia, Madagascar, Malawi, Mali, Mauritania, Mozambique, Myanmar, Nepal, Niger, Rwanda, Sierra Leone, Solomon Islands, Tajikistan, Tanzania, Togo, Uganda, Zambia, Zimbabwe

Middle-income countries: Albania, Algeria, Angola, Antigua and Barbuda, Argentina, Armenia, Azerbaijan, Belarus, Belize, Bhutan, Bolivia, Bosnia and Herzegovina, Botswana, Brazil, Bulgaria, Cameroon, Cabo Verde, Chile, China, Colombia, Republic of Congo, Costa Rica, Côte d’Ivoire, Djibouti, Dominica, Dominican Republic, Ecuador, Egypt, El Salvador, Fiji, Gabon, Georgia, Grenada, Guatemala, Guyana, Honduras, India, Indonesia, Islamic Republic of Iran, Jamaica, Jordan, Kazakhstan, Kiribati, Lebanon, Lesotho, Libya, Lithuania, former Yugoslav Republic of Macedonia, Malaysia, Maldives, Mauritius, Mexico, Moldova, Mongolia, Morocco, Namibia, Nicaragua, Nigeria, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Russian Federation, Samoa, Senegal, Seychelles, South Africa, Sri Lanka, St. Kitts and Nevis, St. Lucia, St. Vincent and the Grenadines, Sudan, Suriname, Swaziland, Syrian Arab Republic, São Tomé and Príncipe, Thailand, Tonga, Tunisia, Turkey, Ukraine, Uruguay, Uzbekistan, Vanuatu, República Bolivariana de Venezuela, Vietnam, Republic of Yemen.

Data on total tax revenue, taxes on goods and services, VAT, income tax revenue, and trade tax revenue are taken from three different sources: the IMF’s Government Financial Statistics (GFS) database, the IMF’s World Economic Outlook (WEO) database, and the Organisation for Economic Co-operation and Development (OECD) Revenue Statistics in Latin America database, relative to GDP. Data for the construction of the dummy variables on IMF program with and without conditionality are taken from the IMF’s Monitoring of Fund Arrangements (MONA) database. Among the economic descriptors for conditionality in the MONA database, considered in this paper are those related to revenue conditionality, which are: revenue measures and revenue administration. Total revenue conditionality and only met revenue conditionality were considered separately. For IMF program without conditionality, the dummy takes the value one if the country has a program with no revenue conditionality in the year t and zero otherwise. The starting year of a program is defined as the year in which it was approved. The end year is the year in which the program expired. For IMF program with revenue conditionality, the dummy takes the value 1 if the country has a program that contains revenue conditionality for a given tax in year t and zero otherwise, as discussed in the Data and Methodology section. In cases in which revenue conditionality cannot be identified with a specific tax in year t (general conditionality), it is assumed that the revenue conditionality applies for each and all of the taxes in that country.

Share of agriculture in aggregate value added, taken from the World Bank’s World Development Indicators (WDI) database. Trade Openness is calculated as imports plus exports in percent of GDP, taken from the IMF’s International Financial Statistics (IFS) database. Per capita GDP is calculated in constant (2000) U.S. dollars, taken from the WDI database, expressed in logs. Inflation is the annual change in the CPI, taken from the IFS database. International reserves, nominal foreign exchange rate to the US dollar is taken from the IMF’s IFS database. The overall fiscal balance, in percent of GDP, is taken from the WDI database. Foreign debt, relative to GDP, is taken from the WDI database. The ICRG corruption scores, produced by Political Risk Services Group, are assessments by staff and relate to actual and potential corruption in the following forms: excessive patronage, nepotism, job reservations, ‘favor-for-favors’, secret party funding and suspiciously close ties between politics and business. The scores range from 0 to 6, where 0 indicates the highest potential risk of corruption and 6 indicates the lowest potential risk for any country.

Other political economy variables include: past IMF program, measured as the lag of a five-years moving average of the IMF program dummy, taken from the MONA database; the KOF index of political globalization as in Dreher (2006b), measured by the number of embassies and high commissions in a country, the number of international organizations of which the country is a member, the number of UN peace missions the country has participated in, and the number of international treaties that the country has signed since 1945. Two indicators from the World Bank’s Database of Political Institutions as in Beck et al., 2001: the index of political plurality, in which legislators are elected using a winner-take-all/first past the post rule, taking the value 1if this system is used or 0 otherwise; and an indicator for chief executive years in office, measured in number of years. Two indicators from Freedom House: one on political rights as a measure of free participation in the political process, including the right to vote freely for distinct alternatives in legitimate elections, compete for public office, join political parties and organizations, and elect representatives who have a decisive impact on public policies and are accountable to the electorate, taking the value 1 (most free) to 7 (least free); and the imputed polity index, which in addition to political rights, also measures civil liberties, allowing for the freedom of expression and belief, associational and organizational rights, rule of law, and personal autonomy without interference from the state. The imputed policy index takes the value 0 for least democratic and 10 for most democratic countries. In addition, Kuncic (2014) indicator on legal institutional quality is considered, taking the value 1 (high) to 0 (low). Finally, Transparency International’s Corruption Perception Index has been considered, which measures the level of corruption in 152 countries, transformed to take the value 0 (high corruption) to 100 (low corruption). All these indicators are available online at www.qog.pol.gu.se, from Dahlberg et al. (2015). Annex Table 19.1.1 summarizes the data.

Annex Table 19.1.1.Descriptive Statistics
Standard
ObservationsMeanDeviationMinimumMaximum
Total Tax Revenue (percent of GDP)2,21115.301.620.3460.94
Low-Income Countries64711.851.590.3436.54
Middle-Income Countries1,56417.091.571.1461.39
Tax on G&S (percent of GDP)2,0515.522.190.1058.26
VAT (percent of GDP)7434.432.090.1019.30
Income Tax (percent of GDP)2,0844.012.140.0450.60
Tax on Corporate Profits (percent of GDP)1,7431.912.300.0324.70
Personal Income Tax (percent of GDP)1,6481.362.440.0013.30
Trade Tax Revenue (percent of GDP)2,0582.292.540.0541.50
IMF Program Variable2,6460.440.500.001.00
Low-Income Countries7760.630.480.001.00
Middle-Income Countries1,8700.350.480.001.00
IMF Program without Revenue Conditionality2,6460.240.430.001.00
IMF Program with Revenue Conditionality:
Total Tax2,6460.190.390.001.00
Low-Income Countries7760.280.450.001.00
Middle-Income Countries1,8700.150.360.001.00
IMF Program with Revenue Conditionality: G&S2,6420.170.370.001.00
IMF Program with Revenue Conditionality: VAT2,6460.160.370.001.00
IMF Program with Revenue Conditionality:
Income Tax2,6420.150.360.001.00
IMF Program with Revenue Conditionality:
Trade Tax2,6210.140.350.001.00
IMF Program with Specific Revenue Conditionality2,6430.060.240.001.00
IMF Program with General Revenue Conditionality2,6390.130.340.001.00
Share of Revenue Conditions in Total2,6470.530.130.001.00
Agriculture Value-added (percent of GDP)2,37920.5514.621.3393.98
Trade Openness (percent of GDP)2,43180.0737.310.31254.61
GDP per capita, US$2,0002,4341,312.903.0349.8914,764.78
Inflation (percent)2,5940.426.84−0.10244.11
Foreign Debt (percent of GDP)2,55159.9857.140.00753.62
ICRG Corruption Score1,7922.310.900.005.00
Political Globalization Index2,26355.8820.708.8694.72
Political Plurality Index2,1020.710.450.001.00
Chief Executive Years in Office2,3377.557.851.0046.00
Polity Index2,5205.762.801.0010.00
Legal Institutional Quality2,1430.470.140.070.90
TI Corruption Perception Index2,14129.5913.410.0079.00
Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.
Source: Authors’ calculations.Note: G&S = goods and services; VAT = value-added tax.
References

    AcemogluD.S.NaiduP. Restrepo and J.Robinson2014Democracy does cause growth,NBER Working Papers 20004National Bureau of Economic Research.

    Acosta-OrmaecheaS. and J.Yoo2012Tax Composition and Economic Growth,IMF Working Paper 12/257 (Washington: International Monetary Fund).

    AndersonT. and C.Hsiao1981Estimation of dynamic models with error components,Journal of the American Statistical Association76 pp. 589606.

    AngristJ.O.Jorda and G.Kuersteiner2013Semiparametric Estimates of Monetary Policy Effects: String Theory Revisited,NBER Working Papers 19355National Bureau of Economic Research.

    ArellanoM. and S.Bond1991Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations,The Review of Economic Studies58 pp. 277297.

    ArezkiR.M.Quintyn and F.Toscani2012Structural Reforms, IMF Programs and Capacity Building: An Empirical Investigation,IMF Working Paper 12/232 (Washington: International Monetary Fund).

    ArnoldJ.B.BrysC. HeadyA.JohanssonC.Schwellnus and L.Vartia2011Tax Policy for Economic Recovery and Growth,Economic Journal Vol.111 pp. 5980.

    AtoianR. and P.Conway2006Evaluating the impact of IMF programs: A comparison of matching and instrumental-variable estimators,Review of International Organizations 1 pp. 99124.

    AuriolE. and M.Warlters2005Taxation base in developing countries,Journal of Public Economics pp. 62546.

    BarroR. and J.Lee2005IMF programs: who is chosen and what are the effects?Journal of Monetary Economics52 pp. 124569.

    BaunsgaardT. and M.Keen2010Tax Revenue and (or?) Trade Liberalization,Journal of Public Economics94(910) pp. 56377.

    BazziS. and M.Clemens2013Blunt Instruments: Avoiding Common Pitfalls in Identifying the Causes of Economic Growth,American Economic Journal:Macroeconomics5(2) pp. 15286.

    BeckT.G.ClarkeA.GroP.Keefer and P.Walsh2001New tools in comparative political economy: The database of political institutions,The World Bank Economic Review15(1) pp. 16576.

    BenedekD.E.CrivelliS. Gupta and P.Muthoora2014Foreign aid and revenue: Still a crowding out effect?FinanzArchiv70(1) pp. 6796.

    BirdG.1996Borrowing from the IMF: the policy implications of recent empirical research,World Development24 pp.175360.

    BirdG.2007The IMF: A bird’s eye view of its role and operations,Journal of Economic Surveys21 pp. 683745.

    BlundellR. and S.Bond1998Initial Conditions and Moment Restrictions in Dynamic Panel Data Models,Journal of Econometrics87 pp. 11543.

    BondS.A.Hoeffler and J.Temple2001GMM Estimation of Empirical Growth Models,” CEPR Discussion Paper 3048.

    BrunJ.G.Chambas and B.Laporte2010IMF programs and tax effort: What role for institutions in Africa?Etudes et Documents E2010.33Centre d’Etudes et de Recherches sur le Developpement International.

    BulirA. and S.Moon2003Do IMF-Supported Programs Help Make Fiscal Adjustment More Durable?IMF Working Paper 03/38 (Washington: International Monetary Fund).

    ChoH.2009Do IMF Programs Discipline Budget Deficit? The Effects of IMF Programs on Government Budget Balance, Expenditure, and Revenue,The Korean Journal of International Relations49(6) pp. 733.

    ChoiI.2001Unit root tests for panel data,Journal of International Money and Finance20 pp. 24971.

    ClementsB.S.Gupta and M.Nozaki2013What happens to social spending in IMF-supported programmes?Applied Economics45(28) pp. 402233.

    ConwayP.1994IMF lending programs: participation and impact,Journal of Development Economics45365391.

    ConwayP.2003Endogenous IMF conditionality: Theoretical and empirical implications,mimeoUniversity of North CarolinaChapel Hill.

    ConwayP.2006The International Monetary Fund in a time of crisis: A review of Stanley Fischer’s IMF essays from a time of crisis: The international financial system, stabilization, and development,Journal of Economic Literature44 pp. 11544.

    CopelovitchM.2010Master or Servant? Agency Slack and the Politics of IMF Lending,International Studies Quarterly54(1) pp. 4977.

    CraggJ. and S.Donald1993Testing Identifiability and Specification in Instrumental Variable Models,Econometric Theory9(2) pp. 22040.

    CrivelliE. and S.Gupta2014Resource blessing, revenue curse? Domestic revenue effort in resource-rich countries,European Journal of Political Economy35 pp. 88101.

    DahlbergS.S.HolmbergB. RothsteinF.Hartmann and R.Svensson2015The Quality of Government Basic Dataset: Version Jan15,University of GothenburgThe Quality of Government Institute. Available under: http://www.qog.pol.gu.se.

    Dickes-MireauxL.M.Mecagni and S.Schadler2000Evaluating the effect of IMF lending to low-income countries,Journal of Development Economics61 pp. 495526.

    DreherA.2006aIMF and economic growth: The effects of programs, loans, and compliance with conditionality,World Development34 pp. 76988.

    DreherA.2006bDoes globalization affect growth? Evidence from a new index of globalization,Applied Economics38 (10) pp. 1091110.

    DreherA.2009IMF conditionality: theory and evidence,Public Choice141 pp. 23367.

    DreherA. and N.Jensen2007Independent Actor or Agent? An Empirical Analysis of the Impact of US Interests on IMF Conditions,The Journal of Law and Economics50 pp. 105124.

    DreherA. and R.Vaubel2004Do IMF and IBRD cause moral hazard and political business cycles? Evidence from panel data,Open Economies Review15 pp. 522.

    EasterlyW.2005What did structural adjustment adjust? The association of policies and growth with repeated IMF and World Bank adjustment loans,Journal of Development Economics76 pp. 122.

    EdwardsS. and J.Santaella1993Devaluation controversies in the developing countries: lessons from the Bretton Woods Era,” in: BordoM. and B.Eichengreen (Eds.) A Retrospective on the Bretton Woods System.University of ChicagoChicago pp. 405455.

    EvrenselA.2002Effectiveness of IMF-supported stabilization programs in developing countries,Journal of International Money and Finance21 pp. 565587.

    FatasA. and I.Mihov2003The case for restricting fiscal policy discretion,The Quarterly Journal of Economics11814191447.

    FatasA. and I.Mihov2006The macroeconomic effects of fiscal rules in the US States,Journal of Public Economics90 pp. 101117.

    GhuraD.1998Tax Revenue in Sub-Saharan Africa: Effects of Economic Policies and Corruption,IMF Working Paper 98/135 (Washington: International Monetary Fund).

    GuitianM.1981Fund Conditionality: Evolution of Principles and Practices,IMF Pamphlet Series38 (Washington: International Monetary Fund).

    GujaratiD. and D.Porter2009Basic Econometrics,McGraw-Hill Irwin.

    HeckmanJ.1976The common structure of statistical models of truncation, sample selection and limited dependent variables and a simple estimator for such models,Annals of Economic and Social Measurement5(4) pp. 475492.

    HeckmanJ.1979Sample Selection Bias as a Specification Error,Econometrica47(1) pp. 153161.

    HiranoK.G.Imbens and G.Ridder2003Efficient Estimation of Average Treatment Effects Using the Estimated Propensity Score,Econometrica71(4) pp. 116189.

    ImbensG.2004Nonparametric Estimation of Average Treatment Effects Under Exogeneity: A Review,Review of Economics and Statistics86(1) pp. 629.

    Independent Evaluation Office of the International Monetary Fund2003Fiscal Adjustment in IMF-Supported Programs,Evaluation Report (Washington: International Monetary Fund).

    International Monetary Fund2005Review of the 2002 Conditionality Guidelines,IMF ReportMarch (Washington: International Monetary Fund). Available via the Internet: http://www.imf.org/external/pubs/cat/longres.cfm?sk=18519.0.

    International Monetary Fund20122011 Review of Conditionality,” (Washington: International Monetary Fund) June. Available via the Internet: http://www.imf.org/external/pp/longres.aspx?id=4692.

    International Monetary Fund2013Issues in International Taxation and the Role of the IMF” (Washington: International Monetary Fund). Available via the Internet: http://www.imf.org/external/np/pp/eng/2013/062813.pdf.

    JenkinsG.H.Jenkins and C.Kuo2006Is the VAT Naturally Progressive?” mimeo (Ontario: Queens University).

    JorraM.2012The effect of IMF lending on the probability of sovereign debt crises,Journal of International Money and Finance 31 pp. 70925.

    KeenM. and J.Ligthart2001Coordinating tariff reductions and domestic tax reform,Journal of International Economics56 pp. 40725.

    KeenM. and B.Lockwood2010The value added tax: Its causes and consequences,Journal of Development Economics92(2) pp. 138151.

    KillickT.1995IMF Programs in Developing Countries: Design and Impact,” Routledge London.

    KleibergenF. and R.Paap2006Generalized Reduced Rank Tests Using the Singular Value Decomposition,Journal of Econometrics133(1) pp. 97126.

    KuncicA.2014Institutional quality dataset,Journal of Institutional Economics10(1) pp. 13561.

    LledoV. and M.Poplawski-Ribeiro2013Fiscal Policy Implementation in Sub-Saharan Africa,World Development46 pp. 7991.

    MaddalaG. and S.Wu1999A comparative study of unit root tests with panel data and a new simple test,Oxford Bulletin of Economics and Statistics61 pp. 63152.

    MoserC. and J-E.Sturm2011Explaining IMF lending decisions after the Cold War,The Review of International Organizations6 pp. 30740.

    MumssenC.Y.GunduzC.Ebeke and L.Kaltani2013IMF-Supported Programs in Low Income Countries: Economic Impact over the Short and Longer Term,IMF Working Paper 13/273 (Washington: International Monetary Fund).

    MuñozS. and S.Cho2004Social Impact of a Tax Reform: The Case of Ethiopia,” in Helping Countries Develop: The Role of Fiscal Policy by GuptaS.B.Clements and G.Inchauste (eds.) pp. 35384 (Washington: International Monetary Fund).

    NsouliS.R.Atoian and A.Mourmouras2006Institutions, program implementation, and macroeconomic performance,In: A. Mody and A. Rebucci (Eds.)IMF Supported Programs: Recent Staff ResearchWashington, D.C.: International Monetary Fund.

    OECD2010Tax Policy Reform and Economic Growth,OECD Tax Policy Studies 20 (Paris: Organization for Economic Co-operation and Development).

    PessinoC. and R.Fenochietto2010Determining Countries’ Tax Effort,Hacienda Pública Española/Revista de Economía Pública195 (4) pp. 6587.

    PrzeworskyA. and J.Vreeland2000The effect of IMF programs on economic growth,Journal of Development Economics62 pp. 385421.

    ReichmannT. and R.Stillson1978Experience with programs of balance of payments adjustment: stand-by arrangements in the highest tranches, 1963–72,IMF Staff Papers 25 (Washington: International Monetary Fund).

    RodrikD.1998Why Do More Open Economies have Bigger Governments?Journal of Political Economy106 pp. 9971032.

    RoodmanD.2009aHow to do xtabond2: An introduction to difference and system GMM in Stata,The Stata Journal9(1) pp. 86136.

    RoodmanD.2009bA Note on the Theme of Too Many Instruments,Oxford Bulletin of Economics and Statistics71(1) pp.13558.

    Sen GuptaA.2007Determinants of Tax Revenue Efforts in Developing Countries,IMF Working Paper 07/184 (Washington: International Monetary Fund).

    SteinwandM. and R.Stone2008The Fund: A review of the recent evidence,Review of International Organization3 pp. 12349.

    StockJ. and M.Yogo2005Testing for Weak Instruments in Linear IV Regression,” In: Identification and Inference for Econometric Models: Essays in Honor of Thomas RothenbergAndrewsD. and J.Stock (eds.)New York: Cambridge University Press.

    ThomasA. and U.Ramakrishnan2006The Incidence and Effectiveness of Prior Actions in IMF-supported Programs,IMF Working Paper 06/213 (Washington: International Monetary Fund).

    WeiS. and Z.Zhang2010Do external interventions work? The case of trade reform conditions in IMF supported programs,Journal of Development Economics92 pp. 7181.

    ZeeH.J.Stotsky and E.Ley2002Tax Incentives for Business Investment: A Primer for Policy Makers in Developing Countries,World Development30(9) pp. 14971516.

This chapter is reprinted from International Tax and Public Finance, Vol. 23, Ernesto Crivelli and Sanjeev Gupta, “Does Conditionality in IMF-Supported Programs Promote Revenue Reform?” ©2015, with permission from Springer.

International Monetary Fund, Washington D.C. 20431, United States. Email addresses: ecrivelli@imf.org (E. Crivelli; corresponding author); sgupta@imf.org (S. Gupta). We would like to thank the Editor Ron Davies and two anonymous referees for excellent suggestions. We are grateful to Santiago Acosta-Ormaechea, Celine Allard, Katherine Baer, Sabina Bhatia, Martin Cerisola, Karla Chaman, Francesco Columba, Ruud De Mooij, Jonathan Dunn, Nisreen Farhan, Katherine Ferry, Geoff Gottlieb, Michael Keen, Christina Kolerus, Svitlana Maslova, Masahiro Nozaki, Iva Petrova, Marcos Poplawski-Ribeiro, Saad Quayyum, and Philippe Wingender for many helpful suggestions on an earlier draft of the paper, and to Haoyu Wang for outstanding assistance with consolidating the data. The views expressed herein are those of the authors and should not be attributed to the IMF, its executive board, or its management.

The average targeted fiscal adjustment in 133 IMF programs was 1.7 percent of GDP during the period 1993—2001 (IEO, 2003).

An example of structural tax revenue reforms with a positive revenue impact is the move to replace harmful trade taxes with broad-based consumption taxes (Baunsgaard and Keen, 2010).

Arezki et al. (2012) find that IMF technical assistance and training support structural reforms in the context of IMF programs.

Most of the literature has focused instead on the effects of IMF programs on the balance of payments (Reichmann and Stillson, 1978; Bird, 1996); on inflation (Edwards and Santaella, 1993; Killick, 1995); on public spending (Conway, 1994); social spending (Clements, Gupta, and Nozaki, 2013); on economic growth (Dreher, 2006a; see also Przeworkski and Vreeland (2000) for a review of the earlier literature); on sovereign risk (Jorra, 2012); and on the effect of IMF conditionality on trade openness (Wei and Zhang, 2010). See also Dreher (2009) for a review of conditionality in IMF programs and a discussion on its effectiveness.

Includes VAT, excise taxes, and other consumption-related taxes.

Specific conditionality can also target revenue administration (such as create a large VAT taxpayers unit).

If the revenue conditionality was not met, the dummy variable takes the value zero.

While unmet conditionality is not likely to be equivalent to no conditionality in terms of its effects on tax reform, Annex 19.2 (available online only, at http://link.springer.com/article/10.1007/s10797-015-9379-7#SupplementaryMaterial) formally considers a revenue conditionality variable that equals one for all revenue conditionality regardless of compliance record and another variable measuring compliance with revenue conditionality. The results indicate that only met revenue conditionality has an impact on revenue collection.

Gujarati and Porter (2009) suggest that the log transformation may be of advantage since it may reduce the incidence of heteroskedasticity and skewness of the data. Auriol and Warlters (2005) suggest that the log transformation may help ensure that out-of-sample fitted values of the tax-to-GDP ratio lie in the 0–100% range. The results are, however, qualitatively identical when using the ratio in levels.

The Blundell and Bond (1998) system-GMM estimator is used instead of Arellano and Bond (1991) difference-GMM estimator since the first one has much better finite sample properties in terms of bias and root mean squared error than the latter; the results are not qualitatively different.

Moser and Sturm (2011) provide a very detailed survey of the literature on the determinants of IMF programs, including a description of the main economic and political variables. For further recent reviews, see for instance Steinwand and Stone (2008), Bird (2007), and Conway (2006).

Alternatively, other economic variables were considered as possible instruments without significant differences in the results, such as the level of external debt-to-GDP, inflation, the change in real GDP per capita, the change in the bilateral exchange rate to the US dollar, and the overall fiscal balance.

The Hansen statistic’s p-value should be high enough to reject correlation between the instruments and the errors but not too high because it weakens confidence in the test.

Using, alternatively, Augmented Dickey-Fuller or Phillips-Perron unit-root tests.

Further disaggregation for taxes on corporate profits (CIT) and on personal income (PIT) was performed with no qualitative difference compared to total taxes on income.

Annex 19.2 (online only, at http://link.springer.com/article/10.1007/s10797-015-9379-7#SupplementaryMaterial) presents the results after also controlling for the level of corruption (omitted here because it reduces considerably the number of observations), which are qualitatively identical to that in Table 19.2.

Alternatively, the impact of revenue conditionality on structural revenue performance can be analyzed by using a dummy on revenue conditionality that equals one during and after each IMF program. The results are qualitatively similar to those presented in the text and the coefficients imply similar revenue gains to those computed by the third year after the program started.

Except perhaps for the tax on corporate profits in low-income countries whose share in total revenue can still be significant (IMF, 2013).

Alternatively, for further robustness, selection bias has been addressed using a fixed-effects model including Heckman’s (1976, 1979) proposed two-stage estimation procedure. As a second alternative to GMM and fixed-effects estimators, we tried the inverse probability weight regression-adjustment method (Hirano et al., 2003). Results from these alternative models are not qualitatively different from those using system-GMM and have been dropped here to preserve space but are available from the authors.

As in the Further Analysis section, the diagnostics here are satisfactory, with a tolerable value for the Hansen, Kleibergen-Paap, and Cragg-Donald tests, and with the Arellano—Bond (1991) test for first and second order serial correlation (M1 and M2) suggesting the former is present but the latter is not, which is consistent with the underlying assumptions.

In addition, we have also included a dummy variable for oil exporter countries to capture potential negative influence of natural-resource revenues on domestic tax effort (Benedek et al., 2014). Alternatively, we have also used non-resource tax revenue only as in Crivelli and Gupta, 2014. The results being qualitatively identical to those in Table 19.2 are omitted to preserve space.

Middle-income countries are classified according to the World Bank criterion. Seventy-two low-income countries are now eligible for concessional lending, which the IMF provides via the Poverty Reduction and Growth Trust (PRGT). It currently carries a zero interest rate on its loans. Eligibility for PRGT lending is based on a member country’s annual per capita income and ability to access international financial markets on a sustainable basis. Concessional support credit lines under the PRGT include the Extended Credit Facility (ECF) and the Standby Credit Facility (SCF). Middle-income countries have been supported mainly under Standby Arrangements (SBA), but also under the Extended Fund Facility (EFF), the Flexible Credit Line (FCL), and the Precautionary and Liquidity Line (PLL). Prior to 2001, low-income countries received support under the Extended Structural Adjustment (ESAF) facility and the Poverty Reduction and Growth Facility (PRGF).

This grouping is almost equivalent to considering the 50th percentile of the distribution with less and more corrupt countries, respectively, also on the basis of the ICRG ranking of corruption.

Control variables include the current inflation rate to ensure that the results are not driven by high inflation episodes and a linear time trend.

Overidentifying restriction tests (notably Wooldridge’s 1995 score test) do not reject the validity of the selected instruments.

Alternatively, qualitatively identical results were obtained when using the total number of revenue conditions.

    Other Resources Citing This Publication