IDA grants or loans? : the effect on tax collection

(1)

tax collection

July 15, 2018

Name Debbie Keijser

E-mail address d.keijser@hotmail.com

Student nr. 10525378

Supervisor Ms. N.J. Leefmans

Second reader Dr. D.J.M. Veestraeten

(2)

ii

Statement of Originality

This document is written by Debbie Keijser who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Acknowledgements

I would first like to thank Mr. Veestraeten for helping me find my research topic and for his advice in the first stages of writing my thesis. Furthermore, I would like to thank my thesis supervisor Ms. Leefmans for her advice and feedback.

I would also like to thank Nina Timmer for her support and being there to discuss my ideas. She has made this final year of university a pleasure.

Finally, I must thank my parents and my boyfriend for providing me with their sup-port and continuous encouragement throughout my study and through the process of researching and writing this thesis.

(4)

iv

Abstract

In which form should financial aid be given, in the form of grants or loans? Many researchers have tried to find an answer to this question but found mixed results. This paper contributes to the literature by analysing the effect of the composition of aid provided by the IDA on the tax revenue of the recipient countries. A recent dataset of 45 IDA countries for the period 2002-2014 has been used to conduct a random effects, fixed effects and system-GMM estimation. The RE and FE results indicate that IDA grants have an insignificant negative effect on tax revenue. However, the RE and FE methods do not correct for the potential reversed causality between tax revenue and aid. The system-GMM does correct for this bias and can be assumed to be a more accurate estimation. The system-GMM results show that IDA grants have no significant effect on tax revenue. IDA loans, however, have a significant negative effect on tax revenue if the analysis does not correct for the external debt level. When this significant variable is included, the effect of IDA loans becomes insignificant.

Another analysis showed the effect for low and middle-income countries separately. Low-income IDA countries face an insignificant negative effect of IDA loans and a positive effect of IDA grants on tax revenue. Whereas middle-income IDA countries show an insignificant positive effect of IDA loans and a negative effect of IDA grants. These results should be taken into account when development aid is provided.

(5)

1 Introduction

How to provide low-income countries with funds is a controversial topic. Generally, these

countries lack a well-functioning national capital market and/or access to international capital markets from which the government can borrow (International Finance Corporation, 2017). It is especially difficult to find funds for projects that do not have a short-term economic benefit. For this purpose the International Development Association (IDA) was established in 1960, as part of the World Bank Group. The IDA aims to reduce poverty by providing financial aid to low-income countries for programs that stimulate economic growth, reduce inequalities, and improve people’s living conditions (World Bank, 2017). To accomplish this goal, the IDA provides financial aid in the form of loans and grants. The loans provided by the IDA are referred to as credits to distinguish them from the loans given by the International Bank for Reconstruction and Development (IBRD) (Sanford, 2002). However, to avoid additional clutter this paper will keep referring to IDA credits as loans.

In October 2017, 75 countries had access to IDA resources. These countries had a Gross National Income (GNI) per capita below the established threshold or they lacked creditworthi-ness to borrow on market terms (World Bank, 2017). The loans provided by the IDA are on concessional terms. This means that interest rates are far below the market rates and that each loan has a long grace period in which no repayment needs to be made (World Bank, 2017). On the other hand, the IDA also provides grants which can be seen as donations because they need not be repaid. Grants can be used to finance projects that do not have an economic but rather a social benefit. These projects try to, for example, improve education or health care in the recipient country (Sanford, 2002). Grants can also be provided to countries with unsustainable debt levels (World Bank, 2017). The positive effects of providing grants rather than loans were highlighted by George W. Bush when in 2001 he suggested to convert half of the money the IDA lends annually into grants (Sanford, 2002).

Providing grants can also have negative side-effects. This paper focuses on one possible negative effect of providing grants rather than loans on tax revenue. Ever since the role of the IDA grants became more important, it was feared that governments of recipient countries could be discouraged to improve or would even weaken tax collection efforts when receiving grants rather than loans (Gupta et al., 2003). The reasoning behind the possible negative relation is related to the substitution effect. Financial aid provided by the IDA can be aimed at financing a specific project or as general budget support. In other words it affects directly public expenditure (Lloyd et al., 2009). A consequence of the financial aid in the form of

(7)

grants could be that governments of recipient countries substitute the national budget for the grants to finance a project. This financial budget relief could be translated into a cut in tax collection effort. If this is the case, the provided financial aid will finance tax cuts instead of government spending or investment (Clist, 2016). If this negative consequence of providing grants is confirmed through research, public support for providing aid will decrease. Since loans need to be repaid it is not expected to observe the same substitution effect when governments receive concessional loans. Therefore, the efforts to collect taxes can depend on the form of aid governments receive, grants or loans. There are more possible disadvantages but also more advantages of providing grants rather than loans which are discussed in Chapter 2.

Besides the possible effect of aid on tax revenue, there might also be an effect of the amount of tax revenue a government collects on aid that it will receive. Generally, the largest aid-recipients also have the lowest tax-to-GDP ratio (Clist, 2016). This might cause a bias in the estimation due to endogeneity caused by reversed causality. Different estimation methods will be conducted to solve the reversed causality bias.

Previous research investigating the effect of financial aid on tax revenue has found mixed results. The research done by Gupta et al. (2003), Benedek et al. (2014) and Odedokun (2003) argues that grants have a negative effect on tax collection. However, Clist & Morrissey (2011) show that grants can also have a positive effect on tax collection. Therefore, more research on this topic has to be done. This paper contributes to the existing literature in two ways. First, this paper analyses the most recent data. Since, previous research has found different results it is important to reexamine the effect using more recent data. Second, this paper focuses on countries that have had access to IDA resources. Previous research mostly focused on all countries that have received Official Development Assistance (ODA). The effect of ODA grants might have a different effect on tax revenue than IDA grants.

Considering that in 2017 already 17% of the IDA commitments was provided in the form of grants (World Bank, 2017) and this percentage is still rising, it is important to analyse and monitor the potential negative effects of providing grants rather than loans. To verify which form of aid minimises the potential negative side effects, this thesis conducts an empirical analysis which estimates the effect of the form of IDA aid on tax revenue. To examine this, a panel data regression analysis is done using annual data of 45 countries which have had access to IDA aid between 2002 and 2014. To estimate the effect, a random effects (RE), a fixed effects (FE) and a system generalised method of moments (system-GMM) estimation are conducted. The FE method eliminates the bias caused by country and time fixed-effects. However, as mentioned before there might also be a bias caused by reversed causality. The system-GMM estimation

(8)

8 MSc Thesis Debbie Keijser – Introduction

corrects for this bias and more.

The remainder of this paper is organised as follows. In Chapter 2 the theoretical framework and the main findings of previous research are discussed. In Chapter 3 the methodology is explained and the dataset is discussed. The results of the regressions are shown and analysed in Chapter 4. The paper continues with Chapter 5 which includes the conclusion and the discussion.

(9)

2 Literature review

In this chapter the main findings in the existing literature on the effect of aid on tax revenue will be discussed. The discussion will start with the advantages and disadvantages of providing grants rather than loans in Section 2.1. This will be followed by the question on how aid distribution can influence tax collection in Section 2.2. Furthermore, this section also discusses the existing literature and its main findings.

2.1 Advantages and disadvantages of grants

In the introduction some advantages and disadvantages of grants have already been briefly discussed. However, this section elaborates more on the benefits and costs of providing grants rather than concessional loans.

An important advantage of providing grants is related to positive externalities. Grants could be provided when the total welfare gain is bigger than the economic gain from a project. Projects for which total welfare exceeds economic gain would not be undertaken if the government can only access loans, since it is not an optimal choice for the country (Odedokun, 2003). For instance, if a certain river is very polluted it is beneficial for all the countries through which this river flows to undertake anti-pollution measures. However, if the benefits for each individual country are below the costs they will not undertake these anti-pollution measures.

Second, grants could also be provided for projects with long term and social benefits. These projects cannot be financed by loans since they lack direct economic returns. As mentioned in the introduction, these projects could aim at improving education or health care in a country (Sanford, 2002).

Furthermore, grants do not increase debt levels. When debt levels are high it can be benefi-cial to provide grants rather than loans, to prevent the debt levels from becoming unsustainable (Sanford, 2002). Sanford (2002) also mentions that the donor countries might be able to require that the recipient countries allow them to monitor programs funded by grants more closely than programs funded by loans, because the grants are a donation and the the loans are not. The extra opportunities to monitor the projects funded by grants can be seen as an extra benefit of providing grants. Lastly, the compassion factor is often mentioned when grants are provided. In the case of, for example, a natural disaster it is more appropriate to provide grants rather than loans.

However, the fact that the IDA still provides the majority of its resources in the form of loans, shows that grants might not always be optimal. This paper examines the possible

(10)

10 MSc Thesis Debbie Keijser – Literature review

negative effect of providing grants rather than loans on tax collection. However, there are more disadvantages which will be shortly discussed.

The direct disadvantage of providing grants for the IDA is related to the amount of aid it can provide. That is, giving grants has a negative effect on the disposable resources of the IDA. Grants are not repaid, therefore, the IDA will have to receive more money from its contributors. This will make the IDA less independent and this might affect the amount of aid it can provide to low-income countries (Sanford, 2002).

Another supposed disadvantage is that aid in the form of a grant might be used less ef-ficiently. When governments know that they have to repay their loans they will finance the most profitable project and will also have to build financial discipline (Odedokun, 2003). This economic pressure is not present when governments receive grants.

2.2 The effect of aid on tax collection

The aid that different multilateral development banks provide to countries is usually directly transferred to the government. The provided aid can be aimed at financing a specific project or as general budget support. Therefore, aid has a direct effect on government expenditures which can influence the fiscal policy and borrowing behaviour (Lloyd et al., 2009).

As mentioned in the introduction, due to the substitution effect it is feared that governments will be discouraged to improve or will even weaken tax collection if they receive grants rather than loans. This hypothesis was examined and confirmed by Gupta et al. (2003) who found a negative relation between grants and tax revenues. Their dataset includes cross-country data from 1970 to 2000 for 107 developing countries. Fixed and random effects estimators are used to examine the effect of aid on the tax to GDP ratio, where they distinguish between grants and loans. They find that the effect of aid on tax collection depends on the form of aid; loans have a positive effect on tax collection whereas grants have a negative effect (Gupta et al., 2003). They also state the importance of corruption; high levels of corruption lead to a greater negative effect of grants on tax revenue. No effect of corruption is however observed when loans are provided (Gupta et al., 2003).

The paper of Clist & Morrissey (2011) concerning the effect of aid on tax collection, however, finds that grants can have a positive effect on tax collection. This study uses data from 82 developing countries from 1970 to 2005. They use the framework suggested by Gupta et al. (2003) and a fixed effects estimator with lagged values for aid to investigate the effect of the aid composition on the tax revenue. Clist & Morrissey (2011) did not find evidence of a negative effect of total aid (loans and grants) on tax collection. They did find evidence of a positive

(11)

effect of grants on tax revenue in the period 1985-2005. However, this positive effect was only observed for middle-income countries. In low-income countries grants did not have an adverse nor a positive effect on tax revenue. A possible explanation for this difference is related to the better fiscal systems that were implemented in middle-income countries after the mid-1980s (Clist & Morrissey, 2011). Nevertheless, they recommend to provide low-income countries with grants rather than loans because grants might prevent debt levels from becoming unsustainable. Benedek et al. (2014) also use the framework from Gupta et al. (2003) to reexamine the effect of aid on tax revenue using a more recent dataset. The paper of Benedek et al. (2014) uses the data of 118 low and middle-income countries for the period 1980-2009. Using FE, difference-GMM and system-GMM, they estimate the effect of ODA on tax revenue. They find that financial aid has a negative effect on domestic tax collection, however, the relationship seems to have weakened compared with the results from Gupta et al. (2003). The decreased negative impact of aid could be explained by the increased use of tax revenue benchmarks in low-income countries. Furthermore, the effect of the composition of aid has been examined. Benedek et al. (2014) state that grants are associated with lower tax revenues, whereas this is not the case with loans. They also conclude that the negative effect of aid (including loans and grants) on domestic tax revenues is strongest in low-income countries and in countries with relatively weak institutions (Benedek et al., 2014).

The paper of Odedokun (2003) examines the data of 72 low-income countries from 1970 to 1999. The paper analyses the response of recipient government tax revenue effort to aid using a FE method. Odedokun (2003) finds that grants have a negative effect on tax collection. The paper also states that loans stimulate budgetary discipline, which is not the case with grants. However, Odedokun (2003) also explains situations in which grants should be preferred over loans. For situations with for example positive externalities or long-lasting downturns grants are usually more appropriate than loans.

In sum, previous research does not show a clear result on whether grants should be preferred over loans concerning tax revenues. Gupta et al. (2003), Benedek et al. (2014) and Odedokun (2003) state that grants negatively affect tax collection. However, the analysis of Clist & Morrissey (2011) shows that grants can also have a positive effect on tax collection. These differing results show that more research has to be done on this topic. To contribute to this discussion, this paper focuses on the effect of the composition of IDA aid on domestic tax collection using the data of 45 IDA countries. Previous research mostly focused on Official Development Assistance (ODA) to both low and middle-income countries. However, research by for example Clist & Morrissey (2011) has shown that the effect of aid on tax revenue is

(12)

12 MSc Thesis Debbie Keijser – Literature review

different for low and middle-income countries. Therefore, this thesis contributes to the research on the effect of aid on tax revenue by focusing only on IDA countries.

The dataset consists of the countries that have had access to IDA aid from 2002 to 2014. This time period is chosen because the first IDA grants were provided in 2002 and the data on tax revenues is available until 2014. Therefore, the most recent data is used to complement the previous research on this topic. This thesis adds five more years of observations compared to the most recent study on this topic, namely Benedek et al. (2014). Using RE, FE and system-GMM regression analysis, it will be tested whether the form of aid affects the tax collection effort.

(13)

3 Methodology and data

In this chapter the methodology and the sample dataset are discussed. In Section 3.1 the regression equation that is used in the analysis and the corresponding variables are explained. To estimate the effect of IDA aid on tax revenue, multiple estimation techniques will be used. Section 3.2 explains the fixed effects and system-GMM estimations that will be applied. Besides discussing the fixed effects estimator, Subsection 3.2.1 also touches upon the estimation biases that can arise. In Subsection 3.2.2 system-GMM is discussed. This is followed by Section 3.3 which discuses the data and show a summary of the statistics. This chapter finishes with the limitations of the data which are discussed in Section 3.4.

3.1 Empirical specification

To examine how financial aid may affect tax revenue, an estimating equation is constructed. The basis for the equation is drawn from the framework suggested by Gupta et al. (2003). Their regression model captures the variables that influence tax revenue. In this regression the dependent variable is tax revenue, the independent variable is financial aid (in the form of grants and loans) and the control variables include agricultural and industry value-added, trade openness and GDP per capita. The baseline regression equation used by Gupta et al. (2003) takes the form:

ln(T ax/GDP )it = β0+ β1AGRit+ β2IN Dit+ β3T RADEit+ β4IN Cit (1)

+ β5GRAN T Sit+ β6LOAN Sit+ ξit

All independent variables, except for IN C, are expressed relative to GDP. i and t are the country and time subscripts, respectively. In Equation 1, AGR stands for agricultural value-added and IN D for industry value-value-added. T RADE represents the sum of exports and imports of the recipient country and IN C presents its GDP per capita. GRAN T S and LOAN S stand for development aid in the form of grants and loans, respectively.

Equation 1 controls for the structure of the economy which is measured by agricultural value-added (AGR) and industry value-value-added (IN D). These variables are viewed as key determinants of the tax base (see Gupta et al. (2003), for example). A large agricultural sector reduces the ability to raise taxes since it makes monitoring and taxing the production more difficult. This is especially the case when the agricultural market is dominated by small and/or subsistence farmers. If the economy is dominated by the industrial sector, the taxable capacity will increase

(14)

14 MSc Thesis Debbie Keijser – Methodology and data

due to easy monitoring and the possibility of a direct tax on industrial products (Clist & Morrissey, 2011). Therefore, a larger agricultural value-added is expected to have a negative effect on tax revenue, whereas industrial value-added is expected to have a positive effect. Other control variables include trade openness (T RADE), which is defined as the sum of exports and imports as percentage of GDP, and GDP per capita (IN C). Trade openness is included because trade taxes often form a big part of total tax revenue, especially for low-income countries. Clist & Morrissey (2011) also show that economic development is positively correlated with tax revenues, which is why GDP per capita is included in Equation 1; GDP per capita is a proxy for the level of economic development.

However, previous research has shown that there are other factors that may influence tax revenue. Benedek et al. (2014) suggest to include a variable which captures the corruption level in the recipient country and a variable that accounts for the level of external indebtedness. Clist & Morrissey (2011) expand the framework suggested by Gupta et al. (2003) by including the variables: Income-squared, Exports and Imports. The variable Income-squared is added to capture non-linear effects. Clist & Morrissey (2011) argue that trade-openness should not be measured by exports and imports together due to the fact that they are taxed differently, therefore, they include them as two separate variables. It is expected that imports have a positive effect on tax revenue because they are relatively easy to tax. On the other hand, exports are usually taxed less than domestic consumption to ensure international trade. This leads to an assumed negative effect of exports on tax revenue (Clist, 2016).

Equation 1 has been used for analysis on all countries that have received ODA (see Gupta et al. (2003), for example). Since this paper focuses on the countries which have had access to IDA aid, there might be other factors that should be taken into account. As mentioned in the introduction, the countries that receive IDA aid either have a GNI per capita below

the established threshold1 or they lacked creditworthiness to borrow on market terms (World

Bank, 2017). Equation 1 already controls for GDP per capita. To capture creditworthiness and the availability of capital the external debt stock to GDP ratio is included as an independent variable. This variable indicates the need to generate tax revenue to repay debt (Benedek et al., 2014). Therefore, a higher level of external debt stock is expected to increase the tax-to-GDP ratio. Furthermore, the alterations suggested by Clist & Morrissey (2011) concerning trade openness and GDP per capita (squared) will be included in the analysis. The level of corruption is not included in this research due to incomplete and unreliable data that is provided for the IDA countries.

(15)

Including the alterations mentioned before, the regression which is used in this thesis takes the form:

ln(T AX/GDP )it= β0+ β1AGRit+ β2IN Dit+ β3Xit+ β4IMit+ β5IN Cit (2)

+ β6IN Cit2 + β7DEBTit+ β8GRAN T Sit+ β9GRAN T Sit2

+ β10LOAN Sit+ β11LOAN Sit2 + ξit

X represents the exports of goods and services as percentage of GDP. Similarly, IM captures the imports of goods and services as percentage of GDP. As mentioned before, exports and imports are included as separate variables as they are taxed differently. The structure of the economy is still measured by AGR and IN D because these industries account for a large share of the economy for countries that receive IDA aid (see Table I). The quadratic variables are added to capture non-linear effects. DEBT indicates the external debt stock as percentage of GDP. Since the debt levels are not constant over time, they have to be included as control variables to avoid a bias when conducting the estimation. The aid variables (GRAN T S and LOAN S) will be lagged variables to capture the dynamic effect of aid on tax revenue. Several lags will be tested for and the most accurate one will be shown in the results.

3.2 Estimation

3.2.1 Fixed Effects estimator and estimation biases

Equation 2 is estimated with a regression that includes country and time fixed-effects (FE). This method corrects for unobserved country-specific characteristics that are constant over time which might influence tax-to-GDP ratios. This paper uses the aid lagged 2 years when conducting the RE and the FE estimations, similar to Clist & Morrissey (2011). This alteration is made to capture the dynamic effect of aid since it is expected that aid does not have a immediate effect on tax revenue (MacGillivray & Morrissey, 2001). It is assumed that it takes 2 years for the effect to be noticeable in tax revenues. Therefore, 2 year lagged variables for grants and loans are used. Other lags have been tested for but the results where similar to the results that included the second lag of the aid variables. Since, using the second lag is advised in the theory this lag is used for the empirical research (Clist & Morrissey, 2011). As a robustness test the results of the RE and FE estimations using the third lag of aid variables can be found in Appendix F.

(16)

Although the FE estimation does correct for the unobserved country-specific characteris-tics, it does not correct for the potential endogeneity problem caused by reversed causality. The level of domestic tax revenue might influence the composition and volume of aid that a country receives, since lower tax-to-GDP ratios in poor countries will lead to more financial aid. Therefore, aid might not only influence tax revenue but it is also influenced by the tax revenue itself. Carter (2010) argues that endogeneity may have driven the negative relationship between grants and tax revenue found in previous research.

To deal with endogeneity, the instrumental variable (IV) approach can be used. This ap-proach uses a instrument for the independent variable (aid in this case). The instrument should

correlated with the endogenous variable to make it relevant. Furthermore, the instrument

should be uncorrelated with the dependent variable to ensure exogeneity (Stock & Watson, 2012). Finding such a variable is considered to be difficult. Therefore, another methodology, namely system-GMM, is used. This method is discussed in Subsection 3.2.2.

Another factor that can cause an estimation bias is reported by Clist & Morrissey (2011) who find a system break in 1985. Their analysis shows that in the period 1970-1985 the relation between aid and tax revenue was negative and it was positive between 1985-2005. This might explain the difference between the results of Gupta et al. (2003) and Clist & Morrissey (2011). However, this break is not relevant for this paper because the time period does not include the ‘break year’, since data for the period 2002-2014 are used.

3.2.2 System Generalised Method of Moments

The FE method corrects for endogeneity caused by omitted variables that are time and/or country specific and do not vary over time. However, the problem of reversed causality is not addressed by this method.

The endogeneity problem caused by reversed causality can be solved with difference-GMM and system-GMM. These methods have been developed by Arellano & Bond (1991) and by Blundell & Bond (1998). Difference-GMM takes the first differences of the regressors to remove fixed effects and uses lagged variables of aid as instruments for first-differenced aid. However, this is only valid if there is no serial correlation (Benedek et al., 2014). In addition to this, the system-GMM differences the instruments to make the instruments exogenous to the fixed effects in the original undifferenced equation. The system-GMM estimator thus consists of two equations, namely the original undifferenced equation in levels and the transformed equation in first differences (Leefmans, 2011). System-GMM is considered more efficient than difference-GMM because it allows the use the information on cross-country differences. Furthermore,

(17)

the addition of the second equation in system-GMM allows for more instruments which also increases efficiency (Mileva, 2007). Therefore, the system-GMM estimator is used instead of the difference-GMM estimation.

The GMM estimator also solves the problem of heteroscedasticity. The error term is het-eroskedastic if the variance of the conditional distribution is not constant and depends on other variables (Stock & Watson, 2012). The Breusch-Pagan test for heteroscedasticity has been con-ducted for Equation 2 using an Ordinary Least Squares (OLS) estimation on the data of all IDA countries. The results show the presence of heteroscedasticity in the error term2. This gives an extra reason to use the GMM estimator rather than the FE estimator (Stock & Watson, 2012). Econometric problems can also arise if the dataset has a time dimension that is shorter than the country dimension (Mileva, 2007). This is the case with the panel dataset that is used in this thesis, since the time dimension includes 13 years and the country dimension covers 45 countries. When using a panel with a small time dimension a shock to the fixed effects of a country will not decline over time, which it does when the time dimension is large (Roodman, 2006). Therefore, it is recommended to use a GMM estimation for the analysis in this thesis.

All the above reasons have led to the inclusion of a system-GMM in this thesis. A two-step system-GMM estimation is conducted using the xtabond2 command in STATA. The two-step system-GMM estimator estimates an equation in first-differences and an equation in levels, simultaneously. To conduct a system-GMM estimation, the variables need to be divided into endogenous and exogenous variables. The aid variables (grants and loans) are assumed to be endogenous due to the possible reversed causality. The external debt level of the recipient country will also be assumed endogenous. It is possible that if the tax-to-GDP ratio decreases, the external debt stock has to increase to pay government expenses. The exogenous variables include: Agricultural and Industry value-added, Exports, Imports and GDP per capita. These variables are assumed to not be affected by changes in the tax-to-GDP ratios.

When conducting the system-GMM estimation in STATA the Arellano-Bond test for serial auto-correlation in the first-differences equation is performed. This test has a null-hypothesis of no serial correlation in the error terms. Therefore, a rejection of the null-hypothesis indicates auto-correlation thus invalid instruments. The first-difference equation is expected to be order 1 auto-correlated because both the difference of the standard errors and the lagged difference of the standard errors contain the lagged standard errors (Roodman, 2006). However, second-order serial correlation should not be present to ensure unbiased results. Furthermore, the Hansen test will be done to test for over-identifying restrictions. The Hansen test tests whether

(18)

the instruments are correlated with the error term. If the null-hypothesis is not rejected, it indicates that the instruments are valid (Roodman, 2006).

The rule of thumb for the number of instruments that could be used is to keep it under the number of groups (countries in this case) (Mileva, 2007). Using too many instruments can overfit endogenous variables (Roodman, 2006). This paper uses the collapse option in STATA to keep the number of instruments to a minimum. The second lag of the endogenous variables are used as instruments to keep the number of instruments below the number of countries (Mileva, 2007).

3.3 Data and summary statistics

There have been 63 countries that have received at least once an IDA grant between 2002 and 2014. For 17 countries the data was not complete, therefore they have been dropped out of the analysis. The missing data has not been completed with data from other databases to avoid high measurement errors. Clist (2016) argues that Benedek et al. (2014) use variables from different data sources as interchangeable which has led to high measurement errors.

Furthermore, Liberia has also been excluded from the dataset because it is considered to be an outlier. The Second Liberian Civil War between 1999 and 2003 caused for extreme circumstances in Liberia (Kieh Jr, 2009). These circumstances lead to abnormal debt levels and a substantial increase in development aid.3 Conclusively, the data sample that is used in this paper consists of 45 countries instead of 63. All 45 countries have data for all years which gives a strongly balanced dataset. Of the 45 countries, 20 are low and 25 are middle-income countries. A list of all countries can be found in Appendix A.

Equation 2 is estimated using panel data regression analysis. The dataset contains 45

countries which have received IDA aid in the period 2002-2014. The data on IDA aid are drawn from the World Bank’s Development Indicators dataset (World Bank, 2018). This dataset also provides the data for the following variables: Agricultural and Industry value-added, Exports, Imports, External debt stock, GDP per capita and GDP (used to calculate the ratios to GDP). The IMF’s World Revenue longitudinal dataset has been used to extract the tax revenue of all 45 countries (International Monetary Fund, 2018). This IMF’s dataset includes all governmental tax revenues from 189 countries for the time period 1990-2014. The data provided by the World Bank’s Development Indicators has been downloaded in absolute numbers in current US dollars. Using the corresponding GDP’s in current US dollars the percentages have been calculated. The

3_{Liberia’s debt levels reached a maximum of 873% of GDP, roughly 638%-points higher than the} maximum excluding Liberia of 235.17%

(19)

IMF’s dataset already provided tax revenue as percentage of national GDP.

Table I Summary statistics

Obs. Average Std. Dev. Min Max

Agricultural value-added 585 25.13 12.04 1.89 55.77

Industry value-added 585 22.59 10.86 5.46 77.41

Exports 585 31.62 17.06 4.69 95.79

Imports 585 46.21 18.59 13.65 119.21

GDP per capita 585 1224.75 1405.81 112.85 8796.69

External debt stock 585 54.12 36.96 10.55 235.17

IDA Loans 585 13.17 14.01 0.00 93.26

IDA Grants 585 0.42 0.78 -0.03 5.50

Tax revenue 585 13.78 4.59 3.35 25.39

Note: All variables except GDP per capita are divided by GDP and multiplied by 100 to make interpretation easier. GDP per capita is measured in US dollars.

Benedek et al. (2014), being the paper that uses the most recent dataset, is used to compare Table I with. Benedek et al. (2014) use annual data of 118 low and middle-income countries for the period 1980-2009. Comparing the statistics, IDA countries show a lower average in Industry value-added and a higher Agricultural value-added. This is in line with the expectations, since, many IDA countries focus on agriculture as a source of income. Exports and Imports cannot be compared because Benedek et al. (2014) have included the variables as a combined variable. GDP per capita is much lower, which is expected for IDA countries compared to ODA countries. The average External debt stock is higher in the sample used in this paper than in that of Benedek et al. (2014), because IDA countries are poorer. This could be a sign of excessive debt taking, which could lead to unsustainable debt. Unsustainable debt levels increase the likely-hood of receiving IDA aid. ODA grants exceed ODA loans, while in our case with IDA aid it is the other way around. The dependent variable T ax revenue, as expected, shows a lower average ratio-to-GDP under IDA countries than under ODA countries. This could be due to the lower GDP per capita but it can also be a sign of weaker institutions.

The correlation between these variables can be found in Appendix C. The signs of the correlation coefficients show that Agricultural value-added, Industry value-added, External debt stock, IDA loans and IDA grants are negatively correlated with T ax revenue. Exports, Imports and GDP per capita are positively correlated with tax revenue. The fact that External debt stock, IDA loans and IDA grants are negatively correlated with T ax revenue could be due to the reversed causality mentioned before. If this is the case, the system-GMM estimation will correct for these biases.

(20)

Figure I

Averages of key variables

Note: The figures shows the average development of key variables over time (bold) and include the underlying country data for reference (light gray).

Figure I shows the averages and the individual country data for the variables T ax revenue, External debt stock, IDA loans and IDA grants for the period 2002-2014. The small increase in the average of tax-to-GDP between 2002 and 2014 can be partly explained by the increase in overall income but there might be other factors that have influenced this change. External debt as percentage of GDP shows a downward moving average. As mentioned in the introduction, the IDA loans to GDP ratios have decreased whereas there is a slight increase in the IDA grants to GDP ratios. Both Tax revenue and IDA grants show an upward moving average, however, the correlation found in Appendix C between these two variables is negative. The graphs do confirm the negative correlation found between Tax revenue and IDA loans and External debt stock. Furthermore, all variables have been tested for non-stationarity. None of the graphs show signs of non-stationarity which is confirmed by results of the unit-root tests4_.

(21)

3.4 Limitations of data

The countries that are eligible for IDA aid are low-income countries and countries with low creditworthiness, therefore, the availability and quality of the data provided is usually low. Several countries had to be excluded from the analysis due to a lack of available data. However, even the data that was available might not be very reliable due to for example weak institutions. Another issue concerning the data is caused by the unequal number of IDA grants and IDA loans that have been given. Especially in 2002, when the first IDA grants have been provided, many countries received IDA loans but only a few countries received IDA grants. Therefore, the number of observations unequal to 0 are larger for IDA loans than for IDA grants. This might influence the results of the empirical analysis. Also because it is easier to receive a loan than a grant, the number of observations is higher for IDA loans than for IDA grants. An extreme example is Honduras which only received one IDA grant between 2002 and 2014 and 12 IDA loans in the same period.

As Table I shows, the average of IDA grants-to-GDP is 0.42% and the average of IDA loans-to-GDP is 13.17%. Even though, the low average of IDA grants is also driven by the data-points that are equal to 0, the maximum grant is also much smaller than the maximum loan. This indicates that the volume of IDA grants is relatively low compared to IDA loans. This might have a negative effect on the reliability of the results. The average percentage of IDA grants is also very low when compared to the 4.47% of GDP that ODA grants account for in the research of Benedek et al. (2014).

(22)

22 MSc Thesis Debbie Keijser – Results

4 Results

Section 4.1 illustrates the results from the panel data regressions for all 45 IDA countries using random effects (RE), fixed effects (FE) and system-GMM estimations. Furthermore, in Section 4.2 it is tested whether the effect of IDA aid on tax revenue is different for low and middle-income countries.

4.1 Estimation results

Table II presents the results of the estimation of Equation 2 using random effect, fixed effect

and system-GMM for all IDA counties. In these regressions the dependent variable is the

natural logarithm of tax-to-GDP. Both the dependent, and independent variables are discussed in Chapter 3. Column (1), (2) and (3) show the results for the random effects estimation, where Column (1) and (2) exclude the variable External debt stock and Column (3) includes this variable. Similarly, Column (4), (5) and (6) present the results for the fixed effects estimation. But since these methods do not correct for the possible endogeneity biases, Column (7), (8) and (9) show the results of the system-GMM estimations, which do correct for these biases.

Although this thesis aims at finding the separate effect of IDA grants and IDA loans on the tax revenue of recipient countries, Column (1), (4) and (7) include the variable IDA total rather than the separate IDA grants and IDA loans variables. IDA total captures the total IDA aid that a countries has received, therefore including grants and loans. These regressions have been performed to show the overall effect of total IDA aid on tax revenue. The other Columns show how this overall effect is divided into the effect of grants and loans.

As mentioned in the Subsection 3.2.2, to conduct the system-GMM estimation the variables should be divided into exogenous and endogenous variables. The endogenous variables are: IDA grants, IDA loans and External debt stock. The exogenous variables are: Agricultural value-added, Industry value-value-added, Exports, Imports and GDP per capita. This distinction makes it possible to create an appropriate instrument for the endogenous variables. Furthermore, the collapse option in STATA has been used to keep the number of instruments to a minimum. The system-GMM estimation also includes a full set of time year dummies which corrects for possible trends in the data. The results of the system-GMM estimation can only be analysed if there is no second-order serial correlation and the Hansen test value is insignificant. Table II shows that there is no second-order serial correlation (AR(2) shows an insignificant value) and that the statistic from the Hansen test is not significant, therefore, the instruments are valid.

(23)

that the variable External debt stock has a significant impact on tax revenue. Higher levels of external debt are expected to increase the tax-to-GDP ratio because it will generate the need of collecting tax revenue to repay debt (Benedek et al., 2014). Therefore, the inclusion of External debt stock in some regressions can be seen as a robustness test.

With the RE and FE estimations, two observations are lost due to the fact that the aid variables are two year lagged. This alteration is made to capture the dynamic effect of aid since it is expected that aid does not have an immediate effect on tax revenue (MacGillivray & Morrissey, 2001). With the system-GMM estimation current aid levels are used because the lagged variables of aid already serve as instruments and these will capture the dynamic effect. Similar to the FE and RE estimations, the system-GMM estimations uses the second lag of the aid variables as instruments. The second lag has been chosen for consistency and due to more efficient results5. All regressions include a full set of year dummies to correct for possible trends in the data. This does, however, not solve all problems with non-stationarity. Therefore, the non-stationary variable (GDP per capita) are included as its natural logarithm which are stationary6.

All regressions, except for the FE estimations, have been conducted with the robust option in STATA. The robust standard errors have been used to correct for heteroskedasticity. The FE estimation has clustered standard errors to correct for heteroskedasticity and for arbitrary auto-correlation within a country (Stock & Watson, 2012).

5_{Appendix F presents the results of the estimation using the third lag as a robustness test} 6_{Results of the unit-root test can be found in Appendix D}

(24)

Table II All IDA countries

Random effects Fixed effects SGMM

(1) (2) (3) (4) (5) (6) (7) (8) (9)

IDA total 0.0004 0.0011 -0.0113***

(0.002) (0.002) (0.004)

IDA total, squared 0.0000 0.0000 0.0002**

(0.000) (0.000) (0.000)

IDA grants -0.0190 -0.0219 -0.0161 -0.0163 0.0821 0.0086

(0.031) (0.031) (0.032) (0.031) (0.122) (0.119)

IDA grants, squared -0.0018 -0.0015 -0.0026 -0.0025 -0.0027 0.0150

(0.007) (0.007) (0.007) (0.007) (0.025) (0.027)

IDA loans 0.0015 0.0016 0.0021 0.0022 -0.0091** -0.0007

(0.002) (0.002) (0.002) (0.002) (0.004) (0.006)

IDA loans, squared -0.0000 -0.0000 -0.0000 -0.0000 0.0002*** 0.0002***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

External debt stock -0.0003 -0.0000 -0.0049**

(0.001) (0.001) (0.002)

Agricultural value-added -0.0009 -0.0020 -0.0021 0.0028 0.0014 0.0014 -0.0119*** -0.0120*** -0.0077* (0.004) (0.004) (0.004) (0.005) (0.005) (0.005) (0.004) (0.004) (0.004) Industry value-added -0.0098** -0.0101*** -0.0104*** -0.0062 -0.0069 -0.0069 -0.0159*** -0.0162*** -0.0163***

(0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.003) (0.004) (0.003) GDP per capita, log 0.7491* 0.8064* 0.7700* 0.7800* 0.8484** 0.8454** 0.3954 0.7782 1.0521**

(0.422) (0.417) (0.410) (0.429) (0.420) (0.407) (0.646) (0.632) (0.507) GDP per capita, squared -0.0439 -0.0495 -0.0469 -0.0484 -0.0550* -0.0548* -0.0256 -0.0499 -0.0661**

(0.031) (0.030) (0.030) (0.033) (0.033) (0.031) (0.043) (0.042) (0.033) Exports 0.0024 0.0023 0.0024 0.0018 0.0018 0.0018 0.0012 0.0021 0.0037* (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Imports 0.0037** 0.0037** 0.0038** 0.0035** 0.0034** 0.0034** 0.0067*** 0.0059*** 0.0076*** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Observations 583 583 583 583 583 583 585 585 585 R-squared 0.4677 0.4641 0.4811 0.3208 0.3265 0.3290 Instruments 25 29 31 AR (1) 0.238 0.118 0.074 AR (2) 0.291 0.430 0.247 Hansen test 0.065 0.211 0.171

Note: The dependent variable is tax-to-GDP in % and in natural logarithm. All regressions include a full set of year dummies. The aid variables IDA Grants and IDA Loans are lagged two years in both the random and fixed effect regression models. Robust and clustered standard errors in parentheses. *, ** and *** indicate significance at 10%, 5% and 1%, respectively.

As mentioned before, Columns (1), (4) and (7) include the variable IDA total rather than

IDA grants and IDA loans. The results show that the overall effect of IDA aid on tax

revenue is insignificant for the RE and FE estimations. However, the more appropriate system-GMM estimation shows a significant negative effect of IDA aid on tax revenue. If a country receives more IDA aid the need to collect tax revenue decreases. Benedek et al. (2014) also find a significant negative coefficient for total ODA, which indicates that this problem is not only present for IDA countries. The positive and significant coefficient of IDA total squared, however, shows that the negative effect of IDA aid decreases when the volume of aid increases. The rest of the analysis will focus on the regressions that include the IDA grants and IDA loans as separate variables.

Comparing the results from the RE and the FE estimations shown in Columns (2), (3), (5) and (6), the results are much alike. The Hausman test has been conducted to see whether the RE or the FE estimator is more appropriate for this analysis. The null-hypothesis states that

(25)

the addition of fixed effects does not significantly affect the results, which makes the random effects model more appropriate. Similarly, if the null-hypothesis is rejected, fixed effects should be included in the regression. The results of the Hausman test, presented in Appendix G, indicate that for all specifications the inclusion of fixed effects is insignificant. Therefore, the random effects model is analysed below.

The important independent variables are IDA grants and IDA loans. A difference in coefficient could indicate that one form of financial aid should be preferred over the other, concerning the effect on tax revenue. Table II shows that IDA grants have an insignificant negative coefficient in all RE and FE regressions. Whereas, IDA loans present an insignificant positive coefficient in Columns (2),(3), (5) and (6). A reasoning behind these results, could be that recipient countries use grants and loans for different purposes. If loans are used to improve weak institutions, for example, it can be expected that tax revenues will increase (Benedek et al., 2014). However, since these coefficients are insignificant, no conclusions can be made. The results from the RE and FE estimations are however possibly biased by reversed causality. Therefore, the results from the system-GMM estimation are compared with the RE results in Column (2) and (3). The most notable difference between the RE and the system-GMM results is that the signs of IDA aid variables have changed. IDA grants have a negative coefficient when estimated with RE and positive coefficient when estimated with system-GMM. The reverse is the case with IDA loans. The system-GMM estimations show a significant negative effect of IDA loans on tax-to-GDP ratio when External debt stock is not included in the regression. The positive significant coefficient of IDA loans squared shows that the negative effect is weakened when the amount of IDA loans increases. The system-GMM estimations also show that IDA grants have an insignificant positive effect on tax revenue. However, both IDA aid variables have an insignificant effect on tax revenue when the external debt level of the recipient country is taken into account.

Although not all coefficients are significant, the different coefficients for the IDA aid variables that the RE and the system-GMM estimations produce can explain the mixed results from previous research. The possible endogeneity problem in the FE estimation may have driven the negative relationship between grants and tax revenue found in previous research (Carter, 2010). External debt stock has been added in Column (3), (6) and (9) as a robustness test. A higher level of external debt is expected to increase the tax-to-GDP ratio, therefore, the coefficient is expected to be positive. However, the tax-to-GDP ratio might also influence external debt stock itself which would cause an endogeneity problem in the RE and FE estimators. The system-GMM estimators uses the second lag of this variable as an instrument for External debt stock

(26)

to eliminate the bias. Therefore, the insignificant negative coefficient of External debt stock found in the RE and FE estimations might not be the most accurate estimation. The negative coefficient found with the system-GMM estimation is significant at 5% level. This finding is in line with the results of Gupta et al. (2003), although they use a RE and FE estimation. The negative effect of External debt stock on tax revenue is not in line with the theory, but it can be explained by the fact that higher debt levels might be a sign of bad institutional management which could also lead to less tax collection efforts.

Table II also shows the results of the coefficients of the control variables that are included in the regressions. For the RE estimation, the coefficient of Agricultural value-added is negative, which is in line with expectations. However, the coefficient is insignificant for all RE estima-tions. The system-GMM estimation for this coefficient also shows a significant negative effect of Agricultural value-added on tax revenue. Industry value-added is expected to have a positive effect on tax revenue but this is contradicted by the results. Both the RE and the system-GMM estimation show a significant negative effect of Industry value-added on tax revenue. A larger industrial sector might lead to lower tax revenues if the government wants to stimulate the investments in this sector, which is often the case in developing economies.

The variable GDP per capita reflects the economic development of the recipient country. The coefficient is expected to be positive since higher economic development is associated with higher tax revenue levels. Table II demonstrates that this is also the case for IDA countries. Trade-openness is divided into Exports and Imports, where Imports are expected to have a positive coefficient while exports are expected to have negative coefficient. Results show a positive coefficient for both variables. The variable Exports has an insignificant positive effect in all regressions except for the system-GMM estimation that includes External debt stock. In Column (9) Exports have a significant positive effect on tax revenue, which is possible for countries that focus more on domestic production rather than international trade. These countries could see exports tariffs as an extra source of income instead of a threat to their competitiveness on the world market. The import-to-GDP ratio has a significant positive effect on tax revenue in all regressions, which is in line with the expectations.

4.2 Difference low and middle-income countries

Although the reason why a country receives IDA aid can differ per country, the main two reasons are a low GDP per capita, and/or the inability the borrow; albeit due to lack of creditworthiness to borrow on market terms, or the absence of a well-functioning national capital market (World Bank, 2017). Therefore, it is possible that the individual circumstances have

(27)

an impact on the relation between tax revenue and IDA aid. To investigate this claim, the IDA countries have been divided into low and middle-income countries. It is assumed that low-income countries receive the IDA aid mainly because of their low level of GDP per capita and middle-income countries due to a lack of capital market options. However, for some countries both circumstances may hold. The decision to divide the countries into low and middle-income countries is also driven by the results of Clist & Morrissey (2011) which show that the effect of aid on tax revenue is different for low and middle-income countries.

(28)

Table III

Difference low and middle-income IDA countries

Random effects Fixed effects SGMM

(1) (2) (3) (4) (5) (6)

IDA grants -0.0352 -0.0362 -0.0313 -0.0311 0.0185 -0.0279

(0.030) (0.030) (0.032) (0.032) (0.055) (0.073)

IDA grants, squared -0.0039 -0.0034 -0.0046 -0.0047 -0.0175 0.0068

(0.008) (0.008) (0.008) (0.008) (0.025) (0.023)

IDA loans 0.0023 0.0024 0.0026 0.0026 -0.0093** 0.0005

(0.002) (0.002) (0.002) (0.002) (0.005) (0.006)

IDA loans, squared -0.0000 -0.0000 -0.0000 -0.0000 0.0002*** 0.0002***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Low-income dummy -0.1207 -0.1218 0.0723 0.0727 -0.1553 -0.1080

(0.128) (0.128) (0.072) (0.075) (0.183) (0.145)

Dummy × IDA grants 0.0300 0.0285 0.0289 0.0289 0.1363* 0.0617

(0.029) (0.029) (0.032) (0.033) (0.081) (0.108)

Dummy × IDA loans -0.0009 -0.0009 -0.0006 -0.0005 -0.0004 -0.0031

(0.001) (0.001) (0.001) (0.001) (0.003) (0.003)

External debt stock -0.0003 0.0000 -0.0050**

(0.001) (0.001) (0.002)

Agricultural value-added -0.0016 -0.0016 0.0015 0.0015 -0.0102*** -0.0064

(0.004) (0.004) (0.005) (0.005) (0.004) (0.004)

Industry value-added -0.0102*** -0.0104*** -0.0068 -0.0068 -0.0164*** -0.0170***

(0.004) (0.004) (0.005) (0.004) (0.004) (0.003)

GDP per capita, log 0.7768* 0.7449* 0.8473** 0.8514** 0.5664 0.6046

(0.409) (0.396) (0.434) (0.414) (0.744) (0.539)

GDP per capita, squared -0.0485 -0.0463 -0.0544 -0.0547* -0.0368 -0.0375

(0.030) (0.029) (0.034) (0.032) (0.047) (0.034) Exports 0.0022 0.0023 0.0019 0.0018 0.0013 0.0029 (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Imports 0.0035* 0.0036* 0.0034* 0.0034* 0.0062*** 0.0078*** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Observations 583 583 583 583 585 585 R-squared 0.4607 0.4740 0.8529 0.8526 Instruments 34 36 AR (1) 0.056 0.027 AR (2) 0.385 0.204 Hansen test 0.276 0.207

Note: The dependent variable is tax-to-GDP in annual % and in natural logarithm. All regressions include a full set of year dummies. The aid variables IDA Grants and IDA Loans are lagged two years in both the random and fixed effect regression models. Robust and clustered standard errors in parentheses. *, ** and *** indicate significance at 10%, 5% and 1%, respectively. Dummy always refers to Low-income dummy.

Table III shows the estimated results of Equation 2 using RE, FE and system-GMM for all IDA counties. The difference compared to Table II can be found in the Low-income dummy and the interaction variables that are included in Table III. The effect for middle-income countries can be found in the IDA grants and IDA loans coefficients. The effect of the aid variables on

(29)

tax revenue for low-income countries can be found by adding the coefficient of the interaction variables to the coefficient found for middle-income countries. Low-income dummy presents the effect of being a low-income country on tax revenue compared with the effect for middle-income countries.

The Hausman test has been performed to see whether the RE or the FE estimation is more appropriate for this analysis. Appendix G shows that the null-hypothesis of the Hausman test is not rejected, therefore, the RE estimation results are considered to be more appropriate. Concerning the results from the system-GMM estimation, the presence of first-order correlation and the absence of second-order serial correlation are consistent with the assumptions (Arellano & Bond, 1991). Furthermore, the insignificant coefficient found with the Hansen test proves that the instruments are valid.

The results in Column (1) and (2) show that the effect of IDA grants on the tax revenue of middle-income countries is negative but insignificant. This result corresponds with the RE results found for all IDA countries. The system-GMM estimation for this coefficient takes the same form as the RE estimation, but only when the variable External debt stock is included. The negative coefficient found for IDA grants can be explained by the substitution effect of grants. Since grants do not need to be repaid, they can decrease the need of collecting tax revenue.

The effect of IDA loans on the tax revenue of middle-income countries is insignificant but positive in all columns but Column (5). The system-GMM estimation that excludes the External debt stock shows a significant negative effect of IDA loans on tax revenue for middle-income countries. As mentioned before, the system-GMM estimation corrects for several biases which makes it a more reliable estimator than the RE and FE estimators. However, the coef-ficient for External debt stock is negative and significant in Column (6) and should therefore be included in the regression. This concludes that the system-GMM also gives an insignificant positive result for IDA loans. The positive coefficient might be driven by the fact that tax revenue is needed to repay the loans in the future.

The interaction variable Dummy × IDA grants shows a positive coefficient meaning that the negative effect of IDA grants found for middle-income countries is less severe for low-income countries; the effect of IDA grants on tax revenue approaches zero for low-income countries in Column (1) and (2). The system-GMM estimation shows a positive coefficient for IDA grants in Column (5), and this becomes even more positive for low-income countries. However, this regression does not correct for the external debt level of the recipient country. Column (6) shows the results when External debt stock is included into the regression. The coefficient of IDA

(30)

grants is negative again but the larger positive coefficient of Dummy × IDA grants makes the overall effect of IDA grants on tax revenue positive for low-income countries.

The RE estimations show a positive coefficient for IDA loans which becomes less positive for low-income countries. The difference between low and middle-income countries is however not significant. In Column (6) the system-GMM estimation also shows a positive coefficient for IDA loans but this coefficient becomes negative for low-income countries.

Although the results are not significantly different from zero, this concludes that the co-efficients for IDA grants and IDA loans found with the system-GMM estimation, have the opposite effect for low and middle-income countries. The substitution effect that might drive the effect of IDA grants for middle-income countries, is not expected to be present for low-income countries. Certain projects will only be realised in low-income countries if they receive a grant. Therefore, the grant does not substitute the national budget it only complements it.

Low-income countries are expected to be able to collect less tax revenue than middle-income countries, therefore, the coefficient of Low-income dummy is expected to be negative. The RE and system-GMM estimations do show a negative coefficient for Low-income dummy, however, the insignificance of the coefficients indicate that the income category does not have a significant impact on tax revenue.

The results of the control variables have already been discussed in the Section 4.1, since the coefficients of the control variables in Table III hold for all IDA countries.

(31)

5 Conclusion and discussion

This chapter contains the conclusion and the discussion of this paper. In Subsection 5.1.1

a summary of this paper and the main findings are given. This is followed by the policy

implications of these results in Subsection 5.1.2. This chapter finishes with the discussion in Section 5.2.

5.1 Conclusion

5.1.1 Summary

The IDA tries to reduce poverty by providing certain countries with financial aid. But what are the side-effects? To tackle a part of this question this paper examines the effect of the form of IDA aid on tax revenue. Previous research did not find a clear answer to the question whether financial aid should be provided in the form of grants or loans. Gupta et al. (2003), Benedek et al. (2014) and Odedokun (2003) conclude that grants negatively affect tax revenues of recipient countries. However, the analysis of Clist & Morrissey (2011) shows that grants can also have a positive effect on tax revenue. The mixed results make further research necessary. This paper contributes to the existing literature by using a more recent dataset, adding five more years of observations, to reexamine the effect of financial aid on tax revenue. Furthermore, the focus on only IDA aid rather than all ODA is new.

The methodology in this paper is largely based on the framework of Gupta et al. (2003); they constructed the basic regression model. This paper expands the model and in addition uses a more advanced, and arguably more appropriate, regression technique known as ‘system-GMM’ on top of their RE and FE regression method. The annual data of 45 countries, which have had access to IDA aid between 2002 and 2014, has been examined using random and fixed effect models, and system-GMM estimation.

First, the effect of the separate IDA aid variables on tax revenue for all IDA countries was estimated. The results of the RE and FE estimations are in line with the results from Gupta et al. (2003) and Benedek et al. (2014) who also find a negative coefficient for grants and a positive coefficient for loans. However, none of these results where significant and therefore no conclusions can be made. Furthermore, these results can be biased due to reversed causality between the aid variables and tax revenue. The system-GMM estimation corrects for the pos-sible endogeneity problem and can be assumed to be a more accurate estimation. The results of the system-GMM estimation show a significant negative effect of IDA loans on tax revenue

(32)

32 MSc Thesis Debbie Keijser – Conclusion and discussion

if External debt stock is not included into the regression. Since the External debt stock coeffi-cient shows a significant negative effect on tax revenue it is advised to include this variable into the regression. The system-GMM estimation also shows that IDA grants have an insignificant positive effect on tax revenue. This concludes that both IDA aid variables have an insignificant effect on tax revenue when the external debt level of the recipient country is taken into account. The analysis continued with a different regression that included a dummy for low-income countries and interaction variables. This made it possible to differentiate the effect of IDA aid on tax revenue for low and middle-income countries.

For middle-income countries the system-GMM estimation shows that IDA grants have an insignificant negative effect on tax revenue when External debt stock is included into the regres-sion. In this same regression, IDA loans have an insignificant positive effect on the tax revenue of middle-income countries. The dummy for low-income countries and the interaction variables capture the effect for low-income IDA countries compared to middle-income IDA countries. The effect of IDA grants on tax revenue is positive and insignificant for low-income countries when the regression corrects for the external debt level. IDA loans have an insignificant negative effect on the tax revenue of low-income IDA countries.

5.1.2 Policy implications

The research done in this paper has clear policy implications, if grants have a negative effect on tax revenue, donors should be more cautious when providing aid in the form of grants. And the same holds for loans.

The system-GMM results of Table II correct for many possible biases and with the exclusion of second order serial correlation and over-identifying restrictions it can be assumed to be the most accurate estimation. In contrast to the results of Benedek et al. (2014) and Gupta et al. (2003), the system-GMM results show an insignificant negative effect of IDA loans on tax revenue. The positive effect of IDA grants is also insignificant. Although the coefficients are not significant, these results indicates that the IDA should be more cautious when providing loans rather than grants, concerning the effect on tax revenue.

Instead of drastically reducing the volume of loans the IDA provides, loans could also be given under certain conditions. Loans should for example still be provided for projects with short-term economic benefits but under strict conditions concerning tax policy. It should also be taken into account that increasing the volume of grants might make countries more aid dependent (Gupta et al., 2003).

(33)

middle-income IDA countries. Although more research should be done, the IDA should take the income category of the recipient country into account when deciding on the form of aid that will be provided.

5.2 Discussion

The research done in this paper answers several questions concerning IDA aid and its effect on tax revenue. However, there are plenty more topics concerning development aid that need further research.

Future research could focus on collecting more data. One could expand the research on IDA aid with more control variables, as has been done for other sources of financial aid (e.g. ODA). Factors that determine political stability can be included, for example. External debt stock has shown to have a significant impact on the tax revenues of IDA countries. Hence, this variable should be included in all future research.

Furthermore, the conclusions based on the results could be reevaluated. The negative effect of IDA grants on tax revenue found in the RE and FE estimations might not be as disadvan-tageous as expected. If the tax reduction is realised through reduction in tax rates rather than through a reduction in tax collection efforts, the tax relief might stimulate economic growth (Gupta et al., 2003). A lower tax rate leads to more resources in the private sector which can stimulate consumption. Therefore, the national tax rates could be included in future research. As mentioned in Chapter 4, the form of aid can influence tax revenue differently. The effect could depend on what the aid is used for. For example, if loans are used to improve weak institutions, tax revenues are expected to increase (Benedek et al., 2014). On the other hand, if grants are used for projects that improve health care, there might not be a direct effect on tax revenue. Therefore, future research could examine where the different aid forms are specifically used for and if the projects where a success.

This paper focuses on the effect of aid on tax revenue. There are however more potential side-effects that are worth examining. Expanding the research might give a more complete overview of the drawbacks or additional benefits of providing development aid. It could be researched, for example, what the effects of aid are on income equality in recipient countries.

(34)

34 MSc Thesis Debbie Keijser – Conclusion and discussion

Appendices

Appendix A: Countries in the sample

Low-income countries: Burundi, Benin, Burkina Faso, Central Afrikan Republic, Democratic Republic of Congo, Comoros, The Gambia, Guinea-Bissau, Madagascar, Mali, Mozambique, Malawi, Niger, Nepal, Rwanda, Sierra Leone, Chad, Togo, Tanzania, Uganda.

Middle-income countries: Bhutan, Cote d’Ivoire, Cameroon, Republic Congo, Georgia, Ghana, Guyana, Honduras, Kenya, Kyrgyz Republic, Cambodia, Lao People’s Democratic Republic, St. Lucia, Sri Lanka, Moldova, Mauritania, Nicaragua, Pakistan, Tajikistan, Tonga, Uzbekistan, St. Vincent and the Grenadines, Vietnam, Republic Yemen, Zambia.

(35)

Appendix B: Data definitions and sources

Code Variable Definition Source

AGR Agricultural

value-added

Net output of the agricultural sector after adding up all outputs and subtracting intermediate inputs (annual as % of GDP) World Development Indicators (World Bank) IN D Industry value-added

Net output of the industrial sector after adding up all outputs and subtracting intermediate inputs (annual as % of GDP) World Development Indicators (World Bank) X Exports

Value of all goods and other market services provided to the rest of the world (annual as % of GDP) World Development Indicators (World Bank) IM Imports

Value of all goods and other market services received from the rest of the world (annual as % of GDP) World Development Indicators (World Bank) IN C GDP per capita

Gross domestic product (GDP) divided by midyear population (annual in US dollars)

World Development Indicators (World Bank)

DEBT Total external

debt stock

The total debt owed to nonresidents repayable in currency, goods, or services (annual as % of GDP)

LOAN S IDA loans

Public and publicly guaranteed debt extended by the World Bank Group at concessional rates (annual as % of GDP)

GRAN T S IDA grants

Net disbursements of grants from the International Development Association (annual as % of GDP) World Development Indicators (World Bank)

T OT AL IDA total The sum of IDA grants and IDA loans (annual as % of GDP)

T AX Tax revenue Governmental revenue from taxation (annual as % of GDP)

IMF’s World Revenue longitudinal dataset (IMF)

IDA grants or loans? : the effect on tax collection

tax collection

July 15, 2018

Statement of Originality

Acknowledgements

Abstract

Contents

1

Introduction

2

Literature review

2.1

Advantages and disadvantages of grants

2.2

The effect of aid on tax collection

3

Methodology and data

3.1

Empirical specification

3.2

Estimation

3.3

Data and summary statistics

3.4

Limitations of data

4

Results

4.1

Estimation results

4.2

Difference low and middle-income countries

5

Conclusion and discussion

5.1

Conclusion

5.2

Discussion

Appendices

Appendix A: Countries in the sample

Appendix B: Data definitions and sources