A Quantitative Analysis on the Determinants of Sovereign Credit Ratings and Their Role during the Recent European Government Debt Crisis of 2009 Peter Aalvink

(1)

A Quantitative Analysis on the Determinants of Sovereign Credit Ratings and Their

Role during the Recent European Government Debt Crisis of 2009

Peter Aalvink

January, 2012

Supervisor: dr. J.P.A.M. (Jan) Jacobs

Abstract

Recently, much attention is put on credit rating agencies (CRAs) and their rating methodology. It is said that they have not been able to predict the crisis and have aggravated the crisis by downgrading sovereign credit ratings during the crisis. Using ordered probit and ordered logit models I have found that Fitch has not significantly modified their ratings methodology. Nevertheless Fitch’s sovereign credit ratings were not impeding the government debt crisis either because Greece’s government debt was overrated before the crisis. However, I have found that during the crisis credit ratings have not exacerbated the crisis by suddenly underrate individual credit ratings.

Keywords: credit ratings, government debt crisis, Greece

(2)

1 – Introduction

According to Fitch, “sovereign issuer default ratings [and debt issue ratings] are a forward-looking assessment of a sovereign’s capacity and willingness to honor its existing and future obligations in full and on time.” 1 In other words, credit ratings reflect the likelihood of a default, where default could mean a cancelling of the payments, or a delay of the payments.

In the economic landscape, credit rating agencies (CRAs) provide credit risk information to financial markets. CRAs construct methodologies to measure credit risk and set up a framework of credit ratings to value that risk. These ratings are used by bond issues, investors and also by regulators. At bond issues CRAs can give advice on setting interest rates and on debt-structuring. In that fashion, investors use credit ratings to find attractive investment opportunities to optimize their investment portfolio. As well, credit ratings play an important role in the Basel III framework to set minimum reserve requirements and they are also used by the SEC for similar purposes. In general, credit ratings provide help in steering the allocation of capital, therefore credit ratings can make sure that the financial market behaves rather stable.

However, sharp downgrades of credit ratings following the subprime mortgage crises and the recent downgrades resulting from weakened sovereign balance sheets have put much attention on credit ratings agencies and their rating methodologies. Even more, analysts of the IMF state that “ratings have inadvertently contributed to financial instability – in financial markets during the recent global crisis and, more recently, with regard to sovereign debt.”2 This had the result that credit ratings (of sovereigns) have been subject of an intense discussion. In Europe, political leaders now join a chorus calling for a new European ratings agency.

The critique often is focused on that CRAs have not been able to predict a crisis in the first place and then when the crisis is running, CRAs aggravated the crisis by downgrading ratings in the midst of the crisis. They did not foresee the default of Lehman brothers or the Asian crisis of 1997. Even more, rating downgrades worsened the Asian crisis by downgrading ratings when

1

See www.fitchratings.com and click on definitions or search for “sovereign ratings methodology”

2

(3)

the crisis already shattered the economy. In Europe, the downgrading of credit ratings are said to have exacerbated the crisis in Portugal, Ireland, Greece and Spain (PIGS).

The Global Financial Stability Report of the IMF recommends that “regulators decrease their reliance on credit ratings and increase their monitoring task on the agencies that assign the ratings used in regulations.” 3 Also, the report concludes that given the importance of credit ratings, CRAs also should provide default probabilities or expected losses. Furthermore, the report endorses that negative “cliff effects” in prices and spreads that ratings cause should be reduced by eliminating buy or sell decisions to ratings. As well, CRAs should continue providing additional information on the accuracy of their ratings, the underlying data, and their efforts to diminish the moral hazard dilemma that is associated with their “issuer pay” model of charging issuers for their ratings.

In this thesis, I will analyze empirically whether CRAs have 1) been able to predict a crisis and 2) aggravated the crisis by downgrading credit ratings during the crisis. Therefore, the research question will be: ‘what role have credit ratings played in the recent government debt crisis?’. This research question will be divided into two sub-questions: 1) “have credit ratings impeded the crisis?” 2) “did credit ratings downgrades exacerbate the crisis ?” For the first sub-question, credit ratings could impede a crisis when their credit ratings were not overrated before that crisis. When they are overrated, the price of borrowing for that sovereign will be low, causing inefficiently much capital to flow into that economy, creating a bubble and a future crisis. When a country is overrated (or underrated) the aggregated sum of the underlying determinants are lower (or higher) than for other countries with the same rating. And for the second sub-question, credit ratings have exacerbated the crisis when these ratings got underrated during the crisis. When a country gets underrated it is more expensive to borrow money at the financial market. Therefore, its economic situation could get worse. Assuming that the macro-economic effects are lagging underrated credit ratings, than rating downgrades, which are quickly followed by another rating downgrade should be found to be underrated.

3

(4)

This thesis is organized as follows: section 2 reviews the relevant literature and section 3 introduces the economic model. Section 4 sets forth the empirical model, followed by a description of the data and its limitations in section 5. Section 6 than presents and interprets the empirical results. And finally, in section 7, the conclusions will be set out.

2 – Previous Literature

Partnoy states that: “Credit ratings pose an interesting paradox. On the one hand, credit ratings purport to provide investors with valuable information they need to make informed decisions about purchasing or selling bonds; credit rating agencies seem to have impressive reputations and most bond issues are rated by multiple agencies. On the other hand, particularly since the mid-1970s, the informational value of credit ratings has plummeted; credit rating agencies, faced with the challenges of globalized, technologically innovative markets and with competition from providers of more current, detailed, and accurate information, have become reactive rather than proactive, and some evidence indicates they have maintained accurate credit ratings (i.e., ratings correlated with actual default experience) due more to after-the-fact corrections than to predictive power.” (1999, page 621, line 7-18) Partnoy, hereby, says that credit ratings are more reacting to macro-economic news these days, than 40 years ago. In Partnoy’s view, credit ratings will be less and less proactive as globalization and technological innovation continues.

Several studies have examined the relationship between sovereign credit ratings and the Credit Default Swaps (CDS) spreads. The aim of these studies is to find whether credit ratings have a market impact. If sovereign credit ratings impacts the market for CDS spreads, than sovereign credit ratings give significant additional information to the market and seem to have predictive power. If sovereign credit ratings do not impact the market for CDS spreads, than their information is stale and purely reacting to macro-economic news.

(5)

the market. Ismailescu and Kazemi (2010) also found that credit rating events on sovereigns which are rated below investment grade have greater impact on the CDS market. Reinhart (2002) finds similar results. In a study on developing countries in a currency crisis, Reinhart finds that negative credit rating events are already anticipated. Therefore, negative credit rating events are following macro-economic news rather than having predictive power and deliver significant information to the market.

However, Larraín et al. (1997) find contradicting results. On a panel event study from 1987 onwards to 1996 they find that for 78 events, out of which 42 affected emerging markets, negative events have a negative significant impact on yield spreads. However, they did not find a significant relation between positive events and yield spreads. Moreover, Afonso et al. (2011) observe a significant relationship between ratings notion and outlook and government bond yield spread, especially in the case of negative announcements.

A more fundamental discussion on credit ratings is provided by Mora (2006). Mora reacts on a paper posed by Ferri et al. (1999), who state that CRA’s amplified the Asian crisis of 1997 by excessively downgrading Asian sovereigns more than their economic fundamentals would justify. The amplification can be seen in an increasing cost of borrowing and a declining pool of investors. This paper was motivated by the question whether credit ratings have “tremendous power to influence market expectations on a country” or whether they are simply reacting to news. Ferri et al. created predicted ratings that were lower for the Asian countries prior to the crisis, but the predicted ratings are not found to be higher during the crisis. Mora finds that these predicted ratings are higher after the crisis. Therefore she states that there is inertia in credit ratings, but she rejects the view that CRAs aggravated the east-Asia crisis by excessively downgrading those sovereigns. She states that credit ratings are rather sticky than pro-cyclical. Therefore, she states that credit ratings appear to lag financial markets.

(6)

an event study analysis that credit rating announcements have an impact on the market for bond spreads.

3 – Economic Model

Recall that the research question was: ‘what role have credit ratings played in the recent government debt crisis?’ and that this research question was divided into two sub-questions: 1) “have credit ratings impeded the crisis?” 2) “were credit ratings downgrades exacerbating the crisis ?”. To test the sub-questions of the research question, the credit ratings of Fitch will be used as the dependent variable. The independent variables are obtained from the Fitch ratings methodology plus labor productivity- and unemployment rates. The variables obtained from the Fitch ratings methodology are summarized in Table 1. However, since I could not find sufficient data for some variables, some variables (interest payments; public foreign currency debt; commodity dependence; gross sovereign external debt; external interest service and financial market depth) are excluded from the model. Moreover, after a test for correlation between these variables I excluded some extra variables from the model, these variables neither significantly improved the model either (these are: GDP; GDP per Capita; GDP volatility; composite governance indicator and reserve currency status). The variable ‘years since default’ is dropped because none of the sample sovereigns has defaulted on its debt since 1980. The variables unemployment and labor productivity (per person) are included because the situation on the labor market has real economic effects as well as government budgetary effects. Also, unemployment en labor productivity could catch some of the effect of the omitted variables. Consequently, the economic model will be:

(7)

away part of their debt. There is a chance that inflation is surprisingly higher than expected, when the first difference of inflation is high.

Second, gross government debt as a % share of GDP reflects the amount of debt a sovereign has outstanding. The likelihood that a sovereign can meet its debt obligations in full and in time diminishes when its government debt to GDP ratio increases. In general, it is stated that sovereigns are in trouble if they pay larger rates of interest on their debt than their GDP growth rates. Such a sovereign needs positive budget balances otherwise it would get indebted. As well, such a sovereign is more sensitive to both negative aggregate demand shocks and a decreasing social willingness to pay taxes. Therefore, higher debt levels are risky and would normally be negatively related to credit ratings.

Third, the relationship between the budget balance and credit ratings is expected to be negative, if there is any. The budget balance does not always need to have a negative impact on credit ratings. If a negative budget balance is the result from an attempt to encounter a negative temporal demand shock, than Keynesian economics would state that this is at the benefit of the sovereign economy; therefore, it should have a minor effect in the credit rating. This is especially the case for sovereigns which already have low gross debt levels. Though, a sovereign’s bad budget balance may reflect a higher risk to repay debt when budget deficits are run to counter permanent negative demand shocks or when they are run in combination with high gross debt levels or because of lower social willingness to pay taxes. Hence, for some sovereigns a negative budget balance may be interpreted as a higher risk to meet its debt obligations, therefore the budget balance might be negatively related to credit ratings.

(8)

expenditures on defense are increasing GDP. Or the critique that important natural resources are left out in calculating GDP, such as nature or other assets. These assets are hard to translate in relative prices. Hence, GDP does not necessarily measure ‘wealth’ or the ‘standard of living’, but it gives an idea how rich a sovereign is and how much it is capable of spending.

Fifth, from the Fitch ratings methodology (see Table 1) the ratio between official international reserves and imports is important as well, with a negative relationship with respect to the credit rating. The economic ratio behind this is that a sovereign that has relatively high imports and low stock of international reserves runs a risk of having to devalue its currency. The sovereign cannot print foreign currency money, so it has to buy that currency in exchange of their own currency. Therefore, the value of its currency decreases and the value of the sovereign’s bonds outstanding is lowered. In case of such devaluation prospects a lower credit rating is a logic result. On the other hand, one might say that a devaluation of Greece and Ireland is less likely, because they are in the European Monetary Union. The ECB’s primary task is to keep inflation below but close to 2% in the euro area. The effect of selling euro’s and buying foreign currency by Greece and Ireland has a smaller effect on the exchange rate of the euro, since these sovereigns are relatively small within the euro area. However, the ascribed above is not a risk of defaulting in debt; CRAs do not necessarily grade ratings on a scale of default likelihood. As can be seen in Table 2, some ratings are used to give a warning not to invest in it.

Sixth, Fitch recognizes that a sovereign’s ability to withstand adverse balance-of-payments shocks is in part a function of the existing stock of external assets and liabilities. Current account deficits must be met by either more external liabilities or less external assets. Hence, the current account balance negatively relates to the net international investment position.

(9)

And finally, higher labor productivity increases sovereign tax-income, therefore their budget balance improves and their credit rating can be graded upwards. Labor productivity is also expected to act as a replacement indicator of some variables from Table 1 which were needed to be excluded from data-stringent motives.

4 – Empirical Model:

I will use the ordered logit and the ordered probit model to estimate the coefficients of explanatory variables on a credit rating and the coefficient of the interval in which a credit rating not changes. The ordered probit and logit models both will be used because than we can see which distribution of errors fits better to the data. First, this will be done for the whole panel, than the panel will be split up in two different ways. First, it will be split up over time periods, where there is one period ranging from 2004Q2 up to 2007 Q2 and the other from 2007Q2 up to 2010Q4. The second split up is across sovereigns, where the first group consists of Greece, Iceland and Ireland (GII) and the second group consists of Estonia, Hungary, Latvia and Lithuania (EHLL).

The dependent variable (y), can take eleven values (1 = AAA; 2 = AA+; 3 = AA; 4 = AA-; 5 = A+; 6 = A; 7 = A-; 8 = BBB+; 9 = BBB; 10 = BBB-; 11 = BB+), these observed values (y) are modeled so that:

-1,& 2 3 4 3 51 -_2 μ 7,&∗ 9 μ; ; = -;,&∗ 9 μ> 3 μ> = ->,&∗ 9 μ@ A μ# = -#,&∗

where M is the number of values y can take (eleven) and y* is the estimated variable equal to, -_1,&∗ _{B ) C}

7,&D1 E)C;,& )C>,&

)C@,& )CF,&G/_+,H)CI,&D' ( ) E

)CJ,&*+,-+ )CK,&./- ) L

(10)

The main difference between the ordered logit and the ordered probit model lies in the assumption on the distribution of the error term. The probit model assumes the standard normal distribution of errors:

∅DLE 1 √2OPQ

R

;

The logit model assumes a standard logistic distribution of errors:

SDLE _{D1 )}Q _Q_E_;

The logistic distribution has wider tails than the normal distribution, or assumes a wider kurtosis. The probabilities for each of the outcomes are:

T-1,& 0VW1,&, C, XY TX7Z W1,&[ CY

T-1,& 1VW1,&, C, XY TX;Z W1,&[ CY Z TX7Z W1,&[ CY

T-1,& 2VW1,&, C, XY TX>Z W1,&[ CY Z TX;Z W1,&[ CY

…

T-1,& AVW1,&, C, XY 1 Z TX#Z W1,&[ CY

where F is the cumulative distribution of ε and the threshold values µ and β are estimated by using maximum likelihood.

The mode is 1 (AAA), except for the EHLL group there the mode is 5 (A+) and all explanatory variables are demeaned, i.e. all variables are subtracted by its sample mean. Finally, the hypotheses to be researched are:

H1 βpanel – βperiod i = 0 and i = 1, 2 the coefficients of the explanatory variables of the sample period i equal the coefficients of the full sample. The periods are split up in two, one from 2004Q2 onto 2007Q2 and the other from 2007Q2 onto 2010Q4.

(11)

sample is split up in to a group including Greece, Iceland and Ireland (GII) and the other group consisting of Estonia, Hungary, Latvia and Lithuania (EHLL).

H3 µpanel, - µperiod t = 0: the interval limits of period t equal the interval limits of the full sample of sovereigns.

H4 µpanel, - µgroup g = 0: the interval limits of the sample group g equal the interval limits of the full sample.

H3 and H4 are run when the interval of the limit points of a rating does not overlap. For example, H3 will only be run when μ_{;,]^_`a}b μ_>,]`cdef or when μ_{>,]^_`a}= μ_;,]`cdef, for the interval of y_;∗. There is one special case for y₇∗ because it has no lower limit. Therefore

μ7,]^_`aZ μ7,]`cdefh 0 will be tested. When this is a positive number the period’s rating could

be underrated while it has an AAA rating, this is not possible a possible outcome since AAA is the highest rating.

To test these hypotheses statistically, I will use the standard deviations of each of the β’s or the µ’s. Since it is not possible to measure the covariance, I will translate the covariance to a function of the standard deviations and their correlation. As the correlation is bounded, I am able to measure the maximum standard deviation of the difference in the β’s or the µ’s. Mathematically, it would look like this:

assume:

Y β$klm Z β&,1

H0: Y = 0

The variance will be:

VarDYE VarTβ$klmY ) VarDβ&,1E q 2 ∗ CovDβ$klm; β&,1E

(12)

CovDX, YE CorrelationDX, YE ∗ SDDXE ∗ SDDYE,

where SD is the standard deviation. The correlation is not known but we know that it has an upper limit of 1 and a lower limit of -1. Therefore, the term q2 ∗ '/Dβ_$klm; β_&,1E has a limit so that:

)2 ∗ CovDX, YE 9 SDDXE ∗ SDDYE Z2 ∗ CovDX, YE ~ ZSDDXE ∗ SDDYE Since the two formulas are the same, we have:

VarDYE 9 VarDβE ) VarDβ&,E ) 2 ∗ DSDDβE ∗ SDTβ&,1YE

SDDYE 9 VarDβE ) VarDβ&,E ) 2 ∗ DSDDβE ∗ SDTβ&,1YE

According to Mudholkar and George (1978), the logistic distribution is very closely related to the t-distribution with nine degrees of freedom. In Table 3 the critical values for a normal distribution and a student’s t distribution with 9 degrees of freedom are given. Now, it is possible to measure whether coefficients (and interval limits) are significantly different from each other for the logit and the probit model.

5 – Data Description

(13)

sometimes, multiple steps are remarkable and raise the question whether they are grounded by macro-economic determinants.

Now I will explain six limitations to this research out three are related to variables, one is related to the sample cross sections’ and two are related to modeling. First of all, recall that the variables Fitch uses in its ratings methodology are summarized in Table 1. Not all of these variables are modeled in this thesis, due to not being capable of finding all the data, or due to problems with correlation. Therefore, the model potentially could suffer from an omitted variable bias, decreasing the model’s fit to the data and affecting other coefficient’s values and significance. To counter part of this negative effect, the variables labor productivity and unemployment are included. Second, not all variables are quarterly. Data on labor productivity and FDI were only available on a yearly basis. Data on the budget balance was only available for Iceland, Ireland and Latvia at the yearly frequency. All other variables are available on a quarterly basis. The non-quarterly series are made quarterly by using the same values on a quarterly basis for labor productivity and budget balance and dividing FDI by four, for each quarter. Third, the relationship between variables and their rating might not always be that straightforward. As can be seen in Figure 3, gross government debt, according to Fitch, is the biggest for sovereigns with AAA rating and the lowest for sovereigns with AA rating. However, in section 2, a negative relationship is assumed between gross government debt and the corresponding credit rating. This is contradicting to the finding that sovereigns with the highest credit rating have the highest amount of debt, as well.

(14)

Fifth, the logistic distribution has the same critical values for the student’s t distribution with nine degrees of freedom is used. Because Mudholkar and Olusegun George (1978) found that the logistic distribution is very closely related to the t-distribution with nine degrees of freedom. However, it is not exactly the same; therefore significance levels are expected to be slightly different. And sixth, when comparing coefficients the significance is measured assuming perfect positive or negative correlation. In reality, correlations might be close to 1 or -1, but they are probably smaller than 1 or bigger than -1. This made significance levels bigger and may have caused significant differences to appear not significant. However, results that are significant, will be significant for certain.

Descriptive statistics, statistics on heteroskedasticity, stationarity, autocorrelation and correlation are given in the Tables 4 and 5. As can be seen tests for heteroskedasticity and autocorrelation are rejected in favor of homoskedasticity and no autocorrelation. Moreover, tests for a unit root are not rejected, after taking the first difference of inflation. The sources of data are: Fitch Ratings, Eurostat, Worldbank, IMF, OECD, Citigroup, Central Banks of Estonia, Greece, Hungary, Iceland, Ireland, Latvia and Lithuania.

6 – Empirical Results

The empirical results are presented in Tables 6 – 12. In Table 6, the outcomes for the full sample are shown; thereafter, the outcomes for the pre – and after financial crisis period and for the GII and EHLL countries are presented. Table 11 summarizes the results for hypotheses one and two and Table 12 sums up the results for hypotheses three and four. When looking at the log likelihood values in Tables 6 – 10, it can be concluded that the ordered logit model is preferred to the ordered probit model except for the EHLL countries (Table 10). Hence, models with a logistic distribution of errors have a better fit than models with a normal distribution of errors. As well, the pseudo R-squared and the information criterions show similar signs of model fits and preferences to either the ordered probit or ordered logit model.

(15)

lower labor productivity are associated with greater risk to default of – or delayed payment of debt. A higher reserves to imports ratio not technically affects the default probability, but it could be seen as a warning sign not to invest in bonds valued in that specific currency. Furthermore, FDI plus current account balance, budget balance and inflation show no relation to credit ratings. These variables therefore, probably do not play an important role in the Fitch ratings methodology. However, the budget balance affects credit ratings through gross government debt. It could be that the budget balance has a significant relation with credit ratings when gross debt is high and GDP growth is low. In such a situation CRAs might more closely look at the policy decisions of sovereigns. However, in such a situation it is complicated to choose what is best for the economy and a sovereign’s credit rating. Does a sovereign seriously have to concern about Keynes’ savings paradox or is obliging budget balances surpluses the best economic remedy? According to Table 6 it can be stated that Fitch in its ratings methodology does not take a significant position in this matter.

Before analyzing Tables 11 and 12 and the hypotheses, I will first go through the outcomes of Tables 7-10. In Table 7 the results are shown for the sample period 2004Q2 until 2007Q2, the pre- financial crisis period. The first thing to note is that gross debt (% of GDP) and labor productivity are the only variables with a significant relationship. In both cases, the coefficient moves away from zero, suggesting that its impact on credit ratings may be bigger than the regressions in Table 6. This could be a sign of rating behavior which is different from the rating behavior in Table 6, the full sample period. However, these results may be misleading. As can be seen in Figures 2b and 2g, gross debt and labor productivity are very steady, stable variables for the pre crises period. As well, in Figure 1 it could be seen that the dependent variable also shows little deviation. The significance of little deviating explanatory variables increases when there is little deviation in the dependent variable, as well. In other words, the effect of gross government debt and labor productivity on credit ratings could be possibly overestimated.

(16)

the logit model) for GDP growth has increased compared with the full sample in Table 6, both causing the significance to diminish to a 95% significance level (and a 90% significance in the probit model). Even the limit points do not deviate a lot from those presented in Table 6.

Considering the possible biasedness of the results in Table 7 and the results of Table 8, which did not differ a lot from those of the full sample period, it can be said that CRAs did not change their rating behavior before or after the recent financial crisis bursts. Furthermore, when looking at the columns for pre-crisis and post-crisis in Tables 11 and 12, it can be seen that neither the coefficient nor the interval limits significantly changed except for the AAA rating limit points in the pre-crisis sample of the probit model. This coefficient is negative stating that the AAA ratings of Iceland and Ireland were overrated. However, in the ordered logit model there is no significant difference. Moreover the ordered logit model has a better fit to the data than the ordered probit model. Moreover, the accusation that credit ratings are exacerbating the crisis seems not too strong in this light.

Now, the outcomes for the GII and the EHLL country groups will be analyzed. First, the GII countries regressions results, in Table 9, show some deviation from the full sample period regressions. The coefficient for unemployment has increased and that there is a significant relationship between the budget balance and the credit rating. Therefore, it could be cautiously stated that budget balance surpluses improve credit ratings. However, the significant relationship of the budget balance is not very strong in the logit model (90 %), but it stronger in the probit model (95%).

(17)

for a long period and that Iceland and Ireland, despite being downgraded, were overrated, as well. When sovereigns are overrated during a crisis, one could not speak of CRAs exacerbating the crisis. Greece was overrated before the crisis, which probably caused too much capital inflow. Such an inefficient allocation of capital certainly caused a bubble to arise. CRAs, therefore, cannot be blamed of exacerbating the crisis during the crisis, but they did not impede the bubble in Greece by overrating Greece before the crisis.

The regression results of the other country group, EHLL, are shown in Table 10. In both the logit and the probit model the coefficient reserves to imports almost tripled. However, the coefficients are not very significantly different. As can be seen Table 11, the probit model shows that reserves to imports for EHLL countries is significantly different from that of the full sample at 10%. An explanation for the higher coefficient is that EHLL countries have an own currency and are not part of a monetary union (Estonia Joined EMU since January 2011). Greece and Ireland joined the euro and face a smaller risk of a decreasing value of the currency (and a decreasing value of outstanding government debt) because these countries are relatively small compared to the euro area.

Interestingly, the interval limits of the EHLL countries for the ratings (A = 6 and A- = 7) are positively, significantly different from those of the full sample. This means that those ratings were underrated, relative to the sample. Estonia got these ratings from 2008Q4 to 2010Q4; Hungary from 2005Q1 to 2008Q3; Latvia from 2004Q1 to 2008Q2 and Lithuania from 2004Q1 to 2006Q3 and in 2008Q4 and 2009Q1. These ratings were given before, as well as after the start of the recent financial crisis. So these mismatches between the full sample and the EHLL countries, for these ratings, did not appear suddenly at the start of the crisis. In fact three out of the EHLL countries were already underrated before the crisis. Only for Estonia one could say that credit ratings might have worsened the crisis there. Hence, these countries were relatively underrated compared with the sample, which is not beneficial for these countries, because money could have been more efficiently allocated in the EHLL countries. Though, it could not be stated that credit ratings downgrades were exacerbating the crisis.

(18)

In this thesis Fitch sovereign credit ratings have been regressed against several macro-economic variables using the ordered probit and ordered logit model. Gross debt, GDP growth, reserves to imports, unemployment and labor productivity are significantly related to sovereign credit ratings. The sample has been split up twice, once across sovereigns (GII – and EHLL countries4) and once across time (2004Q2 – 2007Q2 and 2007Q2 – 2010Q4). The outcomes of these regressions are compared with the outcomes of the full sample. In general, no significant deviation can be shown in the variables’ coefficients and its significance. So it seems that credit ratings did not significantly change their methodology. Nevertheless, there have been significant differences in the interval limits of the ordered logit and probit models. This indicates that for a specific credit rating the underlying macro-economic basis significantly differed from that of the full sample of credit ratings. I have found an indication that Greece was overrated before the recent financial crisis up until the first half of 2009. Fitch therefore did not impede the financial bubble in Greece’s government debt. However, Fitch neither could be accused of exacerbating the crises because they have downgraded Greece until Greece was not overrated anymore. And Greece did not get underrated during the crisis. Moreover, downgrades of Iceland and Ireland during the crisis were not severe enough, since the lower credit ratings still were overrated. Hence, an aggravation of the crisis by the CRA’s is not likely. Moreover, I found that the EHLL countries have been underrated during the sample period. Therefore, these countries would rather have underperformed macro-economically, but have not got into a bigger crisis through credit ratings, since they were not specifically underrated during the crisis.

All in all, credit ratings were not impeding the government debt crisis by overrating Greece’s government debt before the crisis. However, during the crisis credit ratings have not exacerbated the crisis by suddenly underrate individual credit ratings. However, there is an indication that Estonia has become underrated during the crisis, but Estonia has too little debt to speak of an aggravation of the crisis.

4

(19)

The empirical results are obtained from a sample of seven sovereigns. The choice of the sample may have had a great impact on the empirical outcomes. As well, not all variables of Fitch’s ratings methodology are included, so the model could suffer from an omitted variable bias. I have tried to capture this bias by including labor market variables. These are strong weaknesses of this thesis and therefore, I could recommend anyone to do more research on this topic because the results are interesting.

(20)

Table 1: Fitch’s Sovereign Credit Ratings Methodology Macroeconomic

Consumer Price Inflation 3 year average (centered on current year) of annual change in consumer price index (CPI). The forecast at time t rather than the actual outturn is used, signified by 'HF'.

Real GDP Growth 3 year average (centered on current year) of annual change in real GDP. The forecast at time t rather than the actual outturn is used, signified by 'HF'

Real GDP Growth Volatility Natural log of the trailing 10 year standard deviation of average annual change in real GDP.

Public Finances (General Government)

Budget Balance 3 year average (centered on current year) of general government (budget) balance (GGB) as a percent of GDP. The forecast at time t rather than the actual outturn is used, signified by 'HF'.

Gross Debt 3 year average (centered on current year) of gross (general) government debt (GGD) as a percent of GDP. The forecast at time t rather than the actual outturn is used, signified by 'HF'.

Interest Payments 3 year average (centered on current year) of gross government interest payments (GGI) as a share of general government revenues (REV).

Public Foreign Currency Debt 3 year average (centered on current year) of public foreign currency denominated (and indexed) debt (PFCD) as a share of gross (general) government debt (GGD).

External Finances

Commodity Dependence Non-manufactured merchandise exports as a share of current account receipts (CXR).

Current Account Balance plus net Foreign Direct Investment

3 year average (centered on current year) of current account balance (CAB) plus net foreign direct investment (FDI) as a percent of GDP.

Gross Sovereign External Debt 3 year average (centered on current year) of gross sovereign external debt (GPXD) as a share of gross external debt (GXD).

External Interest Service 3 year average (centered on current year) of external interest service expressed as a share of current external receipts (CXR).

Official International Reserves Year-end stock of international reserves (including gold) expressed as months' cover of import payments (CXP).

Structural

Financial Market Depth Natural log of financial assets (sum of the outstanding stock of public and private sector debt securities, market capitalization of the domestic stock market, private sector credit and official international reserves) relative to GDP.

GDP per Capita Percentile rank of GDP per capita in US dollars at market exchange rates.

Composite Governance Indicator Average percentile rank of World Bank governance indicators: 'Rule of Law'; 'Government Effectiveness'; 'Control of Corruption'; 'Voice & Accountability' and 'Political Stability'.

Reserve Currency Status Reserve currency status: 3 = 'strong'; 2 = 'medium'; 1 = 'low'; 0 = none.

(21)

Table 2: Fitch’s Issuer Credit Rating Scales

AAA 'AAA' ratings denote the lowest expectation of default risk. They are assigned only in cases of exceptionally strong capacity for payment of financial commitments. This capacity is highly unlikely to be adversely affected by foreseeable events.

AA 'AA' ratings denote expectations of very low default risk. They indicate very strong capacity for payment of financial commitments. This capacity is not significantly vulnerable to foreseeable events.

A 'A' ratings denote expectations of low default risk. The capacity for payment of financial commitments is considered strong. This capacity may, nevertheless, be more vulnerable to adverse business or economic conditions than is the case for higher ratings.

BBB 'BBB' ratings indicate that expectations of default risk are currently low. The capacity for payment of financial commitments is considered adequate but adverse business or economic conditions are more likely to impair this capacity.

BB 'BB' ratings indicate an elevated vulnerability to default risk, particularly in the event of adverse changes in business or economic conditions over time; however, business or financial flexibility exists which supports the servicing of financial commitments.

B 'B' ratings indicate that material default risk is present, but a limited margin of safety remains. Financial commitments are currently being met; however, capacity for continued payment is vulnerable to deterioration in the business and economic environment.

CCC Default is a real possibility.

CC Default of some kind appears probable.

C Default is imminent or inevitable, or the issuer is in standstill. Conditions that are indicative of a 'C' category rating for an issuer include:

- a. the issuer has entered into a grace or cure period following non-payment of a material financial obligation;

- b. the issuer has entered into a temporary negotiated waiver or standstill agreement following a payment default on a material financial obligation; or

- c. Fitch Ratings otherwise believes a condition of 'RD' or 'D' to be imminent or inevitable, including through the formal announcement of a coercive debt exchange.

RD 'RD' ratings indicate an issuer that in Fitch Ratings’ opinion has experienced an uncured payment default on a bond, loan or other material financial obligation but which has not entered into bankruptcy filings, administration, receivership, liquidation or other formal winding-up procedure, and which has not otherwise ceased business. This would include:

- a. the selective payment default on a specific class or currency of debt;

- b. the uncured expiry of any applicable grace period, cure period or default forbearance period following a payment default on a bank loan, capital markets security or other material financial obligation;

- c. the extension of multiple waivers or forbearance periods upon a payment default on one or more material financial obligations, either in series or in parallel; or

- d. execution of a coercive debt exchange on one or more material financial obligations. D 'D' ratings indicate an issuer that in Fitch Ratings' opinion has entered into bankruptcy filings,

(22)

Table 3: critical values

Student’s t distribution with 9 degrees of freedom

Normal distribution

90% Significance 1.833 1.645

Table 4: descriptive statistics and tests for heteroskedasticity, stationarity and autocorrelation 1st Difference

of inflation

Gross Debt Budget Balance GDP GDP per Capita GDP Growth GDP Volatility Mean 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Median 0.17 -1.76 1.06 -1.17 -1.42 1.59 0.16 Maximum 9.24 10.59 11.86 2.90 6.98 11.89 1.68 Minimum -7.06 -4.66 -28.13 -1.58 -3.12 -20.71 -.199 Std. Dev. 2.47 4.00 6.87 1.72 2.94 6.85 0.99 Skewness 0.12 0.83 -1.37 0.63 1.14 -1.04 -0.42 Kurtosis 4.70 2.61 6.80 1.64 2.87 3.87 2.23

Heteroskedasticity: Jarque-Bera (null: heteroskedasticity)

Statistic 12.97 23.92 177.43 27.71 42.69 41.12 10.31

Probability 0.00 0.00 0.00 0.00 0.00 0.00 0.01

Stationarity: Levin, Lin & Chu t (null: unit root)

Probability 0.102 1.00 0.103 0.16 0.27 0.38 1.00

Statistic (LLC) -1.27 5.29 -1.28 -1.01 -0.62 -0.32 3.01

Autocorrelation: Ljung-Box (lag = 1; null: autocorrelation)

Q-stat 36.27 185.99 85.50 193.30 190.20 164.39 175.50

(23)

Table 4: descriptive statistics (continued) Reserves to

Imports

FDI + CA Balance Unemployment Labor Productivity Reserve Currency Status (dummy) CGI Mean 0.00 0.00 0.00 0.00 0.00 0.00 Median 0.17 3.22 -0.15 -12.07 -0.29 -0.15 Maximum 4.33 221,86 9.99 55.43 0.71 0.84 Minimum -0.99 -120.60 -6.51 -36.27 -0.29 -0.60 Std. Dev. 0.92 55.64 3.71 28.09 0.45 0.41 Skewness 1.99 1.64 0.40 0.63 0.93 0.68 Kurtosis 7.92 9.20 2.55 2.15 1.87 2.05

Heteroskedasticity: Jarque-Bera (null: heteroskedasticity)

Statistic 323.33 401.96 6.88 18.44 22.08

Probability 0.00 0.00 0.03 0.00 0.03

Stationarity: Levin, Lin & Chu t (null: unit root)

Statistic (LLC) 1.21 0.43 0.78 -0.59 0.54

Probability 0.89 0.67 0.78 0.28 0.71

Autocorrelation: Ljung-Box (lag = 1; null: autocorrelation)

Q-stat 160.09 90.95 175.55 184.96 185.14

(24)

Table 5: correlation matrix

1st Difference inflation

Gross Debt Budget Balance GDP GDP per Capita GDP growth GDP Volatility 1st Difference of inflation 1.000 Gross Debt -0.009 1.000 Budget Balance -0.041 -0.364 1.000 GDP 0.019 0.651 -0.189 1.000 GDP per Capita 0.010 0.058 0.054 0.483 1.000 GDP Growth 0.524 -0.243 0.394 -0.067 0.025 1.000 GDP Volatility 0.107 -0.602 -0.023 -0.632 -0.028 0.013 1.000 Reserves to Imports -0.131 -0.020 -0.160 -0.578 -0.304 -0.342 0.392 FDI + CA Balance 0.220 0.115 -0.309 -0.184 -0.355 -0.140 0.181 Unemployment -0.156 0.445 -0.503 0.258 -0.252 -0.513 -0.084 Labor Productivity 0.038 0.389 -0.120 0.703 0.910 -0.066 -0.210

Reserve Currency Status 0.016 0.556 -0.172 0.931 0.585 -0.056 -0.487

CGI 0.023 -0.190 0.106 0.011 0.809 0.062 0.252

(continued)

Reserves to Imports (% GDP)

FDI + CA Balance Unemployment Labor

Productivity Reserve Currency Status CGI 1st Difference of inflation Gross Debt Budget Balance GDP GDP per Capita GDP Growth GDP Volatility Reserves to Imports 1.000 FDI + CA Balance 0.390 1.000 Unemployment 0.150 0.137 1.000 Labor Productivity -0.303 -0.213 -0.066 1.000

Reserve Currency Status -0.572 -0.196 0.215 0.777 1.000

(25)

Table 6: Ordered Logit and Probit full sample results Dependent Variable: Fitch Rating

Method: ML – Ordered Logit (Quadratic hill climbing) Sample: 2004Q2 – 2010Q4

Dependent Variable: Fitch Rating

Method: ML – Ordered Probit (Quadratic hill climbing) Sample: 2004Q2 – 2010Q4

Variable Coefficient Standard error Variable Coefficient Standard error

D(Inflation) 0.082 0.103 D(Inflation) 0.059 0.054

Budget Balance (% of GDP) -0.012 0.029 Budget Balance (% of GDP) -0.023 0.017

Gross Debt (% of GDP) 0.102*** 0.011 Gross Debt (% of GDP) 0.048*** 0.005

GDP Growth -0.201*** 0.038 GDP Growth -0.096*** 0.020

Reserves to Imports 1.664*** 0.260 Reserves to Imports 0.832*** 0.132

FDI + CA Balance -0.003 0.004 FDI + CA Balance -0.001 0.002

Unemployment 0.704*** 0.090 Unemployment 0.363*** 0.044

Labor Productivity -0.263*** 0.027 Labor Productivity -0.124*** 0.011

Limit Points (observations)

Interval Limit Std. Error Interval Limit Limit Points (observations)

Interval Limit Std. Error Interval Limit

Limit 2 (8) -10.338 1.153 -8.968 Limit 2 (8) -5.033 0.524 -9.598 Limit 3 (1) -7.336 0.878 -8.354 Limit 3 (1) -3.462 0.379 -9.140 Limit 4 (3) -7.040 0.849 -8.290 Limit 4 (3) -3.306 0.364 -9.095 Limit 5 (32) -6.260 0.776 -8.069 Limit 5 (32) -2.940 0.338 -8.710 Limit 6 (50) 0.071 0.399 -0.177 Limit 6 (50) -0.146 0.207 -0.705 Limit 7 (28) 4.311 0.515 8.375 Limit 7 (28) 2.021 0.223 9.046 Limit 8 (23) 7.219 0.740 10.162 Limit 8 (23) 3.463 0.294 11.794 Limit 9 (3) 11.400 1.097 10.389 Limit 9 (3) 5.609 0.459 12.219 Limit 10 (10) 12.303 1.192 10.323 Limit 10 (10) 6.056 0.497 12.194

Pseudo R-squared 0.63 Akaike info criterion 1.60 Pseudo R-squared 0.61 Akaike info criterion 1.70 Schwarz criterion 1.89 Log Likelihood -133.49 Schwarz criterion 1.99 Log Likelihood -142.68 Hannan-Quinn crit. 1.72 Restr. Log likelihood -364.72 Hannan-Quinn crit. 1.82 Restr. Log likelihood -364.72 LR statistic 462.46 Avg. log likelihood -0.71 LR statistic 444.09 Avg. log likelihood -0.76

Prob (LR statistic) 0.00 Prob (LR statistic) 0.00

(26)

Table 7: pre start financial crisis results Dependent Variable: Fitch Rating

GDP Growth -0.121 0.199 GDP Growth -0.019 0.098

Reserves to Imports 1.039 0.933 Reserves to Imports 0.831 1.047

Unemployment -0.019 0.299 Unemployment 0.071 0.157

Limit 2 (2) -9.969 2.841 -3.509 Limit 2 (2) -5.422 1.466 -3.697 Limit 3 Limit 3 Limit 4 Limit 4 Limit 5 (21) -6.802 2.408 -2.825 Limit 5 (21) -3.871 1.254 -3.017 Limit 6 (37) 4.061 2.063 1.969 Limit 6 (37) 1.953 1.031 1.895 Limit 7 (7) 9.966 2.414 4.129 Limit 7 (7) 5.015 1.113 4.504 Limit 8 Limit 8 Limit 9 Limit 9 Limit 10 Limit 10

(27)

Table 8: post start financial crisis results Dependent Variable: Fitch Rating

Budget Balance (% of GDP) -0.028 0.037 Budget Balance (% of GDP) -0.034* 0.021

GDP Growth -0.106** 0.046 GDP Growth -0.048* 0.024

Limit 2 (7) -12.698 1.819 -6.982 Limit 2 (7) -5.945 0.785 -7.574 Limit 3 (1) -8.046 1.291 -6.279 Limit 3 (1) 3.498 0.497 -7.043 Limit 4 (3) -7.594 1.233 -6.161 Limit 4 (3) -3.256 0.470 -6.935 Limit 5 (13) -6.370 1.116 -5.711 Limit 5 (13) -2.676 0.422 -6.336 Limit 6 (15) -0.493 0.587 -0.839 Limit 6 (15) -0.401 0.297 -1.350 Limit 7 (22) 2.967 0.627 4.731 Limit 7 (22) 1.223 0.279 4.340 Limit 8 (23) 6.840 0.919 7.443 Limit 8 (23) 3.058 0.361 8.478 Limit 9 (3) 11.956 1.520 7.867 Limit 9 (3) 5.577 0.573 9.727 Limit 10 (10) 12.861 1.606 8.011 Limit 10 (10) 6.029 0.607 9.926

(28)

Table 9: results for the GII countries Dependent Variable: Fitch Rating

Budget Balance (% of GDP) -0.203* 0.111 Budget Balance (% of GDP) -0.119** 0.059

Gross Debt (% of GDP) 0.091** 0.045 Gross Debt (% of GDP) 0.051** 0.025

GDP Growth -0.020 0.108 GDP Growth 0.000 0.058

Labor Productivity -0.352*** 0.134 Labor Productivity -0.168** 0.069

Limit 2 (8) -6.361 1.819 -3.498 Limit 2 (8) -3.477 0.986 -3.527 Limit 3 (1) 0.0596 1.454 0.410 Limit 3 (1) 0.589 0.757 0.779 Limit 4 (3) 1.492 1.426 1.046 Limit 4 (3) 1.068 0.728 1.468 Limit 5 (3) 1.329 1.579 2.742 Limit 5 (3) 2.480 0.791 3.134 Limit 6 (19) 6.289 1.526 4.122 Limit 6 (19) 3.266 0.769 4.248 Limit 7 (6) 11.080 1.957 5.661 Limit 7 (6) 5.627 0.881 6.387 Limit 8 (7) 13.412 2.280 5.883 Limit 8 (7) 6.776 1.021 6.636 Limit 9 Limit 9 Limit 10 (3) 17.257 2.880 5.993 Limit 10 (3) 9.034 1.395 6.478

(29)

Table 10: results for the EHLL countries Dependent Variable: Fitch Rating

GDP Growth -0.233*** 0.061 GDP Growth -0.132*** 0.034

FDI + CA Balance 0.005 0.028 FDI + CA Balance 0.002 0.015

Limit 2 Limit 2 Limit 3 Limit 3 Limit 4 Limit 4 Limit 5 Limit 5 Limit 6 (31) -4.399 0.697 -6.310 Limit 6 (31) -2.404 0.343 -7.007 Limit 7 (22) -0.028 0.387 -0.073 Limit 7 (22) -0.021 0.223 -0.092 Limit 8 (16) 3.681 0.687 5.358 Limit 8 (16) 2.066 0.357 5.793 Limit 9 (3) 9.389 1.468 6.397 Limit 9 (3) 5.068 0.695 7.290 Limit 10 (7) 11.171 1.710 6.534 Limit 10 (7) 6.149 0.897 6.857

(30)

Table 11: results for hypotheses 1 and 2

Ordered Logit Pre-Crisis Post-Crisis GII EHLL

Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error

D(Inflation) 0.003 0.413 0.076 0.220 -0.049 0.259 -0.117 0.288 Budget Balance (% of GDP) 0.024 0.121 0.016 0.066 0.191 0.140 0.030 0.078 Gross Debt (% of GDP) -0.119 0.689 -0.007 0.266 0.012 0.566 0.011 0.323 GDP Growth -0.080 0.237 -0.095 0.084 -0.181 0.146 0.032 0.099 Reserves to Imports 0.625 1.193 -0.014 0.579 -0.094 0.914 -2.398 1.451 FDI + CA Balance 0.016 0.034 0.000 0.008 0.001 0.008 -0.008 0.032 Unemployment 0.723* 0.389 -0.165 0.224 -0.408 0.404 0.048 0.218 Labor Productivity 0.207 0.145 0.032 0.067 0.089 0.161 -0.024 0.096

Ordered Probit Pre-Crisis Post-Crisis GII EHLL

Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error

D(Inflation) -0.015 0.230 0.032 0.117 -0.52 0.140 -0.044 0.160 Budget Balance (% of GDP) 0.005 0.065 0.011 0.038 0.096 0.076 -0.002 0.045 Gross Debt (% of GDP) -0.060* 0.313 -0.002 0.109 -0.003 0.301 -0.003 0.163 GDP Growth -0.077 0.118 -0.048 0.044 -0.096 0.078 0.036 0.054 Reserves to Imports 0.001 1.179 0.053 0.281 -0.169 0.497 -1.451* 0.763 FDI + CA Balance 0.004 0.017 0.000 0.004 0.001 0.004 -0.003 0.017 Unemployment 0.292 0.201 -0.047 0.100 -0.198 0.202 -0.001 0.112 Labor Productivity 0.099 0.063 0.009 0.026 0.044 0.080 0.014 0.050

(31)

Table 12: results for the hypotheses 3 and 4

Ordered Logit Pre-Crisis Post-Crisis GII EHLL

Interval of limit:

∗ Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error

∗ -0.371 3.994 2.358 2.972 -3.979 2.972 ∗ -3.979 2.972 ∗ _-7.396** _2.332 ∗ _-8.086*** _2.275 ∗ -7.589** 2.355 ∗ -1.898 2.462 0.316 0.986 -4.051* 1.925 4.222*** 1.096 ∗ 4.339*** 0.902 ∗ 1.861 1.427 ∗ 1.000 2.565 ∗

Ordered Probit Pre-Crisis Post-Crisis GII EHLL

Interval of limit:

∗ Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error

∗ -4.176** 1.990 -2.024 2.251 -6.071*** 1.510 ∗ _-1.506 _1.510 ∗ -4.051*** 1.136 ∗ -4.374*** 1.092 ∗ _-5.420*** _1.129 ∗ _-2.041* _1.238 _-3.412*** _0.976 _1.699*** _0.550 ∗ 2.042*** 0.446 ∗ ∗ ∗

***Significantly different from zero at 99%; ** significant at 95%; * significant at 90% . The coefficients are measured as follows: µpanel, - µperiod t

(32)

Figure 1: Fitch Rating Figure 1a: Fitch Rating (1 = AAA)

Figure 1b: Fitch Rating (1 = AAA)

(33)

Figure 2: Explanatory variables 2a: Budget Balance (demeaned)

2b: Gross Debt (% of GDP, demeaned)

(34)

2d: Reserves to Imports (demeaned)

2e: FDI – CA deficit (% of GDP, demeaned)

(35)

2g: Labor Productivity (demeaned)

Figure 3:

Gross Government Debt, per rating

Source: Fitch ratings, considering medians of 2001 – 2010 for all sovereigns rated by Fitch during this period.

(36)

References

Afonso, A., Furceri, D., Gomes, P., 2011, “Sovereign credit ratings and financial markets linkages: application to European data”, ECB working paper, No 1347.

Cantor, R. and Packer, F., 1996, “Determinants and Impact of Sovereign CreDit Ratings”, Economic Policy Review, 2, p. 37 – 54.

Ferri, G., Liu, L., Stiglitz, J.E., 1999. “The procyclical role of rating agencies: Evidence from the east Asiancrisis”, Economic Notes 28, 335–355.

De Haan, J., Amtenbrink, F., 2011, “Credit Rating Agencies”, DNB working paper, No. 278 / January 2011.

IMF, 1998, “The Asian Crisis: Capital Markets Dynamics and Spillover”, IMF International Capital Markets Report, Chapter 2, September 1998.

IMF, 2010, “Reducing Role of Credit Ratings Would Aid Markets”, IMF Survey Magazine, IMF Research, September 29th 2010.

IMF, 2010, “The Uses and Abuses of Sovereign Credit Ratings”, IMF Global Financial Stability Report, Chapter 3, October 2010.

Ismailescu, l., Kazemi,H., 2010, “The Reaction of Emerging Market Credit Default Swap Spreads to Sovereign Credit Rating Changes”, Journal of Banking and Finance, 34, 2861 – 2873.

Lannoo, K., 2010, “What Reforms the Credit Rating Agency Industry?”, ECMI Policy Brief, No. 17.

Larraín, G., Reisen, H., and von Maltzan, J., 1997, “Emerging Market Risk and Sovereign Credit Ratings”, OECD Working Paper, No. 124.

Mora, N., 2006, “Sovereign Credit Ratings: Guilty Beyond Reasonable Doubt”, Journal of Banking and Finance, 30, 2041 – 2061.

(37)

Partnoy, 1999, “The Siskel and Ebert of Financial Markets?: Two Thumbs Down for the Credit Rating Agencies”, Washington University Law Quarterly, Volume 77, Number 3, p. 619 – 722.