The effect of IDA loans versus IDA grants on tax revenues

(1)

The effect of IDA loans versus IDA grants on tax

revenues

Master Thesis

Group NL2 Student C. S. Ajamlou Student number 6076076

E-mail address Chloe.ajamlou@gmail.com

Date July 10, 2016

Supervisor dhr. B. Thio

Second reader dhr. D.J.M. Veestraeten

Number of words 11.625

Master Economics

(2)

Statement of Originality

This document is written by Chloé Sara Ajamlou, who declares 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)

Abstract

The International Development Association (IDA) was focused on providing loans since its founding in 1960. Since the beginning of this century, the IDA has changed its strategy and has started to use grants in addition to loans. This thesis investigates the effects of IDA loans versus IDA grants tax revenues. This is done by a fixed effects estimation based on panel data. The data comprises 59 IDA-eligible countries for the years 2000-2012. IDA loans tend to have a negative effect on tax revenues, whereas the grants tend to have a positive effect on tax revenues. But, the effects of neither loans nor grants were significant. For further research on this topic, a country-specific study is advised instead of a cross-country study.

(4)

Abbreviations

Abbreviation Explanation

DAC Development Assistance Committee

FE Fixed Effects estimation

GDP Gross Domestic Product

HIPC Initiative Heavily Indebted Poor Countries Initiative

IBRD International Bank for Reconstruction and Development IDA International Development Association

IFIAC International Financial Institution Advisory Commission MDRI Multilateral Debt Relief Initiative

ODA Official Development Assistance

OECD Organization for Economic Cooperation and Development

RE Random effects estimation

USA United States of America

(6)

I. Introduction

The International Development Association (hereafter: IDA) has increased the role of grants since the beginning of this century. The IDA is the association of the World Bank aiming at reducing poverty in the poorest developing countries in the world. The World Bank is one of the main providers of official development assistance (hereafter: ODA). As illustrated in figure 1, IDA-grants increased from zero in the year 2000 up to a yearly amount of around 2000 million USD from 2008 onwards. Besides grants, the IDA disburses funding in the form of loans and debt reliefs, as shown in figure 1. These grants and debt reliefs are provided to countries which are at risk of debt distress, because massive debt levels can hamper economic growth (International Development Association, Debt Sustainability & Grants, 2016).

The discussion on how ODA could best be provided, disbursing a loan or a grant, is complex. The effect of loans versus grants on tax revenues is an important issue in this discussion. Yet no consensus has been reached on whether the effect of disbursing a grant or a loan is more favourable in terms of tax revenues. Not only theoretical views indicate incompatible arguments comparing the effect on tax revenues of a loan or a grant, but the results of previous empirical research are ambiguous too. Whereas some previous empirical studies showed a positive effect of loans on tax revenues to GDP and opposing result were retrieved for grants (Benedek et al., 2012; Gupta et al., 2003), other studies suggested this effect not being straightforward (Morrissey et al., 2006) or even contradicting, indicating grants result in an increase in tax revenues (Clist, 2016). In these studies tax revenues to GDP refers to the actual performance of the recipients in collecting tax. Although, no consensus has been reached about the effect of aid on tax revenues, tax revenues are considered to be an important element in collecting domestic resources (McGillivray and Morrissey, 2001, p.21), which help to overcome aid-dependency problems in poor countries.

The goal of the increased role of IDA grants is to prevent unsustainable debt levels and eventually promote development. However, if these grants reduce tax revenues in developing countries and induce these countries to become aid-dependent, it might be not the best strategy to continue providing development financing through grants. Whereas, the differential effects of ODA loans and grants from all donors have been studied before, research on the effect of loans versus grants on tax revenues from the specific institution IDA is lacking. Since the IDA increased its focus on grants from the beginning of this century, increasing knowledge on the effects of IDA loans versus IDA grants is needed. Therefore, the central question of this study is:

(7)

To answer the above question a panel data analysis will be performed. This thesis builds upon the research of Gupta et al. (2003), who investigate the effect of loans versus grants by means of a fixed effects estimation. The estimation in Gupta et al. (2003) contains a comprehensive set of control variables to control for the structure of the economy and several macroeconomic variables. Also, potential simultaneous causality problems of aid and tax revenue are addressed by using one year lagged values of aid. Since IDA decided to spend a significant part of their development aid on grants from the beginning of this century, the sample includes annual data of 59 IDA-eligible countries over the period 2000 to 2012. Most data is collected from the WDI and additionally, data on disbursed aid is collected from the OECD. This thesis will contribute to the understanding of the effectiveness of the changed strategy of the IDA towards allocating grants.

The structure of the thesis is as follows. The following section provides a literature overview, after which the methodology and data are described. Next, the results of the empirical analysis are discussed and this thesis ends with concluding remarks.

Figure 1: IDA disbursements expressed in million USD, per year

Note: The amount of IDA loans and debt forgiveness are excluded in this graph for the year 2006, as this year contains extreme values which would disturb the visual understanding of the graph. Source: graph constructed by author based on data from OECD retrieved at April 16, 2016. 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14

IDA grants (excluding debt forgiveness)

IDA loans

(8)

II. Literature review

This section starts with background information about ODA and the role of IDA within aid architecture. Then we discuss theoretical arguments of the supposedly-different effect of loans versus grants on tax revenues. The last section presents results of previous empirical studies that studied the tax response on development loans versus grants.

Background information ODA

The Development Assistance Committee of the OECD defines ODA as flows to countries and multilateral institutions which are provided by official donors (OECD, 2008)1_{. Official donors are state} and local governments, private funds, bilateral organizations and multilateral organizations. A multilateral organization is formed by international member governments, which collectively govern the organization and provide funds (AidWatch, 2008). The International Development Association (IDA), which is part of the World Bank Group, is one of the main multilateral organizations in aid architecture, in terms of their disbursed funds for the years 1960-2005 (International Development Association, 2007, p.4). Other main multilateral organizations are Regional Development Banks, the United Nations, the European commission and the other institutions of the World Bank Group. Provided that development aid by official donors can be listed as ODA, two additional criteria should be met. The first criterion is that ODA’s main objective should be to promote economic development and welfare in developing countries (OECD, 2008). The IDA complies with this criterion as its main objective is to support development activities that promote economic growth, job creation, higher incomes, and better living conditions (International Development Association, What is IDA?, n.d.). The work of IDA covers primary education, basic health services, clean water and sanitation, agriculture, business climate improvements, infrastructure, and institutional reforms. The third and last criteria to be listed as ODA is that these flows are provided on concessional terms and more specifically that the grant element should be at least 25% (OECD, 2008). A grant element of 100% means that the funding is disbursed as pure grant. A loan contains a grant element too, since ODA loans cost the recipient much less than borrowing on commercial terms. The interest rates on loans are zero or lower than market interest rates. Additionally a grace period and long payback period are provided. A grace period is a period of no repayment obligations; right after the loan is obtained. Thus the benefit compared to lending on commercial terms is the grant element (Odedokun, 2003, p.2).

The IDA distinguishes itself within the wider aid architecture by two matters. The first one is its scope. The IDA was founded to make funding available for the poorest countries in the world, since it appeared these could not afford lending of the International Bank for Reconstruction and

(9)

Development, the other lending arm of the World Bank (Sanford, 2002, p.743). Still, the poorest countries in the world2_{and countries which lack creditworthiness to borrow on market terms are} eligible for IDA only (International Development Association, What is IDA?, n.d.). The second matter relates to its terms of funding, although IDA’s funding terms evolve over time as the IDA aims at adapting to new insights. Since its founding the IDA was focused on providing aid in the form of loans, while there was no intrinsic motivation for that (Sanford, 2002, pp.741-747). Even before the IDA was created there was discussion around whether it should provide loans or grants. Eventually it was agreed that the IDA would provide loans, because it was expected that prospective IDA donors would be more willing to contribute to an association that disburses loans rather than grants. IDA’s Articles of Agreement state that is should provide funding in the form of loans, but that it was able to provide grants in special circumstances (Sanford, 2002, p.743). The discussion had revived again after a proposal of the International Financial Institution Advisory Commission (IFIAC) and a speech by former president Bush of the USA in 2000 and 2001 respectively. It was proposed to increase the use of grants to improve the lives of people in the poorest countries. The IDA has responded to those calls and has started to disburse grants since 2002. Whether the IDA disburses a grant, loan or mix of both, depends on the level of debt distress in the recipient country. Pure grants are disbursed to countries with high levels of debt distress to prevent debt accumulation in poor countries (International Development Association, Debt Sustainability & Grants, n.d.). In the past, excessive lending has led to debt accumulation in poor countries and this holds back development (Gupta, 2003, p.3). Although the IDA provides both loans and grants, it should be noted IDA loans are highly concessional, so these contain a substantial grant element too.

In addition to providing loans and grants, the IDA participated in two debt relief initiatives. Several ODA donors, among which the IDA, initiated together the Heavily Indebted Poor Countries (HIPC) Initiative and the Multilateral Debt Relief Initiative (MDRI) (International Development Association, What is IDA?, n.d.). The HIPC initiative was launched in 1996, to contribute toward growth and debt sustainability in the poorest and most heavily indebted countries. The countries that are covered with the HIPC initiative are IDA-eligible countries only. On top of the HIPC initiative, the MDRI was launched in 2006 to provide additional support to HIPCs to reach the Millennium Development Goals3_(World Bank, 2013). The goals were set up by world leaders to fight together against extreme poverty during

2_{Countries with a Gross Net Income (GNI) per capita of lower than $1,215, are eligible for funds from the IDA} (International Development Association, Borrowing Countries, n.d.).

3_{This program is currently replaced by Sustainable Development Goals (A new sustainable development} agenda, n.d.).

(10)

2000 till 2015. MDRI-eligible countries are those that participated in and completed the HIPC initiative, but still experience massive debt levels (World Bank, 2013).

Figure 2 illustrates disbursed IDA funds from 1990 until and including 2014. It is visible IDA’s focus has been on disbursing loans and that debt relief and grant disbursements have started since 1998 and 2002 respectively. Figure 3 illustrates that ODA disbursements from all donors have mainly been in the form of grants. This visualizes that the IDA has a different strategy in disbursing aid when compared to the common ODA trend. IDA’s distinguishing terms of funding do not relate to the modality of aid only, but relate to the conditions accompanied with that funding too. The IDA requires the recipient country to comply with certain conditions, before aid is disbursed. This is done to ensure that aid contributes to the recipient’s development goals and to ensure that the aid is used for the intended purpose (World Bank, 2005). These conditions apply to both loans and grants. The IDA reviews and adapts it conditionality policy in response to new insights continuously.

In sum: the IDA is one of the official donors that provides aid to the poorest countries. It disburses funds in the form of loans since its creation in 1960 and in addition grants and debt relief since 1998 and 2002 respectively. IDA-eligibility requires complying with stronger conditions than ODA-eligibility requires.

(11)

Figure 2: IDA disbursements expressed in million USD, per year

Figure 3: ODA (by donors including the IDA) disbursements expressed in million USD, per year

Notes: For year 2006 IDA net loans and IDA debt forgiveness are excluded in figure 2, since these are extreme values (-27.190 and 32270 mill. USD resp.) and disturb the visual understanding of the graph. Net loans by both all donors and just IDA are negative in year 2006, because loan repayments were higher than new loan disbursements. Source: graph constructed by author based on data from OECD retrieved at April 16, 2016.

Theoretical argumentation

The discussion in which form ODA could best be provided is complex. Some arguments in favour of providing a loan are: political economic reasons (Odedokun, 2003, p.3); ownership feeling of the recipient government; and efficient use of development aid (Schmidt, 1964, p.388, Odedokun, 2003, p.14). On the contrary, some arguments in favour of providing a grant are: the humanitarian character of the intended project (Gupta, 2003, p.3); and preventing massive debt levels in developing countries.

0 1000 2000 3000 4000 5000 6000 7000 8000 9000

IDA net loans

IDA grants (excl. debt forgiveness)

IDA debt forgiveness

-60000,00 -40000,00 -20000,00 0,00 20000,00 40000,00 60000,00 80000,00 100000,00 120000,00 140000,00 160000,00 1990199219941996199820002002200420062008201020122014

ODA net loans

ODA grants (excl. debt forgiveness)

(12)

The effect of aid on tax revenues is an important issue too when it comes the discussion of loans versus grants. Yet no consensus has been reached on whether the effect of disbursing a grant or a loan is more favourable in terms of tax revenues.

Gupta et al. (2003, pp. 6-7) demonstrate the effect of aid on tax revenues in terms of a government’s budget constraint. Investment together with consumption determines the expenditure side of the budget constraint (G), whereas tax revenue (T), aid (A) and borrowing (B) determine the revenue side. This results into the definition equation ‘G = T + A + B’. The authors assume a behavioral effect of aid disbursements within the framework of a budget constraint. Like tax revenue and domestic borrowing, aid is a kind of governmental revenue. The different parts of governmental revenue may be substitutes for each other. The budget constraint does not distinguish the effects of loans versus grants, however Gupta et al. (2003, p.7) state that it is plausible that loans and grants have a different effect on tax revenues. It is argued that interest and repayment costs of a loan force a government to spend the loaned money carefully (Schmidt, 1964, p.388) and to mobilize taxes (Khan and Hoshino, 1992, p.1486, Gupta et al., 2003, p.4) in order to eventually be able to repay the loan. Inversely, provided that grants lack repayment obligations, it is feared that grants substitute tax revenues in recipient developing countries. This fear supposes a different behavioral response of loans versus grants on tax revenues, but this is not an established fact. Gupta et al. (2003. p.4) acknowledge that policy makers are nowadays less likely to perceive grants and loans differently, because earlier disbursed loans are frequently forgiven. Examples are the cases of the HIPC initiative and the MDRI. In addition, even if a behavioral response of loans versus grants exists, this does not imply that this affects tax revenue in particular. Instead of tax revenues, other parts of the budget constraint might be affected. Morrissey et al. (2006, p.14) state that the relationship between grants and tax revenues is unlikely to be causal at all, since there are many reasons why poor countries have difficulties in increasing tax revenues.

Combes et al. (2016, pp.2-3) state that the effect of aid on tax revenues is not straightforward at all, apart from the modality of aid. ODA disbursements are more and more bundled with conditions on policy reforms. Depending on the policy conditions, tax revenues could in turn increase or decrease. If a donor requires the recipient country to open up trade conditional on aid, aid will most likely decrease tax revenues due to lower tariffs. On the other hand, if conditionality bundled with aid covers minimum requirements on tax revenues, this will lead to an increase in tax revenues. A theoretical situation in which aid strictly increased tax revenues is mentioned too. If aid is used to improve the tax system and tax collection, aid positively affects tax revenues. But in practice, increasing tax revenues can be very difficult to implement due to the political costs for politicians (McGillivray and

(13)

Morrissey, 2001, p.13). Policy makers might fear that the public will object if it announces its desire to increase tax revenues, whereas a decrease in tax revenues will be applauded.

An argument in favor of grants is the debt overhang problem. Debt overhang exists when the net present value of repayment obligations is higher than the net present value of all potential returns in a country (Krugman, 1988, p.5-15). Therefore a government has no incentive to make investments, because it knows that all the returns will flow to the creditors. To prevent the debt overhang problem, grants and debt relief programs are preferred over loans. Morrissey et al. (2006, p.13) links a country’s indebtedness to its fiscal policy. The authors argue that grants do not add to the country’s indebtedness and therefore grants may be more likely to spur economic growth than loans. This in turn strengthens their argument that grants support a stable fiscal policy in a country.

The important question arises whether it is bad if development aid inflows reduce tax revenues. Development aid is provided to stimulate development and a reduction in tax revenues as an effect of aid is intuitively not associated with development. McGillivray and Morrissey (2001, p.19) confirm this intuition and found that the likelihood of aid to promote growth in a developing country is higher when aid disbursements do not encourage a lower tax effort. The countries that are dependent on aid have generally a low tax to GDP ratio. Increasing the tax to GDP ratio is an important element of collecting domestic resources (2001, p.21) and to overcome aid-dependency problems. Yet, the authors state that the reasons behind decreasing tax revenues should be known, in order to criticize decreased tax revenues. If aid is used as a substitute for tax revenues due to rent-seeking behavior of the recipient government (Gupta et al., 2003, p.7), this is undesirable for both the recipient country and donor, since aid-dependency problems remain. A donor’s disbursement is supposed to be a contribution when it comes to development of developing countries, but if disbursed aid encourages rent-seeking behavior it is a waste of money. Another adverse behavioral effect of aid is at issue if the recipient country keeps tax revenues strategically low (McGillivray and Morrissey, 2001, p.14). The recipient might do this, because it knows it will be allocated aid if tax collection is not sufficient to meet the budget constraint.

However, a reduction in tax revenues will not always be undesirable. A reduction in tax revenues could be desirable when a recipient government decides to pass (a part of) the aid benefit through to the private sector (Gupta, 2003, p.7). Also in the example of conditions on opening up for trade, one should be cautious when drawing conclusions when a there is a reduced tax revenues situation. Although lower tax revenues are regarded as undesirable at first glance, the reason behind it, opening up for trade is regarded as a positive event (2001, p.18).

(14)

The key point of the previous section is to show that the effect of loans versus grants on tax revenues is theoretically not straightforward. Even if loans or grants affect tax revenues, it is not always an undesirable effect, because it depends on the situation.

Findings of empirical literature

Various studies estimated the effect of aid on tax revenues empirically. One of the main issues in estimating this effect is the potential existence of simultaneous causality between aid and tax revenues. The reasoning is as follows. If allocated aid is substituted for tax revenues in a recipient country, aid affects tax revenues. On the other hand, it could be that aid is provided to countries which are unable to collect sufficient taxes to finance their development goals. In this case low tax revenues determine the allocation and amount of aid. In econometrics the phenomenon of simultaneous causality is defined as an endogeneity problem (Stock and Watson, 2012, p.462). This endogeneity problem could be addressed in different ways. In the case of aid and tax revenues the use of lagged aid as an instrument in a fixed effects estimation is widely used. Although the use of lagged aid addresses the endogeneity problem, it is not solved completely. For each empirical study mentioned in this section, it is indicated how endogeneity is addressed. This section focusses on cross country analysis only.

According to Gupta et al. (2003, p.7) the then existing fiscal response studies on aid lacked modeling a distinguished effect of loans versus grants on tax revenues. And if a study did distinguish this effect, the sample was limited4_{. Therefore Gupta et al. (2003) examine empirically whether the composition} of ODA matters for the tax revenue response based on a broad sample. Both loans (L) and grants (F) are included in percentage of GDP in a fixed effects estimation. In addition squared aid variables are included to allow for a non-linear causation. To following four control variables are included to control for the structure of the economy: both agriculture (AGR) and industry (IND) value added and a measure of openness of a country (TRADE), all in percentage of the recipient’s country GDP; and an approximation for the status of development of a country, real GDP per capita (SIZE). The sample is comprehensive: the dataset comprises 107 countries for the period 1970-2000. Data on aid are retrieved from the OECD, tax data are retrieved from the IMF and data on control variables are from the WDI and IMF. The baseline estimation is as follows:

[_𝐺𝐷𝑃𝑇 ] 𝑖,𝑡 = 𝛽0+ 𝛽1. [ 𝐴𝐺𝑅 𝐺𝐷𝑃]_𝑖,𝑡+ 𝛽2. [ 𝐼𝑁𝐷 𝐺𝐷𝑃]_𝑖,𝑡+ 𝛽3. [ 𝑇𝑅𝐴𝐷𝐸 𝐺𝐷𝑃 ] + 𝛽4. 𝑆𝐼𝑍𝐸 + 𝛽5. 𝐹 + 𝛽6. 𝐿 + 𝜀𝑖,𝑡 [1]

4_{For an example see Khan and Hoshino (1992), who found that grants reduce taxation effort while loans} increase it. This is based on a sample of five South and Southeast Asian countries.

(15)

Since the dependent variable of the estimation is non-negative and skewed, a log transformation of the dependent variable is used. The independent variables are not transformed into logarithmic values. The β’s are then an estimate of the impact of each independent variable in the regression on the growth rate of tax/GDP (p.8). The results of the estimation, using one-year lagged aid values, indicate that loans have a positive effect on tax revenues and grants have a negative effect on tax revenues. These results are significant at 1% (2003, p.13). Squared loans and squared grants show a negative and positive effect respectively. This indicates a dampening effect of loans versus grants as the amount of aid becomes larger. The effects of the aid variables remain unaffected when other control variables are added to test for robustness.

Morrissey et al. (2006), Clist and Morrissey (2011) and Benedek et al. (2012) elaborate on the study and method of Gupta et al (2003), to estimate the effect of loans versus grants on tax revenue. The main addition of the study of Morrissey et al. (2006) is the use of a balanced panel rather than an unbalanced panel. As there are many missing observations in especially the tax/GDP ratio of countries, the balanced panel (omitting countries with missing observations) is smaller than the unbalanced panel (Morrissey, p.3). The advantage of using a balanced panel is that it allows observing the statistics of a specific country in every time year. The aim of Morrissey et al. is to check whether the results of Gupta et al. (2003) are robust (2006, p.3). One test is done assessing 46 developing countries for the years 1980-1990. One year lagged values of aid address the endogeneity issue. The results of the random effects specification show that loans have a significant and positive impact when regressed on tax/GDP ratio, while grants have a negative but not significant impact (2006, p.7). In another test data are averaged over four 5 year periods (1976-1980, 1981-1985, 1986-1990 and 1991-1995) to smooth annual variability and expand the sample. The results of their fixed effects estimation show the same sign as in the random effects specification, but it is not significant for neither loans nor grants. The sample consists of developing low and middle income countries, but it is not clear which countries are specifically included. Morrissey et al. (2006, p.13) state that they did not find robust evidence and significant evidence that grants reduce tax revenues.

Clist and Morrissey (2011) use two-year lagged aid variables to address the endogeneity problem5_. Their fixed effects estimation for the period 1970-2005 shows the same results as Gupta et al. (2003), which is a significant and positive effect of loans on tax revenues and a significant and negative effect of grants on tax revenues (Clist and Morrissey, 2011, p.170). To smooth out annual variability, four year sub-period averages of the data over the period 1976-2005 are used as a robustness check. The outcome of this differs to the baseline regression, in the sense that the effect of grants is not significant

(16)

anymore (p.172). Remarkably, when a sample covering the years 1985-2005 is used, the effect of grants changes into a positive sign. This incites suspecting a changed relationship between grants and tax revenue around the middle of 1970-2005 and is further researched in additional robustness checks. Those robustness checks show that grants become less significant after 1984 and lagged grants even positive and significant in medium term estimates with differenced data (p.174). All in all, Clist and Morrissey (2011) find no evidence that aid, either in the form of loans or grants, reduce tax revenue. There is even some evidence that grants tend to increase tax revenues since the mid-1980s. Benedek et al. (2012) attempt to improve the model specification of Gupta et al. (2003) in the way endogeneity is addressed. This is done by the difference and the system generalized method of moments (2012, p.9). The authors claim that they used a broader and more diversified sample than Gupta et al. (2003). It can be confirmed that the sample is indeed more balanced regarding the income levels of the included countries6_{. The time span covers the years 1980-2009. In general the results of} the estimations are similar to that of Gupta et al. (2003), but it appeared that the results are stronger in low-income countries. In low-income countries grants appeared to have a significant negative effect, but loans have an insignificant positive effect7_.

Both Carter (2013)8_{and Clist (2016) replicate the research of Benedek et al. (2012). However, both are} unable to retrieve the same results as Benedek et al. (2012) even though nearly the same dataset was used. Clist’s main critiques on existing empirical studies concern the endogeneity problem as well as data quality of tax revenue. Clist (2016, p.9) argues that tax data is measured differently across different data sources and that the effect of aid depends on which tax data source is used. Therefore Clist runs five different regressions covering the years 1980-2011 and using lagged aid to address endogeneity, in addition to the replication estimations. Each regression is based on a different tax data definition and/or a different tax data source, to ensure consistency of tax data within each regression. The results show that the sign of the effect of both loans and grants is ambiguous (p.13). The only significant effects that are found are positive for grants and negative for loans.

Odedokun (2003) studies the effect of the ratio grants to total aid on tax revenue, measured in tax to GDP. The panel consists of 72 developing countries, which are split into 42 low-income developing countries and 30 middle-income developing countries9_{. The results of a fixed effects estimation,}

6_{The author found that the sample of Benedek et al. (2012) includes relatively more low and middle income} countries and relatively less high income countries than Gupta et al. (2003).

7_{The results of previously mentioned studies are displayed in appendix I.}

8_{This is based on the abstract of Carter (2013) only, since the paper itself was not available.}

9_{The threshold for the division between low- and middle-income developing countries was set at 1000 USD} per capita per annum, averaged over the period 1970-99.

(17)

assessing the period 1970-1999, show that the grants to total aid ratio has a significant negative effect on the tax to GDP ratio in low income developing countries. Contrary, this effect is significant positive in high income developing countries. The estimation is less sophisticated than that of Gupta et al (2003), since it does not contain variables that control for the concerning economy and the effect of loans specifically are not taken into account in the estimation. The low income countries outcome is of interest, since the IDA focusses on low income countries. Odedokun used ODA data from the OECD. In the above mentioned empirical studies ODA is split up into net loans and grants. These studies are done with data on aid disbursed by all official donors. The specification ‘net’ is important for loans, since this is a measure of newly disbursed loans minus loan repayments in a specific year. This means that the net loan amount could be negative if a country’s loan repayments are higher than a newly-disbursed loan. For grants, is it not specifically mentioned whether the amount of grants included grants as debt relief or not. It is most likely that at least Gupta et al. (2003) and Odedokun (2003) used data on grants which includes debt relief, because their aid data are drawn from the OECD. Since the OECD provides data on grants including debt relief only and nowhere in text is referred to the adjustment of this data, we assume that the data used on grants includes debt relief. In the section Model and Data we come back to this.

In sum: there is some empirical evidence that loans increase tax revenues and grants decrease tax revenues. However, some follow-up studies found that the effect of loans versus grants on tax revenues is ambiguous. Clist (2016) even retrieved a significant negative effect of loans on tax revenues and a significant positive effect of grants on tax revenues. Whereas the differential effects of ODA loans and grants from all donors have been studied before, research on the effect of loans versus grants on tax revenues from the specific institution IDA is lacking. Since the IDA increased its focus on grants from the beginning of this century, increasing knowledge on the effects of IDA loans versus IDA grants is needed. This thesis addresses this gap by estimating the sign and sizes of the effects of IDA loans versus IDA grants on tax revenues. It will be assessed whether these effects are significant. It is not clear whether IDA-aid disbursements affect tax revenues, since the IDA imposes conditions on aid to ensure aid effectiveness. In addition no distinguished effect in sign and size of loans versus grants is expected, because the IDA imposes the same conditions on its loans and grants. Also, it is not clear whether the effect of IDA-aid deviates from the effect of aid from other donors. In the following section we present our empirical model and data.

(18)

III. Model and data

Model

The model estimation in this research is built up on the estimation of Gupta et al. (2003). That means that the dependent variable, tax revenues in percentage of the recipient country’s GDP, is regressed in a fixed effects estimation10_{. Since the dependent variable is nonnegative and skewed, its logarithmic} values are used in the estimation. The independent variables, loans and grants are regressed as separated variables on tax revenues by a fixed effects estimation, all measured in percentage of the recipient’s country GDP. The reason to include loans and grants as separate variables is that we are interested in the sign and size of loans versus grants on tax revenues. In addition, non-linear effects of loans and grants are allowed for by including squared aid variables in the regression, like Gupta et al. (2003). Alternatively, since Gupta et al. (2003) provide no clear argument why squared aid variables would be the best way to deal with non-linearity, a substitute of squared aid variables is proposed. The logarithmic values of loans versus grants will be regressed as substitute for the aid and squared aid variables. To address simultaneous causality11_{between aid and tax revenues, aid will be lagged for} one year, similar to previous studies12_{. As mentioned in the empirical overview section, previous} studies used data on grants which included debt relief or debt forgiveness. Previous research showed that debt relief could have a different effect on tax revenues than grants. For example, Cassimon and van Campenhout (2008, p.438) found that debt relief positively affects tax revenues, whereas grants negatively affects tax revenues. Since the IDA is one of the participants of the HIPC initiative and MDRI, and these debt relief programmes apply to IDA eligible countries, it is considered to be appropriate to make a distinction between grants including debt relief and grants excluding debt relief. Therefore estimations on the effect of grants will be reported for grants both including and excluding debt relief in this thesis.

Earlier research13_{estimated crucial determinants of both tax ratio and effort in developing countries.} Gupta et al. (2003) use four variables to control for the structure of the economy from those studies which will be included in this thesis as well. The first two control for the sector composition of the economy. The first one is agriculture value added (AGR). The reasoning is that the agricultural sector is hard to tax directly, so when a country has a large agricultural sector, which is mostly the case in low income countries, this reduces the tax base (Morrissey et al., 2006, p.3). Agriculture value added is expected to negatively impact tax revenue. Contrary, the second control variable, industry value

10_{A fixed effects regression controls for omitted variables in panel data, when the omitted variables vary} across countries but not over time (Stock and Watson, 2012, p.396).

11_{See section II}

12_{Gupta et al. (2003), Morrissey et al. (2006) and Odedokun (2003)} 13_{See Ghura (1998) and Gupta (2007)}

(19)

added (IND), is expected to positively affect tax revenue, because this sector is relatively easy to tax. The third control variable is real GDP per capita, which is meant as approximation for the level of economic development in a country. Higher developed countries are assumed to be more likely to collect more tax than lesser developed countries. The fourth control variable is trade, which is a measure of openness in a country, calculated by the sum of imports and exports (Trade). It is included since trade taxes are an import source of tax in developing countries. AGR, IND and Trade are expressed as a percentage of the recipient’s country GDP.

The baseline regression equation is as follows:

𝑙𝑜𝑔 [𝑇𝑎𝑥 𝐺𝐷𝑃]_𝑖,𝑡 = 𝛽0+ 𝛽1. [ 𝐿𝑜𝑎𝑛 𝐺𝐷𝑃]_{𝑖,𝑡−1}+ 𝛽2. [ 𝐿𝑜𝑎𝑛 𝐺𝐷𝑃] 2 𝑖,𝑡−1+ 𝛽3. [ 𝐺𝑟𝑎𝑛𝑡 𝐺𝐷𝑃 ]_{𝑖,𝑡−1}+ 𝛽4. [ 𝐺𝑟𝑎𝑛𝑡 𝐺𝐷𝑃 ] 2 𝑖,𝑡−1+ 𝛽5. [ 𝐴𝐺𝑅 𝐺𝐷𝑃]_𝑖,𝑡+ 𝛽6. [ 𝐼𝑁𝐷 𝐺𝐷𝑃]_𝑖,𝑡+ 𝛽7. [ 𝐺𝐷𝑃 𝑐𝑎𝑝𝑖𝑡𝑎]_𝑖,𝑡𝛽8. [ 𝑇𝑟𝑎𝑑𝑒 𝐺𝐷𝑃 ]_𝑖,𝑡+𝛼𝑖 + ℇ𝑖,𝑡 [2]

The subscripts ‘i’ and ‘t’ refer to a specific country and a specific year respectively. The regressions in this thesis can be clustered in three sections and will be presented in tables 1-3 of the results section accordingly.

1. Regressions based on disbursed loans and grants provided by all donors as an attempt to reproduce the regression results of Gupta et al. (2003). This enables us to compare the results of the baseline regression with the results of Gupta et al. (2003). In addition a distinction between two samples will be made to check whether the results differ depending on the countries included in the sample: one including all ODA-eligible countries and one including IDA-eligible countries only.

2. Regressions in which both disbursed loans and grants from the IDA only are included. The sample includes IDA-eligible countries only. These regressions provide us results on the effect of IDA loans versus IDA grants and can be compared with the regression results mentioned in point 1. As an alternative to allow for non-linear effects, logarithmic aid values replace all the aid variables of the baseline regression in additional regressions.

3. Regressions in which loans from the IDA versus loans from all donors but the IDA are included as separate variables into one estimation. This is done to check robustness of the effects of loans and to enable a comparison between the effects of IDA-aid and aid provided by other donors than the IDA. As an alternative to allow for non-linear effects, logarithmic loan values replace all the loan variables of the baseline regression in additional regressions. The sample includes IDA-eligible countries only. Everything mentioned here in point 3 will be done for grants as well.

(20)

Scope and data

The dataset comprises 100 ODA eligible14_{developing countries, of which only 59 are IDA-eligible}15_{, for} the period 2000-201216_{. The starting point of the timespan is determined by the fact that IDA increased} its use of grants since the beginning of this century. The year 2012 is determined by the fact that tax data is limited available after 2012. The countries included in the dataset are determined by data availability too.

The following data are drawn from the WDI: tax revenues; value added of the agricultural sector; value added of the industrial sector; and trade, which is the sum of imports plus exports of goods and services. All these data were already expressed in percentage of recipient’s country GDP. Real GDP per capita expressed in constant 2005 USD, and GDP expressed in current USD, were drawn from the WDI as well. Data on aid expressed in current USD were collected from the OECD, and were subsequently expressed in percentage of the recipient’s country GDP by the author. The aid figures present disbursed loans, grants including debt relief and debt relief, from both all donors and the IDA only. All data are annual statistics.

The panel has some missing observations, which means it does not have all the observations for all countries in every year. This is called an unbalanced panel (Stock and Watson, 2012, p.390). The missing values will be excluded from statistical calculations. For descriptive statistics on the dataset and a correlation table see appendix IV and V, respectively.

IV. Regression results and discussion

Table 1 provides regression results that can be compared to those in Gupta et al. (2003, p.10). All the aid figures are disbursed amounts provided by all donors and aid is lagged for one year. A country fixed effects estimation with the dependent variable being tax revenues to GDP in logarithmic form is used for these and all following results. The first and third columns are based on the sample with all ODA-eligible countries, while the second and the fourth column include IDA-eligible countries only. The first and second models provide results for grant figures including debt relief, whereas the third and fourth models provide results for grant figures excluding debt relief. Both loans and squared loans show a negative effect on tax revenues in the estimations in columns 1-4. Especially in the estimations in which grant figures exclude debt relief, the effects of loans and squared loans are significant at the 5 and 10 percent levels respectively. Two aspects are remarkable in the case of the effect of loans.

14_{Eligible for ODA at 2013}

15_{Eligible for IDA from 2000 until and including 2012. Although Albania, Azerbaijan, Indonesia, Macedonia and} Serbia graduated during this period, these countries are included in the sample, since aid disbursements from IDA continued during that time.

(21)

First of all, the sign of the effect of loans is contradicting to the results of Gupta et al (2003)17_{. Gupta} et al (2003) found that loans increase tax revenues, but the results of our estimations tend to show a decrease in tax revenues. The second remarkable aspect is that no dampening effect of loans is observed, since the results for squared loans are negative too. Grants show a negative effect and squared grants a positive effect on tax revenues. These effects do not differ if grants exclude debt relief. The sign of grants and the dampening effect of grants are consistent with Gupta et al. (2003), although we were not able to produce significant results.

Regarding the control variables, real GDP per capita and trade show a significant positive effect on tax revenue while agriculture show a negative non-significant result. The signs of agriculture and trade are as expected, namely negative and positive respectively. The sign of real GDP per capita, which is positive, is also as expected, since it was expected that countries with a higher real GDP per capita are better able to collect taxes. Authors mentioned in the empirical literature overview were not consistently able to produce a positive and significant effect for real GDP per capita, but in these estimations it does show a consistent positive effect and the effect appears to be stronger when IDA only countries are taken in the sample. Remarkable is that the sign of industry value added switches depending on which countries are taken in the sample. When all countries are included in the sample, the sign of this variable is similar to previous studies, namely positive, but when a sample of only IDA eligible countries is used the sign changes to negative. It is not clear why this happens. The estimated effects of the control variables by previous studies can be looked up in appendix II.

In short it can be stated that the sign of the effects of squared loans, grants and squared grants match the findings of Gupta et al. (2003). Contrary our results for loans do deviate from Gupta et al. (2003) but not to Clist (2014), since he estimated a negative effect of grants in some of his estimations too. An explanation why our results for loans correspond to a larger extend to that of Clist than to Gupta et al. could be that our timespan overlaps to a larger extent the timespan of Clist18_{. Although, this} argument is weak, since the effects of the other aid figures do match the results of Gupta et al. (2003). Furthermore no large differences in effects are observable when we switched from the larger sample including all ODA-eligible countries to the smaller sample including IDA-eligible countries only, with the exception of the effect of IND. The sign of industry value added switches from positive to negative accordingly. The results presented in table 2 and 3 are based on the sample including IDA-eligible countries only, so these can be compared to estimations in column 2 and 4 in table 1.

17 Contradicting to Morrissey et al. (2006), Clist and Morrissey (2011) and Benedek et al. (2012) as well. 18 Clist (2014) covers the years 1980-2009, whereas Gupta et al. (2003) cover the years 1970-2000. This thesis addresses the years 2000-2012.

(22)

Table 1: Effects of aid by all donors on tax revenues

Notes: clustered standard errors are given in parentheses. Aid is lagged for one year and results are based on fixed effects estimations. (***), (**), and (*) denote significance at the 1, 5 and 10 percent levels, respectively.

Table 2 provides results of the effects of IDA-aid alone, and again distinguishes loans and grants. IDA grant figures include debt relief in the estimations in column 5 and 6, whereas it is excluded in the estimations in column 7 and 8. The effect of the variables IDA loan and squared IDA loan in estimation in column 5 and 7 are comparable in sign and size with estimations in column 2 and 4, however these effects are not significant at the 10 percent level. In the estimation in column 5 the effects of IDA grants and squared IDA grants including debt relief are similar to the results in model 2 and 4 too. Grants show a negative effect on tax revenues while squared grants show a positive effect. This might

To all ODA-eligible Grant incl. debt relief (1) To IDA-eligible Grant incl. debt relief (2) To all ODA-eligible Grant excl. debt relief (3) To IDA-eligible Grant excl. debt relief (4) Loan to GDP -0.00256 (0.00160) -0.00265 (0.00159) -0.00177** (0.00088) -0.00185** (0.00088) [Loan to GDP]² -0.00002 (0.00001) -0.00002* (0.00001) -0.00001* (0.00001) -0.00001* (0.00001) Grant -0.00085 (0.00163) -0.00088 (0.00160) -0.00276 (0.00395) -0.00232 (0.00398) [Grant to GDP]² 0.00000 (0.00001) 0.00000 (0.00001) 0.00002 (0.00003) 0.00002 (0.00003) AGR -0.00330 -0.00275 -0.00331 -0.00277 (0.00280) (0.00298) (0.00331) (0.00300) IND 0.00009 -0.00101 0.00004 -0.00104 (0.00192) (0.00289) (0.00192) (0.00289) GDPPC 0.00002*** 0.00007** 0.00002*** 0.00007** (0.00001) (0.00003) (0.00001) (0.00003) TRADE 0.00105*** 0.00115*** 0.00104*** 0.00115*** (0.00031) (0.00040) (0.00032) (0.00042) Constant 1.05548*** 0.97014*** 1.06433*** 0.97971*** (0.08112) (0.11876) (0.08622) (0.12549) Observations 829 485 829 485 Countries 96 57 96 57 R squared 0.10944 0.15017 0.11060 0.15039 F statistic 10.79786 6.22921 28.42924 22.36624 P value 0.00000 0.00001 0.00000 0.00000

(23)

indicate that if grants are large, the effect on tax revenues reverses from negative to positive. On the contrary, if debt relief is excluded as in the estimation in column 7, then the sign of the effect of both IDA grants and squared IDA grants reverse with respect to the estimated effects presented in column 5. This might be an indication that grants including debt relief have a different effect on tax revenues than grants excluding debt relief. Furthermore specifications presented in column 6 and 8 are an attempt to address non-linearity in a different manner than the baseline specifications presented in column 5 and 7, namely by the use of logarithmic values of aid. The effect of loans is then negative and significant and at a 5% level in both specifications. The effect of grants is significant and positive at the 10% level when grants include debt relief, while positive but not significant when debt relief is excluded. Across different regressions in column 5-8 the effect of grants is ambiguous. One might question what the effects of loans and grants are when only linear effects are allowed for. These estimations (not presented in the table), showed a negative result for loans and a positive result for grants, both in- and excluding debt relief. These results were not significant at the 10% level.

Although the aid coefficient estimates in column 5 are not significant, it is interesting to have an interpretation of the rather abstract semi-logarithmic regression outcomes. For the interpretation we assume an increase of one percent point in aid disbursements to GDP. For this increase in loan disbursements it implies a decrease of 0,00327 percent in tax revenues to GDP in the recipient country. For the same increase in grants this implies a decrease of 0.00088 percent in tax revenues. So a one percent point increase in the form of a loan implies a larger decrease in tax revenues than a similar increase in grants implies. However, the sizes of both effects are considered to be small. Gupta et al. (2003, p.10) demonstrate an interpretation of their results in which a doubling of average aid levels is assumed. To be able to compare their interpretation straight away, we provide an interpretation like them as well19_{. A doubling of average loan disbursements to GDP of 0.136% implies} a decrease in tax revenues with 0.00062 percentage point of GDP20_{, whereas a doubling of average} grant disbursements to GDP of 1.032% imply a decrease of tax revenues with 0.013 percentage point of GDP. When glancing at the latter interpretation method, it seems that grants have a larger impact on tax revenues than loans, but this originates from the fact that the average amount of loans is substantially lower than of grants and in the interpretation we assumed doubling of average aid amounts.

19_{The aid averages are retrieved from the sample underlying the regressions in table 2. Furthermore average} tax revenues to GDP are 14.397%.

20_{Calculation of this number is done according to the formula: [x(β}

1+ 2* β2*x)]*average tax revenues to GDP. X represents average loan disbursements. For loans the following is calculated: [0.136 *(-0.000313 +

(24)

Table 2: Effects of IDA-aid on tax revenues

Notes: clustered standard errors are given in parentheses. Aid is lagged for one year and results are based on fixed effects estimations. (***), (**), and (*) denote significance at the 1, 5 and 10 percent levels, respectively. All regressions include country fixed effects. Column 8 additionally includes time fixed effects.

Table 3 provides results of estimations in which loans, grants including debt relief and grants excluding debt relief are estimated in isolation. Again non-linear effects are allowed for and this is estimated in two different ways: by including squared aid values and by including logarithmic aid values. The different forms of aid are regressed in isolation to check robustness of the results in table 2. Another specific characteristic of the results in table 3 is that aid disbursements from the IDA are regressed versus all ODA excluding IDA-aid. This is done to be able to compare the results for IDA-aid with aid

Grant incl. debt relief (5) Grant incl. debt relief (6) Grant excl. debt relief (7) Grant excl. debt relief (8) Loan to GDP -0.00313 -0.00211 (0.00474) (0.00186) [Loan to GDP]² -0.00007 -0.00004 (0.00092) (0.00005) log(loan to GDP) -0.04356*** -0.02107** (0.01577) (0.00788) Grant to GDP -0.00092 0.00809 (0.00430) (0.00671) [Grant to GDP]² 0.00002 -0.00012 (0.00007) (0.00012) log(grant to 0.01314* 0.01287 GDP) (0.00763) (0.00811) AGR -0.00305 0.00285 -0.00286 0.00325 (0.00300) (0.00261) (0.00303) (0.00288) IND -0.00081 -0.00110 -0.00089 0.00107 (0.00291) (0.00315) (0.00289) (0.00428) Real GDP/ capita 0.00007** 0.00009 0.00007** -0.00002 (0.00003) (0.00008) (0.00003) (0.00010) Trade 0.00108** 0.00194*** 0.00111** 0.00134** (0.00043) (0.00054) (0.00042) (0.00065) Constant 0.94915*** 0.81156*** 0.94132*** 0.88090*** (0.12040) (0.13859) (0.11923) (0.16471) Observations 485 219 485 182 Countries 57 41 57 39 R squared 0.14288 0.27429 0.14609 0.25168 F statistic 118.69355 6.58331 645.92029 P value 0.000 0.00007 0.00000

(25)

from other donors and it is achieved by including IDA-aid and aid from other donors in the same regressions. It is understandable that there is some correlation between aid from other donors and IDA-aid, since those countries are the ones which need aid and do not receive aid from the IDA only. The statistics show that there is indeed imperfect multicollinearity between aid from the IDA and aid from all donors (minus IDA)21_{, but according to Stock and Watson this does not prevent estimation of} the regression nor does it imply a logical problem with the choice of the aid variables in one estimation. However, it could be that one or more regression coefficients are estimated imprecisely (2012, p.241). As a check the IDA-aid only is regressed too, and the results were similar for IDA-aid in sign, sizes and significances.22_{The results for IDA loans (see column 9 and 10) are similar to those in} table 2: negative for loans, squared loans and logarithmic values of loans. When the effect of IDA loans are compared with aid by other donors, both in logarithmic values in column 10, it is remarkable that IDA loans have a significant negative effect on tax revenue while loans by other donors show an insignificant positive effect on tax revenue. It is not clear why logarithmic values of IDA loans have a negative effect on tax revenues, whereas ODA loans a positive effect, although the latter is insignificant.

The results for IDA grants show consistently a positive effect and squared IDA grants a consistently negative effect, for both grants including and excluding debt relief. This indicates that IDA grants tend to have a positive effect on tax revenue, but when the amount of IDA grants becomes larger, it has a negative effect on tax revenue. In addition IDA grants in logarithmic value show a positive effect on tax revenues, but it is not significant at the 10 percent level. When we compare the results of IDA-aid to aid by all donors but the IDA we can observe that the effect of grants excluding debt relief in logarithmic value is not significant for IDA, but it significant for grants by all donors (column 14). Grants excluding debt relief are not significant for IDA but it is for grants by all donors. In addition different specifications are regressed, where the dependent variable tax revenue is not in logarithmic value, but just in linear value. The signs of the effects of loans and grants do not change. Furthermore it should be noted that the effects of aid in non-linear estimations were similar to the results reported in column 9, 11 and 13.

21_{Collinearity statistics between aid from the IDA and aid from other donors are: 0.5818, 0.3761 and 0.2808} for loans, grants including debt relief and grants excluding debt relief, respectively.

22_{With the only exception of an estimation as in column 9, but then with IDA-aid only: IDA loans were} significant at the 5% level when aid by all donors was excluded.

(26)

Table 3: Effects of IDA-aid versus all ODA excluding IDA-aid on tax revenues Loan (9) Loan (10) Grant incl. debt relief (11) Grant incl. debt relief (12) Grant excl. debt relief (13) Grant excl. debt relief (14) IDA-aid to GDP -0.00240 (0.00258) 0.00188* (0.00108) 0.00600 (0.00791)

All ODA excl. IDA to GDP -0.00244** (0.00100) 0.00020 (0.00183) -0.00009 (0.00013) [IDA-aid to GDP]² -0.00008 -0.00002 -0.00009 (0.00010) (0.00002) (0.00013)

[All ODA excl. IDA to GDP]² -0.00002 (0.00002) -0.00000 (0.00001) -0.00005 (0.00012) log(IDA-aid to -0.02079* 0.01053 0.00707 GDP) (0.01098) (0.00715) (0.00908) Log(all ODA 0.00479 0.04273 0.09585** excl. IDA to GDP) (0.01052) (0.02975) (0.04287) AGR -0.00274 -0.00393 -0.00310 0.00207 -0.00270 0.00176 (0.00298) (0.00388) (0.00304) (0.00222) (0.00307) (0.00224) IND -0.00103 0.00185 -0.00086 0.00218 -0.00095 0.00256 (0.00288) (0.00357) (0.00289) (0.00330) (0.00287) (0.00358) Real GDP/ capita 0.00007** 0.00008** 0.00008** -0.00002 0.00007** -0.00001 (0.00003) (0.00004) (0.00003) (0.00011) (0.00003) (0.00010) Trade 0.00115*** 0.00102* 0.00108** 0.00087** 0.00121** 0.00069* (0.00040) (0.00052) (0.00043) (0.00033) (0.00046) (0.00036) Constant 0.96157*** 0.91808*** 0.96409*** 0.81840*** 0.96709*** 0.82212*** (0.11497) (0.13581) (0.05010) (0.13662) (0.13378) (0.14118) Observations 485 314 485 264 485 220 Countries 57 53 57 43 57 41 R squared 0.14823 0.21209 0.14125 0.28167 0.14550 0.21984 F statistic 3.20233 2.79899 8.63550 6.48121 448.39224 P value 0.00460 0.01954 0.00000 0.0000 0.00000

Notes: clustered standard errors are given in parentheses. Aid is lagged for one year and results are based on fixed effects estimations. (***), (**), and (*) denote significance at the 1, 5 and 10 percent levels, respectively. All regressions include country fixed effects. Columns 12 and 14 additionally include time fixed effects. All ODA variables are excluding IDA-aid.

Across all regressions in tables 1-3 our main findings are that IDA loans have a negative effect on tax revenues. The functional form in which IDA loans are regressed does not alter the sign of the effect. The estimations in which loans showed a significant effect, were in those were the logarithmic value was regressed only. Grants have the tendency to have a positive effect on tax revenues, but this is not consistent across all estimations in tables 1-3. Also the results for grants were significant in the estimation results presented in column 6 (logarithmic value) and 11 (linear value) only. In addition the sign of the effect of IDA grants did not depend on its in- or exclusion of debt relief. Although we were not able to retrieve clear significant results for both IDA loans and IDA grants, the results indicate

(27)

that IDA loans tend to have a negative effect on tax revenues, whereas IDA grants tend to have a positive effect on tax revenues. These results connect to the reasoning of Morrissey et al. (2006), who state that grants do not increase a country’s indebtedness and this in turn might support a stable fiscal policy. The opposite can be argued for loans. All in all, there is no robust evidence that IDA-aid, in the form of either loans or grants have a significant effect on tax revenues. Furthermore, the effect of IDA-aid on tax revenues deviates from the effect of IDA-aid by other donors in some specifications, while in other specifications the effects of IDA-aid and aid by other donors did not differ. Therefore we postulate that it is ambiguous whether the effect of IDA-aid on tax revenues is different from the effect of aid by other donors.

V. Concluding remarks

This study aimed at estimating the effects of IDA loans versus IDA grants on tax revenues. Authors of several studies feared that grants reduce the incentive to collect taxes of the recipient country. Some empirical research has showed that grants indeed negatively affect tax revenues. But, this causal relationship between aid inflows and reduced tax revenues was questioned by some studies as well. Morrissey et al. (2006) even argue that grants reduce indebtedness of recipient countries, which promotes economic growth and this in turn helps a country to build a stable fiscal policy. These studies focussed on ODA in general, but not for a specific institution like the IDA. This potential differential effect in sign and size of IDA-loans versus IDA-grants was estimated by means of a fixed effects estimation. The dataset comprises 59 IDA-eligible countries for the years 2000-2012. Potential simultaneous causality between aid and tax revenues was addressed by the use of one year lagged aid values. Overall, IDA loans show a negative effect on tax revenue, while IDA grants show a positive effect on tax revenue for grants both in- and excluding debt relief23_{. This was not in line with our} expectations, since no differential effect in sign and size of IDA-loans versus IDA-grants were expected. However, for most of the performed estimations we are not able to produce significant effects, except for estimations including logarithmic values of aid to allow for non-linear effects instead of squared aid values. Since we did not produce significant and robust results, there is not enough evidence that IDA-loans versus IDA-grants have an effect on tax revenues. The reason why we were not able to produce significant results remains guessing. One possibility is that the IDA imposes conditions on its funding and these are the same for loans and grants. If IDA-conditions cover tax revenues, this could be an explanation. In line with Clist and Morrissey (2011, p.168) we argue that one should be cautious to drawing conclusions on policy implications when assessing the above described empirical results.

23_{Only one estimation, in which both loans and grants including debt relief are regressed showed a negative} effect for IDA grants, but this was not significant.

(28)

McGillivray and Morrissey (2001, p.18) warn for attaching to much weight to cross-country studies, because the impact of aid might differ by country. We agree with McGillivray and Morrissey that one should be careful interpreting the results of this study at this stage. When one wants to estimate more precise the effect of IDA-loans versus IDA-grants it is better to study a specific country (Gupta et al., 2003, p.4). Another improvement for studying this topic is addressing endogeneity in another way than the use of one year lagged aid variables. One example is the estimation method of Benedek et al. (2012), who use GMM. Alternatively, aid could have been lagged for more years. To the knowledge of the author there is no study were committed aid amounts are used, instead of disbursed aid, so using committed aid instead of disbursed aid could be an improvement too.

(29)

References

AidWatch. (2008). What is multilateral aid? Retrieved May/20, 2016, from

http://www.aidwatch.org.au/stories/what-is-multilateral-aid/

Benedek, D., Crivelli, E., Gupta, S., Muthoora, P. (2012). Foreign aid and revenue: Still a crowding out effect?(WP/12/186), May/20.

Carter, P.W., (2013). Does Foreign Aid Displace Domestic Taxation?. Journal of Globalization and

Development, vol. 4., pp. 1-47

Cassimon, D., van Campenhout, B. (2008). Comparative Fiscal Response Effects of Debt Relief: An Application to African HIPCS, South African Journal of Economics 76(3), pp. 427-442

Clist, P. (2016). Foreign Aid and Domestic Taxation: Multiple Sources, One Conclusion, Development Policy Review 34(3), pp. 365-383

Clist, P., Morrissey, O. (2011). Aid and tax revenue: Signs of a positive effect since the 1980s. Journal

of International Development 23, 3(2), 165–180.

Combes, J.-L., Ouedraogo, R., Tapsoba, S. J.-A. (2016): Structural shifts in aid dependency and fiscal policy in developing countries, Applied Economics 1-21, DOI: 10.1080/00036846.2016.1158920

OECD (2008). Is it ODA? Retrieved May/20, 2016, from

http://www.oecd.org/dac/stats/34086975.pdf

Ghura, D. (1998). Tax revenue in Sub-Saharan Africa: Effects of economic policies and corruption.

IMF Working Paper no. 98/135,pp. 1-25.

Gupta, A. S. (2007). Determinants of tax revenue efforts in developing countries. IMF Working Paper

07/184

Gupta, S., Clements, B., Pivovarsky, A., Tiongson, E.R. (2003). Foreign aid and revenue response: Does the composition of aid matter? IMF Working Paper 03/17,

(30)

International Development Association. (2007). Aid architecture: An overview of the main trends in

official development assistance flows. Retrieved May/20, 2016, from

http://www-wds.worldbank.org/external/default/WDSContentServer/WDSP/IB/2007/02/23/000310607_20 070223093528/Rendered/PDF/387500idasecm200710103core.pdf

International Development Association. (No date). Borrowing countries. Retrieved May/20, 2016, from http://ida.worldbank.org/about/borrowing-countries

International Development Association. (No date). Debt sustainability & grants Retrieved May/20, 2016, from http://ida.worldbank.org/financing/debt-sustainability-grants

International Development Association. (No date). What is IDA? Retrieved May/20, 2016, from

http://ida.worldbank.org/about/what-ida

Khan, H. A., Hoshino, E. (1992). Impact of foreign aid on the fiscal behavior of LDC governments.

World Development, 20(10), pp. 1481-1488.

Krugman, P.R. (1988). Financing vs. Forgiving a Debt Overhang. Journal of Development of Economics, 29, 253-268.

MacGillivray, M., Morrissey, O. (2001). Fiscal effects of aid. WIDER Discussion Papers, World Institute

for Development Economics (UNU-WIDER), (No. 2001/61)

Morrissey, O., Islei, O., M’Amanja, D. (2006). Aid loans versus aid grants: Are the effects different?

CREDIT Research Paper, no. 06/07, Centre for Research in Economic Development and

International Trade University of Nottingham,

Odedokun, M. (2004). Economics and politics of official loans versus grants: Panoramic issues

and empirical evidence. World Institute for Development Economics (UNU-WIDER), no. 2003/04,

(31)

Schmidt, W. E. (1964).

The economics of charity: Loans versus grants Journal of Political Economy, 72(4), pp. 387-395.

Stock, J. H., Watson, M.M. (Ed.). (2012). Introduction to econometrics (Third Edition ed.). Harlow: Pearson.

United Nations Development Programme. (No date). A sustainable development agenda. Retrieved May/20, 2016, from http://www.undp.org/content/undp/en/home/sdgoverview.html

World Bank. (2005). Review of World Bank Conditionality Retrieved June/28, 2016, from

http://siteresources.worldbank.org/PROJECTS/Resources/40940-1114615847489/ConditionalityFinalDCpaperDC9-9-05.pdf

World Bank. (2013). HIPC & MDRI Retrieved May/20, 2016, from

http://siteresources.worldbank.org/INTDEBTDEPT/Resources/468980-1256580106544/HIPC_Fall2013_ENG_CRAweb.pdf

The effect of IDA loans versus IDA grants on tax revenues