University of Amsterdam
Faculty of Economics & Business
The political economy of banking crises and aid disbursements
A LSDV and PVAR approach
Master’s Thesis Wessel van Proosdij 10244670
15/08/16
MSc Economics: Monetary Policy, Banking & Regulation Supervisor: prof. dr. S.J.G. van Wijnbergen
Statement of originality
This document is written by Wessel van Proosdij 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.
Table of Contents
Abstract ... 4
1. Introduction ... 4
2. Aid overview ... 6
3. Existing studies ... 10
4. Theoretical framework and hypotheses ... 14
5. Data ... 18
6. Econometric approach ... 19
6.1. LSDV approach ... 19
6.2. PVAR approach ... 21
7. Results and analysis ... 24
7.1. LSDV approach ... 24
7.2. PVAR approach ... 37
8. Conclusions ... 41
References ... 42
Abstract
This study analyses the effects of banking crises on the amount of aid supplied by donors, using both a Least Squares Dummy Variable (LSDV) and a Panel Vector Autoregression (PVAR) approach. We estimate how banking crises in donor countries negatively affect their aid disbursements, using panel data from 24 countries between 1960 and 2014. We find that banking crises in donor countries cause substantial falls in aid flows, between 15 and 40 per cent relative to the counterfactual. The negative effects of a crisis last until around twelve years after the onset of the crisis, while the amount of aid is only restored relative to the counterfactual after more than twenty years. We argue that fiscal costs are the main driver behind these negative effects. Furthermore, our results imply that the global financial crisis so far has a larger negative effect on aid than previous crises had. In addition, we find no support that the political ideology of a donor’s government affects aid supply, that donors protect social and humanitarian aid, and that donors protect aid to the poorest of countries during a banking crisis, in contrast to most of the literature. Our findings are robust to changes in sample, period, and the addition of political and macroeconomic variables.
1. Introduction
The supply of development aid is primarily considered from the perspective of recipient countries. Less analysis has been devoted to supply-side factors that influence aid disbursements made by the donor countries. However, since the emergence of the global financial crisis of 2008-09 and pressures on government debt the supply-side of aid has taken on renewed importance. (Jones, 2015: 31) The goal of this study is to identify the effects of banking crises on these so-called supply-side factors that affect aid budgets in donor countries. In order to successfully identify the effects of banking crises, we use two approaches. First, using panel data, we apply a least squares dummy variable (LSDV) method to obtain the effects of financial crises on the amount of aid supply. In the second part a panel vector autoregression (PVAR) approach is adopted to forecast the dynamics of several macroeconomic shocks on aid supply during normal and difficult economic times.
This paper extends the literature in several important respects. First of all, a completely new dataset is constructed with a longer time-period (1960 – 2014) to cover more information, including more recent data of the global financial crisis. The used dataset is among others, composed of an extension of the banking crisis dataset of Laeven & Valencia (2013). Second, the effects of the supply-side factors are not only researched on total aid
volumes, but also on various sectors and destinations of aid. This is useful since donors may protect certain types of aid during a crisis and the various supply-side factors may also have distinctive effects on the different sectors of aid. Third, since the decision how much aid to distribute is also a political one, several political indicators will be analysed to research the relation between these indicators, banking crises and aid supply. Fourth, the panel regressions will be subjected to several robustness tests by adding additional control variables and dropping potentially influential observations (donors or years). Lastly, the panel vector autoregression (PVAR) approach is seldom used before to simulate shocks to aid supply. This should therefore lead to new insights in the way shocks affect aid supply.
Several disclaims are necessary to explicate what is not analysed. This study does not address questions regarding where aid is allocated or whether it is effective. Furthermore, no attempt is made to analyse trends in the aid supply of non-DAC donors (such as China), philanthropic organisations or other private institutions and multilateral institutions. These are excluded for reasons of data limitations. Lastly, the main focus of this research is on the aggregate behaviour of donors rather than donor-recipient relations.
This research is structured as follows. First, a short overview of the relative aid literature, concepts and history is given to act as introduction to the topic. The second section is devoted to existing studies that are relevant for our own for reasons of findings or methodology. Third, a theoretical framework is constructed to aid us in testing our hypotheses. Fourth, a brief explanation of the used data is given. The fifth section is dedicated for our methodology in which we describe our econometric methods. The rest of our research is devoted to analysing the obtained results and concluding our research.
Our findings show that a financial crisis reduces aid supply between 15 and 40 per cent relative to the counterfactual. These negative effects last until around twelve years after the onset of the crisis, while the amount of aid is only restored relative to the counterfactual after more than twenty years. We argue that fiscal costs are the main driver behind these negative effects. We also find that the global financial crisis has a larger negative effect on aid supply than previous banking crises had. Furthermore, we find that political ideology has no effect on the amount of aid supply during a crisis using both the LSDV and PVAR approach. Our estimates suggest further that social aid has declined more than other sectors of aid, which goes against most of the literature. Finally, we find no evidence that donors protect aid for the poorest countries, which again contradicts what is found in most of the literature. We therefore strongly recommend further research regarding our findings.
2. Aid overview
Development aid, foreign aid, development assistance or simply ‘aid’ can be defined in several ways. The most comprehensive definition is the one of the Development Assistance Committee (DAC) of the Organisation of Economic Cooperation and Development (OECD). This definition is the most accepted and shall also be used in this study. The DAC uses the definition of Official Development Assistance (ODA), which is defined as “aid provided by official agencies that is administered with the promotion of the economic development and welfare of developing countries as its main objective”. The DAC was founded in 1960 in order to “consult on the methods for making national resources available for assisting countries and areas in the process of economic development and for expanding and improving the flow of long-term funds and other development assistance to them”. It consists of 28 member states plus the European Union, who is an official member. (OECD, 2016a) ODA itself can be divided into several channels: bilateral aid (flows directly from donors to recipients) and multilateral aid (flows provided via multilateral agencies (e.g. the World Bank or the United Nations) who then distribute it to developing countries). Thus, ODA does not contain aid flows between private actors (corporations, citizens) and recipients and therefore private aid is not researched in this study.
ODA can be divided into net and gross ODA disbursements. Gross disbursements are the amount a donor actually spends in a given year. This figure becomes net once repayments of the principal on loans made in prior years (without interest) are taken into account, as well as offsetting entries for forgiven debt and recoveries made on grants. (OECD, 2016a) Since net ODA reflect the true impact of aid, the use of net ODA is more widespread. In order to fairly compare countries with each other, ODA as percentage of gross national income (GNI) is a common measure. The ODA to GNI ratio connects the national income with the amount of aid disbursements. Since this method leads to fair comparisons, a target level has been proposed. DAC members accepted a proposal on a target of 0.7% of GNI within the United Nations in 1970, at least as long-term objective. (Clemens et al, 2007) Due to a revision of the System of National Accounts in 1993, GNI has been the most used concept for national accounting, but a ratio of ODA and gross domestic product (GDP) is similar.
The allocation of ODA can be broken into different sectors. In figure 1 the sectorial allocation of Dutch ODA over the period 2010 – 2014 is showed. This figure illustrates that the Dutch provide more aid towards social sectors (e.g. health and education) and less towards economic infrastructure and the production sector. The Netherlands probably stimulate the social sectors in order to support the living conditions in recipient countries.
Figure 1: Sectorial allocation of ODA of the Netherlands (2010 – 2014)
Source: Aidflows.org
As can be seen in figure 2, net ODA rose steadily since 1960 until the start of the 1990s, where a period of ‘aid fatigue’ appeared. Since then, net ODA have been rising
steadily until 2007 when it stabilised. Summarizing, the ODA to GNI ratio has been declining since the start of the DAC from 0.5% down to 0.2% during the ‘aid fatigue’ period with an increase to 0.3% during the start of the 2000s.
Figure 2: DAC ODA over the period 1960 - 2014
Source: Author’s graph, data from OECD (2016)
0 20 40 60 80 100 120 140 160 0 0.1 0.2 0.3 0.4 0.5 0.6 DA C O DA in bi lli on s US D DA C O DA as pe rc en tag e of G NI
DAC ODA over the period 1960 - 2014
Although total ODA flows always consisted of dozens of billion US dollars (USD), the initial idea was that donors should achieve the 0.7% UN target by the mid-1970s, but the donors have never been near the target level of 0.7%, with a current level of 0.3% being very far off. The target has therefore been repeatedly postponed over the course of history.
During the 1990s, several factors have contributed to the deviation from trend: the aid fatigue period mentioned before. One of these factors was the frustration with the poor record of the developing and poverty-reducing objectives of aid. (Bird, 1999) According to Boone (1994 & 1996), foreign aid does not lead to economic growth nor benefits the poor nor increases investment. These ideas, supported by low growth rates for many developing countries and the increase in importance of private capital markets led to a reduction of aid supply of many DAC members. (Bird, 1999: 1 & 15) Another important factor was the end of the Cold War and the fall of communism. Western countries did not feel the need any longer to provide aid as a means to discourage the spread of communism. This changing political economy further contributed to the aid fatigue period of the 1990s. (Bird, 1999: 12 - 13) In the new century, aid quickly grew to unprecedented levels on account of the Millennium Agreement. However, the aid to GNI ratio only increased slightly.
Even though the main focus of this study is on the aggregate behaviour of the DAC members, this small section is dedicated to cross-country differences. When these differences between DAC members are analysed, it is clear that the United States (US) has the largest aid budget of all DAC members (see figure 3). Together with other large economies such as the United Kingdom, Germany and France they provide more than 50% of all net ODA flows. Small economies such as Iceland, Slovenia and Slovakia only provide several millions. The total ODA budget of the DAC members was 137.2 billion in 2014, which was the largest amount to date. (OECD, 2016a) Interestingly, if the ODA to GNI ratio is used (see figure 4) the US falls to the nineteenth place, comparable with European countries such as Italy and Spain that had strong sovereign debt crises. Only five countries have reached the target level of 0.7% in 2014; Sweden, Luxembourg, Norway, Denmark and the United Kingdom, with the Netherlands, Finland, Switzerland, Belgium and Germany as the others who reached above the average of 0.39%. Concluding, the variations in aid supply across time and countries is large and very interesting to analyse. As stated, these cross-country differences are not the main focus of this study and will only return in the form of small details. The cross-time differences however are something that will get analysed in the remaining of this research.
Figure 3: Net ODA in 2014 – in billions USD
Source: Author’s graph, data from OECD (2016)
Figure 4: Net ODA in 2014 – as percentage of GNI
Source: Author’s graph, data from OECD (2016)
33.10 19.31 16.57 10.62 9.27 6.23 5.57 5.09 4.38 4.24 4.013.52 3.00 2.45 1.881.861.63 1.23 0.82 0.510.450.430.420.250.210.08 0.06 0.04 0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 40.00 Net O DA in bi lli on s US D
Net ODA in 2014 - in billions USD
137.2 1.09 1.06 1.00 0.86 0.70 0.64 0.59 0.50 0.46 0.42 0.38 0.37 0.31 0.28 0.27 0.24 0.22 0.19 0.19 0.19 0.19 0.13 0.13 0.12 0.110.110.09 0.09 0.30 0.00 0.20 0.40 0.60 0.80 1.00 1.20 O DA as % of G NI
Net ODA in 2014 - as percentage of GNI
UN target 0.70
3. Existing studies
The literature regarding development aid is incredibly extensive. A lot of research has been conducted regarding the motives of distributing aid, to whom it is distributed and, extremely popular, its effectiveness. A great part of this literature is focused on the recipient side of development aid. A common question is whether aid leads to economic growth in recipient countries. Chenery & Strout (1966), Papanek (1973), Burnside & Dollar (2000), Hansen & Tarp (2001) and Minoiu & Reddy (2010) find in their research that aid can lead to economic growth. In contrast, Friedman (1958), Bauer (1971), Levy (1988), Boone (1994 & 1996), Adam & O’Connell (1999) and Rajan & Subramanian (2011) all find that there are serious impediments to be overcome before aid can lead to sustainable economic growth.
One important impediment for such sustainable economic growth is the volatility of aid flows. Aid volatility is the degree in which aid flows fluctuate on a year-to-year basis and this volatility is very large. (Bulíř & Hamann, 2003) The literature regarding the impact of aid volatility on aid recipients is massive. Lensink & Morissey (2000), Bulíř & Hamann (2003, 2008), Kharas (2008), Hudson & Mosley (2008), Fielding & Mavrotas (2008), Chervin et al (2013) all researched the volatility of aid flows and found that it negatively affects economic growth in recipient countries. However, some scholars have researched donor-recipient dynamics, such as past colonial relationships, that play a role in the dynamics of aid flows. Alesina & Dollar (2000), Rajan & Subramanian (2011) and Chervin et al (2013) find that donors are willing to distribute more aid if a recipient country is a former colony and if the donor’s macroeconomic conditions are more favourable. However, the main focus of these studies is on the recipient side of aid.
Nonetheless, a small part of the literature has been researching the donor’s decision-making process regarding the allocation of aid. Beenstock (1980) was one of the first to adopt an econometric analysis regarding donor’s aid budgets. Beenstock analyses the effect of changes in several macroeconomic variables on the amount of total world aid. By using panel data, Boonstock finds that if GDP increased by $1000, aid would increase with $3.37, leading to a marginal propensity of 0.33%. If the government balance would increase by $1000 aid would increase by $9.51, but if the government balance would decrease by $1000, aid would fall by $17.20. Furthermore, Beenstock finds negative effects of the unemployment rate and total population on aid budgets, but the effects seem overestimated.
A follow-up study was conducted by Mosley (1985). He argues that Beenstock’s results are not based on any theoretical framework, making it difficult to interpret the robustness of the results. Mosley’s approach is to build a model which treats foreign aid as a
public good on a market. The demand for this good is based on donor countries taxpayers who pay for the aid with a part of their tax bills. The supply is based on the government’s desire to obtain strategic or trading benefits from aid. Mosley adopts this model and shows that it is consistent with the data. Aid budgets are dependent on government budget processes which are path dependent and based on information of earlier years and on the interaction between governments and their people. Furthermore, donor countries may deviate from their desired level of foreign aid due to budgetary pressures and the behaviour of other donors. This last one, the bandwagon effect, means that other donors may pressure individual members into increasing their aid budgets.
Round & Odedokun (2004) conduct a panel regression of 22 DAC members over the period 1970 – 2000 and find more or less the same results as Beenstock. However, Round & Odedokun argue that the decision how much aid to distribute is a mere political one. Therefore, it is important to control for political indicators. In their research, Round & Odedokun use the political orientation of the government, the amount of checks and balances of the government and the amount of polarization within the government itself. They find significant effects of left-wing governments that increase their aid budgets, of checks and balances leading to more aid, and polarization to a decrease in aid efforts. Furthermore, they find a positive relationship between government military spending and aid effort.
Since the emergence of the financial crisis of 2008-09 scholars have shifted increasingly attention towards donor countries and the effects of macroeconomic shocks and financial crises on their aid budgets. One of the first that expressed his warnings that aid is expected to fall during the recent financial crisis was Roodman (2008). Roodman calculates that after every financial crisis aid budgets fell (sometimes substantively). For example, after the Nordic financial crisis of 1991, aid from Norway, Sweden and Finland fell by 10%, 17% and 62% respectively. Roodman therefore expected that the financial crisis that emerged in 2008 would also lead to a fall in aid disbursements.
Mendoza et al (2009) research the effects of adverse economic and financial conditions on the amount of aid supply of the United States. They find that US ODA tends to decline as its economic conditions worsen. A 1 percentage point increase in unemployment and inflation leads to a 0.01 percentage point decline in US aid. Furthermore, an increase in financial volatility from 1 to 2 per cent is associated with a decrease in US ODA by about $2.78 billion. Mendoza et al argue that the financial crisis from 2007 could lead to a potential decline in US aid of anywhere between 13 to 30 per cent, depending on the severity of the economic conditions in the financial crisis.
A more complete study was conducted by Frot (2009). Frot uses two methods to analyse the effects of financial crises on aid budgets. First, he uses past financial crises to divide aid donors in two groups; countries who experienced financial crisis in the past (the treatment group), and countries who did not (control group). Using this method, he finds the short-run effects of financial crises on aid budgets, which is that a banking crisis in a donor country decreases aid by 13% (level effect) or that aid falls by 5% (trend effect) per year after the crisis. Second, he uses a vector autoregression (VAR) with GDP, unemployment and budget deficits to research the long-run effects of crises on aid budgets. He finds that five years after a negative GDP shock of 1 per cent aid budgets fall by 8 per cent.
Gravier-Rymaszeska (2012) uses a panel VAR (PVAR) approach to review the short-run effects of macroeconomic shocks on the aid supply of donors. The main finding is that aid strongly falls on mid-term after a GDP shock. Furthermore, aid budgets tend to fall not immediately during a recession, but only after a while. After a crisis, aid grows back to pre-crisis levels. Crisis are found to affect aid budgets and their trend through two channels: directly through lower revenues and indirectly by increasing fiscal costs through exchange rates and financial volatility. In addition, Gravier-Rymaszeska argues that the donor’s internal political orientation also plays an important role. Right-wing and centrist governments tend to decrease aid when the economy is affected by a shock, while left-wing governments tend to protect aid.
Dang et al (2013) use a panel regression of 24 donor countries between 1977 and 2010 to research the results of banking crises on aid flows. They find that (in accordance with Frot (2009)), aid tends to fall a lot after a crisis. However, they even find a stronger effect of banking crises on aid flows of 28%. Furthermore, they find that a crisis bottoms out only after a decade after the onset of the crisis, which sounds remarkably long. Their explanation is based on research of Reinhart & Rogoff (2013). Their research finds that banking crises have large negative fiscal effects. Not only do government revenues fall, governments also have to bail out banks and additional expenditures to address the economic effects of the crisis. As a result, public debt increases which leads to large debt overhang. These debt pressures lead governments to cut aid budgets for years after the emergence of the crisis.
Jones (2015) argues that within the existing literature of donor’s aid decisions an important number of gaps remain. First, many studies focus on one or multiple specific aspects of aid supplies with very little cross-fertilization. Therefore, the interaction between longer- and shorter-term factors have been ignored. It is important to disentangle these interactions since aid decisions often involve multi-year commitments. Consequently, in line
of Mosley (1985), aid flows are likely path dependent and reflect information sets of previous budget allocations. Existing commitments limit the extent to which aid volumes might fully adjust to macroeconomic shocks. Second, there is heterogeneity in aid supply behaviour between donors and years. Different countries may place different weights on their long- or short-term goals. Also, since global geopolitical conditions are constantly changing, they may affect countries differently over time. Third, aid supplies are often treated as stationary, which is not necessarily the case. In order to resolve the last two issues Jones develops a framework that allows for heterogeneity between countries in their long-run aid targets, variations through time, and short-run variation around these trends. Jones manages this by adopting an error-correction model with an explicit distinction between long and short-run determinants. The long-term effects are determined by deviations of aid of the periods before. Short-run movements are driven by difference terms and random noise. By using this clear distinction, the model can account for heterogeneity. Furthermore, variables are made stationary by using first differences. (Jones, 2015: 32 – 34)
Several attempts have been made to employ the VAR method to the aid literature. One of the first were Osei et al (2005). They apply VAR to 34 years of annual data in Ghana to model the effects of aid on fiscal behaviour. They find that aid to Ghana has been associated with reduced domestic borrowing and increases in tax efforts. Their research therefore provides evidence that aid has been used sensibly since the fiscal performance of Ghana improved. In a comparable study, Morrissey et al (2006) find that government spending has a positive long-run influence on growth and that there is no evidence that taxes retard growth. It is therefore necessary that aid is associated with fiscal discipline.
Hansen & Headey (2010) apply a PVAR to investigate the short-run macroeconomic impact of aid in small developing nations on imports and domestic demand. They divide the developing nations in two groups to create a counterfactual: aid-dependent countries and countries that depend on natural resources or tourism. Their results show that in aid-dependent countries imports equal domestic demand, while only half of the aid flow is adsorbed and spent. In the non-aid-dependent group, aid is not spent in any systematic fashion at all.
Juselius et al (2011) use a cointegrated VAR (CVAR) approach to study the long-run effects of aid on several macroeconomic variables in 36 sub-Saharan African countries. The CVAR approach is chosen since it provides broad confidence intervals within which empirically relevant claims can be made. Their results show that aid has a positive effect on investment in 33 of the 36 researched countries.
4. Theoretical framework and hypotheses
Before the data and methodology can be described, it is necessary to construct a theoretical framework in which the government’s decision to supply aid is operationalized. In this framework, the decision of a government ! to supply aid in period " can be seen as a function where # stands for aid supply that is dependent of a vector $ which reflects a list of endogenous economic and political factors that determine the amount of aid supplied:
#%,' = ) $%,' (1)
This function can further be specified since donors follow some kind of target level of aid. (Jones, 2015) Current aid supply #%,' is therefore also dependent on the aid supply #%,'*+ of previous periods and whether a banking crisis ,-%,' is present. Furthermore, since economic and political factors can have long-lasting effects as argued by Jones (2015), the vector $ from previous periods " − / affects current aid supply as well. Lastly, several exogenous factors 0%,' such as aid supply of other donors (bandwagon effect) or the economic growth in
recipient countries can additionally affect aid supply:
#%,' = #%,'*+,-%,' $%,' $%,'*+ 0%,' 2,3 %45 '45 2,3 %45 '45 2,3 %45 '45 (2) Per factor within $%,', $%,'*+ and 0%,', the partial derivative can be specified to find the
expected effect of that factor on the amount of aid supply of the donor. For example, the partial derivative of GDP on $%,' (and therefore also aid supply #%,') can be expressed as:
6$%,'
6789%,' > 0 (3)
It is thus expected that higher levels of a country’s GDP lead to higher levels of aid supply. This is logical, since a higher GDP means that more resources can be spend on aid. Aid supply is therefore expected to be procyclical: drops in GDP lead to a consecutive decline in aid. In addition, an increase in unemployment can also affect aid supply. Higher unemployment rates reflect economies in distress, leading to governments investing in their domestic economy instead of foreign economies. It is expected that higher values of unemployment lead to less aid supply. The partial derivative is therefore negative:
6$%,'
6<=>?@ABC?>="%,' < 0 (4)
For every independent variable the partial derivative can be specified to find the sign of their hypothesized effects on aid supply. The results can be found in appendix C. The signs of the partial derivatives are based on theory and previous literature.
The dynamics of aid supply during a financial crisis are important to operationalize, since they influence the necessary econometric approach. One might expect that aid falls as depicted in figure 5: an immediate decline in aid supply during a crisis, after which aid supply quickly returns to the old slope (trend). This would require testing for structural breaks in the amount of aid supply.
Figure 5: Immediate shock to aid supply
Source: Author’s graph
The above graph might not fit the data sufficiently looking at the literature. (Frot, 2009; Dang et al, 2013) It rather looks like a crisis changes the trend from positive to negative or flat, and that it only recovers to its pre-crisis slope after a couple of periods. This is depicted in figure 6. Since aid supply is heavily dependent on its previous levels, large changes are not observed during the course of a financial crisis.
Figure 6: Expected dynamics of aid supply during a crisis
Source: Author’s graph
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 1 2 3 4 5 6 7 A id S up pl y Time Crisis -> 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 1 2 3 4 5 6 7 A id S up pl y Time <- Crisis
Besides the empirics in the literature, there are certain theoretical arguments to why aid falls in certain ways, which are based on Reinhart & Rogoff (2013). Assume a government that runs a balanced budget with a sustainable level of debt. A government’s income is given by non-lump sum taxes EF%,' and a moderate amount of borrowing ,%,'. A government’s expenditure consists of two types: aid supply #%,' and all other non-aid expenditures G%,' (e.g. education or unemployment benefits). For these goods the government gains utility, depending on the preferences for both goods that is captured by the parameters H and I, and the elasticity of substitution J (similar to Chong & Gradstein, 2008):
< = 1
1 − JH#%,'5*L+ 1
1 − JIG%,'5*L (5)
A government’s primary deficit in a certain period " can be described as the difference between government expenditure (#%,'+ G%,') and income by taxes and borrowing (,%,' + EF%,') :
8>)!P!" = (#%,'+ G%,') − (,%,'+ EF%,') (6)
A banking crisis leads to a drop in productivity: national income F%,' declines, leading to a
decrease in tax revenue EF%,'. Furthermore, parameter I is increased due to a greater preference for non-aid goods in order to smother the effects of the crisis, which leads to more government spending for non-aid goods G%,'. The government’s primary deficit therefore
increases and in order to run a balanced budget, borrowing ,%,' needs to increase. This in turn
increases public debt:
8%,' = (1 + Q)8%,'*5+ (#%,'+ G%,') − (,%,'+ EF%,') (7) Depending on the severity of the crisis, this pattern can continue for a certain amount of periods, leading to an increasing amount of debt. This increase in debt raises the interest rate (1 + Q) on that debt, leading to a possible negative debt spiral and debt overhang. However, in order to reduce the budget deficit and consecutively the amount of debt, the government can decide to introduce certain austerity measures to reduce government spending #%,'+ G%,'. Since the parameter I has increased due to the economy’s distress, in order to maximize utility governments will reduce the amount of aid supply. How much aid is reduced depends on parameter H, the increase in G%,' and decrease in EF%,'.
This theoretical framework shows that it is unlikely that aid falls immediately after the emergence of a banking crisis: productivity falls after which a government is likely to stimulate the domestic economy. This leads to increases in the budget deficit and public debt in the following periods. In order to restrain debt becoming unsustainable, aid supply is being
cut. This happens smoothly since aid supply is strongly based on previous levels and can only change slowly over time. Furthermore, governments are not forward-looking and hope that the upcoming crisis is moderate so that no budgets cuts are necessary.
Having explored the literature and operationalized the theory behind the relation between financial crises and aid supply, the hypotheses for this research can be formulated. Our first hypothesis is that aid supply falls as depicted in figure 6 with percentages similar to the studies of Frot (2009) and Dang et al (2013), who find that aid supply falls with 10 to 30 per cent during a financial crisis:
H1: Aid supply falls with around 10 to 30 per cent relative to the counterfactual during a banking crisis, but only after a certain period.
The second hypothesis focuses on the expected differences between the global financial crisis of 2008-09 and banking crises before that. Hypothesized is that, even when controlling for GDP loss, the global financial crisis had a larger negative effect on the amount of aid supplied. This is due to the severity of the financial crisis that led to several sovereign debt crises and financial volatility in donor countries.
H2: When controlling for GDP loss, the great financial crisis of 2008-09 has a stronger negative effect when compared with previous banking crises.
Our third hypothesis is a more political one and based on Round & Odedokun (2004) and Gravier-Rymaszeska (2012) who find that left-wing governments protect aid during a crisis. When budgets cuts are necessary in order to stimulate the economy, left-wing governments tend to cut other services than aid, such as military or infrastructure.
H3: Left-wing governments will protect aid during a financial crisis, in contrast to right-wing and centrist governments who will not.
The fourth hypothesis is that a banking crisis has a smaller effect on aid supply towards former colonies and poor regions. Donors protect aid supply to regions that form historical ties, in order to protect these ties and countries in need, since they depend strongly on aid.
H4: A financial crisis has a smaller negative effect on aid supply to regions that are strongly in need of aid and on countries that are former colonies.
The fifth hypothesis is that a banking crisis has a less negative effect on social (e.g. health and education) and humanitarian (e.g. emergency aid and disaster relief) aid sectors. Since these sectors are more urgent than aid sectors that focus more on long-term structural goals such as economic and infrastructure aid, expected is that these will be protected.
H5: A financial crisis has a larger negative effect on non-social aid sectors than social and humanitarian aid sectors.
5. Data
The sample used in this research is composed of 24 countries within the Development Assistance Committee (DAC) from 1960 until 2014. Czech Republic, Poland, Slovakia and Slovenia are dropped due to a considerable amount of missing data (see appendix A for a complete list of countries included). Aid is measured as ODA, which, as stated before, refers to official assistance to developing countries and to multilateral institutions with regards to flows to these developing countries only. In order to successfully research the effects of banking crises on aid supply, an entirely new dataset is constructed. This dataset consists of data from a variety of sources, such as World Bank (2016), OECD (2016b & 2016c), the comparative political data set of Klaus et al (2015), and data on banking crises from the dataset of Laeven & Valencia (2013). Appendix B provides a detailed list of variables and sources for the data used in this research. In appendix D the summary statistics of the included variables are given.
Since the data of the comparative political data set runs up to 2013, an extension was made for the year 2014. For the government debt and government deficit variables this is realized by extrapolating using data from the OECD. Since the data of the OECD is extremely similar to the data in the comparative political data set, no large changes are observed. This extrapolation is accomplished by calculating the percentage change between 2013 and 2014 in the OECD data. This percentage change is then applied to the 2013 data of the comparative political data set to create the extrapolated data for 2014. For all political variables (electoral fractionalization, institutional constraints, and the share of cabinet posts (left/centre/right)) the extension is realized by simply continuing the values when no elections took place in 2014 since no change is expected in the division of cabinet posts in that case. In the one case of elections (i.e. New Zealand) no significant electoral swings were observed. (Elections, 2014) The variable could therefore simply be continued for New Zealand as well.
One of the most important indicators in this study is the banking crisis dummy from the Laeven & Valencia (2013) dataset. This dummy is used to research the effects of a banking crisis on the amount of aid supplied and measures the aid losses for crisis countries and the counterfactual of non-crisis countries. According to Laeven & Valencia a banking crisis is defined as those meeting two conditions:
1. Significant signs of financial distress in the banking system (as indicated by significant bank runs, losses in the banking system, and bank liquidations)
2. Significant banking policy intervention measures in response to significant losses in the banking system
Policy intervention in turn is deemed to be significant if at least three (or two on a large scale) of the following measures have been used:
1. Deposit freezes and/or bank holidays 2. Significant bank nationalizations
3. Bank restructuring gross costs (at least 3 per cent of GDP)
4. Extensive liquidity support (5 per cent of deposits and liabilities to non-residents) 5. Significant guarantees in place
6. Significant asset purchases (at least 5 per cent of GDP)
This banking crisis dummy measures the amount of years since the emergence of a banking or financial crisis, with the first year of the crisis taking a value of 1. Since aid is expected to fall only after a certain period, this dummy captures the effect per year of crisis. (Laeven & Valencia, 2013: 226 – 229) Additionally, a squared version of the dummy is used to account for non-linearity in the dynamics of a banking crisis.
Since the Laeven & Valencia dataset is only updated until 2011, an extension is made to complete the data until 2014. This is accomplished by further counting the years since a banking crisis emerged in an affected country. Also, a division is made to separate the great financial crisis (financial crisis from here on) of the late 2000s and banking crises (earlier banking crises from here on) before that. This is realized by giving each crisis year a separate dummy (1, 2, 3 and so on). This crude way of dummy creation provides a way to compare the effects and course of earlier banking crises and the great financial crisis. Examples of earlier banking crises are the Finish, Norse, Swedish and Japanese crises that occurred in the 90s. A complete list of all crises included in this study can be found in appendix A.
6. Econometric setup
6.1. LSDV approach
In the first part of this paper the goal is to estimate versions of the following equation using a least squared dummy variable (LSDV) estimator:
#%,' = IR+ I5#%,'*++ IS,-%,'+ IT$%,'+ IU$%,'*++ IV0%,'*+ + W%+ X'+ Y%,' (8) where #%,' is some measure of aid supply from donor country ! in year ", which is dependent on previous levels of aid supply #%,'*+, whether a banking crisis ,-%,' is present, endogenous vectors $%,' and $%,'*+ and the exogenous vector 0%,'*+. The endogenous vectors consist of several (lagged) domestic explanatory variables, for example per-capita GDP, unemployment and government balance, where the exogenous vector consists of external (foreign) factors,
such as economic growth in recipient countries. Given the heterogeneity among countries and the likely importance of unobserved factors, donor fixed effects W% will be used to capture
time-invariant country-specific influences on aid disbursements. Year dummies X' are also added to account for common shocks to aid disbursements in any given year. Normally, adding both year dummies and a banking crisis dummy would lead to possible misspecification. However, since not every country included in this research has observed a banking or financial crisis (the counterfactual), misspecification due to dummies is no issue in our models. The addition of the fixed effects makes it a LSDV estimator, rather than the conventional ordinary least squares (OLS). The LSDV estimator is preferred above the similar instrumental variables (IV) estimator since Bruno (2005) shows that IV is more appropriate for ‘small t, large n’ data sets. This does not hold in this study: the number of years is larger than the amount of countries researched.
As argued by Jones (2015), several common issues consist within aid studies that need taken care of. First of all, a clear division between short- and long-run factors is necessary. The main focus of this study is on the short to medium run. By adding lagged values of aid supply in the regression, the model can only measure short to medium-run impacts. Generating longer-run effects in our models would require additional assumptions about the dynamic mechanisms between aid and other variables. Second, heterogeneity in aid supply behaviour between countries and years needs to be accounted for. This is solved by adding the country- and time-fixed effects that account for unobserved differences between donors and years. Third, aid is often treated as stationary while it is not the case which may lead to biased estimates. This is no issue for the LSDV estimator, but becomes one for our PVAR models. The issue will be duly solved for that part.
Versions of equation (8) are estimated by conducting both static (without lagged aid) and dynamic (lagged aid included) regressions. Using LSDV as estimator, there will be multiple regressions models. First of all, net ODA disbursements will be used as dependent variable to find the effects of banking crises on aid supply and to compare the effects and dynamics of previous banking crises and the global financial crisis. Second, several robustness checks will be conducted such as dropping influential countries and periods. Third, several macroeconomic variables within endogenous vectors X[,\ and X[,\*] and the exogenous vector φ[,\*] will be added to find their effects on aid supply. Fourth, the effects political indicators on aid supply will be analysed. Fifth, the ODA to GNI and ODA to GDP ratios will be used to research how these ratios develop during a crisis. Sixth, the effects of
banking crises on different sectors of aid will be analysed. Lastly, the effects of banking crises on different regions and former colonies will be researched.
Since a linear regression method is chosen and the dynamics of a financial crisis are hypothesized to not be linear, the banking crisis dummy and its squared variant account for the change in slopes, since every year of crisis has its own slope. (Laeven & Valencia 2013, 226 – 229) Furthermore, Wald tests are conducted to check whether all variables are significant. Also, in order to correct for heteroskedasticity and to produce unbiased and consistent standard errors of the estimators, bootstrapped standard errors will be used. Bootstrapping is an easy-to-use nonparametric estimation method that is appropriate when there is doubt about the distributional assumptions of the data, which is the case in our research.
6.2. PVAR approach
In the second part of this paper the goal is to employ a panel vector autoregression (PVAR) approach in line of Gravier-Rymaszeska (2012) in order to estimate the effects of several macroeconomic shocks on aid supply.1 PVAR extends the traditional VAR approach popularized by Sims (1980) with a panel-data dimension. This flexible method treats all variables in the system as endogenous and independent, without worrying about the direction of causality. Each variable is explained by its own lags and by lagged values of the other variables. PVAR allows for unobserved individual heterogeneity and the results show the importance of different shocks. The PVAR will be estimated by some version of the following equation, is exemplified by Canova & Ciccarelli (2004):
C%,' = #R_%,'+ `5C%,'*5+ ⋯ + `bC%,b*5+ c%+ d'+ Y%,' (9)
Vector C%,' consists of a e f 1 vector of e panel data variables, vector _%,' consists of deterministic terms such as linear trends, dummy variables or a constant with #R as the
parameter matrix, ` are the parameter matrices associated to the lagged variables C%,' with lag order @, c% representing country specific effects, d' the time specific effects, and Y%,' the disturbance term.
This specification imposes an important restriction: it generates an average response based on all countries in the sample. Once the unknown parameters are estimated, the
1
We use a modified version of the Stata pvar and helm programs by Abrigo & Love (2015). The original programs are available at: https://sites.google.com/a/hawaii.edu/inessalove/home/pvar. In order to suite these programs to our needs, we have modified the parameters in the programs regarding the amount of simulations of the Monte Carlo standard errors and the time horizon of the output graphs.
reduced form PVAR allows utilization of dynamic simulations based on these average responses. The results come in the form of impulse response functions (IRFs) and the analysis of the forecast error variance decompositions (FEVDs). IRFs show the dynamics of the shock where variance decompositions give information about the variation in one variable due to shocks in the others. These responses correspond to shocks in one variable holding all others constant, which requires orthogonalization of the variables in the system. In order to successfully achieve orgonalized shocks, the Cholesky decomposition method of causal ordering is employed. This method implies that variables that appear earlier in the model are more exogenous than the endogenous variables that appear later: the variables that appear earlier affect the following variables both contemporaneously and lagged, while the variables that appear later only affect the previous variables with lag. It is furthermore assumed that positive and negative shocks are symmetrical, with the focus of this study on the shocks that are negative for the economy (e.g. negative shocks in GDP, positive shocks in unemployment and fiscal deficit).
The following causal ordering is chosen and consists of four variables: GDP as share of the population (per capita), unemployment, government deficit and aid as share of GDP.
789
@B@ → <=>?@ABC?>=" → 7Bh>Q=?>=" 8>)!P!" → #!i
789 (10)
This order implicates that shocks to GDP per capita, unemployment and the government deficit impact the amount of aid supplied. The rationale behind this ordering is chosen on account of the theoretical framework. A macroeconomic shock reduces the amount of GDP which leads to changes in unemployment, after which the government has to act to counter these changes. Since the government’s expenditure increases, it is expected that its deficit will increase in turn, which eventually leads to a change in the amount of aid supplied. Aid only affects GDP, unemployment and government deficit with some lag as public spending on aid positively affect the government deficit, which in turn affects unemployment and GDP. Of the 24 possible orderings, this one seems the most plausible.
The amounts of lags within the models are extremely important. Too many lags cause over-parameterisation by wasting degrees of freedom while too few leave the estimations potentially misspecified and may cause autocorrelation. Most models in the literature use two or three-lag models, so we follow the literature and use three.2 In order to produce unbiased estimators, our standard errors are 5% on each side generated by Monte Carlo simulation with 500 replications.
The aforementioned PVAR model can be also be represented as: 1 _5S _S5 1 _5T _5U _ST _SU _T5 _TS _U5 _US _1UT _1TU ∆789 @B@ %,' ∆<=>?@ABC? %,' ∆7Bh8 %,' ∆ #!i 789 %,' = _5R _SR _TR _UR + `55 `5S `S5 `SS `5T `5U `ST `SU `T5 `TS `U5 `US `TT `TU `UT `UU ∆789 @B@ %,'*b ∆<=>?@ABC? %,'*b ∆7Bh8 %,'*b ∆ #!i 789 %,'*b + c5 cS cT cU (11)
In this equation, C%,' is a 4-variable vector including 4 endogenous variables: GDP per capita, unemployment, government deficit and the aid to GDP ratio (as aid supply). The 4 f 4 matrix ` contains the coefficients of contemporaneous relationships between the four variables that have @ amount of lags. The aim of this model is to achieve insights in the impulse responses of shocks in GDP per capita, unemployment and the government deficit on the amount of aid supply.
Before the PVAR technique can be applied several data transformations are required. Since PVAR models assume that the underlying cross-sectional structure is similar, we remove both the constant and time fixed effects d' by time demeaning: the mean of each calculated variable is subtracted for each country-year. However, this procedure creates a bias when using lags. (Arellano & Bond, 1991: 278). In order to solve this problem, a Helmert (forward mean-differencing) transformation is conducted. This transformation keeps the orthogonality between the variables and their lags in such a way that the lags can be used as instruments in order to estimate the coefficients. These coefficients are then estimated by a GMM estimator. Lastly, since only stationary data can give consistent PVAR results (as argued by Jones) we run our model in first differences.
The main drawback of this approach is that it only focuses on the short-run dynamics of macroeconomic shocks on aid supply. As argued in the theoretical framework, we expect that banking crises can have long-lasting effects. We cannot control for these effects in our models, since our PVAR models can only be used to calculate ten years after the shock. The main focus of our research is therefore on the short-run effects of these macroeconomic shocks .
7. Results and analysis
7.1. LSDV approach
The first part of the analysis is devoted to estimations of equation (8) using the LSDV estimator. Table 1 displays the results of our baseline models. The first column (model 1.1) consists of a model without fixed effects and with all banking crises added in the banking crisis variable. This variable implies that aid supply declines on average with 4,4 per cent per year of crisis, relative to the year before. This means that five years after a crisis hits, aid is about 15 per cent lower than it would have been in the absence of a crisis. This effect continues until around 12 years after the onset of the crisis, at which point aid supply has decreased by around 25 per cent relative to the counterfactual. Model 1.1 further shows that aid supply is highly dependent on previous levels due to the large and extremely significant lagged dependent variable: around 89 per cent of aid supply can be explained by its previous level. Furthermore, capita income and population emerge as positive and significant predictors of aid supply. Larger and richer donors are expected to provide more aid than smaller and poorer donors, even when controlling for the effects of a banking crisis.
Table 1: Baseline models
Model 1.1 1.2 1.3 1.4 1.5
Variations All banking
crises, no fixed effects All banking crises, with fixed effects Only earlier crises Only financial crisis Both earlier crises and fin. crisis Lagged dependent variable 0.8918*** (0.0264) 0.8065*** (0.0346) 0.8008*** (0.0344) 0.8063*** (0.0345) 0.8016*** (0.0346) No. of years from start
banking crisis
-0.0444** (0.0206)
-0.0434* (0.0234) No. of years from start
of earlier crises
-0.0149*** (0.0056)
-0.0166*** (0.0057) No. of years from start
of earlier crises squared
0.0007*** (0.0002)
0.0008*** (0.0003) No. of years from start
of financial crisis
-0.0672*** (0.0189)
-0.0767*** (0.0185) No. of years from start
of financial crisis sq.
0.0101*** (0.0031)
0.0116*** (0.0030) Log capita income 0.2688***
(0.0786) 0.6365*** (0.1200) 0.6843*** (0.1191) 0.6403*** (0.1175) 0.6735*** (0.1194) Log population 0.1034*** (0.0306) -0.0758 (0.1114) -0.0632 (0.1081) -0.0832 (0.1094) -0.1065 (0.1110) Wald test p value 0.0000 0.0000 0.0000 0.0000 0.0000
N 1085 1085 1085 1085 1085
Adj. R2 0.974 0.979 0.979 0.979 0.980
Notes: Dependent variable is log of net ODA disbursements. Standard errors are bootstrapped with 200
However, when both country and time fixed effects are introduced (model 1.2), two large changes can be observed due to resolving possible omitted variables bias: the capita income coefficient increases immensely while the population coefficient becomes both negative and insignificant. This would signify that the size of a country does not affect the amount of aid supply, which goes against theoretical expectations (see appendix C). Perhaps surprisingly, the banking crisis coefficient remains stable when introducing fixed effects.
In model 1.3, the financial crisis is omitted from the regression, showing the effects of only previous banking crises in the sample. This banking crisis coefficient is much smaller than the ones in the previous models. On average, previous banking crises caused a decline in aid supply by around 1,5 per cent a year, which continued until around 15 years, at which aid supply was 15 per cent lower.
In model 1.4, all previous crises are omitted from the regression in order to obtain the effects of the global financial crisis. The coefficient is much larger than in the previous models and show that aid has declined by around 6,7 per cent per year so far which means that in the eight years after the crisis (2007 until 2014), aid has declined with 30 per cent, while the effects of the crisis not completely disappeared. This also explains the differences between models 1 through 3. In the first model an average effect is calculated, which consists of the average of all the effects of all banking crises in the dataset.
Model 1.5 represents the main model in this study, with coefficients added for both earlier banking crises and the financial crisis. The coefficient for the earlier crises is only slightly larger than the one in model 1.3, but the coefficient for the financial crisis has increased to 7,7 per cent a year. Since the difference in severity between the financial crisis and previous crises is so large, the separation between the two will continuously be made in the remaining part of this research in order to further analyze the differences between the two. For all models the p value of the Wald test (or F test) is very close to zero, meaning that all included variables are jointly significant. The adjusted R squared value is also very high for all regressions in table 1, showing that a large part of variance in aid supply is explained by the included independent variables. This is mainly due to the lagged dependent variable, which has large explanatory power looking at the high estimated coefficients in all models.
The quadratic specification in models 1.3, 1.4 and 1.5 approximate a relationship that can also be estimated by adding the dummy variables for each year of crisis. Instead of reporting the results of each dummy, the coefficients are depicted graphically in figure 7 where the light blue line shows the coefficients of earlier banking crises and the dark blue one
the coefficients of the global financial crisis (so far). This graph shows that previous banking crises had their nadir (on average) in the 11th year after the onset of the crisis and that aid supply is only equal to the counterfactual again after nineteen years. Analyzing the dark blue line that reflects the coefficients of the financial crisis, a similar trend is observed relative to the banking crises coefficients. Since only eight years after the onset of this crisis is observed, the full course can obviously not be depicted. However, since both lines are very similar for these eight years, suggested is that all banking crises follow a similar course. As a crisis hits, aid supply will not fall immediately due to previous values (commitments and budgets) and due to governments possibly hoping to contain the crisis before cutting aid budgets. Only after about four years aid supply is lower relative to the counterfactual.
Figure 7: Estimated dynamics of earlier banking crises and the financial crisis
Notes: The figure shows the coefficients of both earlier banking crises and the great financial crisis on the
individual-year counter dummy for years after the onset of a crisis. These coefficients are estimated in LSDV regressions controlling for GDP per capita and population, as well as both time- and donor fixed effects.
Source: Author’s calculations using data from Laeven & Valencia (2013)
Furthermore, note that this graph is very much alike figure 6 that depicted the theoretical dynamics of aid supply during a banking crisis. This confirms the hypothesis about the expected course of a banking crisis. However, this graph shows that the financial crisis had a similar course relative to previous crises, where the results of the linear regressions in table 1
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 E sti m ate d cr is is c oe ff ic ien t
Years after onset crisis
show that the financial crisis had a much larger negative effect. This could have two reasons. First of all, it could be due to the counterfactual (countries without a banking crisis). During previous banking crises, the counterfactual consist of multiple countries who were unaffected by a crisis. During the financial crisis the counterfactual is reduced since a lot of countries in our sample experienced the financial crisis. If the counterfactual countries (who were often also affected by the financial crisis in a way), did not increase or even decreased their aid supply, the effects in figure 7 are biased upwards. Another possibility for biased results is that our models can only successfully estimate short-run effects of a banking crisis. This may decrease the effects of previous banking crises in figure 7.
Rather than graphing the dummy variables for each regression model, the quadratic specification is used for the remaining regressions for convenience. A higher value would imply a heavier and longer crisis, where a lower value would imply a less substantial one.
To summarize, the results show that banking crises reduce aid supply between 15 and 40 per cent relative to the counterfactual, depending on which crises are added and the estimation method. This is in line with hypothesis 1 (H1) and previous literature. Hypothesis 2 (H2) also seems correct. Our linear regression methods in table 1 show a much stronger effect for the financial crisis relative to previous crises. Despite, our graphical method from figure 7 showing similar courses and severities, we argue that this is due to the counterfactual. This hypothesis will further be analyzed using PVAR models in the second part of this analysis.
Since it is implausible that lower aid can cause a banking crisis, the estimated negative effects imply causality. However, it could be that several important observations bias our results. To check whether this is the case, several robustness checks are executed, which can be found in table 2. For convenience, the baseline model (1.5) has been added to this table to make comparisons less complicated. In model 2.2 Iceland is dropped. Since every country in the panel is weighted equally in the regressions, Iceland might influence the results due to its modest size in terms of population and GDP.3 However, dropping Iceland
has only small effects on the regression results.
Another country that could influence results is Japan, which had economic troubles from the start of the 1990s which are still noticeable nowadays. Model 2.3 shows indeed that the coefficient of earlier crises is somewhat depressed while the financial crisis coefficient have increased when Japan is omitted.
3 Since the US is also somewhat of a outlier in terms of population and GDP, the same could be applied. We
Model 2.4 drops all years after 2006. Since some countries had banking crises whose effects overlapped with the global financial crisis, this model is added to omit effects of possible overlapping or following crises. Since the financial crisis started after 2006, this variable is removed. The banking crises coefficient shows no large change relative to the baseline model, except a reasonable loss of significance.
As stated, the employed dataset consists of data from 1960 until 2014. It is possible that data before 1975 is less reliable than the data after, since economic measuring methods have improved drastically since then. Dropping pre-1975 years show that banking crises have had slightly stronger negative effects relative to the baseline, where the financial crisis had a slightly less negative effects, although the differences are small.
Table 2: Robustness checks
Model 2.1 2.2 2.3 2.4 2.5 Variations Baseline model (1.5) Iceland dropped (smallest donor) Japan dropped (longest crisis) 1960 – 2006 only 1975 – 2014 only Lagged dependent variable 0.8016*** (0.0346) 0.8079*** (0.0379) 0.7987*** (0.0365) 0.7948*** (0.0383) 0.7690*** (0.0354) No. of years from start
of earlier crises -0.0166*** (0.0057) -0.0161*** (0.0063) -0.0120** (0.0059) -0.0142* (0.0082) -0.0205*** (0.0053) No. of years from start
of earlier crises sq. 0.0008*** (0.0003) 0.0008*** (0.0003) 0.0006** (0.0003) 0.0007 (0.0005) 0.0009*** (0.0003) No. of years from start
of financial crisis -0.0767*** (0.0185) -0.0717*** (0.0211) -0.0810*** (0.0198) -0.0695*** (0.0186) No. of years from start
of financial crisis sq. 0.0116*** (0.0030) 0.0108*** (0.0035) 0.0115*** (0.0031) 0.0108*** (0.0031) Log capita income 0.6735***
(0.1194) 0.6468*** (0.1321) 0.6984*** (0.1273) 0.7106*** (0.1321) 0.8164*** (0.1381) Log population -0.1065 (0.1110) -0.0909 (0.1098) -0.1452 (0.1098) -0.1490 (0.1509) 0.0730 (0.1444) Wald test p value 0.0000 0.0000 0.0000 0.0000 0.0000
N 1085 1061 1031 893 885
Adj. R2 0.974 0.980 0.979 0.974 0.962
Notes: Dependent variable is log of net ODA disbursements. Standard errors are bootstrapped with 200
replications and are given in parentheses. * p < .10 ** p < .05 *** p < .01
The above models show that the results of table 1 are quite robust to changes in the sample period and included donors. Except robustness to these changes, the results should also be robust to the inclusion of macroeconomic and political variables. As argued in the theoretical framework, there are variables that determine the amount of aid supply besides the presence of a banking crisis. In table 3 several macroeconomic variables are added to our baseline model to check whether these variables affect aid supply.
Model 3.1 adds GDP growth, gross domestic savings, economic openness, unemployment and inflation to the regression. Against expectations, all coefficients have both very small and insignificant effects. Especially in the case of unemployment this is quite interesting. Expected was that higher levels of unemployment would decrease aid supply, since this would induce governments to invest in their domestic economies instead in foreign ones. The results of this regressions suggest that this effect is absent.
In model 3.2 two other variables are added: military spending which is based on Round & Odedokun (2004) and exchange rate which is based on the results of Gravier-Rymaszeska (2012). The results show that both variables are (extremely) insignificant, while the banking crisis variables remain consistent. It can therefore be concluded that both military spending and exchange rate do not affect aid supply. However, the population coefficient becomes extremely significant in this model. This is probably due to military spending, what is of course much higher in larger countries (especially the US).
Model 3.3 includes all variables of models 3.1 and 3.2 to see if certain interaction effects are present. Two variables become significant in this model: economic openness and unemployment, albeit merely. Since the coefficient for military spending has also changed (but remains insignificant) this variable probably causes economic openness and unemployment to change.
In model 3.4 three government budget variables are added: government expenditures, government deficit and government debt. All three variables are significant, and have their expected results. Since aid disbursements are a part of government expenditure, it make sense that higher expenditures also lead to higher aid supply. As argued in the theoretical framework, higher government deficit and debt lead to cutting aid supply as a form of austerity measure, but only after a certain period. Therefore both variables are lagged to correct for this delayed effect. When the contemporary variables are added, the effect disappears. The effects of government deficit on aid supply seem clear-cut but will be researched further in the second part of this analysis.
Model 3.5 has two exogenous variables added to the model: GDP growth in aid receiving countries, and the amount of ODA distributed by other donors. The coefficient of world GDP growth is largely significant and negative, which means that if recipient countries do well (large economic growth), donors are inclined to lower aid disbursements. The coefficient of world ODA (all ODA supplied by other donors) is practically zero, leading to the fact that there is no evidence of a bandwagon effect. Lagged values of world ODA have also been used without success.
Table 3: Robustness to macroeconomic variables Model 3.1 3.2 3.3 3.4 3.5 Variations Macro-economic variables 1 Macro-economic variables 2 Total macro-economic variables Gov. budget variables Endogenous economic variables Lagged dependent variable 0.7545*** (0.0327) 0.6542*** (0.0482) 0.6343*** (0.0483) 0.7216*** (0.0527) 0.8016*** (0.0334) No. of years from start
of earlier crises -0.0203** (0.0089) -0.0192*** (0.0061) -0.0154** (0.0073) -0.0170*** (0.0041) -0.0166*** (0.0059) No. of years from start
of earlier crises squared
0.0009** (0.0004) 0.0008*** (0.0003) 0.0008** (0.0003) 0.0008*** (0.0003) 0.0008*** (0.0003) No. of years from start
of financial crisis -0.0699*** (0.0243) -0.0538*** (0.0202) -0.0604*** (0.0202) -0.0706*** (0.0235) -0.0767*** (0.0185) No. of years from start
of financial crisis sq. 0.0108*** (0.0032) 0.0083** (0.0034) 0.0093*** (0.0033) 0.0113*** (0.0037) 0.0116*** (0.0030) Log capita income 0.9949***
(0.1834) 1.1020*** (0.1668) 1.0377*** (0.2097) 1.0146*** (0.1996) 0.6735*** (0.1163) Log population -0.2115 (0.2556) 0.6258*** (0.2281) 0.6740*** (0.2502) -0.2839** (0.1318) -0.1065 (0.1096) GDP growth rate 0.0041 (0.0051) 0.0003 (0.004) Savings 0.0017 (0.0028) -0.0014 (0.0030) Economic openness 0.0001 (0.0006) 0.0010* (0.0006) Unemployment -0.0013 (0.0041) -0.0095* (0.0050) Inflation (log) 0.0095 (0.0152) 0.0064 (0.0150) Military spending -0.0041 (0.0318) -0.0313 (0.0318) Exchange rate 0.0002 (0.0002) 0.0002 (0.0002) Government expenditures 0.0289*** (0.0105) Government deficit (lagged) -0.0058*** (0.0022) Government debt (lagged) -0.0010*** (0.0004) World GDP growth -0.5066* (0.0302) World ODA 0.0000 (0.0001) Wald test p value 0.0000 0.0000 0.0000 0.0000 0.0000
N 769 617 587 1016 1085
Adj. R2 0.960 0.939 0.945 0.978 0.980
Notes: Dependent variable is log of net ODA disbursements. Standard errors are bootstrapped with 200
