Macro, institutional and political influence on CDSs
An exploratory research of the sovereign credit default swaps of 13 Latin American countries.
Leon Piening
Master thesis Msc. Finance and Msc. IFM University of Groningen
Supervisor: Dr. J. O. Mierau June 2015
Abstract
In this study an OLS regression is conducted to explorer whether sovereign credit default swaps (CDSs) are primarily influenced by macro variables or by country-specific variables. The results are promising and evidence is found that institutional and political variables mainly influence CDSs. This research provides a foundation for more elaborate research to the country-specific variables that explain the CDS. Finally, CDSs may be a good, market-based, alternative to sovereign credit ratings for investors in order to calculate the country risk premium.
Keywords: CDS; macro influence; country-specific risk; institutional risk;
political risk; PCA
JEL-Classification: F34; G15; G17
1. Introduction
Globalization led to increased integration across financial markets over the last decades. Because of globalization, country risk can arguably be reduced through diversification. As Driessen and Laeven (2007) stated: “The gains from international portfolio diversification appear to be largest for countries with high country risk”. However, this does not explain the different equity risk premiums used by practitioners when investing in other countries, as indicated by Fernandez, P., Linares, and Fernandez A. (2014). Their survey indicates that investor do take country risk into account when investing across boarders.
Sovereign credit ratings are frequently used and examined to reflect country-specific risk (Damodaran, 2015). These ratings are lagging markets, since rating agencies, e.g. Fitch, Moody’s or S&P, first have to process public information before sovereign ratings are changed.
In this research the focus is on another country-specific risk measure, the credit default swap (CDS). The CDS may be used as a market-based alternative for sovereign credit ratings. Since the availability of CDSs is growing, it becomes an increasingly applicable measure to calculate the country risk premium (CRP) (Damodaran, 2015). A market-based alternative of the sovereign credit ratings to calculate the CRP is preferable, as this gives investors the opportunity to respond more quickly to changes in a country’s risk level.
The determinants of CDSs are frequently investigated, e.g. by Benzoni Collin-Dufresne, Goldstein, and Helwege (2012), Longstaff, Pan, Pedersen and Singleton (2011) Pan and Singleton (2008), Wang and Moore (2012). Interesting are the results of Longstaff et al. (2011), who found that more than half of the CDSs of 26 countries can be explained by macro variables. However, I think that country-specific variables are under-represented in their research.
One of the papers that did include a variety of country-specific variables is the paper by Alexe, Hamme, Kogan & Lejeune (2003). Alexe et al. (2003) used institutional and political variables to explain the sovereign credit ratings of 69 countries. It is important that no lagging variable for the sovereign credit rating was included in their model. Lagging variables are able to predict future values of sovereign credit ratings reliably, since ratings are very stable over time. The model that Alexe et al. (2003) estimated for their in-sample period was able to explain 99.1% of the out-of-sample sovereign credit ratings.
define the determinants of the CDS. Firstly, the statement of Longstaff et al. (2011) that CDSs are mainly explained by macro variables is tested. Several OLS regression models are built, first with macro variables, then with country-specific variables, and finally combining all variables. The results show that macro variables do explain some of the CDSs. Though, when combined with country-specific variables, the institutional and political state definitely explain some of the CDSs. This result provides evidence that the CDS may be an applicable market-based measure for the calculation of the CRP.
Secondly, Alexe et al. (2003) built a model that can reliably predict sovereign credit ratings for 69 countries. Probably these variables are able to explain the CDS as well. Therefore, a selection of the institutional and political risk variables, that were defined by Alexe et al. (2003), are included in this paper. The results show that the institutional and political variables are significant for multiple countries.
The focus of this research is on Latin America, where thirteen countries have CDSs available. This is an interesting area to explorer CDSs, since it is known to be a risky area with varying political stability (“Political risks in Latin America”, 2012).
Two important results emerge from the analysis. First, when macro and country-specific variables are combined in one model, not only macro variables explain the CDS. Institutional and political variables were highly significant in my analysis, showing promising results for future research and the applicability of CDSs to calculate the CRP.
Second, the significant values of the country-specific variables indicate that CDSs represent risks, comparable to the risks represented by sovereign credit ratings.
Thus, this paper contributes to the existing literature in the sense that evidence is found that institutional and political risks influence the CDS, so CDSs are not solely influenced by macro variables.
2. Literature Review
This section is constructed as follows. First, the country risk premium is described. Next, the main advantages and disadvantages are discussed. Hereafter, empirical evidence is provided for the actual use of the CRP. Finally, several studies are described that try to explain different estimators for country risk and how these studies provide the basis of our research.
Country Risk Premium
For several decades the country risk is a subject of discussion, and around the 1980s the determinants of country risk and country risk analysis were frequently examined, by e.g. Burton and Inoue (1987), Merrill (1982), Saini and Bates (1984). Damodaran (2015) wrote a paper that is updated annually about country risk and the CRP.1 In his paper Damodaran (2015) elaborately discusses whether country risk it is applicable for investors to use CRPs. He suggests that every country has its own country risk premium, influenced by individual country risk. Diverse arguments exist in for and against the use of CRPs, of which the most important are: country risk is diversifiable, a global asset-pricing model exists, historical equity premiums, survey premiums. These will be more elaborately discussed.
Country risk diversifiable
Some say that a CRP does not exist, because country risk is diversifiable. For country risk to be diversifiable, it should be idiosyncratic risk, since systematic risk is the only important risk for estimating the risk premium, as explained by e.g. Sharpe (1964) and Lintner (1965) in the Capital Asset Pricing Model. All, or almost all of the risk should be country-specific to be idiosyncratic and thus, diversifiable. This means that the correlation across markets has to be low, because when the correlation of market returns across countries is significantly positive/negative, the risk at least partly exists of non-diversifiable risk for which investors should require a risk premium. Multiple studies found evidence that the correlation between capital markets is significant, especially when downside volatility is high (Bekaert & Harvey, 1997; Longin & Solnik, 2001; Yang, Tapon & Sun, 2006). Secondly, in case the risk is idiosyncratic, the risk should indeed be diversified. In 1999 Stultz researched
1 The original version of this paper by Damodaran is from 2008.
2 In-sample and out-of-sample testing is more clearly defined in the methodology section.
whether a CRP exists by investigating segmented and open markets. On the one hand, in a segmented market, the investor is diversified within the segmented market, though he cannot diversify in other markets, thus different risk premiums may exist for every market. On the other hand, in open markets the marginal investor has access to all markets across the world and can diversify globally, leading to one global risk premium. Though recent globalization led to more integrated markets and the possibility to diversify globally, the effect of a more global market is lower than expected. One of the reasons for the low effect of more global capital markets on the risk premium that is confirmed through research is that significant home bias still exists for investors (Kang & Stultz, 1997; Lewis, 1999; Ahearne, Griever, & Warnock, 2004). Hence, country risk appears not to be fully diversifiable risk and investor’s home bias of shows that investors do not utilize the full potential diversification.
Global asset-pricing model
Nowadays capital markets are highly interconnected, arguably leading to one global market to which every investor has access. When this global market exists, one global asset-pricing model could be estimated, where country-specific risk is incorporated in the beta of each asset. However, in practice, little evidence exists that betas capture country-specific risk in global asset-pricing models. Since the global equity indices are market weighted, companies in developed markets get higher betas than companies in emerging markets when using these models (Damodaran, 2015). This is not what one would expect: emerging markets are assumed to be riskier than developed markets. Damodaran (2015) states that there are creative fixes that practitioners have used to get around this problem, but they seem to be based on little more than the desire to end up with higher expected returns for emerging market companies. Lewis (2011) confirms this and in her research she stated that: Despite decades of increased globalization, the prices of many internationally traded assets continue to depend upon local risk factors.
Historical equity premiums
historical equity risk premiums of emerging markets were also found in a research conducted by Salomons and Grootveld (2003).
Survey premiums
Recently, Fernandez et al. (2014) published their survey, which they conduct yearly, regarding the market risk premium (equity risk premium) that is used in daily business by companies, professors and theorist for different countries, The reported average premiums vary widely across markets and are higher for riskier emerging markets. Though these survey results do not exclusively prove that a CRP should be used, clearly CRPs are used frequently by a variety of practitioners.
Based on the arguments given for and against using CRPs, it can be stated that most arguments and empirical evidence are tilted in favor of using a CRP.
Methods to estimate the CRP
Multiple methods exist to estimate the CRP, of which two are described in this paper: sovereign credit ratings and CDSs.
Using sovereign credit ratings published by rating agencies (e.g. Moody’s, Standard & Poor’s (S&P) and Fitch) is the easiest method. The magnitude of the spread, indicated by the sovereign credit rating, is used as an estimate of country risk. There are 2 disadvantages of using the sovereign credit spread: first, rating agencies lag markets as the market circumstances have to be evaluated and processed before sovereign credit ratings are adapted. Second, sovereign credit ratings are focused on default risks, other risks exist that influence equity markets, resulting in risk for investing in a country.
that they are primarily focused on default risk, this may obscure other risks that could still affect equity markets.
This study explorers the factors that explain the CDS and whether it is indeed only influenced by default risk, or that other country-specific risks are also captured in the CDS. The CDS is a very popular subject to investigate, probably because its relative short existence. The determinants of companies’ CDSs are studied frequently by various authors, e.g. by Alexander and Kaeck (2008), Aunon-Nerin, Cosin, Hricko and Huang (2002), Di Cesare and Guazzarotti (2010), and Ericsson, Jacobs and Oviedo (2009). However, for this study sovereign CDSs are examined. The influence of macro variables or interdependence between market is studied by, for example, Benzoni et al. (2012), Longstaff et al. (2011) Pan and Singleton (2008), Wang and Moore (2012). Especially the study by Longstaff et al. (2011) uses a broad variety of variables to explain the magnitude of the CDSs of 26 countries worldwide. The main finding of the research by Longstaff et al. (2011) is that over half of the magnitude of the CDSs is explained by macro variables, representing the global market state. Longstaff et al. (2011) built a regression model for every individual country, consisting of variables, representing the local and global situation of capital markets. Multiple of their institutional variables as well as macro variables were highly significant. Their conclusion that over 50% of the CDS is explained by macro variables is striking with the meaning of CDSs, which is to reflect the risk of default of a sovereign. This should be influenced mainly by the local state of the country, and less by macro variables.
tested the model out-of-sample2 with a high R-squared of 99.1%, which is quite impressive without including the lagged value of the sovereign credit rating. Since sovereign credit ratings are fairly stable over time, in many models authors include the lagged value to get high R-squares. As the variables used by Alexe et al. (2003) appear to be good predictors of sovereign credit ratings, in this study these variables are tested on the market-based CDSs.
In their model, Alexe et al. (2003) included a variety of institutional variables that describe the economic state of a country. Beside the financial variables, the World Bank governance indicators were included as a proxy for the political stability of a country. The governance indicators are yearly published and consist of six dimensions: voice and accountability, political stability and absence of violence, government effectiveness, regulatory quality, rule of law, and control of corruption. Governance is measured by these dimensions and used as an approximation of political risk, while originally defined by Kaufman et al. (1999a) as: “The traditions and institutions by which authority in a country is exercised.” As emphasized by Kaufmann et al. (1999, a&b) and by Alexe et al. (2003), these dimensions, per couple, measure the same instrument. Therefore, Alexe et al. (2003) chose to use one dimension per instrument. This led to the inclusion of the following three dimensions: Political stability, government effectiveness and corruption. When the political variables are excluded from the model, the R-square and adjusted R-square are lower than with the political variables included, indicating a loss in predictive power resulting from the omission of the three political variables (Alexe et al. 2003).
In this study, a combination of the models of Alexe et al. (2003) and Longstaff et al. (2011) will be used. On the one hand, Alexe et al. (2003) focus on a variety institutional and political variables for explaining the less responsive measure to market fluctuations; sovereign credit ratings. On the other hand, Longstaff et al. (2011) focus on the macro economic influences on the market-based measure; CDSs. This research merges these two models into one model that explorers the market-based CDSs. The goal of this paper is to determine whether CDSs are to a bigger extent explained by macro variables or by institutional and political variables.
3. Data
This section describes the variables that are used in this study. A variety of macro and institutional variables are identified, of which most are defined similar to the variables of either Alexe et al. (2003) or Longstaff et al. (2011). The data is collected from several sources: the governance indicators and domestic credit to the private sector provided by banks are retrieved from the World Bank, information about Treasury yields is retrieved from the H.15 Federal Reserve Statistical Release, and all other data is retrieved from Thomson Reuters Datastream. A number of variables are defined slightly different than the ones used by Alexe et al. (2003) and Longstaff et al. (2011), because Bloomberg terminals are not available at the University of Groningen. For analyzing the data and conducting tests, the statistical software package of StataSE 13 is used. The focus of this study is on the Latin American area, since strong economic growth is present here, while still experiencing significant political risk and populist policy making (“Political risks in Latin America”, 2012). Besides, this area is an interesting research subject, because CDSs for these countries are still relatively young market derivatives. The dataset used is composed of six years of data: from January 1st, 2008 until December 31st, 2013.3 In Appendix 1.1 the descriptive statistics of all variables and the correlation matrix of the independent variables can be found.
In the remainder of this section, the dependent and independent variables are defined.
Dependent variable
Credit default swap (CDS)
As previously mentioned, the market-based CDSs are used as the dependent variable. In Latin America, data on CDSs is available from January 2008 onwards for thirteen countries.4 These countries are: Argentina, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, El Salvador, Guatemala, Jamaica, Panama, Peru, Uruguay and Venezuela. The CDSs are quoted in basis points that the protection buyer has to pay over the nominal amount of a loan to the protection seller. The magnitude of the CDS is an indicator of default risk and reflects the country risk for that specific country. The larger the magnitude of the CDS the more basis
3 The research period is constrained by the data availability. First the CDS data is available from January 2008 onwards, second most recent data on the governance indicators, published by the World Bank (Worldwide Governance Indicators, 2015), is from 2013.
points an investor has to pay over the nominal amount of a loan to secure it. Thus, a higher magnitude of the CDS indicates a higher risk of default for a country.
Subsequently, several macro and institutional variables are defined, which are likely to explain the magnitude of the CDSs of the thirteen Latin American countries.
Independent variables
The remaining of this section describes all independent variables and provides the expected relation to the CDS.
Macro variables
In accordance with Longstaff et al. (2011), some of the variables are used to reflect the state of the global economy. As sovereigns typically have extensive economic relationships with other countries, this state influences the ability of sovereigns to repay their debts, and thus the magnitude of CDSs (Longstaff et al., 2011). Dooley & Hutchison (2009), Eichengreen et al. (2012), and others, found evidence that financial markets are correlated; shocks in the U.S. market spread to other financial markets, especially in the event of a crisis. Therefore, U.S. financial market variables are applicable to reflect the state of the global economy, especially because the U.S. is not one of the countries included in this study. To capture the state of the global economy, two variables are defined, one reflects the equity market state, and the other reflects the fixed market state.5
Dow Jones Industrials Average (DJIA)
The DJIA is used to represent the U.S. equity market state. Given the high correlation6 between U.S. stock markets, the DJIA is assumed to be applicable to reflect movements in the state of the global economy. First, the log-return of the DJIA is calculated, and then this variable is defined as the excess return of the DJIA over the one-month Treasury-Bill return. This is similar to the definition by Longstaff et al. (2011). A positive excess return signals that the global economy improves. When the global economy’s state improves, the magnitude of the CDS is likely to decrease. Hence, a negative coefficient is expected.
Treasury Yield (Yield)
For presenting the state of the fixed market, the percentage return on the five-year constant maturity Treasury (CMT) rate is included. This is in accordance with the variable used by Longstaff et al. (2011). The expectations for the yield equal those of the DJIA variable: a positive return on the yield indicates an improvement of the global economy, causing the magnitude of the CDS to decrease. Thus, a negative coefficient is expected.
Institutional Variables
Both, Alexe et al. (2003) and Longstaff et al. (2011) used a variety of variables that reflects the country-specific risk. A combination of their variables is used in this study.7 The country-specific variables are split in two main categories: institutional variables, representing the economic state of a country, and political variables, representing the political stability of a country. Institutional risk and political risk are important indicators for country-specific risk, as stated by Moffett, Stonehill and Eiteman (2008). Eventually, four institutional variables are used and one principal component8, reflecting the political state of a country.
Exchange rate (Xrate)
Both, Alexe et al. (2003) and Longstaff et al. (2011), use the exchange rate. For the Xrate variable I use the definition of Longstaff et al. (2011), which is more straightforward. The Xrate variable represents the exchange rates, expressed as units of the local currency per U.S. dollar. An increase in this variable indicates a depreciation of the currency relative to the U.S. dollar. In accordance with the findings of Longstaff et al. (2011), positive coefficients are expected, since a depreciation of the currency is likely to increase the CDSs.
GDP per capita (GDP)
This variable is calculated as the Gross Domestic Product per capita for each country and converted into international dollars.9 The international dollar has the same purchasing power for each country as the dollar in the U.S., so this makes the variable comparable across
7 Initially the trade balance as percentage of GDP was included. However, some ambiguity exists whether a trade surplus should be interpreted as favorable or unfavorable, since a positive trade balance can be highly negative for consumer (Bhala, 2008). Thereby, the trade balance variable led to multicollinearity. For the same multicollinearity reason “Debt as a percentage of GDP” is excluded.
8 How a principal component analysis is conducted and what variables are included in this component is defined later in the data section.
countries. GDP per capita is an indicator of the relative wealth of a country and its level of development (Alexe et al., 2003). GDP was highly significant and positive in the study by Alexe et al. (2003). This result indicates that an increase in the GDP per capita, calculated using the purchasing power parity, leads to a higher rating by agencies. Thus, in this study a negative coefficient is expected: the CDS of a country decreases, when GDP increases.
Local index (local)
As a measure for the state of the local financial market, the log-return of the local stock index is included. This variable was highly significant in the tests that Longstaff et al. (2011) conducted. Several countries included in this study have no stock market, or no data is available on Thomson Reuters Datastream about the country’s stock market.10 For the countries missing this variable, it is excluded from the regression analysis, leading to one less explanatory variable. This has no further consequences for the analysis. All significant coefficients of the local index that were found by Longstaff et al. (2011) are negative, indicating that a positive return on the local stock market reduces the CDS. Therefore, in this study a negative coefficient is expected as well.
Political variables
The macro variables that account for the state of the global economy and institutional variables that account for the state of the local economy are definded. Now, a variety of institutional variables are defined that account for the political situation of a country. In accordance with Alexe et al. (2003) political variables are included to show whether the political stability of a country impacts the CDSs, because they found evidence that political variables add value in explaining the CDS of a country. To account for the political state of a country, the six governance dimensions11, which are yearly published by the World Bank, are included (Worldwide Governance Indicators, 2015).
Only three of the six dimensions of governance were included by Alexe et al. (2003) in order to avoid or at least limit multicollinearity. As described above, all governance indicators are supposed to capture information about the political stability of a country, leading to high correlation across them. The governance indicators are highly correlated per pair (Kaufmann et al., 1999, a&b), as both dimensions of each pair measure the same, one of the two is said to
be redundant. Alexe et al. (2003) avoided this redundancy by excluding one variable for each pair.
Contrary to their research, in this study all the six dimensions of governance are included, in this way all information captured by the governance indicators remains included. In order to decrease redundancy in the data and to calculate the common factor that influences the CDSs, a principal component analysis (PCA) is used. 12
In the remainder of this section, the definitions of the governance indicators, which reflect the political state of a country, are provided. Further, one extra variable is defined: domestic credit provided to the private sector by banks as a percentage of GDP. This variable is also included by Alexe et al. (2003) and is highly correlated with the dimensions of governance.13
Voice and Accountability (voice)
This variable reflects to which extent a country's citizens are able to participate in selecting their government, as well as freedom of expression, freedom of association, and a free media (Worldwide Governance Indicators, 2015).
Political Stability and Absence of Violence (stability)
This captures the perceptions of the likelihood of political instability and/or politically-motivated violence, including terrorism (Worldwide Governance Indicators, 2015).
Government Effectiveness (effectiveness)
The quality of public services, of the civil service and the degree of its independence from political pressures, of policy formulation and implementation, and the credibility of the government's commitment to such policies is captured by the government effectiveness (Worldwide Governance Indicators, 2015).
Regulatory Quality (quality)
12 PCA is used in case of redundancy, Shlens (2014) provides an annually updated tutorial for PCA and states that it is advantageous to use PCA when a common basis, which is a linear combination of the original basis, best re-expresses the dataset. In the “Methodology” section PCA is described more elaborately.
Regulatory quality reflects perceptions of the ability of the government to formulate and implement sound policies and regulations, which permit and promote private sector development (Worldwide Governance Indicators, 2015).
Rule of Law (law)
This variable captures the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence (Worldwide Governance Indicators, 2015).
Control of Corruption (corruption)
This variable captures the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as "capture" of the state by elites and private interests (Worldwide Governance Indicators, 2015).
All together, these six dimensions reflect the political state of a country, which may influence the likelihood of default and thus, the magnitude of a country’s CDS. The higher the values of these variables, the less likely the country is to default. The variables are measured on a [-3.5; 3.5] interval and: “the governance indicators are based on a large number of different data sources, capturing the views and experiences of survey respondents and experts in the public and private sectors, as well as various NGOs.” (World Governance Indicators FAQ, 2014).
Domestic credit to private sector provided by banks to GDP (credit)
Alexe et al. (2003) claim to be the first in using this variable to predict country risk ratings and find significant positive results for it. The variable is defined as the ratio of the domestic credit provided by the banking sector to the GDP. Correlated to the development of the economy; the indirect lending by savers to investors becomes more efficient and gradually increases assets relative to the GDP (Alexe et al., 2003). The ratio reflects the financial depth and efficiency of the country’s financial system.
investment. Hence, the credit variable is likely to be dependent on the political situation and is included in the PCA.
All these seven variables increase when the political state of a country is more stable, thus, the less likely a country is to default. Therefore, the expectation for the principal component is that the coefficient is negative, indicating that a better political state reduces the magnitude of the CDS.
In the following section the PCA is described and conducted on the political variables and credit variable in order to reflect the influence of the political state of a country on the magnitude of CDSs, hereafter the regression model is defined.
4. Methodology
In this section the PCA is conducted. Hereafter, the OLS regression model is defined, which is used to explorer the determinants of the CDSs.
Principal Component Analysis
The PCA is used to form the political variables and credit variable into one or a few principal components. This principal component reflects the common basis of the political variables that influences CDSs the most. Brooks (2008) stated that: “PCA is primarily used as a dimensionality reduction in situations with a large number of closely related variables and where I wish to allow for the most important influences from all of these variables at the same time.” The main argument to use PCA is the level of redundancy: in case of high correlation between independent variables (multicollinearity), redundancy is high. When this occurs it is better to use a principal component to reflect the common basis (Shlens, 2014). In Appendix 2.1 the correlation matrix of the originally defined independent variables is provided. The high correlations between the governance indicators and the credit variable are an indication that PCA may reduce the redundancy. The principal components are calculated as linear combinations of the original data in the following way (Brooks, 2008)
𝑝! = 𝛼!!𝑥!+ 𝛼!"𝑥!+ ⋯ + 𝛼!!𝑥!
𝑝! = 𝛼!"𝑥!+ 𝛼!!𝑥!+ ⋯ + 𝛼!!𝑥! (2) … … … …
where 𝛼!" are coefficients to be calculated, representing the coefficient on the jth explanatory variable in the ith principal component. Thereby, it is required that the sum of the squares of the coefficients for each principal component is one
! 𝛼!"!
!!! = 1 ∀ 𝑖 = 1, … , 𝑘 (3)
These components are constructed through constrained optimization, leading to uncorrelated principal components. Thus, the problem of multicollinearity between the independent variables is resolved through PCA. The principal components can be understood as the eigenvalues of (X’X), where X is the matrix of observations on the original variables (Brooks, 2008).
To confirm whether the use of PCA is justified, the Kaiser-Meyer-Olkin14 (KMO) measure for sampling adequacy is used as described by Kaiser (1974). The results of the sampling adequacy test are provided in Appendix 2.2, all scores are high with an overall result of 0.8161, indicating that the use of PCA is justified.15
The model should only include the principal components with enough explanatory power. Since the goal of principal components is both, to reduce redundancy and to reduce the number of variables, the applicable number of components should be determined. Despite sometimes being designated as arbitrary, the Cattell’s scree test (Cattell, 1966) and Kaiser’s K1 method (Kaiser, 1960) are the most often used tests to determine the applicable number of principal components. In this case both tests clearly indicate that only one of the six principal components should be used.
First, according to Cattell’s scree test the eigenvalues all component are plotted and a line is drawn through their values. From the point where the line levels off, the principal components should be excluded. As a result only the first component should be included according to this test.16 Second, Kaiser’s K1 method confirms that only the first principal component should be included. According to this method only principal components with an eigenvalue above one should be used. The reasoning behind this method is that the eigenvalues an eigenvalue of one accounts for as much variance as one single variable, thus a
14 The KMO measure is a PCA post estimation tool, which shows sampling adequacy of the component-loadings.
component with an eigenvalue greater than one provides more summarizing power than an original variable (Zwick & Velicer, 1986).
Finally, in appendix 2.4 it is shown that the correlation of the original political variables and the credit variable with the principal component is very high, besides the correlation diagram of all independent variables is giving, with the principal component included instead of all political variables. Thus, all these methods verify the use of a PCA for these data.
Since all variables and the principle component are determined, the research model is defined next. The data is analyzed, using an Ordinary Least Squares (OLS) model, which is defined equally to the models used by Alexe et al. (2003) and Longstaff et al. (2011). In order to confirm whether the independent variables are able to explain the magnitude of the CDS, the following OLS regression equation is fitted
𝑌!" = 𝛼!+ 𝛽!" !
!!!
∗ 𝑋!"+ 𝜀!" , 1
where the dependent variable Y is the CDS, stated in basis points, 𝛼 is the constant, the independent variables 𝛸 are the independent variables, including the principal component, 𝑖 represents eacht country and 𝑡 represents time. The coefficient of each independent variable on the CDS is represented by 𝛽 and 𝜀 is the error term.
Alexe et al. (2003) used this OLS model, as it is non-recursive, meaning that it is built without using lagging variables. The R-squared and adjusted R-squared are definitely high when including the lagged value of the CDS, therefore, non-recursive models are better for exploring the relationship between the independent variables and the dependent variable, since.
The subsequent section covers the results of the OLS models and discusses whether the outcome of the model corresponds to the expectations. Afterwards, the predictive power of the in-sample estimated model for the out-of-sample data is discussed.
5. Results and discussion
When analyzing the panel data it appeared not to be applicable for this data. Almost no significant results were found, indicating that not one unique equation exists that predicts the CDS for the Latin American countries. In Appendix 3 the panel data results, using fixed effects17, are provided. Instead of using panel data, the regression model is built step by step for each country individually. Firstly, the macro variables are regressed on the CDSs. Secondly, the institutional and political variables are regressed on the CDSs, and finally, all independent variables are regressed on the CDSs. By gradually building the model, it becomes apparent what type of variables have more influence on the CDSs, macro or institutional. The data is analyzed on a significance level, α = 0.10.
In table 1 the results are provided for the first regression model, using only macro variables. The coefficient of the world index (DJIA) is significant for six of the thirteen countries. All signs are negative, according to the expectations. This indicates that a positive return on the global capital market leads to a reduction of the CDS of a country. This result coincides with the findings of Longstaff et al. (2011).
Table 1: Regression results of the macro variables on the CDSs
Macro variables
Country DJIA Yield R-Sq. Argentina -18.27 9.453 0.017 Brazil -4.232*** 0.406 0.123 Chile -2.671*** 0.187 0.105 Colombia 1.385 -0.310 0.014 Costa Rica -5.185*** 0.387 0.178 Dominican Rep. -7.634 0.472 0.022 El Salvador -4.240* 0.472 0.041 Guatemala -2.595 0.379 0.030 Jamaica -2.424 1.346 0.014 Panama -4.608*** 0.251 0.108 Peru -4.506*** 0.0864 0.130 Uruguay -4.594 0.292 0.039 Venezuela -16.58 1.297 0.036 *** p<0.01, ** p<0.05, * p<0.1
The second macro variable, the percentage return on the five-year constant maturity Treasury yield is in no case significant. This confirms the findings of Longstaff et al. (2011) that the CDS of a country is only rarely influenced by the fixed income market return.
Hence, it appears that from the macro variables only the state of the global market influences the CDS of these thirteen Latin American countries. The R-squared for each regression is still very low. This is caused by the low number of macro variables included, but also indicates that institutional variables may contain much information about the CDSs.
Table 2 provides the regression results of the institutional variables on the CDSs. The first variable, the return on the local index, is only significant for Panama. For the other six countries with a local stock index, the coefficient is not significant. The sign of the coefficient of Panama’s stock return is as anticipated: a positive return on the local stock market leads to a reduction in the CDS.
The exchange rate is significant for five of the twelve countries, the Panamanian Balboa is pegged at par to the U.S. dollar, leading to the exclusion of this variable. Four of the five significant components are positive, indicating that a depreciation of the currency relative to the U.S. dollar has a negative impact on the magnitude of the CDS.
Table 2: Regression results of the institutional variables on the CDSs
Institutional Variables
Country Local Xrate GDP Component R-Sq. Argentina 1.832 475.9*** 0.0480 139.5 0.298 Brazil -0.936 -0.235 -0.0760*** 26.72** 0.177 Chile -1.664 1.068 -0.00753** 0.608 0.104 Colombia 0.401 -1.403 0.0692*** -8.365* 0.465 Costa Rica 13.90* 0.0843*** -38.03*** 0.200 Dominican Rep. 24.12 -0.594*** 172.9*** 0.312 El Salvador -4,631* 0.112 9.577 0.197 Guatemala 26.02*** -0.147*** -3.882 0.260 Jamaica 45.07*** -0.639*** -23.63** 0.324 Panama -5.503** -0.0101** -9.753** 0.333 Peru -1.120 -3.359 -0.0580* 14.90 0.247 Uruguay 2.943 -0.102*** 47.84*** 0.500 Venezuela -0.103 -2.398 -0.238*** 34.38 0.208 *** p<0.01, ** p<0.05, * p<0.1
variables, nine are negative, indicating that an increase in the GDP per capita leads to a reduction of the CDS.
The last variable, the principle component, is significant for most of the countries: seven out of thirteen. Surprisingly, some ambiguity exists in the signs, of the significant countries, three coefficients are positive, while only three are negative. Obviously, all coefficients were expected to be negative, since an improvement of the political situation of a country should reduce the CDS.
The R-squared for each of the regressions in this model is higher than in the first model. Apparently, these four institutional and political variables explain a lot more of the CDSs than the two macro variables. The results of these regressions provide evidence that CDSs are at least partially explained by institutional variables.
In the third model the macro, institutional and political variables are regressed on the CDSs, the results are shown in table 3. In this model the excess return on the DJIA is significant for 5 countries, one country less than the model where only macro variables were tested on the CDSs. Still, all five significant coefficients are negative, in accordance with what is anticipated.
The return on the five-year constant maturity Treasury rate is still significant for none of the countries included in this sample. Thus, the CDSs of the Latin American countries included in this study seem not to respond on the return in the U.S. fixed income markets.
The local variable is again only significant for Panama, with a negative coefficient. This is in accordance with the expectation for the coefficient, since a positive return on the Panamanian local market is expected to reduce the CDS.
The exchange rate is significant for four of the twelve countries with currencies that are not pegged to the U.S. dollar. Three of the four significant coefficients are positive, as expected: a depreciation of the local currency relative to the U.S. dollar leads to an increase in the CDS.
The fifth variable, GDP per capita, is again the most significant variable. Eleven of the thirteen coefficients are significant, of which nine are negative. A negative coefficient is in accordance with expectations: an increase in GDP per capita should lead to a decrease of the CDS.
Table 3: Regression results of the macro and institutional variables on the CDSs Macro variables Institutional variables
Country DJIA Yield Local Xrate GDP Component R-squared Argentina 5.069 6.870 -1.064 478.7*** 0.0356 134.3 0.306 Brazil -4.986** 0.230 1.940 -0.0255 -0.0648*** 23.89** 0.258 Chile -2.251* 0.287 -0.482 -2.665 -0.00608* -0.563 0.154 Colombia 0.293 -0.551 0.347 -1.234 0.0689*** -7.825 0.477 Costa Rica -4.933*** -0.199 10.40 0.0944*** -41.30*** 0.374 Dominican Rep. -8.655 0.0317 25.40 -0.658*** 204.1*** 0.337 El Salvador -9.144*** 0.118 -7,356*** 0.0464 25.41** 0.346 Guatemala 0.213 0.731 28.06*** -0.153*** -3.973 0.275 Jamaica 1.134 0.674 46.56*** -0.633*** -23.03** 0.329 Panama -2.099 0.339 -4.643* -0.0103** -8.240* 0.352 Peru -4.588** 0.435 0.341 -3.469 -0.0671** 23.07 0.319 Uruguay -3.259 0.780 1.064 -0.0986*** 45.66*** 0.521 Venezuela -15.12 2.802 -0.356 -2.518 -0.263*** 20.69 0.237 *** p<0.01, ** p<0.05, * p<0.1
Putting macro and institutional variables together increased the R-squared to values between 0.154 and 0.521. This is not as high as the R-squared for the elaborate set of macro variables included in the model of Longstaff et al. (2011), but it does show that the institutional and political variables included in this research have some explanatory power.
The third model is built gradually. This combined model has the highest R-squared for each country, providing evidence that the CDSs cannot be explained, solely by using macro variables. It is favorable for investors that institutional and political variables are able to explain parts of the CDS, because this indicates that CDSs may be applicable as a market-based measure for country risk.
Overall, the outputs of these models show that CDSs of the Latin American countries in this sample are influenced by the state macro variables, as well as the country-specific variables. Two variables, the GDP per capita and exchange rate relative to the U.S. dollar, are most significant in explaining the CDS. Unfortunately, no unambiguous evidence is found for the influence of the political state on a country’s CDS.
Robustness checks
political data is used to test whether this data does give unambiguous results for the influence of the political state on a country’s CDS.
For the model including all independent variables, the coefficients are estimated for the first five years of data: from Jan. 1st 2008 until Dec. 31st 2012. The estimated coefficients are tested on the last year of data: from Jan. 1st 2013 until Dec. 31st 2013. In Table 4 the R-squares are provided for the fit of the predicted values in the out-of-sample period with their true values.
From the R-squares it is clear that the in-sample estimated model is quite bad in predicting the CDS values for the out-of-sample period. The bad fit of the model is possibly caused by the research period, because it is a crisis and after-crisis period. Kaminsky and Schmukler (1999) state that investors over-react to bad news and during crisis periods investors are driven by herd instincts.
Table 4: The R-squares of the out-of-sample estimates with their true values
Country R-sq. Argentina 0.1798 Brazil 0.0403 Chile 0.2485 Colombia 0.0409 Costa Rica 0.4387 Dominican Rep. 0.1114 El Salvador 0.3085 Guatemala 0.0099 Jamaica 0.0006 Panama 0.3498 Peru 0.1797 Uruguay 0.0784 Venezuela 0.1024
component with the largest eigenvalue for each country is included in the analysis. Finally, the regression, using the robustness set of political components instead of the original political component, led to the results provided in table 5. The ICRG principal component is abbreviated as: ICRG.
For this robustness check I solely focus on the coefficients of the ICRG variable. Although the coefficients of the other variables obviously slightly change, because uncorrelated independent variables are unavoidable, no interesting changes occurred. The ICRG variable is significant for four of the thirteen countries. The signs of these significant coefficients are primarily negative, three out of four, which according to the presumptions.
These results confirm that the political state is an important factor in explaining the CDS, at least for some of the countries. Thereby, the sign of the coefficient is more in accordance with expectations when the ICRG data is used to represent the political state.
Table 5: Robustness check including the ICRG principal component Macro Variables Institutional Variables
Country DJIA Yield Local Xrate GDP ICRG R-squared Argentina 8.925 2.560 4.029 471.6*** -0.277*** -35.93 0.407 Brazil -5.759** 0.130 2.305 -0.162 -0.0326** -1.064 0.271 Chile -2.128 0.409 -0.660 -1.841 -0.0184* -5.924 0.168 Colombia -0.185 0.211 -0.168 -0.0916 0.109*** -3.754 0.649 Costa Rica -4.595*** -0.0195 8.594 0.0453 -27.03** 0.448 Dominican Rep. -1.282 -0.569 -14.19 -0.304*** 8.088 0.280 El Salvador -8.047** -0.203 -6,989** 0.00186 0.306 0.201 Guatemala -0.348 0.177 28.35** -0.360*** -4.388 0.436 Jamaica -1.421 -0.797 54.76*** -0.759*** 7.079 0.356 Panama -1.365 0.400 -5.035* -0.0490*** -22.70*** 0.444 Peru -3.126 0.367 0.252 -5.195 -0.0255 17.73** 0.339 Uruguay -4.500 0.0521 1.488 -0.0240 -39.19* 0.609 Venezuela -17.57 1.235 0.469 -3.366 -0.449** 46.18 0.163 *** p<0.01, ** p<0.05, * p<0.1
These results slightly contradict the claim by Longstaff et al. (2011) that CDSs are mainly influenced by macro variables. In this study the significance of institutional variables to explain the CDS is confirmed. Longstaff et al. (2011) regressed primarily macro variables on the CDSs of 26 countries, which explained a large part of the CDSs, while in this study primarily institutional variables were regressed on the CDSs, supported by a few macro variables.
institutional variable reflecting the political state. The second robustness check confirmed the importance of the political state, since multiple coefficients were significant. Moreover, the signs of the ICRG coefficients were more compliant with expectations. Hence, the political state is an important independent variable to explain the CDS.
6. Conclusion and recommendations
In this study I explorer whether CDSs are better explained by macro variables, or by institutional and political variables. It is especially important that CDSs are influenced by the institutional and political variables, since these variables are important to reflect country-specific risk (Moffett et al., 2008). The results show that both, macro and country-country-specific variables are good in predicting CDSs. My results provide a good foundation for more elaborate models that explorer the relationships between institutional variables, political variables and the CDS.
Gradually building a model, based on research by Alexe et al. (2003) and Longstaff et al. (2011), I found evidence that, e.g. GDP per capita, the exchange rate, and the political state of a country influence the magnitude of a country’s CDS.
These results are especially important for investors managing portfolios in these countries. The CDSs can be more reliably predicted, when a model is constructed with unique coefficients for each country. For investors, these unique models can be used as a market-based substitute to sovereign credit ratings. When, eventually, one succeeds to build reliable models that explain CDSs, the disadvantage of sovereign credit ratings, that they lag the market, can be overcome.
I began to build a model that explains CDSs and, in contrast to the claim of Longstaff et al. (2011), I found evidence that CDSs are based on country-specific risk variables.
This research, however, is just the beginning. More research is needed that identifies country-specific risks, influencing CDSs. Thereby, some extra macro variables might be needed to complement these models. As long as country-specific variables can primarily explain the CDSs the inclusion of macro variables is no issue.
years. In crisis periods, investors show herd-behavior (Kaminsky and Schmukler, 1999). Perhaps, this behavior led to the confusing results on political variable.
Future research may overcome the limitations by using an extensive dataset with more developed countries included, which have data available for longer periods.
References
Books and Papers
Afonso, A., Gomes, P., & Rother, P. (2011). Short-and long-run determinants of sovereign debt credit ratings. International Journal of Finance & Economics, 16(1), 1-15. Ahearne, A. G., Griever, W. L. & Warnock, F. E. (2004). Information costs and home bias: an
analysis of US holdings of foreign equities. Journal of International Economics, 62(2): 313-336.
Alexander, C., & Kaeck, A. (2008). Regime dependent determinants o spreads. Journal of
Banking & Finance, 32(6), 1008-1021.
Alexe, S., Hammer, P. L., Kogan, A. & Lejeune, M. A. (2003). A non-recursive regression model for country risk rating. RUTCOR-Rutgers University Research Report RRR, 9: 1-40.
Aunon-Nerin, D., Cossin, D., Hricko, T., & Huang, Z. (2002). Exploring for the determinants of credit risk in credit default swap transaction data: Is fixed-income markets' information sufficient to evaluate credit risk?. FAME Research Paper, (65).
Bekaert, G. & Harvey, C. R. (1997). Emerging equity market volatility. Journal of Financial
economics, 43(1), 29-77.
Benzoni, L., Collin-Dufresne, P., Goldstein, R. S. and Helwege, J. (2012). Modeling credit contagion via the updating of fragile beliefs, Federal Reserve Bank of Chicago
Working Paper 2012-04.
Bhala, R. (2008). International trade law: Interdisciplinary theory and practice, 3rd ed.
Danvers, Mass: LexisNexis.
Brooks C. (2008). Introductroy Econometrics for Finance. New York, NY: Cambridge University Press.
Burton, F. N., & Inoue, H. (1987). A country risk appraisal model of foreign asset expropriation in developing countries. Applied Economics, 19(8), 1009-1048.
Implications. Social Science Research Network. Retrieved from http://ssrn.com/abstract=1274967.
Damodaran A. (2015) Equity Risk Premiums (ERP): Determinants, Estimation and Implications. Social Science Research Network. Retrieved from
http://ssrn.c om/abstract=2581517.
Di Cesare, A., & Guazzarotti, G. (2010). An analysis of the determinants of credit default swap spread changes before and during the subprime financial turmoil. Bank of Italy
Temi di Discussione (Working Paper) No, 749.
Donadelli, M. & Prosperi, L. (2011). The Equity Risk Premium: Empirical Evidence from Emerging Markets, Working Paper. Social Science Research Network. Retrieved from http://ssrn.com/abstract=1893378.
Dooley, M., & Hutchison, M. (2009). Transmission of the US subprime crisis to emerging markets: Evidence on the decoupling–recoupling hypothesis. Journal of International
Money and Finance, 28(8), 1331-1349.
Driessen, J., & Laeven, L. (2007). International portfolio diversification benefits: Cross-country evidence from a local perspective. Journal of Banking & Finance, 31(6), 1693-1712.
Eichengreen, B., Mody, A., Nedeljkovic, M., & Sarno, L. (2012). How the subprime crisis went global: evidence from bank credit default swap spreads. Journal of International
Money and Finance, 31(5), 1299-1318.
Ericsson, J., Jacobs, K. and Oviedo, R. (2009). The determinants of credit default swap premia. Journal of Financial and Quantitative Analysis 44(1): 109-132.
Fernandez, P., Linares, P., & Fernandez A. I. (2014). Market Risk Premium used in 88
countries in 2014: A survey with 8,228 answers. Social Science Research Network. Retrieved from http://papers.ssrn.com/sol3/Papers.cfm?abstract_id=2450452.
Hull, J. (2012). Options, Futures, and other Derivatives (8th ed.). Boston, MA: Pearson
Education, Inc.
Kaminsky, G. L., & Schmukler, S. L. (1999). What triggers market jitters?: A chronicle of the Asian crisis. Journal of international money and Finance, 18(4), 537-560.
Kang, J. K. & Stulz, R. (1997). Why is there a home bias? An analysis of foreign portfolio equity ownership in Japan. Journal of Financial Economics, 46(1): 3-28.
Kaufmann, D., Kraay, A. and Zoido-Lobaton, P. (1999b). Governance Matters. World Bank
Policy Research Department Working Paper 2196.
Lewis, K. K. (1999). Trying to explain home bias in equities and consumption. Journal of
Economic Literature, 37(2): 571-608.
Lewis, K. K. (2011). Global asset pricing. National Bureau of Economic Research Working
Paper 17261.
Lintner, J. (1965). The Valuation of Risk Assets and the Selection of Risky Investments in Stock Portfolios and Capital Budgets. Review of Economics and Statistics. 47(1): 13–37.
Longin, F. and Solnik, B. (2001). Extreme Correlation of International Equity Markets.
Journal of Finance, 56: 649-675.
Longstaff, F. A., Pan, J., Pedersen, L. H., & Singleton K. J. (2011). How Sovereign Is Sovereign Credit Risk? American Economic Journal: Macroeconomics, 3(2): 75-103. Mellios, C., & Paget-Blanc, E. (2006). Which factors determine sovereign credit ratings?. The
European Journal of Finance, 12(4), 361-377.
Merrill, J. (1982). Country Risk Analysis. Columbia Journal of World Business, 17: 88-91. Moffett, M. H., Stonehill, A. I. & Eiteman, D. K. (2008). Fundamentals of
multinational finance. Pearson Addison Wesley.
Pan, J., & Singleton, K. J. (2008). Default and recovery implicit in the term structure of sovereign CDS spreads. The Journal of Finance, 63(5), 2345-2384.
Saini, K. G., & Bates, P. S. (1984). A survey of the quantitative approaches to country risk analysis. Journal of Banking & Finance, 8(2), 341-356.
Salomons, R., & Grootveld, H. (2003). The equity risk premium: emerging vs. developed markets. Emerging Markets Review, 4(2), 121-144.
Schmukler, S. L. (2004). Financial globalization: gain and pain for developing countries. Federal Reserve Bank of Atlanta Economic Review, 89(2): 39-66.
Sharpe, W. F. (1964). Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk. Journal of Finance. 19(3): 425– 42.
Shlens, J. (2014). A tutorial on principal component analysis v. 3.02. Google Research
Mountain View: CA 94043.
Stulz, R. M. (1999). Golbalization, corporate finance, and the cost of capital. Journal of
Wang, P., & Moore, T. (2012). The integration of the credit default swap markets during the US subprime crisis: Dynamic correlation analysis. Journal of International Financial
Markets, Institutions and Money, 22(1), 1-15.
Yang, L., Tapon, F., & Sun, Y. (2006). International correlations across stock markets and industries: trends and patterns 1988–2002. Applied Financial Economics, 16(16): 1171-1183.
Zwick, W. R., & Velicer, W. F. (1986). Comparison of five rules for determining the number of components to retain. Psychological bulletin, 99(3), 432.
Websites
Political risks in Latin America. (2012, July). Retrieved from https://www.worldriskreview.com/publications/articles/viewer?ID=qYAbCl9EY%2F M%3D.
MRCI's Inter-Market Correlations, DJIA & S&P 500 correlations. (n.d.) Retrieved from http://www.mrci.com/special/index.php.
Geary-Khamis method. (2006, February 06). Retrieved from
https://stats.oecd.org/glossary/detail.asp?ID=5528.
International Country Risk Guide (ICRG) (n.d.). The Political Risk Services Group. Retrieved from http://www.prsgroup.com/about-us/our-two-methodologies/icrg.
World Governance Indicators FAQ. (2014). World Bank Group. Retrieved from http://info.worldbank.org/governance/wgi/index.aspx#faq-3.
Appendix
Appendix 1
Appendix 1.1
Table 6: Correlation matrix of all independent variables
DJIA Yield Local Xrate GDP Principal
DJIA 1 Yield 0.2318 1 Local 0.3779 0.0907 1 Xrate -0.0283 0.0247 -0.0357 1 GDP 0.0507 0.0349 0.0507 0.057 1 Principal 0.01 0.0055 -0.1195 -0.0768 0.2168 1
Table 7.1: Descriptive statistics of the variables
Table 7.2: Descriptive statistics of the variables, continued
Variable n Mean S.D. Min Mdn Max
Table 7.3: Descriptive statistics of the variables, continued
Variable n Mean S.D. Min Mdn Max Peru CDS 71.0 156.5 71.2 86.0 138.7 465.5 DJIA 72.0 0.2 5.7 -15.6 1.4 13.9 Yield 72.0 -0.1 14.3 -33.8 -2.6 43.0 Local 72.0 0.0 10.7 -36.9 -0.1 32.5 Xrate 72.0 -0.1 1.4 -3.3 -0.3 3.9 GDP 72.0 10136.0 1074.6 8878.5 10067.1 11763.8 Principal 72.0 -1.1 0.3 -1.6 -1.1 -0.8 Uruguay CDS 61.0 228.8 124.9 131.6 184.7 640.5 DJIA 72.0 0.2 5.7 -15.6 1.4 13.9 Yield 72.0 -0.1 14.3 -33.8 -2.6 43.0 Local 0.0 . . . . . Xrate 72.0 0.0 3.3 -6.1 -0.6 13.1 GDP 72.0 16848.2 1932.1 14361.2 16902.6 19595.5 Principal 72.0 2.7 0.2 2.5 2.7 3.1 Venezuela CDS 70.0 1084.5 491.6 598.0 973.7 3222.9 DJIA 72.0 0.2 5.7 -15.6 1.4 13.9 Yield 72.0 -0.1 14.3 -33.8 -2.6 43.0 Local 72.0 5.6 13.3 -29.4 3.1 36.2 Xrate 72.0 2.0 12.9 0.0 0.0 100.1 GDP 72.0 17172.1 723.4 16201.7 17082.3 18197.8 Principal 72.0 -4.3 0.3 -4.8 -4.3 -3.8
The descriptive statistics in table 7.1; 7.2 and 7.3 show the number of observations (n), the mean, the standard deviation (S.D.), the minimum (Min), the median (Mdn) and the maximum (Max).
It can be seen that Costa Rica, the Dominican Republic, El Salvador, Guatemala, Jamaica and Uruguay have no local stock market, or no data available. For these countries no local index return is included as an institutional variable, this has no further consequences.
Appendix 2
Appendix 2.1
Table 8: Correlation matrix of the independent variables, including the governance indicators (without the principle component)
Appendix 2.2
Table 9: The results of the KMO measure
Variable KMO Corruption 0.7942 Effectiveness 0.9259 Stability 0.7518 Quality 0.8838 Law 0.822 Voice 0.7819 Credit 0.6991 Overall 0.8161
DJIA Yield Local Xrate GDP Corruption EffectivenessStability Quality Law Voice Credit DJIA 1.0000
Yield 0.2318 1.0000 Local 0.3779 0.0907 1.0000 Xrate I0.0283 0.0247 I0.0357 1.0000 GDP 0.0507 0.0349 0.0507 0.0570 1.0000
Corruption I0.0011 I0.0005 I0.1036 I0.0651 0.2847 1.0000
Effectiveness 0.0120 0.0069 I0.1186 I0.0684 0.1998 0.8961 1.0000
Appendix 2.3
Figure 1: A scree plot of the eigenvalues of the principal components
Appendix 2.4
Table 10: Correlation of the governance indicators and the credit variable with the principal component
Principal Principal 1 Corruption 0.9211 Effectiveness 0.9535 Stability 0.7731 Quality 0.8712 Law 0.9748 Voice 0.9429 Credit 0.6275 0 2 4 6 Ei g e n va lu e s 0 2 4 6 8 Number
Appendix 3
Table 11: Panel data results, model one includes the countries with a local stock market, model 2 includes the countries without a local stock market
VARIABLES Model 1 Model 2
DJIA -5.413 -5.809*** (3.695) (2.221) Yield 1.604 1.210 (1.361) (0.877) Local -1.882 (2.363) Xrate 2.118 2.223 (3.696) (2.948) GDP -0.0175 -0.0300*** (0.0142) (0.0105) Component 165.0** 174.2*** (77.18) (56.60) Constant 788.6*** 838.8*** (209.6) (139.8) Observations 568 904 R-squared 0.019 0.024 Number of country 8 13
Country FE YES YES
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
