Political instability and sovereign default

(1)

Political Instability and Sovereign Default

Pyke Polderman

5828058

September, 2014

Faculty of Economics and Business

Department of Economics

Master Thesis

MSc Economics

Monetary Policy & Banking

(2)

2

Abstract

This study empirically examines the effect of political instability on sovereign default. I employ a model where the incumbent political party constrains the choices of a successive party by the strategic use of debt. In the model, political instability increases the probability of default as well as the size of default. To test these features of the model, I use political turnover, elections and wars as indicators of political instability. Using a unique sample of 136 countries over the period 1975 – 2012, I find that the likelihood of default is significantly increased by wars. A country is more like to repudiate its debts by 8.85% when it fights a war or a war is fought on its territory. The effect of wars on the default probability is more pronounced in low-income countries. Political turnover increases the probability of default only in high-income countries. Further, an increase in political instability leads to an increase in the size of default. The magnitude of sovereign default increases significantly in response to political turnover, elections, and wars. The debt in default to GDP ratio increases by 3.98 in response to political turnover in the preceding year. Wars both increase the size of default in high-income countries as well as low-income countries.

(3)

3

1. Introduction

The recent defaults of Greece, Argentina, and Russia highlight the fact that sovereign borrowing is risky and that countries default on large portions of their outstanding debt. Sovereign default was seen as a phenomenon of the developing world but the recent European sovereign debt crisis stressed that the developed world is also prone to sovereign default and debt crises. Sovereign defaults are costly for borrowing countries, as well as for international and domestic creditors. In order to prevent situations of default it is important to identify what determines sovereign default. The theoretical literature on sovereign debt and default relates default to output fluctuations, where countries are prone to default when output is below trend. Besides output contractions, solvency and liquidity problems are typically mentioned as determinants of default. Though, I argue that political instability is a main driver of sovereign default. I empirically test a model of political instability and sovereign default based on Alesina and Tabellini (1990) and Cuadra and Sapriza (2008). Using a unique sample of 136 countries spanning from 1975 to 2012, I examine the effect of political instability on the likelihood of sovereign default. Additionally, I study the effect of political instability on the size of sovereign default. This is the first empirical study that connects political instability and the magnitude of default. Moreover, this study is unique in the use of actual political turnover as an indicator of political instability.

Historically, countries have issued debt for hundreds of years, as sovereign debt was one of the first financial assets ever traded (Tomz & Wright, 2013). External debt held by a foreign entity emerged in the 1820’s with the independence of nations in Latin America and other regions. The newly-independent countries began issuing bonds in foreign currency in European financial centers. Besides issuing sovereign debt, these countries also started to default on their debt from this period onwards (Beers & Nadeau, 2014). To illustrate the widespread occurrence of default, Tomz and Wright (2013) distinguish four major default episodes wherein at least 30% of the world debtors were in default by number. The first period began in the 1820’s when newly independent countries immediately defaulted on their external debt obligations. The second episode was around the 1870’s caused by the wars in Central and Latin America. The defaults in the third period were due to the Great Depression in the 1930’s. The last episode of massive sovereign defaults was in the 1980’s through debt crises in several countries (Tomz and Wright, 2013).

Withstanding the large occurrence of sovereign defaults, sovereign debt is still a very important asset class as it accounted for about 19% of global financial assets in 2010 (Tomz and Wright, 2013). Sovereign debt is a specific asset class because the debt is not enforceable through a third party. In contrast, private debt contracts are enforced through third parties, typically by the state in advanced or developing economies. Private debt contracts are honored because a creditor has power vis-à-vis a debtor. If a debt contract between private agents is breached, the creditor can sue the borrower in court and bankruptcy law provides for a transfer of assets from the borrower to the creditor. Sovereign debt is not enforceable through a third party. Sovereign default occurs because debtor countries are not able to pay or willing to pay their debts. The fact that a sovereign debt market nonetheless exists is because countries suffer from costs of default. These costs consist of capital market exclusion following a default, increased borrowing costs, direct or indirect sanctions, output costs, political costs, and reputational costs. Although large literature is devoted to these costs as they explain why

(5)

5

countries repay their debts, I emphasize the determinants of sovereign default to explore the rationale behind a sovereign default.

It is important to explore the association between political turnover and sovereign default. Countries have repudiated their debts for almost 200 years and will do so in the future. With high government debt levels, low economic growth, and increasing demands for public spending, countries are more prone to default in the future (Buiter and Rahbari, 2013). This makes it important to study the determinants of sovereign default. It is indispensable to study political factors that affect default as political risk is a key characteristic of poor and emerging economies as well as advanced economies. The main question of this study is: does political instability determine sovereign default? The main hypothesis of this study is that there is a positive empirical relationship between political instability and sovereign default.

In models of sovereign default, output contractions are the main driver of sovereign default. Sovereigns borrow to smooth consumption and insure themselves against times of low output relative to trend. Countries thus borrow during bad times and repay during good times (Panizza, Sturzenegger & Zettelmeyer, 2009). After a series of negative output shocks, default becomes more tempting and the probability of default goes up (Eaton & Gersovitz, 1981; Grossman & Van Huyck, 1988; Aguiar & Gopinath, 2006). The theoretical premise that sovereign default is countercyclical and caused by negative output shocks is tested empirically. Empirical studies show that low output relative to the trend can lead to default but the empirical evidence is not conclusive. Levy-Yeyati and Panizza (2006) find that sovereign default is preceded by output contractions. Albeit, Tomz and Wright (2007) find that only 62 percent of the 169 default episodes they investigated from 1820 to 2004 began in bad economic times. In more than 39 percent of the observations, countries managed to avoid default in a situation where output was below trend (Tomz & Wright, 2007). Sovereign defaults are thus not only explained by shocks to economic growth. This study explores the role of political factors as determinants of sovereign defaults.

Incorporation of political factors in the sovereign debt literature relies on a countries’ willingness to pay. The decision to repay sovereign debt is essentially due to the “ability to pay” and the “willingness to pay” of a country. On the one hand, countries that are unable to repay their debts suffer from insolvency or illiquidity. On the other hand, countries that are not willing to repay their debt base this decision on a consideration of the costs and benefits of default (Manasse & Roubini, 2009). Looking at recent default episodes, for instance in Argentina and Peru, it seems that political instability and sovereign default are intertwined. As for Argentina in 2001 and Peru in 1989, the announcement of default coincided with the inauguration speech of newly-elected presidents (Kohlscheen, 2009).

Theoretically, the willingness-to-pay approach is incorporated in models of the strategic use of government debt. Models of strategic accumulation of government debt highlight the importance of political instability in the form of a positive probability of political turnover (see Alesina & Tabellini 1990; Ozler & Tabellini, 1991). These authors show that governments issue larger portions of debt when political instability is higher. Although these authors did not consider the option of default, they laid the foundation for the forthcoming literature. Using Alesina and Tabellini (1990) and Ozler and Tabellini (1991), Cuadra and Sapriza (2008) develop a model where a higher probability of political turnover makes the government more inclined to default. Theoretically, Cuadra and Sapriza (2008), and later on Hatchondo, Martinez and Sapriza (2009), connect the strategic use of government debt, political instability, and sovereign default.

(6)

6

Sovereign default is thus incorporated in models of political instability and the strategic use of debt. The aim of the study is to empirically examine the relationship between political instability and sovereign default. The empirical literature on political instability and sovereign default is not extensive. This study is a contribution to this empirical literature. Although the relationship between political instability and sovereign default is not investigated empirically in a comprehensive way, it is clear from the empirical research that political conditions are connected to the willingness to repay foreign debt (see for instance: Citron & Nickelsburg, 1987; Van Rijckeghem & Weder, 2009; Kohlscheen, 2009). Models employ the probability of turnover as an indicator of political instability which is a determinant of debt accumulation and default. Nevertheless, the existing literature does not take actual political turnover into account as an indicator of political instability. For instance, Kohlscheen (2009) constructs a political instability variable by taking the number of changes of the executive in a period of 10 years. Van Rijckeghem and Weder (2009) use the length of tenure of the chief executive, the number of veto players who drop from the government, and a variable that tracks wars to measure political instability. Instead, I use actual political turnover to denote political instability.

Besides using actual political turnover as proxy for political instability, I employ the size of default as dependent variable to exactly determine the effects of various political variables on sovereign default. In this way, the political triggers of a sovereign default can be identified. This study is unique in its use of the size of default as all other empirical studies on sovereign default employ the probability of default as dependent variable (see for instance Balkan, 1992; Kohlscheen, 2009; Van Rijckeghem & Weder, 2009; Saiegh, 2009). The size of default is extracted from the Bank of Canada’s Credit Rating Agency (CRAG) database. This database includes sovereign debt in default from a very large group of debtors and creditors. The sample used for this study includes 136 countries from 1975 and 2012. This extensive sample allows me to make comparisons between different income groups. All countries that had debt in default in this period are included. Whereas other studies only focused on democratic borrowing countries (see Kohlscheen, 2009), or debt owed to private creditors (Van Rijckeghem & Weder, 2009), this study includes all debtors and a broad group of creditors.

I link political instability and sovereign default in a model that is tested empirically. The model is based on Alesina and Tabellini (1990), Ozler and Tabellini (1991), and Cuadra and Sapriza (2008). In the two period model with two different political parties that compete for power, elections after the first period create political instability. A positive probability of political turnover leads to a build-up of debt in the first period to constrain the choices of a possible successor after the elections. This build-up of debt, which is increasing in the probability of turnover, causes the probability of default as well as the size of default to rise. When policy preferences of the parties fully diverge, the effects of the probability of political turnover are greater than in the case of only differing policy preferences. From the model, two testable implications are derived. The main hypothesis is that increasing political instability leads to a higher probability of default and a higher size of default. The second hypothesis is that increased divergence over policy preferences, which is synonymous for more polarization, causes the default probability and the magnitude of default to rise.

I test the implications from the model regarding the probability of default employing a logit regression model. To test the effects on the size of default, a censored regression model is adopted. I use the tobit model, which is a widely used censored regression model. I find consistent support for the main hypothesis. From the logit model, I find that political instability

(7)

7

increases the likelihood of sovereign default. If a country is engaged in a war or a war is fought on its territory, the country is more likely to default. From splitting the sample, I conclude that if the probability of political turnover increases in a high-income country, the probability that a country will repudiate its debts rises. Contrastingly, wars only increase the likelihood of sovereign default in low-income countries. Furthermore, employing the tobit model, I find that political instability triggers sovereign default and increases the scale of sovereign default. Political turnover causes a raise in the size of defaulted debt while the size of default also rises a year after an election. In addition, wars induce a sovereign default and a country defaults on larger parts of their debt in war years. These results are robust do different specifications.

The remainder of the paper proceeds as follows. Section 2 presents the theoretical and empirical literature on political instability and sovereign default. Section 3 introduces the model and the testable implications from the model. Section 4 describes the data and the methodology used in the empirical analysis. Section 5 contains the results from the logit regression analysis and tobit regression analysis as well as robustness checks. Section 6 concludes.

2. Literature Review

The model of this study is based on the models of Alesina and Tabellini (1990), Ozler and Tabellini (1991), and Cuadra and Sapriza (2008). The models of these authors are based on the willingness-to-pay approach. This approach is fundamental in the sovereign debt literature after Eaton and Gersovitz (1981) emphasized that countries are still able to borrow when there is a positive probability of sovereign default. The model of Eaton and Gersovitz (1981) and the subsequent critique are described to introduce the willingness-to-pay approach. Quantitative models based on the model of Eaton and Gersovitz (1981) highlight the use of this approach in modern dynamic stochastic equilibrium models. Cuadra and Sapriza (2008) incorporate political instability in a willingness-to-pay approach based model. They use Alesina and Tabellini (1990) to model the political and electoral environment that also provides the basis of the model of this study. Besides theoretical literate on sovereign default and political instability, the empirical literature on this subject is described extensively. This literature is mainly focused on the relationship between institutional factors and the probability of default.

2.1 Theoretical Literature

Eaton and Gersovitz (1981) introduce the possibility to default in a model of sovereign borrowing. Sovereign debt markets are built on reputation; it is therefore important to describe reputational models. Countries that are not willing to repay their debt make this decision based on a consideration of the costs and benefits of default. A countries’ reputation is based on the willingness to repay sovereign debt. Reputational models accordingly are essential to the willingness-to-pay approach that is that is central to the model used in this paper. In Eaton and Gersovitz (1981), countries are risk-averse and borrow to smooth consumption between periods of high- and low income. In the model, countries have the possibility to not repay their debts. After a sovereign default, a country is permanently excluded from the capital markets and is faced with an additional penalty. Eaton and Gersovitz (1981) show that if a country earns a low income in the period where loans come due, default is more tempting and the probability of default increases. Eaton and Gersovitz (1981) further show that creditors are willing to extend credit up to a certain ceiling when there is a positive probability of default.

(8)

8

As Eaton and Gersovitz (1981), Grossman and Van Huyck (1988) also build a model where the issuance of sovereign debt is based on reputation of the borrower. If the sovereign has repudiated its debts in the past and the current lenders know this, lenders would not be willing to extend credit and borrowers are in financial autarchy. In the reputational equilibrium, Grossman and Van Huyck (1988) show that the short run gain from repudiation is smaller than the long run loss of a trustworthy relationship. A sovereign debt market can thus exist wherein debt repudiation never occurs, nonetheless excusable default happens.

The fact that permanent exclusion is not observed in data, led to critique on the article of Eaton and Gersovitz (1981). Bulow and Rogoff (1989), criticize the model of Eaton and Gersovitz on the fact that borrowing in international capital markets is not the only way in which countries can smooth consumption. Bulow and Rogoff (1989) emphasize that borrowers can never be excluded from capital markets entirely as they are still able to buy cash-in-advance contracts where countries make a payment in advance which entitles them to receiving a payment in the future. In practice, these contracts could be foreign assets, treasury bills or stocks. They show that reputational contracts can only exist when creditors have direct ways of punishment because borrowers cannot establish a reputation for repayment.

Quantitative models of sovereign debt and default are built around the analysis of Eaton and Gersovitz, and Grossman and Van Huyck. Some of these models include political factors and connect the political instability to sovereign default. In this respect, they are important to describe. Alfaro and Kanczuk (2005) were the first to capture the models of sovereign debt in a quantitative model (Stähler, 2013). Their model is based on the principles of the Grossman and Van Huyck model where lenders cannot verify whether borrowers have a good repayment record and have to form expectations about repayment. Countries do only default when they witness a series of low output outcomes, which makes the default always excusable. Aguiar and Gopinath (2006) improve the quantitative models by introducing a productivity process that is subject to a volatile stochastic trend. They show that the model performs better if a volatile productivity process is assumed instead of transitory fluctuations around a stable trend.

In the literature on reputational borrowing described above, sovereign debt and default are not related to political factors. Political instability is absent both in the model of Eaton and Gersovitz (1981) as in the model of Bulow and Rogoff (1989). Foreign debt problems are intertwined with political crises, as the recent defaults of Argentina and Mexico show (see for instance: Hatchondo & Martinez, 2010; Bussière & Mulder, 2000). Models of sovereign debt and political instability focus on the strategic use of debt to constrain future choices of successors. Government stability, the timing of elections, the ideological orientation of political parties, political fragmentation, and the form of budgetary institutions are all explanations for the existence and strategic use of sovereign debt (Carmignani, 2003). Most of the political factors that are related to the existence and use of debt fall under political instability. In this paper, the probability of political turnover is used to proxy for political instability.

The strategic use of debt in a politically unstable environment is introduced by Alesina and Tabellini (1990). Since countries accumulated debt and faced increasing deficits from the 1970’s onwards, theoretical literature became directed towards explaining this accumulation that could not be explained through standard economic theory in which debt was used to distribute tax distortions over time (Alesina & Tabellini, 1990). In the positive theories of fiscal deficits and debt, political factors act as determinants of fiscal policy outcomes. Ozler and Tabellini (1991) explicitly stressed the relationship between political instability and external debt as they

(9)

9

sought an explanation for the accumulation of external debt at the beginning of the 1980’s and the resulting debt crises in this period.

Alesina and Tabellini (1990) consider an economy where two political parties with different policy objectives alternate in office through elections. Elections are held under a large number of individuals that differ in their preference for two public goods, where both goods are represented by one political party. The government can issue debt and parties can choose to leave this debt to the future. The debt becomes due in the next period. Because every government is committed to honor the debt obligations of its predecessors, default is not possible. An increasing probability of re-election increases the marginal cost of issuing debt because of tax distortions and lower future public consumption. Thus, the higher is the probability of re-election, which means a lower probability of political turnover, the lower is equilibrium debt. By this mechanism, current political parties influence the choices of successive parties.

Alesina and Tabellini (1990) show that if the parties disagree more about the public goods individuals prefer, governments will over-issue debt to constrain their successors in spending on the successors’ preferred public goods. Political instability and polarization consequently lead to over-accumulation of debt. This conclusion forms the basis of the model of this paper. Ozler and Tabellini (1991) specify the model of Alesina and Tabellini (1990) with the strategic use of external debt where Alesina and Tabellini (1990) do not distinguish between domestic and external debt. Ozler and Tabellini (1991) also show that political instability and polarization both increase the demand for loans.

Amador (2003) connects the models of Eaton and Gersovitz (1981), Bulow and Rogoff (1989), and Alesina and Tabellini (1990) in a model where sovereigns are unable to replicate the original debt contract after a default because of political uncertainty. Political turnover leads to overconsumption of a countries’ assets and reduces saving by the impatient incumbent political party. The fact that a political party alternating in power is unwilling to save and the realization that future political parties are impatient as well leads to repayment of debts in the current period. Hence, countries are repaying their debt because of political uncertainty. Hatchondo, Martinez and Sapriza (2009) do not differentiate between the public goods political policymakers consume as Alesina and Tabellini (1990) do, but distinguish the type of policymaker that is in power. Their model is connected to the model of Alfaro and Kanczuk (2005).

In the model of Hatchondo, Martinez and Sapriza (2009), one policymaker is more patient than the other. A default can be triggered by political turnover if a “patient” policymaker is replaced by an “impatient” policymaker. These political parties differ in their willingness to repay the debts. In an environment where political instability is high, the probability of political turnover is high. If, in such an environment, political instability decreases, the patient policymaker would choose debt levels that would lead to a default when an impatient policymaker comes to office. At the other hand, in an environment where political instability is high, patient policymakers already choose debt levels that would lead to default under an impatient policymaker. That is why less political instability, which means a lower likelihood probability, has a decreasing effect on the probability of default. The authors show that default is triggered by political turnover in an environment with enough political stability. Hatchondo, Martinez and Sapriza (2009) use data from Argentina in the run-up to the 2001 default and generate some of the features of the Argentine economy.

(10)

10

The quantitative models of sovereign default described earlier revolve around the willingness-to-pay approach of Eaton and Gersovitz (1981) that characterizes reputational models of sovereign debt repayment. This willingness-to-pay is associated with political factors such as turnover and polarization. Cuadra and Sapriza (2008) employ a dynamic stochastic model where two different households are represented by two different political parties which alternate in power. The two different parties can consume two public goods but care more about the public good that is preferred by their own type of household. Cuadra and Sapriza (2008) hence use the set-up of Alesina and Tabellini (1990) to model political uncertainty but they endogenize the probability of default and connect it to political uncertainty. This link between political uncertainty and the probability of default is also present in the forthcoming model. Cuadra and Sapriza (2008) show that political parties, acting under political uncertainty, are impatient and consume more relative to a situation without uncertainty. Countries are willing to take loans at higher interest rates to be able to consume in the current period. The model of Cuadra and Sapriza (2008) is superior in replicating quantitative features in the data than models without political uncertainty. It generates interest rate spreads that are more in line with the data. Their novelty is the direct link present in the model between political instability and sovereign default (Stähler, 2013).

2.2 Empirical Literature

The theoretical literature of sovereign debt and default has culminated in empirical work on the assumptions and the outcomes of the models. Sovereigns borrow to smooth consumption over time; they borrow in periods when income is low and save when output is high relative to trend. Sovereign borrowing is thus countercyclical. Nonetheless, this feature of the models is not supported by the data (Panizza, Sturzenegger and Zettelmeyer, 2009). Another feature of the models is that sovereign default mainly happens in bad economic times when output is low (see for instance Grossman and Van Huyck, 1988). Similarly to sovereign borrowing, sovereign default is countercyclical. Levy-Yeyati and Panizza (2006) use quarterly data and conclude that economic growth is lower in the years preceding a default but not after a default (Levi-Yeyati & Panizza, 2006).

Tomz and Wright (2007) use an extensive sample to examine the relationship between output and sovereign default for the period between 1820 and 2004. They show with historical data that countries managed to avoid default in bad economic times in 39 percent of all observations and output was above trend in nearly 44 percent of all years in which countries defaulted (Tomz & Wright, 2007). The authors conclude that defaults are associated with lower output but that the relationship is unexpectedly weak. This shows that output contractions do have some explanatory power in explaining defaults. But this research also shows that output contractions are not the only factor determining defaults. This leaves room for other explanations, for instance political factors.

As noted above, political instability and sovereign default are linked in the theoretical literature by Hatchondo, Martinez and Sapriza (2009) and Cuadra and Sapriza (2008). Empirically, this link is also examined, albeit not extensively. This study is complementary to the literature described below. Citron and Nickelsburg (1987) were the first to include a political instability variable into an empirical model of country risk and foreign borrowing. Their research was prompted by the foreign debt crises of the 1980’s, particularly in Latin America. The authors use a logit model with a binary default indicator and data from 5 countries between 1960 and 1983. For political instability, they use a variable that resembles a five year moving

(11)

11

aggregate of the number of changes in government. Citron and Nickelsburg (1987) find a positive, large, and significant effect of the political instability variable on default (Citron & Nickelsburg, 1987). Balkan (1992) acquires similar results using a probit model and a sample with 31 countries stemming from 1971 to 1984. Controlling for macroeconomic indicators, he finds a positive relationship between his political instability index which measured the amount of social unrest in a given year and the probability of debt rescheduling in the next year (Balkan, 1992).

De Haan, Siermann and Van Lubeck (1997) critically examine the results of Citron and Nickelsburg (1987) and Balkan (1992) and derive different results. De Haan, Siermann and Van Lubeck (1997) use a probit model with a rescheduling dummy and data for 65 developing countries from 1984 to 1993. They include an indicator for low political violence and an indicator for high political violence as proxies for political instability and find no significant effect with the correct sign. De Haan, Siermann and Van Lubeck (1997) conclude that political instability is already reflected in economic aggregates. Bordo and Oosterlinck (2005) run probit and logit regressions with a data for 29 countries from the gold standard period from 1880 to 1913. Only 9 of the 29 countries experienced a default during this period which makes the research limited. From their indicators of political instability, they find that coups and revolutions increase the probability of default, while government changes reduce the likelihood of default (Bordo & Oosterlinck, 2005).

Using a different methodology, Van Rijckeghem and Weder (2009) argue that political institutions matter in explaining sovereign default. They execute the CART methodology as Manasse and Roubini (2009) to predict safety from default. This method is not similar to standard regression analysis. Their sample consists of 73 countries between 1974 and 2000. The authors use various political variables to estimate the effect on sovereign default. They use variables that proxy for political instability, constraints, and polarization like the number of veto players, the nature of the political system, the length of tenure of the chief executive, a polarization index, and a variable that tracks election years. Van Rijckeghem and Weder (2009) conclude that constraints on the executive help in avoiding default but only in conjunction with strong liquidity and macroeconomic fundamentals. They also find that democracies and parliamentary systems perform better in avoiding defaults than non-democratic or presidential systems (Van Rijckeghem & Weder, 2009). Manasse, Roubini, and Schimmelpfennig (2003) also find that presidential election years have explanatory power in predicting debt crises.

The outcome that parliamentary regimes are differing from presidential regimes is supported by the empirical work of Kohlscheen (2009). Kohlscheen (2009) argues that the institutional setting is the crucial determinant of debt repayment. He develops a game theory model along the lines of different forms of government and claims that parliamentary countries are less prone to default ceteris paribus. Furthermore, constrained executives and coalition governments are less prone to default, and political turnover increases the probability of default. Kohlscheen (2009) constructs a variable that measures political turnover by taking the number of changes of the executive in the last 10 years. Kohlscheen (2009) takes a sample of 59 democratic countries from 1976 to 2003 and employs probit regression to test his main hypotheses. Kohlscheen (2009) finds that parliamentary democracies are less prone to default than their presidential counterparts. He also finds that veto players and lower political turnover makes rescheduling less likely (Kohlscheen, 2009Saiegh (2009) executes similar research but distinguishes between single party and multiparty coalitions as determinants of sovereign

(12)

12

default. He runs a probit model with a default dummy as dependent variable with cross-national data for 48 countries from 1971 to 1997. His main finding is that debt repudiation is significantly lower in countries that are governed by a multiparty coalition compared to countries with a single-party government (Saiegh, 2009).

Other studies emphasize the effect of political risk and institutions on sovereign spreads in developing countries. The empirical findings suggest that domestic political factors influence the sovereign debt market. Moser (2007) investigates the relationship between changes within the government and sovereign bond spreads in Latin America. He finds that sovereign bond spreads rise with roughly one percentage point on the day of announcement of a cabinet changes involving the economics or finance minister. Furthermore, the average bond-spread level is higher after a change in the cabinet than before. Block and Vaaler (2004) employ political business cycle theory to examine the effect of election dates on sovereign spreads and ratings. They find that the proximity of elections has a negative effect on sovereign ratings. They also find that credit spreads are higher in pre-election years and decrease with the proximity of the election (Block & Vaaler, 2004). Bussière and Mulder (2000) study the ‘tequila’ crisis of 1994 and the Asian crisis of 1997 and examine whether political instability had an effect on the economic vulnerability of countries in these periods. Their main finding is that not only pre-electoral periods but also post-pre-electoral periods are characterized by higher economic vulnerability (Bussière & Mulder, 2000).

3. Theoretical Framework

This theoretical framework provides the link between political instability and sovereign default. The main hypothesis of this paper is that political instability increases the likelihood and the magnitude of a sovereign default. In this model, political instability is resembled by electoral uncertainty that an incumbent political party faces. Electoral uncertainty is denoted by the probability of political turnover, which is the probability that the political party is replaced in office after elections. The incumbent political party uses debt and deficit strategically to constrain the choices of a successive party. If the probability of being out of office in the next period increases, the relative cost of a default is lower. More debt is accumulated, which increases the probability of default.

In the early models of political instability and the strategic use of debt, countries did not have the ability to default on their debt. However, the dynamic stochastic model of Cuadra and Sapriza (2008) includes a default decision where political parties can decide to default on the country’s foreign debt. In Cuadra and Sapriza (2008), the political system is modelled using the modelling of Alesina and Tabellini (1990) and Ozler and Tabellini (1991). Sovereign default is modelled using the willingness to pay approach based on Eaton and Gersovitz (1981). The model of this study is based on Alesina and Tabellini (1990), Ozler and Tabellini (1991), and Cuadra and Sapriza (2008). The full model is drawn from Carmignani (2003) and Yu (2012).

3.1 The Model

Consider a two-period model with a small open economy where two types of domestic agents are represented by a political party. The domestic agent acts as a generic voter i and maximizes the following utility function:

(13)

13

𝑈𝑖= 𝐸 {∑[𝛼𝑖𝑢(𝑥𝑡) + (1 − 𝛼𝑖)𝑢(𝑦𝑡)] 2

𝑡=1

} (1)

where t denotes time, E is the expectations operator, x and y are two types of public goods consumed each period, and 𝛼𝑖 is a non-negative parameter (𝛼𝑖 ≤ 1). The parameter α is the

weight placed on good x. The utility function is concave and increasing with 𝑢′_{(∙) > 0 and}

𝑢′′_{(∙) < 0.}

At the end of the first period, elections take place. There are two political parties, X and Y, can participate in these elections and compete for votes. The objective function of generic party 𝑗 (𝑗 = 𝑋, 𝑌) is given by (1) with 𝛼 = 𝛼_𝑗 and 1 ≥ 𝛼_𝑋 > 𝛼_𝑌≥ 0. Taking the utility function into account, this means party X cares more about public good x than about good y. While using the same logic, party Y cares more about good y than about good x. Hence, the parameter α embodies the disagreement of the individuals on whether political parties should consume good

x or good y. The preferences α of the political parties differ and can either diverge or converge.

Divergence of preferences on the supply of public goods in α is synonymous for increasing political fragmentation or polarization where public goods resemble policy. Voters cast their vote for the party that maximizes their utility given 𝛼𝑋 and 𝛼𝑌, which are assumed to be

exogenously given. The parameter 𝛼_𝑖 completely characterizes the preferences and voting behavior of individuals. This implies that the median voter theorem applies (Carmignani, 2003). The position of the median voter changes with π, with 0 ≤ 𝜋 ≤ 1 . The parameter π thus creates political instability that is characterized by an incumbent party facing a possible loss of power in the next period. That is to say, the incumbent party faces a positive probability of political turnover by which the incumbent loses power the next period.

There is one unit of output available to the government in period 1 and e units in period 2 to finance the purchase of public goods. Output e in period 2 is bounded by e ∈ [1, 𝑒̅] and is characterized by a continuous uniform distribution with a probability density function 𝑓(𝑒) = 1

𝑒̅−1 . The units of output available in period 2 for the incumbent party are more than

output available in period 1. This encourages the incumbent in period 1 to smooth consumption by borrowing. The incumbent chooses a debt level to smooth the supply of public goods in both periods. The value of output e is revealed at the beginning of period 2 and the incumbent maximizes utility given e. The debt issued in period 1 has to be repaid by the incumbent in period 2. The level of debt and interest on the debt that are accumulated by the incumbent in period 1 can be higher than the endowment e in period 2. In this case, the government defaults on the outstanding debt. In the case of 𝜋 = 0 or 𝜋 = 1 , when the outcome of the elections at the end of period 1 is certain, the incumbent does not face uncertainty and borrows up to the level where the utility of each period is equalized.

The incumbent party in period 1 issues debt in the form of bonds. These bonds are issued to foreign risk-neutral creditors facing a real interest rate of zero. The incumbent borrows 𝑞(𝑏)𝑏 in period 1 where b denotes the amount of bond issued and q(b) denotes the price of a bond. The government has to repay b in period 2 to the creditors. The probability of default is defined as follows:

𝑃𝑟(𝑒 ≤ 𝑏) = 𝑓(𝑏) =𝑏 − 1

(14)

14

From the probability of default can be seen that 𝑏 ∈ [1, 𝑒̅]. Equation (2) states that the incumbent party in period 2 defaults if the endowment e in period 2 is lower or equal than b. Furthermore, the probability of default is monotonically increasing in the level of debt. There is no recovery for creditors if the incumbent defaults. Lenders are willing to provide loans if the expected return equals the gross world interest rate, or:

𝑏(1 − 𝑓(𝑏))

𝑞(𝑏)𝑏 = 1 (3)

Solving (3) for 𝑞(𝑏) using (2) gives the price of a bond (𝑏) = 1 − 𝑓(𝑏) =𝑒̅−𝑏_𝑒̅−1 . If the amount of bonds issued increases, the price of a bond strictly decreases to compensate for the default risk that lenders face. Notice that the condition 𝑏 ∈ [1, 𝑒̅] ensures that the price of a bond cannot be negative.

The two political parties do not try to affect the outcome of the elections held after period 1. The incumbent in the first period chooses (𝑥1, 𝑦1, 𝑏) to maximize utility. The incumbent takes

the probability of being in office in period 2, (1 − 𝜋), as given and maximizes its objective function subject to the government budget constraints from the two periods. The incumbent in the second period observes (𝑏, 𝑒) and chooses (𝑥₂, 𝑦₂) to maximize its objective function given b and subject to the government budget constraints. If 𝑒 ≤ 𝑏, the government defaults and 𝑥₂= 𝑦₂ = 0.

The parameter α represents differences in policy platforms. If 𝛼_𝑋 and 𝛼_𝑌 are divergent, preferences about policies differ and the political sphere is polarized. When 𝛼_𝑋 and 𝛼_𝑌 are convergent, preferences of the political parties are more in line with each other. First, the case of full divergence is considered. The model is solved for period 2 when there is no uncertainty about the electoral outcome. Then the model is solved for period 1 when fiscal policy decisions are subject to the instability associated with the election at the end of period 1. Second, the general case is considered with 0 < 𝛼_𝑗< 1. In both cases, the incumbent will over-issue debt in period 1 in order to reduce the resources available for its successor in period 2 when there is a political instability characterized by a positive probability of political turnover.

3.2 Full Divergence Case

When there is full divergence of preferences, 𝛼𝑋 = 1 and 𝛼𝑌= 0. This means that party X

only gets utility from supplying good x, and party Y only prefers to supply good y according to the utility function from equation (1). The model is solved for the case where party X is in office in period 2. Employing the utility function from (1), it then solves the following maximization problem:

max

𝑥2,𝑦2 𝑢(𝑥2)

𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑥2 = (𝑒 − 𝑏 − 𝑦_{0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒}2, 𝑖𝑓 𝑒 > 𝑏)

(4)

In the case of full divergence, party X only cares about supplying good x. This will lead to a solution where 𝑥₂= 𝑒 − 𝑏 and 𝑦₂ = 0 in the case where 𝑒 > 𝑏. In words, party X spends period 2 output on good x after debt repayment, and will not spend anything on good y. The opposite applies for party Y when in office in period 2 as it only cares about supplying good y. The party

(15)

15

in office in period 2 will thus only supply the public good that it prefers in the case of full divergence of policies.

In period 1, the issuance of debt must be taken into account when the political parties decide how much to supply from the public good. Because of the elections at the end of period 1, uncertainty about the electoral outcome also affects the decision-making process. When party X is the incumbent in the first period, it will solve the following maximization problem:

max 𝑥1,𝑦1,𝑏 𝑢(𝑥1) + (1 − 𝜋)𝐸[𝑢(𝑥2)] 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑥₁= 1 + 𝑞(𝑏)𝑏 − 𝑦₁ 𝑥₂= (𝑒 − 𝑏 − 𝑦2, 𝑖𝑓 𝑒 > 𝑏 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 ) (5)

where the probability that party X is in office in period 2 is (1 − 𝜋). With full divergence of policy platforms, party X has no interest in supplying good y in both periods. Therefore, it will supply 𝑥₁= 1 + 𝑞(𝑏)𝑏 of good x and 𝑦₁= 0 of good y in period 1. At the end of period 1, party X has to compete in elections and is replaced in office with probability π. If it stays in office, it will supply 𝑥₂= 𝑒 − 𝑏, and if it loses the election with probability π, 𝑥₂= 0. The problem party X faces now is how much debt to issue in period 1 to smooth the supply of the pubic goods over the two periods. Taking account of full divergence, the distribution of e, assuming no default in period 2, substituting the budget constraints will result in the following maximization problem:

max 𝑏 𝑢(1 + 𝑞(𝑏)𝑏) + 1 − 𝜋 𝑒̅ − 1∫ (𝑢(𝑒 − 𝑏))𝑑𝑒 𝑒̅ 𝑏 (6)

Marginal utility is given by the first order condition: 𝐹𝑂𝐶: 𝑢′_{(1 + 𝑞(𝑏)𝑏)(𝑞}′_{(𝑏)𝑏 + 𝑞(𝑏)) −}1 − 𝜋

𝑒̅ − 1∫ (𝑢′(𝑒 − 𝑏))𝑑𝑒

𝑒̅ 𝑏

= 0 (7)

In the case of 𝜋 = 1 when there is maximum political instability, good x will not be supplied in period 2 and the first order condition will reduce to:

𝑢′_{(1 + 𝑞(𝑏)𝑏)(𝑞}′_{(𝑏)𝑏 + 𝑞(𝑏)) = 0} ₍₈₎

An interior solution for optimal 𝑏∗_{is guaranteed given the concavity of 𝑢(⋅). Furthermore,}

notice that (𝑞′_{(𝑏)𝑏 + 𝑞(𝑏)) can be rewritten as}𝑒̅−2𝑏

𝑒̅−1, using the definition of 𝑞(𝑏) and applying

the quotient rule on 𝑞′_{(𝑏)𝑏. Notice that}𝑒̅−2𝑏

𝑒̅−1 is decreasing in b. Party X is only maximizing utility

in the case of marginal utility being zero. As party X wants to maximize consumption of x in the first period, it borrows to the maximum level in period 1. Party X thus maximizes q(b)b in equation (8). To ensure marginal utility to be zero, (𝑞′_{(𝑏)𝑏 + 𝑞(𝑏)) must be set to zero. Notice}

that this term can be rewritten as 𝑒̅−2𝑏

𝑒̅−1. Setting this term to zero implies that optimal debt issued

in period 1 to smooth consumption is equal to 𝑒̅

2. In the case of 𝜋 = 0, where there is no political

(16)

16

side of equation (7) is maximized. Party X still wants to smooth consumption between period 1 and 2. To ensure that marginal utility is zero with a maximized second term on the left hand side, the first term on the left hand side must be larger than zero. This requires 𝑒̅−2𝑏_𝑒̅−1 > 0. Solving for b leads to optimal debt 𝑏∗_{that is strictly lower than}𝑒̅

2.

When π moves from 0 to 1, moving from no instability to maximum instability, the level of optimal debt will rise towards 𝑒̅

2 to remain the equality of (7) that assures party X is maximizing

utility. When political instability increases, the level of debt issued will increase which reduces the resources available in period 2 for the party that won the elections. A higher level of debt implies a higher probability of default and a higher size of default. In summary, in the case of diverging preferences, increasing political instability leads to more debt being issued which increases the probability of default, as can be seen from (2). The default in absolute terms will be higher when the amount of debt issued rises as the difference between b and e increases. Consequently, this model provides a theoretical explanation for the positive relationship between political instability and the probability of default as well as the size of default.

3.3 General Case

In the general case with 0 < 𝛼𝑗 < 1, preferences over policy differ but both parties will

supply both goods. Polarization is less stringent in the general case. As with the fully diverging case, I describe the solve the model for period 2 first. Consider the case of party j that is in office in period 2 solving the following maximization problem:

max 𝛼_𝑗 𝑥2,𝑦2 𝑢(𝑥₂) + (1 − 𝛼𝑗)𝑢(𝑦2) 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑥₂ = (𝑒 − 𝑏 − 𝑦2, 𝑖𝑓 𝑒 > 𝑏 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 ) (9)

After inserting the budget constraint, the first order condition for this maximization problem becomes:

𝛼_𝑗𝑢′_(𝑥

2) − (1 − 𝛼𝑗)𝑢′(𝑒 − 𝑏 − 𝑥2) = 0 (10)

The above equation explicitly defines the optimal choice of 𝑥2 and 𝑦2, as these are functions of

predetermined variables (𝛼𝑗, 𝑏, 𝑒). When the supply of both goods in period 2 is expressed as

functions of (𝛼𝑗, 𝑏, 𝑒), the supply of good x is 𝑥2= 𝐴(𝑎𝑗, 𝑏, 𝑒) and the supply of good y is

𝑦₂= 𝐵(𝛼_𝑗, 𝑏, 𝑒). Taking partial derivatives gives: 𝜕𝐴/𝜕𝑏 < 0, 𝜕𝐵/𝜕𝑏 < 0, 𝜕𝐴/𝜕𝛼_𝑗 > 0, and 𝜕𝐵/𝜕𝛼_𝑗< 0. The intuition behind this is that the resources available for the incumbent in period 2 decrease with the amount of debt issued in period 1. Furthermore, the party that cares more about good x, represented by a higher value of 𝑎_𝑗, will be inclined to provide more of good x and less of good y.

In period 1, the uncertainty of the electoral outcome must be taken into account. The incumbent’s choice of the optimal level of debt depends on the degree of political instability. Suppose party X is the incumbent in period 1 and there is a positive probability π that party X will lose office at the end of period 1. Taking the distribution of e into account, it chooses (𝑥1, 𝑦1, 𝑏) and solves the following problem:

(17)

17 max 𝑥1,𝑦1,𝑏 𝐸[𝑈₁𝑋_{] = 𝛼} 𝑋𝑢(𝑥1) + (1 − 𝛼𝑋)𝑢(𝑦1) +1 − 𝜋 𝑒̅ − 1∫ [𝛼𝑋𝑢(𝐴(𝛼𝑋, 𝑏, 𝑒)) + (1 − 𝛼𝑋)𝑢(𝐵(𝛼𝑋, 𝑏, 𝑒))]𝑑𝑒 𝑒̅ 𝑏 + 𝜋 𝑒̅ − 1∫ [𝛼𝑋𝑢(𝐴(𝛼𝑌, 𝑏, 𝑒)) + (1 − 𝛼𝑋)𝑢(𝐵(𝛼𝑌, 𝑏, 𝑒))]𝑑𝑒 𝑒̅ 𝑏 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑥₁= 1 + 𝑞(𝑏)𝑏 − 𝑦₁ (11)

Rewrite the budget constraint as a function of 𝑦₁ and substitute the budget constraint in the utility function. The first order conditions with respect to 𝑥1 and b are given by:

𝛼_𝑋𝑢′_(𝑥 1) − (1 − 𝛼𝑋)𝑢′(1 + 𝑞(𝑏)𝑏 − 𝑥1) = 0 (12) (1 − 𝛼𝑋)𝑢′(1 + 𝑞(𝑏)𝑏 − 𝑥1)(𝑞(𝑏) + 𝑞′(𝑏)𝑏) +1 − 𝜋 𝑒̅ − 1∫ [𝛼𝑋𝑢′(𝐴(𝛼𝑋, 𝑏, 𝑒)) 𝜕𝐴(𝛼_𝑋, 𝑏, 𝑒) 𝜕𝑏 𝑒̅ 𝑏 + (1 − 𝛼𝑋)𝑢′(𝐵(𝛼𝑋, 𝑏, 𝑒)) 𝜕𝐵(𝛼_𝑋, 𝑏, 𝑒) 𝜕𝑏 ] 𝑑𝑒 + 𝜋 𝑒̅ − 1∫ [𝛼𝑋𝑢′(𝐴(𝛼𝑌, 𝑏, 𝑒)) 𝜕𝐴(𝛼_𝑌, 𝑏, 𝑒) 𝜕𝑏 𝑒̅ 𝑏 + (1 − 𝛼𝑋)𝑢′(𝐵(𝛼𝑌, 𝑏, 𝑒)) 𝜕𝐵(𝛼_𝑌, 𝑏, 𝑒) 𝜕𝑏 ] 𝑑𝑒 = 0 (13)

Rewrite the first order condition with respect to 𝑥₁ and substitute in the first order condition with respect to b gives:

𝛼_𝑋𝑢′_(𝑥 1)(𝑞(𝑏) + 𝑞′(𝑏)𝑏) +1 − 𝜋 𝑒̅ − 1∫ [𝛼𝑋𝑢′(𝐴(𝛼𝑋, 𝑏, 𝑒)) 𝜕𝐴(𝛼_𝑋, 𝑏, 𝑒) 𝜕𝑏 𝑒̅ 𝑏 + (1 − 𝛼_𝑋)𝑢′_(𝐵(𝛼 𝑋, 𝑏, 𝑒)) 𝜕𝐵(𝛼𝑋, 𝑏, 𝑒) 𝜕𝑏 ] 𝑑𝑒 + 𝜋 𝑒̅ − 1∫ [𝛼𝑋𝑢′(𝐴(𝛼𝑌, 𝑏, 𝑒)) 𝜕𝐴(𝛼_𝑌, 𝑏, 𝑒) 𝜕𝑏 𝑒̅ 𝑏 + (1 − 𝛼_𝑋)𝑢′_(𝐵(𝛼 𝑌, 𝑏, 𝑒)) 𝜕𝐵(𝛼_𝑌, 𝑏, 𝑒) 𝜕𝑏 ] 𝑑𝑒 = 0 (14)

When 𝜋 = 0 there is no political turnover after period 1 and the incumbent is certainly staying in office. With 𝜋 = 0, the third term on the left hand side is 0 and we can reduce equation (14) to:

(18)

18 𝛼_𝑋𝑢′_(𝑥 1)(𝑞(𝑏) + 𝑞′(𝑏)𝑏) +1 − 𝜋 𝑒̅ − 1∫ [𝛼𝑋𝑢′(𝐴(𝛼𝑋, 𝑏, 𝑒)) 𝜕𝐴(𝛼_𝑋, 𝑏, 𝑒) 𝜕𝑏 𝑒̅ 𝑏 + (1 − 𝛼_𝑋)𝑢′_(𝐵(𝛼 𝑋, 𝑏, 𝑒)) 𝜕𝐵(𝛼𝑋, 𝑏, 𝑒) 𝜕𝑏 ] 𝑑𝑒 = 0 (15)

The partial derivatives of 𝑥₂ and 𝑦₂ with respect to b result in 𝜕𝐴/𝜕𝑏 < 0 and 𝜕𝐵/𝜕𝑏 < 0. Notice that the second term on the left hand side will be negative because of the signs of the partial derivatives. Recall that (𝑞′_{(𝑏)𝑏 + 𝑞(𝑏)) can be rewritten as}𝑒̅−2𝑏

𝑒̅−1 which is decreasing in b. To

ensure an interior solution, 𝑒̅−2𝑏

𝑒̅−1 must be strictly positive because 𝑢′ > 0. Solving for 𝑏 results in

optimal debt 𝑏∗_{that is strictly lower than}𝑒̅ 2.

In the case of political instability with 𝜋 > 0, the effect of a positive probability of turnover π on the level of debt b issued is ambiguous. The first effect is similar to the effect in the full divergence case with electoral uncertainty. Even when party X prefers to supply good y because 𝛼𝑌> 0, party X will always prefer good x over good y because 𝛼𝑋 > 𝛼𝑌. With a positive

probability to lose office during elections, party X will over-issue debt in the first period to constrain party Y in supplying good y in the second period with political instability. However there is a second effect in the case of 0 < 𝛼_𝑗< 1. Because of 𝛼_𝑋 > 0, party X knows that party Y would provide some of good x in period 2 if it is elected. The higher resources are in period 2, the higher is the supply of good x in period 2 by party Y. Borrowing in the first period by party X would reduce resources available to party Y in the second period if there is instability. This could even lead to the incentive for party X to run a surplus and issue less debt in period 1. The fact that party Y will spend more on good y than on good x in period 2, does not have to reduce party X’s utility in period 2 if the supply of good x is on the desired level from the point of view of party X.

The first effect is denoted as the “desired composition effect”, whereas the second effect is the “desired level effect”. When the composition effect dominates the level effect, political instability leads to more debt being issued by the incumbent in the first period. Which effect dominates depends on the form of the utility function 𝑢′_{(⋅). Alesina and Tabellini (1990) show}

that a sufficient condition for the composition effect to dominate the level effect is when −𝑢′′_{(𝑐)/[𝑢(𝑐)]}2_{is decreasing in c, where c denotes the consumption of public goods.}

3.4 Testable Implications

Political instability is resembled by the parameter π. When 𝜋 > 0 there is political instability and the incumbent faces a possible loss of power in the next period. In the full divergence case, increasing political instability leads to a higher level of optimal debt. A higher level of debt implies a higher probability of default. Increasing political instability thus raises the likelihood of default. This theoretical result can be tested empirically with data on political instability and the probability of default. Additionally, there is a second mechanism that can be investigated empirically. Increasing political instability leads to a higher level of optimal debt which in turn leads to an increase in the magnitude of the default, if the political party in the second period faces 𝑒 ≤ 𝑏. Political instability thus leads to a higher default probability and a

(19)

19

higher size of default in the model. These two outcomes of the model are examined empirically in the remainder of the paper.

The preferences of the different political parties or political parties are given by the parameter α. When preferences over policy are fully divergent, 𝛼𝑋 = 1 and 𝛼𝑌= 0. The

parameter α fully resembles the preferences and voting behavior of the utility maximizing individuals. Divergence of preferences in α is synonymous for polarization in society. When societies are polarized, the differences over policy preferences differ greatly. In the full divergence case, when preferences over policy differ to the utmost, political instability increases the likelihood of default and the scale of default. In the case where preferences differ but not fully diverge, the effect of political instability on the probability of default is ambiguous. Whether the effect is positive or negative depends on the composition effect and the level effect and the form of the utility function. If there is a level effect, debt issued will always be higher in the full divergence case compared to the general case. Hence, polarization leads to a higher probability of default and a higher size of default. This theoretical premise is also tested empirically.

4. Data and Methodology

4.1 Sample Selection

The sample includes annual data on 136 countries with debt in default from 1975 to 2012. A full list of countries in the sample can be found in the Appendix. The sample selection is based on the Database of Sovereign Defaults of the Bank of Canada’s Credit Rating Assessment Group (CRAG) (Beers & Nadeau, 2014). This database will be mentioned from now on as the CRAG database. All the countries that had debt in default between 1975 and 2012 are included in the sample. For some transition economies, particularly former Soviet Union states, data is only available from the 1990’s onwards. For the Cook Islands, Korea Democratic Republic, and Yugoslavia most data is unavailable.

The CRAG Database contains data from various official and private sector sources and is novel as it was last updated in the beginning of 2014. The CRAG Database includes debt from official and private creditors and includes all sovereign defaults between 1975 and 2013. Data for 2013 is missing for most countries in the CRAG Database so this year is not included in the sample. Thereby, Ireland and Portugal are the two countries that are excluded from the sample because they only defaulted in 2013. The CRAG Database shows that total debt in default increased from 1,860 million current USD in 1975 to 395,893 million USD in 1990 at the height of the debt crises in Latin America. Debt in default declined thereafter until the Russian and Argentinean defaults of 1998 and 2001 when defaulted debt increased again to 290,801 million USD in 2002. After that, defaulted debt declined with the debt cancellation for Heavily Indebted Poor Countries (HIPC). Nevertheless, debt in default surged again in 2012 with the default of Greece that consisted of 312,420 million USD. Total defaulted debt as a percentage of Gross World Product follows the same pattern. This ratio increased during the 1980’s, declined during the 1990’s to increase again in 1998 and 2001. Thereafter it declined further until 2011 when the European sovereign debt crisis emerged on the surface.

(20)

20

4.2 Dependent Variables

A table with definitions and a full description of all the variables used can be found in the Appendix. DEF is the dependent variable that resembles the probability of default. DEF is a default dummy and is constructed from the defaulted debt data of the CRAG Database. Most empirical research uses default data from Standard and Poor’s (see Kohlscheen, 2009). The definition of default that the Bank of Canada and Standard and Poor’s use are essentially the same so the outcomes are comparable to other studies. In a given year the binary indicator DEF takes the value 1 when there was defaulted debt, and the value 0 otherwise. For the size of default, Debt in Default to GDP (DDEBT) is used. This variable is constructed by dividing debt in default by GDP using the debt in default data from the CRAG Database and GDP data from the World Bank. Both defaulted debt and GDP are measured in nominal million US Dollars. Debt in default is scaled by GDP to control for the fact that economically large countries in absolute terms presumably suffer from higher absolute size of defaults.

Debt in default includes the following types of defaulted debt: International Monetary Fund (IMF) lending, official creditors lending, private creditors lending, foreign currency bank loans, foreign currency bonds, and local currency debt. Hence, all debtors are included. While other studies exclude certain debtors from the sample, for example debtors that do not have a have a sovereign credit rating (see Kohlscheen, 2009). It also captures many debt categories where other studies exclude certain debt categories, for instance debt to official creditors (Van Rijckeghem & Weder, 2009). By including many debt categories, general conclusions on sovereign default can be made.

In the CRAG Database, default is defined as an occurrence where debt service is not paid on the due date (or within a specified grace period), payments are not made within the time frame specified under a guarantee, or, absent an outright payment default, in the circumstances where creditors incur material economic losses on the sovereign debt they hold. These circumstances consist of: a reduction of interest rates or the extension of maturities on outstanding debt, an exchange offer where old debt is swapped for new deb on less economic terms, government purchases of debt at substantial discounts to par, redenomination of foreign currency debt into local currency debt on economic terms, swaps of sovereign debt for equity on less-economic terms, and the conversion of central bank notes into new currency of less-than-equivalent face value. This definition is in line with the definition applied by international credit rating agencies and the definitions used in other empirical work on sovereign default (see Saiegh, 2009; Manasse & Roubini, 2009). Notice that not all countries that have debt in default missed payments on their obligations. For example, Greece did not miss any payments on their debt obligations in 2012 but the restructuring consisted of the reduction of interest rates and charges on outstanding debt, and the extension of maturities.

4.3 Independent Variables

The main hypothesis derived from the model is that political instability increases the probability of default as well as the size of default. For political instability, three different indicators are employed. The first indicator is directly derived from the model. It is a binary indicator tracking political turnover (PTURN) which takes the value 1 if political turnover is observed in a given year and 0 otherwise. This is the first study where actual political turnover is used instead of changes in the head of the government or an index indicating political turnover (see Kohlscheen, 2009). The political turnover variable is constructed using the

(21)

21

Database of Political Institutions (DPI) of the World Bank which was last updated in 2013 (Beck et al., 2001). Data on political turnover and the other variables extracted from the DPI is missing for the whole sample period for Antigua and Barbuda, the Cook Islands, Dominica, Sao Tome and Principe, Serbia, Seychelles, St. Kitts and Nevis, St. Vincent and Grenadines, and Tonga.

The political turnover variable is constructed using two indicators from the DPI: the numbers of years that an executive is in office and the number of years the ruling party of the executive is in office. In the DPI, for the years that an executive is in office, years are counted in which the executive was in power as of January 1, or was elected but had not taken office as of January 1. Counting thus starts at 1, a year after the executive came into office. For transition governments, like former Soviet Union countries, tenure starts at independence and counting starts at the year following independence.

When counting starts at 1 in a given year, the actual change of the executive takes place in the year before. The same applies to the change of the party of the executive. In this way, the years in office of the executive and the ruling party are translated into two variables that track the change of the executive and the change of the ruling party. PTURN combines these changes and is a binary indicator which takes the value 1 if change of the executive and/or the ruling party is observed in a given year and 0 otherwise. Generally, the change of the executive and the ruling party coincide. In other instances only the executive changes. This arises when an executive from the same political party is voted into office or political parties do not exist. It is rarely the case that only the ruling party changes when the executive stays the same. This generally happens for transition countries were political parties were non-existent before.

The second political instability variable is a variable capturing election years. An election

(ELEC) in a given year is captured by a binary indicator which takes the value 1 if an election is

observed in a given year and 0 otherwise. This variable includes elections for the executive as well as elections for the legislature. Other empirical studies found that periods around elections are characterized by high economic vulnerability (Bussière & Mulder, 2000). Furthermore, the absence of recent election leads to lower default probabilities (Van Rijckeghem and Weder, 2009).

The third political instability indicator is the binary variable WAR that is given a value 1 if the country was in war in a given year and 0 otherwise. Van Rijckeghem and Weder (2009) also adopt a war variable as a measure of political instability. War is defined as an event of sustained combat, involving organized armed forces resulting in a minimum of 1000 battle-related fatalities in a twelve-month period. The data is from the Correlates of War (COW) Database and includes inter-state wars, intra-state wars and extra-state wars. Inter-state wars are wars between territorial entities that are recognized as states. The invasion in Iraq in 2003 is an example of an inter-state war. Intra-state wars are wars between entities predominantly fought within a state. For example, the civil wars in Rwanda and DR Congo in the 1990’s are included in the intra-state war category. These wars do not necessarily take place on the own territory of the combatant. Extra-state wars are wars that are fought between a state and a non-state entity outside the borders of the state. An example of an extra-state war is the war between Western countries and the Taliban in Afghanistan from 2001 onwards.

The second hypothesis from the model states that the divergence of policy preferences raises the probability of default and the size of default. To examine the effect of policy preferences an index capturing polarization (POLAR) is used. For POLAR, the Competitiveness of Participation Index is employed. This index resembles polarization in a given country and is

Political instability and sovereign default