| Effects of changes in sovereign credit ratings on investors’ behavior | | | | |

University: University Utrecht, the Netherlands
Author: A.D. Hollaar
Project-Coordinator: J.H.J.Lukkezen
Course-Coordinator: dr. C. Remery
Course: Applied Economics Research Course
Date: 13th of November, 2011

University: University Utrecht, the Netherlands
Author: A.D. Hollaar
Project-Coordinator: J.H.J.Lukkezen
Course-Coordinator: dr. C. Remery
Course: Applied Economics Research Course
Date: 13th of November, 2011

Table of Contents Abstract 2 Introduction 3 Section I: Theory 5 1.1 Sovereign bonds and credit rating agencies 5 1.2 Measures for investors behavior 6 1.3 Expected behavior of investors 11 1.4 Related literature 15 1.5 Models 16 Section II: Data & Stylized facts 17 2.1 Data 17 2.2 Stylized facts 20 Section III: Empirical analyses 26 3.1 Effect of rating events on investors’ behavior 27 3.2 Effect of business cycles on investors’ behavior surrounding rating events 33 Conclusion 46 Reference list 48 Appendix 52 Section I: Rating symbols & definitions 52 Section II: Tables 54 Section III: Figures 56 Section IV: Extended theory 57 Section V: Graphs 59 Section VI: Data 67 Section VII: Testing classical assumptions 71

Abstract
Firstly, this paper investigates if investors react to changes in sovereign credit ratings. Hereby rating changes for European, Non-European and European Union countries are considered for the period: 1990-2011. Using both bond spreads and credit default swap (CDS) spreads as measures for investors’ behavior, analysis shows that changes in sovereign credit ratings significantly affect these spreads. Furthermore evidence is found that a rating downgrade of a sovereign country has a bigger impact on the spreads than a rating upgrade. Secondly, it is theoretically examined why investors would react to changes in credit ratings. If they are fully rational they would normally only use credit ratings as the sole information source for estimating the default risk of a country, but irrational behavior, as a result of behavioral biases, might influence this estimation. Thirdly, it is analyzed if business cycles are a key determinant in explaining deviations in sovereign bond and CDS spreads surrounding a sovereign credit rating event. This notion is not fully supported, since these cycles play a role in explaining deviations in CDS spreads, but the same does not hold for bond spreads.

Introduction
The current global financial and economic crisis, which was offset in 2008, worsened the economic situation especially for European countries such as: Greece, Italy and Spain. These countries are losing creditworthiness and that is why credit rating agencies downgraded their credit ratings accordingly. Newspapers wrote extensively about this subject, for example: (Financial Times, 2011) ‘Standard & Poor’s cut Greece’s long-term sovereign credit rating by three notches to triple C’, (Xinua News Agency, 2011) ‘Rating agency Fitch cut the sovereign credit ratings of Italy and Spain’. This paper mainly tries to investigate if this news, about an increase or decrease of a country’s rating, affects investors’ behavior and if an economic crisis could be an additional factor investors consider when estimating the creditworthiness of a country.
Credit rating agencies assign ratings to sovereign countries which can inform investors, who are active in the sovereign bond market, how likely it will be that a country will default on its payments. These ratings can change over time for these sovereign countries, because the likelihood a country will default also changes over time. This research focuses on the possible reaction of investors on these rating changes. Hereby the first research question is: Do investors react to changes in sovereign credit ratings?
If the first question confirms that investors do react to changes in sovereign credit ratings, the next question that is posed is: Why do investors react to changes in sovereign credit ratings? Hereby it is explained why a rational or irrational investor would react to a change in sovereign credit ratings, building upon recent literature. Lastly, this research tries to investigate empirically if credit ratings are the only information investors consider in estimating the probability a country will default. They might also consider other information, such as the business cycle a country is in. To investigate the latter, the third research question is: Do investors react differently on changes in sovereign credit ratings, dependent on the economic business cycle that a country is in?

The relevance of these research questions is first of all scientific. It builds upon existing recent literature to investigate the factors which determine investors’ behavior concerning changes in sovereign credit ratings. Thus it satisfies the scientific curiosity as to why investors’ behavior is affected in existing sovereign bond markets. Secondly, this research is also practically relevant for investors. If behavior is more predictable, investors have more knowledge which leads them to make better investment decisions. Thirdly, if sovereign countries could better understand the impact of an upgrade or downgrade in their credit ratings on investors’ behavior, they could be able to make more accurate predictions about the consequences. (Cantor & Packer, 1996) For example, if investors react to downgrades, this would negatively affect a country’s access to credit and cost of borrowing. If countries are aware if this fact, they could try to prevent their sovereign credit rating from being downgraded by adjusting their policy mix, either by tightening or loosening their fiscal and/or monetary policy.

This paper is structured as follows, first theory is discussed, next the data with stylized facts, then the empirical analyses are executed in order to answer the research questions and these answers are summarized in the conclusion.
The goal of the theory section is to make the reader understand how the econometric models used in the empirical analyses are derived and the underlying assumptions which needed to be made, in order to estimate the effect of changes in credit ratings on investors’ behavior. In this section first the basics are discussed, namely how sovereign bonds, credit rating agencies, sovereign governments and investors relate to one another. Secondly, the measures for investors’ behavior are elaborated upon, which are bond spreads and credit default swap spreads. Thirdly, it is shown how these spreads relate to one another from a theoretical and empirical point of view. Fourthly, the expected behavior of investors in the sovereign bond market and credit default swap market is explained using rational theory as well irrational behavioral theory. Fifthly, findings of earlier research studies are discussed which relate to this paper.
In section II, the data used for the empirical analyses is discussed accompanied with stylized facts, for the following variables: sovereign credit ratings, credit default swap spreads, bond spreads, GDP in constant prices and bank lending tightness. This section is insightful, as to which data is used for the empirical analyses, how these data are collected and how they are constructed.
In section III, the empirical analyses are discussed, which test if sovereign credit rating changes affect investors’ behavior and if this behavior is affected by business cycles.
Lastly, the main results of this paper are summarized in the conclusion.

Section I: Theory
1.1 Sovereign bonds and credit rating agencies
(Brealey et al., 2009) A government can borrow money by selling sovereign bonds to investors. In this case the government is the issuer of the sovereign bonds and the holders of the bonds are the investors. The government who issued sovereign bonds usually pays interest on the loan and the initial amount lent is paid back to the investor in full at maturity. (See Appendix, Section IV, for a more elaborate explanation of bonds.) However, it could be the case that a government is not able or not willing to pay the loaned money plus the interest back to the investor in time or in full. This is a risk the investor faces, also called: default risk. How big this default risk is depends on a number of factors, which include economic, political and social factors. (Cantor & Packer, 1995) It is costly for investors of sovereign bonds to keep track of these different factors in order to estimate the likelihood a sovereign country will default. However, it is important for investors to know how likely it is that they will receive the money back on time and in full, which they lent to the sovereign government in question. This information plays a key role in their investment decision, namely whether to buy, hold or sell a sovereign bond.
Investors do not have to keep track themselves of a country’s default probability, because these are assessed by credit rating agencies. Hereby the three biggest agencies are: Moody’s, Standard&Poor’s and Fitch-Ratings. These agencies give their opinion about a country’s default credit risk which they asses in their own manner. (Baker & Mansi 2002) These credit rating agencies play a major role because issuers, investors and government regulators have increased their reliance on the opinions of credit rating agencies for corporate financing, investment decisions and risk management, since the role and complexity of financial markets have grown tremendously.

Credit rating agencies translate the default risk of sovereign countries into credit ratings. These credit ratings are defined in letters, starting from AAA, which stands for the lowest possible default risk, to C, which stands for a very large default probability. (See Appendix: Section I: Ratings and Symbols, for a full overview of the rating symbols and definitions that the three credit rating agencies use.)
These sovereign credit ratings can also change over time for sovereign countries. They can either be upgraded, which means a decrease in default risk, or downgraded, which means an increase in default risk. When an upgrade or downgrade announcement of a credit agency takes place for a given country, this must add new information to the sovereign bond market. Because as mentioned before, for investors it is costly to estimate the default probability themselves, so theory suggests they rely largely on credit rating agencies and their ratings. Thus these agencies add value to the financial market, which is also empirically supported by research of Cavallo et al. (2008).

(Hull et al., 2004) Investors often use ratings as an indicator of the creditworthiness of sovereign countries, rather than indicators of the quality of the sovereign bonds themselves. This is reasonable because it is rare for two different sovereign bonds issued by the same country to have different ratings. When rating agencies announce rating changes they often refer to the country, not to the individual bond issues.

There is also a distinction in sovereign credit ratings between local currency bonds and foreign currency bonds of sovereign countries. (Packer, 2003) Usually the foreign currency bonds receive a lower rating than domestic currency bonds of sovereign countries. (Cantor & Packer, 1995) The difference is usually justified because governments have broad-ranging powers to tax domestic income and print domestic currency, which makes them better able to fulfill their domestic currency obligations. (Packer, 2003) The countries of the euro area are special cases: hereby the issuance and printing of money is in hands of the European Central Bank (ECB), which has greatly diminished the distinctions drawn between local and foreign currency debt. That is why the sovereign credit ratings hardly differ between foreign or local currency bonds of countries within the euro area.

The sovereign credit ratings are not only important for investors, but also for sovereign governments themselves. Since these credit ratings indicate the costs at which governments can borrow money, by issuing sovereign bonds. More specific, credit ratings indicate the height of the interest payments on the sovereign bonds if the country in question decides to borrow more. This could have repercussions on the fiscal policy a government pursues, if a downgrade or upgrade of the credit rating takes place.
1.2 Measures for investors behavior
Because theory suggests these credit rating agencies add new information to the financial market for investors, in the case of an upgrade or downgrade announcement, the next question that rises is: What is the expected behavior of investors in these two scenarios?

Before this question can be answered, the question that needs to be addressed is: How can this behavior of investors be measured? The behavior after a rating announcement, either being an upgrade or downgrade of the rating, can be measured in the following two ways: either by looking at the credit default swap (CDS) spread or the government bond spread. Each of these will be explained in turn.

Bond spread
The bond spread is the bond yield minus the risk-free rate. The bond yield is the expected rate of return on a government bond if the holder of the bond holds it until maturity. The risk-free rate is the expected rate of return on a riskless bond. Usually a U.S. treasury bond or German bond is taken as a proxy for this, because these bonds are perceived as nearly riskless.

Credit default swap spread
A credit default swap (CDS) for sovereign bonds can be seen as insurance contract for default risk. (Berndt et al., 2005) If the credit event occurs before the expiration of the CDS, the buyer of the insurance receives from the seller the face value of the underlying debt.
Note that with a CDS, the risk that a sovereign country won’t repay in full or on time, is replaced by the risk that the seller of the insurance won’t repay the difference between the market value of the sovereign bond and the face value.
The CDS spread are the payments of the buyer of the default swap to the seller, annualized, as a percentage (in basis points) of the face value. For example if the payments are quarterly, the CDS spread is four times the quarterly premium as a percentage (in basis points) of the face value.

Flows of a sovereign bond and a CDS
To make it more clear how a bond and a CDS work, the flows of a sovereign bond and a credit default swap between the different parties involved can be seen in Figure 3.1, Section III of the Appendix, in case a sovereign government doesn’t default. As can be seen, the investor pays the principal and receives from the sovereign country, bond and interest payments.
Figure 3.2 of the Appendix displays the flows if a country does default and the investor has a sovereign bond and a corresponding credit default swap. The investor receives the principal of the bond back from the third party and the third party is the new legal owner of the bond after the default. For which the third party could receive back the market value.

Deriving the CDS spread and bond spread
Before turning to more advanced models (which will be discussed in a later section) which explain the CDS spread and the bond spread, two simple models will be introduced first.
(Olléon-Assouan, 2004) Let us consider an approximate model for the CDS spread, whereby it depends only on the default probability and the recovery rate. The formula is as follows:

(1) CDSi, t ≈ Pcds x Tcds = µi, t x (1 – RRi,t)
With:
CDSi, t = CDS spread for country ‘i’ at a specific time ‘t’.
Pcds = CDS premium
Tcds = CDS maturity µ = default probability for country ‘i’ at a specific time period ‘t’.
RR = recovery rate for country ‘i’ at a specific time period ‘t’., whereby RR = recovery valueface value
As shown above the CDS spread is a function of the default probability and the recovery rate:
CDSi, t = f (default probability, recovery rate)

The recovery rate is the ratio of the recovery value divided by the initial face value of a bond. For example, if the recovery rate is fifty percent, this means that the government in question is able to only pay back half of the initial loan (also known as the face value).
If a country does not default, the third party who sold the CDS, receives the CDS premium from the investor until the contract matures. If a country does default, the third party receives the recovery value from the sovereign government and pays the face value to the investor.

The next simple model that is considered is for the bond spread:
(2) Si, t = yi, t – rt
With:
Si, t = bond spread for country ‘i’ at a specific time period ‘t’ y = bond yield for country ‘i’ at a specific time period ‘t’ r = risk-free rate at a specific time period ‘t’

As shown above the bond spread is a function of the bond yield and the risk-free rate:
Si,t = f (bond yield, risk-free rate)

The bond yield is the expected return on a government bond, which has default risk and a certain recovery rate. The yield is a compensation for the default risk and the recovery rate. Since investors could also invest in riskless bonds and receive the risk-free interest, the bond yield should therefore also compensate by the amount of the foregone earnings (also known as opportunity costs).
The bond yield thereby implies that it depends on default risk, the recovery rate and the opportunity costs. Thus:
(3) yi, t = µi, t x (1 – RRi,t) + rt

Now it is possible to fill in equation (3) into equation (2), which results in:
(4) Si, t = µi, t x (1 – RRi,t)

Equation (4) shows that the bond spread is a function of the default risk and recovery rate:
Si,t = f (default probability, recovery rate)

As can be seen, the equations (4) and (1), of the bond spread and the CDS spread respectively, depend on the same determinants. The next subsection explains why this relation holds in theory. ere Theoretical comparison of the CDS spread compared to the bond spread
(Coudert & Gex, 2010) As shown in the previous section, the CDS spread is roughly equal to the bond spread for the same borrower and maturity. To see this, let us consider a portfolio made up of a bond and a CDS. As the CDS is meant to hedge the default risk, a long position in this portfolio is roughly equivalent to holding a risk-free asset. Therefore, the return on the portfolio, which is equal to the bond yield ‘yt’ minus the CDS premium ‘ct’, must be close to the risk-free rate rt. This equality can be written as: yt – ct ≈ rt. In other words, the CDS spread is approximately equal to the bond spread: ct ≈ yt - rt.
(Hull et al., 2004) If ‘ct’ is greater than ‘y – r’, an arbitrageur will find it profitable to buy a riskless bond, short a sovereign bond and sell the credit default swap. If ‘ct’ is less than ‘y – r’, the arbitrageur will find it profitable to buy a sovereign bond, buy the credit default swap and short a riskless bond.
(Coudert & Gex, 2010) As well as the bond spread as the CDS spread are meant to compensate for the investor’s loss in the event of the borrower’s default. They thus depend on the same main determinants. Although they have same main determinants, the CDS spreads and bond spreads tend to differ from one another in practice. The causes of these differences will be explained in the following subsection.

Empirical comparison of the CDS spread compared to the bond spread
(Coudert & Gex, 2010) Although CDS spreads generally approximate the spreads of the underlying bonds, there are several reasons why the two need not be identical. I will not discuss every possible reason, since it would be too extensive for this paper. However some of them will be discussed below.

(Zhu, 2004) One of the reasons is that there are cash flow differences between bonds and CDS contracts that can induce differences in spreads. (Blanco et al., 2005) In order to execute a trade for a CDS, neither cash nor the underlying bonds need to be immediately sourced and exchanged. This is different in the bond market, whereas an immediate exchange, of cash and a bond, is present.

(Zhu, 2004) These cash flow differences can result in different liquidity risk between CDS spreads and sovereign bond spreads. (Alquist, 2008) Liquidity risk is the risk that the price of a security will decline when the market as a whole becomes illiquid. Liquidity describes the ease with which an investor can anonymously trade large quantities of a security quickly, at low cost, and without altering the security’s price. (Schwarz, 2010) The effect of liquidity can take different forms. Investors may demand a premium for holding bonds that are less liquid at present, to compensate them for higher transactions costs, or they may demand a premium for the risk that bonds will become less liquid in the future.

(Alquist, 2008) Understanding the relationship between a security’s return and market liquidity is especially important during financial crises when market liquidity becomes scarce. During such crises market participants tend to value liquidity more highly and shift into more liquid sovereign bonds in so-called “flights to liquidity”. A flight to liquidity is a type of flight to quality, where the quality an investor values is the ease with which he can convert the security into cash during periods of market distress.

This reasoning suggests that the price of a more liquid asset will be less sensitive to abrupt changes in market liquidity during periods of financial stress. In this case default swap rates are less likely to be affected by market illiquidity than are bond yield spreads. Blanco et al. (2005) show that CDS rates, because of a smaller likelihood of illiquidity, represent somewhat fresher price information than bond yield spreads. Put in other words, CDS spreads will probably adjust faster to new public information than bond spreads. Thus if a credit rating change takes place, which is new public information, my research will test if the CDS spreads indeed adjust faster than bond spreads.
Calice et al. (2011) also show that the bond yield liquidity spreads have increased substantially over the period 2007-2010, while CDS liquidity spreads have fallen dramatically within the Eurozone.

Bongaerts et al. (2010) argue that although researchers and practitioners often use the CDS spread as pure measures for default risk, there is also a liquidity risk premium. Their research finds that part of the CDS spread reflects liquidity effects. Whereby they conclude that CDS spreads cannot be used as frictionless measures of default risk.

Relevance of CDS spreads and bond spreads
It is relevant to look at the CDS spreads as well as the bond spreads, because although they are the same in theory, as discussed before they tend to differ in reality. (Hull et al., 2004) Furthermore CDS spreads are an interesting alternative to bond spreads for two reasons: the first is that CDS spread data provided by a broker consists of firm bid and offer quotes from dealers. Once a quote has been made, the dealer is committed to trading a minimal principal (usually $10 million) at the quoted price. The second attraction of CDS spreads is that no adjustment is required: they are already credit spreads. Bond yields require an assumption about the appropriate benchmark risk-free rate before they can be converted into credit spreads. The benchmark often used which is either a U.S. Treasury bond or a German government bond, is highly questionable.
1.3 Expected behavior of investors
Now that the ways to measure investors’ behavior are described surrounding sovereign credit rating upgrades or downgrades, the question which now can be examined is: what is the expected behavior of investors on rating upgrades and downgrades?

If investors are rational, they respond to upgrade or downgrade announcements of credit rating agencies, because this is new public information. Which leads to an adjustment of the government bond spreads and CDS spreads. Theory suggests when a downgrade takes place, the bond spread and CDS spread should increase. Because the default risk has increased, investors would like to be compensated for the fact they are now running more risk on their sovereign bond. The opposite should hold for an upgrade, namely that the bond spread and CDS spread should decrease. Because default risk has decreased, investors should demand a lower return because the investment is less risky.

Assuming that credit rating agencies can calculate default probabilities fairly well, other factors such as business cycles or a country’s fundamentals, which are economic, political and social factors, should not have an additional impact on investor’s behavior after a rating announcement is made. Since a change in these factors are already absorbed in the change of the sovereign credit rating at the time of an upgrade or downgrade. Also it is still assumed that rational actors do not keep track of a country’s fundamentals, because it is too time-consuming, and therefore treat credit rating announcements of upgrades or downgrades as the sole information source about changes in these fundamentals.

1.3.1 Rationality
The hypothesis that investors behave rational is captured in the Efficient Markets Hypothesis (EMH). (Barberis & Thaler, 2003) This hypothesis states that actual prices reflect fundamental values. Put simply, under this hypothesis, “prices are right”. In an efficient market, there is “no free lunch”: no investment strategy can earn excess risk-adjusted average returns or average returns greater than are warranted for its risk. (Edwards, 1984) According to the EMH, investors are rational and have powerful incentives to exploit all the available information and to discriminate among borrowing sovereign countries. As a result, asset prices always reflect the information publicly available, as evidenced by the yield differential on bonds issued by sovereign borrowers with different credit ratings and macro characteristics.

Rating drift
If investors are rational, they should be aware of the fact that credit rating agencies could have a certain pattern in their rating behavior, which is suggested by Cantor and Packer (1995). If there is such a path dependency by credit rating agencies, then it is assumed that ratings that have been downgraded before are less frequently upgraded in the next period, while ratings that have experienced prior upgrading are prone to further upgrading. Therefore, two-period changes like “Down-Down” or “Up-Up” are generally considered to be more probable than alternating rating changes like “Down-Up” or “Up-Down” – the former is the so called rating drift. Rational investors might anticipate this rating drift by credit rating agencies and demand a higher (lower) risk premium than an initial decrease (increase) in the rating suggests.

(S&P, 2011) Table 2.1 in the Appendix shows Standard & Poor’s historical correlation between two subsequent identical rating actions over a two-year horizon. Between 1975 and 2010, there were 216 foreign-currency sovereign credit rating upgrades. Of this set, 37% of ratings were raised again over the course of the next two years, 57% remained unchanged, and 6% were lowered. Similarly, during these 36 years, S&P lowered foreign-currency sovereign credit ratings 148 times. Of this set, they lowered 52% again within two years, 39% remained the same and raised 9%. The data suggest that there has typically been a small correlation within two years between an upgrade and a subsequent upgrade of 0.30, and there has typically been a moderate correlation between a downgrade and a subsequent downgrade of 0.47.

Anticipated changes in default risk
If a government of a sovereign country publically announces that a default on their debt obligations is imminent, this will informs investors that the potential default risk has changed. Credit rating agencies should downgrade the sovereign credit rating of the country accordingly, but this may lag behind the government announcement about the default. This results in the fact that the credit rating change adds no new information about a country’s default risk for investors, because this new information was already added by the government in question. It is thus rational for investors in such a scenario to use the information provided by the government about a country’s default, next to relying solely on information provided by credit rating agencies. This means that the sovereign bond and CDS spreads will increase even before the credit rating downgrade has taken place. Such an example will be discussed in more detail in Section 2.2.

Skepticism about rationality
(Ferrucci, 2003) Some people question the rationality of investors and are more skeptical about market efficiency. They emphasize that market failures and imperfect information may cause distortions in the way assets are priced. They argue that the information necessary to forecast returns on sovereign bonds is costly to acquire and process, and that asset prices are often determined on the basis of incomplete knowledge of a country’s economic and financial position.

1.3.2 Irrationality
(Barberis & Thaler, 2003) If it is the case that investors do not react rationally towards sovereign credit ratings, this can be explained using theories borrowed from behavioral economics. Behavioral economics argues that some financial phenomena can be better understood using models in which some agents are not fully rational. In this case the EMH does not hold. Theories which can support irrationality of investors’ behavior surrounding changes in credit rating announcements are: home bias, availability bias and risk aversion which cause deviations from the rational response predicted by the EMH. Each will be discussed in turn.

Home Bias
Barberis and Thaler (2003) argue that investors can have insufficient diversification of their portfolio as a result from a home bias. There is a possibility that this effect is present for investors in evaluating sovereign bonds. Investors could more likely to buy bonds from their country of residence. This could be the case because investors can feel more familiar and (think) they have more knowledge about their own country’s economic situation. In my opinion this can lead to possible bias by letting their investment decision not only be decided by the sovereign credit rating but also by their human capital. In economic ‘bad’ times this can lead to a possible overestimation of the risk and in economic ‘good’ times to an underestimation of risk.

Availability Bias
Another theory which seems plausible is that investors have an ‘availability bias’, which states: (Kahneman & Tversky, 1979) when judging the probability of an event, people often search their memories for relevant information. While this is a perfectly sensible procedure, it can produce biased estimates because not all memories are equally retrievable or available.
For example, it could be the case that investors rely too much, when evaluating the default probability of a country, on recent memories which are more readily available. For example, if a country receives bad publicity in the media, this could bias the investment decision of many investors by selling the sovereign bond or buying a credit default swap which causes an increase in the requested yield-to-maturity.

Risk Aversion
(Pindyck & Rubinfeld, 2009) Risk-averse individuals prefer a certain given income to a risky income with the same expected value. With respect to sovereign bonds, which entail risk, this means risk-averse investors require a risk premium, in order to be extra compensated for the risk they face when holding such a bond. Risk aversion changes over time and a proxy for this which is often used are either U.S. Treasury bonds or German government bonds. Investors usually perceive these bonds as risk-free. If there is an increased demand for these risk-free assets, the price goes up and the corresponding yield goes down. In other words, investors become more risk-averse, because they demand more risk-free assets (U.S. Treasury bonds or German government bonds). The same holds in opposite direction, if the demand decreases for risk-free assets, the price goes down and the corresponding yield goes up. With an economic downturn, investors in practice tend to become more risk-averse, which decreases the yield of riskless bonds, whereas in times of economic growth they usually become less risk-averse, which increases the yield of riskless bonds.
In principal the yield of a riskless bond depends on the expected yearly inflation rate and country’s yearly real growth. Since, if an investor would diversify his portfolio in every sector of a country’s economy, his return would be equal to the country’s real growth. Furthermore the inflation rate increases the general price level which makes it more expensive to buy an item within one year from now, compared to buying it today. Because of these mentioned foregone earnings, when investing in riskless government bonds, investors would at least want a return equal to a country’s real growth plus the inflation rate. Hereby the expected inflation rate and the real growth rate within a year are fairly constant, but both tend to fluctuate over the years. That is why within a year, changes in the riskless bond yields can be seen as a good proxy for risk aversion. But yields across years cannot be used as a proxy for risk aversion since inflation and real growth change over time.

To illustrate this, the daily yields from the 1st of January, 1990, up until 12th of October, 2011, for the German government bonds and the U.S. Treasury bonds are shown in graphs 5.1 and 5.2 in the Appendix: Section V. In these figures it can be shown that at the end of the year 2008, assuming that inflation and real growth expectations are fairly constant within this year, after the collapse of Lehman Brothers, there was a sharp decline in the riskless bond yield, meaning an increase in risk aversion. This example shows that in economic crisis, which was triggered by the collapse of Lehman Brothers, investors tend to become more risk-averse than in periods of economic growth.
1.4 Related literature
There are several studies which relate to this research and can be roughly categorized into two groups. Firstly, several studies take sovereign credit ratings as levels and estimate the impact of these ratings on bond spread or CDS spread. For example, Dell’Ariccia et al. (2006) found evidence that credit ratings are crucial, even when controlling for a country’s fundamentals.
Secondly, there are studies that focus on the effect of changes in sovereign credit ratings on changes in bond spread or CDS spread. Hereby Alfonso et al. (2011) found a significant effect of rating changes, for European Union countries, on bond spreads and CDS spreads, whereby downgrades tend to have a bigger effect than upgrades on the spreads. A similar result was found by Cantor and Packer (1996), who show that rating changes affect bond yields, even when controlling for a country’s fundamentals. They conclude that ratings add additional information beyond a country’s macroeconomic statistics. Whereas Cavallo et al. (2008) studied whether bond spreads and credit ratings are imperfect measures of the unobservable fundamentals of the economy, and therefore ratings provide information above and beyond what the spreads reflect. They show that not all the information of a rating change is reflected by the bond spread and that the rating explains some of the variation within nominal exchange rates and stock market prices. Thus they argue that ratings add value beyond what is captured by the bond spread.

This paper addresses both rating changes and a country’s fundamentals, as business cycles, for several areas using data from 1990 up to 2011 with respect to bond spreads and CDS spreads.
This makes it complementary to the earlier literature that was either lacking in: the chosen timespan, the number of areas considered or they did not consider business cycles as a possible explanatory variable.
1.5 Models
This subsection tries to close the gap between the theory section and the empirical analyses, of which the latter will be dealt with in Section III. Theories discussed in this section tried to explain factors that can drive investors’ behavior within the bond and credit default swap market. These factors are discussed in order to be able to understand why changes in sovereign credit ratings should add new information within these markets, but that there are also other possible information sources investors might consider.

First it was assumed that the bond and CDS spreads are equal in theory, and that both spreads depend only on default risk and the recovery rate. Secondly, it was shown that in practice liquidity risk also plays a role, whereby sovereign bonds usually bear a larger liquidity risk than credit default swaps. Thirdly, it was discussed that if investors are rational, the bond and CDS spread would be determined by: default risk, the recovery rate, liquidity risk and investors would take into account possible rating drift by credit rating agencies. Furthermore, rational investors would normally only consider credit ratings as a sole information source, as a measure for default risk, but in some cases they might also consider other reliable public information about anticipated changes in a country’s default probability.
Fourthly, if investors show irrationalities they could tend to over- or underestimate default risk as a result of behavioral biases such as: home bias, availability bias or risk aversion.

All these factors should theoretically be included within a full blown model in order to measure the precise effect of changes in sovereign credit ratings on investors’ behavior, measured as deviations in bond and CDS spreads. However the models used in the empirical analyses (see Section III) will be rather simple because it is rather difficult to incorporate all these factors and to use the right methods when estimating such a full blown model. Within this simple model I will first use credit rating changes as a single measure for changes in default risk, which should explain deviations in bond or CDS spreads. Later on I will add business cycles as an explanatory variable, because irrational investors might also consider the cycle a country is in when estimating default risk. Another reason for the use of the relatively simple models (used in Section III) is that, from a pragmatic point of view, these models ought to be reliable enough to answer the research questions dealt with in this paper.

Section II: Data & Stylized facts
2.1 Data
This subsection explains which data is used and the way it is obtained in order to perform the analyses. Data includes: sovereign credit ratings, CDS spreads, bond spreads, GDP per capita in constant prices, bank lending strictness and the countries involved for each variable respectively.

Sovereign Credit Ratings
The data that is used in my analysis consists of rating announcements of the three major credit rating agencies: Moody’s, Standard & Poor’s and Fitch Ratings. For Moody’s credit rating announcements from 1990 up until end 2010 are used. For S&P the data consists of rating announcements from 1990 up until 31st of May, 2011. For Fitch ratings the data ranges from 1990 up until 28th of July, 2011.

CDS spreads
For the CDS spreads before the 30th of September, 2010, data is used from the Credit Market Analyst (CMA). For CDS spreads after the 30th of September, 2010, data is used from Thomson Reuter DataStream. For both datasets it contains end-of-day data, in which the CDS spread is the average of the last bid and last offer on a specific trading day.

The countries which are used for the analysis are countries with available data on CDS spreads surrounding a change in a sovereign credit rating. These countries are the following: Bahrain, Belgium, Bolivia, Brazil, Bulgaria, Chile, China, Colombia, Costa Rica, Croatia, Cyprus, Czech Republic, Ecuador, Egypt, El Salvador, Estonia, Greece, Hong Kong, Hungary, Iceland, Indonesia, Ireland, Israel, Italy, Japan, Kazakhstan, Korea, Latvia, Lebanon, Lithuania, Malaysia, Malta, Morocco, Panama, Peru, Philippines, Poland, Portugal, Romania, Russia, Saudi Arabia, Serbia, Slovakia, Slovenia, South Africa, Spain, Thailand, Tunisia, Turkey, Ukraine, Uruguay, Venezuela and Vietnam.

Bond Spreads
In order to calculate the bond spreads first the risk-free rate was approximated by using German government bond yields. Then these German bond yields are subtracted from the bond yields of EU-countries which have undergone changes in sovereign credit ratings over the period: 1990-2011. The bond yield data was acquired from the Eurostat Statistics Database. Hereby the bond yields are daily interest rates over the timespan from the 1st of January, 1990 up until 2011 taken for EU-countries (for more details see Eurostat, 2011a).

The countries which are used for the analysis are all the EU-countries for which bond spreads where available surrounding a change in a sovereign credit rating. These countries are the following: Belgium, Bulgaria, Cyprus, Czech Republic, Denmark, Finland, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Malta, Poland, Portugal, Romania, Slovakia, Slovenia, Spain and Sweden

GDP in constant prices
In order to estimate the quarterly change in a country’s gross domestic product (GDP) in constant prices, the quarterly GDP in current prices (in millions of national currency) which is seasonally adjusted, is discounted by the quarterly inflation rate. In order to calculate this quarterly inflation rate, the yearly annual average inflation rate is used. Hereby the inflation rates are the ‘Harmonized Indices of Consumer Prices’ (HICPs), for more details see Eurostat (2011b). HICPs give comparable measures of inflation between countries. They are economic indicators that measure the change over time of the prices of consumer goods and services acquired by households. In other words they are a set of consumer price indices (CPIs) calculated according to a harmonized approach and a single set of definitions.
The data for the quarterly GDP in current prices and the yearly average inflation rate are retrieved from Eurostat Statistics Database.

The GDP in constant prices is constructed as follows:

GDP in constant prices = Quarterly GDP in current prices(1+(Yearly inflation rate10014))

The countries that are taken into the analysis are EU countries which had available data (from the Eurostat Statistics Database) on yearly inflation rates and quarterly GDP. These countries are the following: Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden and the United Kingdom.

Bank lending tightening/easing
(European Central Bank, 2011a) The variable bank lending tightening/easing is based upon an index, which is the result of an extensive survey from the ECB, handed out to European banks, which generates an overall score for the strictness with which banks issue money. The survey addresses issues such as credit standards for approving loans as well as credit terms and conditions applied to enterprises and households. It also asks for an assessment of the conditions affecting credit demand. The survey is addressed to senior loan officers of a representative sample of euro area banks and is conducted four times a year. The sample groups, participating in the survey, are banks from all euro area countries and the characteristics of their respective national banking structures are taken into account. The questions distinguish between three categories of loans: loans or credit lines to enterprises; loans to households for house purchase; and consumer credit and other lending to households. For all three categories, questions are posed on credit standards for approving loans, credit terms and conditions and credit demand and the factors affecting it. The responses to questions relate to the difference (‘net percentage’) between the share of banks reporting that credit standards have been tightened and the share of banks reporting that they have been eased. A positive net percentage indicates that a larger proportion of banks have tightened credit standards (‘net tightening’), whereas a negative net percentage indicates that a larger proportion of banks have eased credit standards (‘net easing’).

The countries that are considered are within the euro area. (European Central Bank, 2011b) These are the following: Belgium, Germany, Estonia, Ireland, Greece, Spain, France, Italy, Cyprus, Luxembourg, Malta, the Netherlands, Austria, Portugal, Slovenia, Slovakia and Finland.

2.2 Stylized facts
In this section stylized facts are mentioned for: sovereign credit ratings, CDS spreads, bond spreads and bank lending tightness/ease.

Sovereign Credit Ratings
Within the time period 1990-2011 the three credit rating agencies: Moody’s, Standard&Poor’s and Fitch Ratings had a combined total of 1882 sovereign credit rating changes, from which where 1014 upgrades and 868 were downgrades (see Table 2.2).

Table 2.2 | Upgrade | Downgrade | Total | Moody’s | 325 | 211 | 536 | Standard&Poor’s | 364 | 380 | 744 | Fitch Ratings | 325 | 277 | 602 | Total | 1014 | 868 | 1882 |

CDS spreads
The change in CDS spreads in basis points surrounding a credit rating downgrade or upgrade for local currency bonds and foreign currency bonds for the areas: Non-European countries, European-countries and EU-countries, can be seen in the graphs 5.3 until 5.8 in the Appendix, Section V. More details about the data used to construct the graphs can be seen in the tables in the Appendix below the graphs. Furthermore in Section II in the Appendix, Table 2.3, an overview of all the European and EU countries can be seen.
The graphs represent a twenty-day window, ten days before and up to ten days after a rating announcement is made, with the corresponding average difference and median difference in the CDS rates for the several areas. Hereby the rating announcement is made at ‘t=0’.
For the average CDS rates not the level is taken, but the average difference of these rates. Because it is interesting to see whether or not upgrade or downgrade announcements of sovereign credit ratings, are followed by an incremental increase or decrease on average of the CDS rate for Europe, the EU or Non-European countries. Thereby the average CDS rate, one day before the rating upgrade or downgrade was announced, is taken as a reference rate, for which the average CDS rates per day are discounted for. Because theory suggests that one day before the rating no new information is added to the CDS market about a country’s default risk.

The formula for the average difference in CDS rates, at a certain day within the twenty-day window, for one of the three areas, is as follows:

∆CDS(t, i) = CDS(t,i) – CDS(t-1, i)

Hereby ∆CDS(t, i) stands for the average difference of a group of countries (i), either being; non-European countries, Europe or the EU, on a certain day (t), whereby the day (t) is within the twenty day window during a rating announcement. CDS(t, i) stands for the average CDS rate of a group of countries ‘i’, on a certain day (t). Lastly CDS(t-1, i) stands for the average CDS rate, a day before the credit rating announcement was made, for countries ‘i’.

The non-European countries shows a very large average initial increase after a rating announcement for the foreign currency downgrades compared to Europe and the EU. This could be the effect of an outlier, which biases the results. That is why also the median is taken for the differences in CDS rates within a twenty-day window of the rating announcement.

The formula for the median difference in CDS rates at a certain day within the twenty-day window, for a group of countries, is as follows:

∆CDS (t, i) = CDS(t,i) – CDS(t-1, i)

Hereby ∆CDS (t, i) stands for the median of a group of countries (i), either being; non-European, European or EU countries, on a certain day (t), whereby the day is within the twenty day window during a rating announcement. CDS(t,i) stands for the median CDS rate of a group of countries ‘i’, on a certain day (t). Lastly CDS(t-1, i) stands for the median CDS rate, a day before the credit rating announcement was made, for countries ‘i’.

The figures of the average and median difference of CDS spreads are indeed informative as they show that after a downgrade announcement is made by a credit rating agency, the CDS spread increases substantially. This holds for all three areas. However for an upgrade announcement the effect on the CDS spread appears to be less significant. For Europe and the EU the effect of a downgrade or upgrade between local and foreign currency sovereign bonds appears to be the same when looking at the average ∆CDS. However the EU has a much lower spread increase (decrease) after a downgrade (upgrade) announcement of a sovereign country. The results obviously are a lot less striking for the median of Non-EU than for the average differences in CDS rates. Thus the graph of the median is more reliable in the case of the non-European countries.

The earlier discussed figures for non-European countries show that the average difference in CDS spreads, surrounding a credit rating event, is very different from the median difference in CDS spreads. This is the result of possible outliers and this notion will now be examined in more detail.
The scatterplot below shows, with the CDS spread on the vertical axis and the rating change on the horizontal axis, that there are five non-European countries which contain CDS spreads that are outliers, namely: Venezuela, Lebanon, Ecuador, Brazil, and Bolivia.

Outliers of non-European countries with ∆CDS during ∆Ratings

When analyzing the scatterplot more closely, it seems very odd that a large downgrade of Ecuador’s credit rating decreases the CDS spread when one would expect an increase in the spread. When analyzing the constructed dataset it shows that this outlier is the rating change by Moody’s, who downgraded Ecuador at the 16th of December, 2008, from CAA1 to a CA status. Moody’s (2011) reason for this was because the government of Ecuador announced on the 15th of December, 2008, it would default on two of its sovereign debt securities.
The next question that rises is if this outlier can be dropped from the dataset. It is argued by Studenmund (2006) that the mere existence of an outlier is not a justification for dropping that observation from the sample, because a regression needs to be able to explain all the observations in a sample, not just the well-behaved ones.
However, it will now be discussed why it is in my opinion applicable in this case to drop the outlier of Ecuador. The increase in Ecuador’s probability to default was anticipated by investors, because of the government announcement at the 15th of December. During this day the CDS spread increased from 4473 to 6065 basis points. Thus the rating change by Moody’s, the day after the government announcement, at the 16th of December, did not have the impact it would have had if investors, within the CDS market for Ecuador, would not already be aware of the news about the government default. That is why a decrease in the sovereign credit rating of Ecuador is not accompanied by an increase in the CDS spread, but actually after the rating announcement by Moody’s the CDS spread decreased from 6065 to 5899 basis points.
This particular case illustrates that the assumption used throughout this paper: ‘Changes in sovereign credit ratings add new information about a country’s default risk’, is clearly violated.
When including this outlier in the dataset, it biases the results in a major way. If this single outlier of Ecuador is added into the regression analysis, there appears no significant effect of a change in a credit rating on the CDS spread for non-European countries, whereas, if the outlier is dropped, there is a significant effect present, with a significance level of 0.05. There are 238 other rating changes with an accompanied CDS spread which would thereby be biased by this one outlier. The regression analysis in Section III therefore uses the dataset of CDS spreads excluding the previous mentioned outlier of Ecuador.

Bond spreads
The change in bond spreads surrounding a credit rating downgrade or upgrade for local currency bonds and foreign currency bonds for EU-countries can be seen in the graphs 5.9 and 5.10 in the Appendix, Section IV. More details about data used to construct the graphs can be seen in the tables below the graphs.
The figures represent a twenty-day window, ten days before and up to ten days after a rating announcement is made, with the corresponding average difference and median difference in the bond spreads for the EU. Whereby the rating announcement is made at ‘t=0’.
Thereby the average bond spread, one day before the rating upgrade or downgrade was announced, is taken as a reference rate for which the average bond spreads per day are discounted by. Because theory suggests that one day before the rating, no new information is added to the sovereign bond market about a country’s default risk.

The formula for the average difference in bond spreads, at a certain day within the twenty-day window for the EU, is as follows:

∆s(EU, t) = s(EU, t) – s(EU, t-1)

Hereby ∆S(EU, t) stands for the average difference for the EU countries on a certain day (t), whereby the day is within the twenty day window during a rating announcement. S(t, i) stands for the average bond spreads of the EU, on a certain day (t). Lastly S(t-1, i) stands for the average bond spreads, a day before the credit rating announcement was made, for the EU.

The formula for the median difference in bond spreads, at a certain day within the twenty-day window, for the EU, is as follows:

∆s (EU, t) = s(EU, t) – s(EU, t-1)

Hereby ∆s (EU, t) stands for the median bond spread for the EU-countries on a certain day (t), whereby the day is within the twenty day window during a rating announcement. s(EU, t) stands for the median bond spreads of the EU, on a certain day (t). Lastly s(EU, t-1) stands for the median bond spread, a day before the credit rating announcement was made, for the EU.

The graphs 5.9 and 5.10 in the Appendix, Section V, illustrate that a change in a sovereign credit rating appears to affect the average and median bond spread for downgrades but hardly for upgrades.

Bank lending tightness/easing
In graph 4.11 (see Appendix, Section V) the change in the bank lending tightness index, based upon the quarterly survey, can be seen for the period 2003-2011, for the euro area (see Appendix, Section II, Table 2.3, for an overview of the countries belonging to the euro area). A positive net percentage indicates that a larger proportion of banks have tightened credit standards (‘net tightening’), whereas a negative net percentage indicates that a larger proportion of banks have eased credit standards (‘net easing’). It shows that roughly from 2003 up until the 1st of July, 2005, the change per quarter in net tightening was increasing fairly moderate. From the period from the 1st of October, 2005 up until the 1st of July, 2007, the change in net tightening per quarter increased almost constant. Whereas from the period: the 1st of October, 2007, up until the 1st of January, 2009, there is a relatively large increase in the net easing per quarter of credit standards, with the exception of the first of September, 2008, which was during the time Lehman Brothers declared bankruptcy where there is a small increase in net tightening. After the first of January, 2009 there is an increase in the net tightening again up to the first of October, 2011.

Section III: Empirical analyses
In this section the two following research questions are examined: * Do investors react to changes in sovereign credit ratings? * Do investors react differently on changes in sovereign credit ratings, dependent on the economic business cycle that a country is in?

Before the empirical analyses is elaborated upon, first a full description of the variables used in the analyses and some general specifications of these variables are dealt with, secondly an explanation of rating transformations and thirdly how rating changes are calculated.

A full description of the variables used for the empirical analyses are summarized in the Appendix, Table 6.1. Furthermore table 6.2 in the Appendix shows the following characteristics of the variables: number of observations, mean, standard deviation, minimum and maximum.

In order to perform an analysis of the effect of changes in credit ratings on CDS spreads and bond spreads, the ratings of Moody’s, Standard & Poor’s and Fitch Ratings are transformed by using a linear scale. Hereby the highest rating receives the number 1, the second highest rating the number 2, and so on. For a full overview of the linear scaling, see table 2.1, Appendix, Section II. Note that Fitch Ratings is missing in the table, because they use the same rating system as S&P, therefore the linear transformation of credit ratings is the same in both instances.

By transforming credit ratings to a linear scale it is possible to calculate the change in the credit rating, either being an upgrade or downgrade, for a sovereign country. Using an example, an upgrade in a sovereign country’s credit rating is calculated as follows: if a country is upgraded from AA+ to AAA, this means a (linear transformed) rating change from 2 to 1, the change in a credit rating is: 1-2 = -1. Thus an upgrade in a sovereign credit rating is represented by a negative number. The opposite holds for a downgrade, which is represented by a positive number.

3.1 Effect of rating events on investors’ behavior
Let us turn to the first research question first: Do investors react to changes in sovereign credit ratings?
As discussed before the graphs 5.3 until 5.8 in the Appendix, Section IV, suggested that there is an effect of a change in a sovereign credit rating on the CDS spread for the EU, European and non-European countries. Furthermore graphs 5.9 and 5.10 implied that the impact of a change in a sovereign credit rating affected the bond spread of EU-countries only noticeably for downgrades. However the graphs didn’t differentiate between the sizes of the changes in sovereign credit ratings on the impact on CDS spreads and bond spreads. This however can be tested by using the OLS (Ordinary Least Squares) method. First the models used for the analyses will be described before the regressions are executed.

Literature in the previous theory section of this paper suggested that the following relation might hold for a change in the spread of a CDS and bond:
CDSi t, = µi, t x (1 – RRi,t) si, t = µi, t x (1 – RRi,t)

However it is also mentioned in the theory section that the CDS and bond spread may depend on several behavioral biases, liquidity risk and risk aversion. In the following analyses the very strong assumptions are made that: the recovery rate, liquidity risk, risk aversion, home bias and availability bias are kept constant surrounding a sovereign credit rating change. This means that a change in a sovereign credit rating, as a measure for default risk, is the only explanatory variable for a change in the CDS spread for country ‘i’ at time period ‘t’. Whereby ‘t’ is the date of the change in the sovereign credit rating, either being an upgrade or downgrade.

With the other explanatory variables kept constant, the following two bivariate population models are constructed: * (1) ∆CDSi, t = β0 +β1 * ∆Ratingi,t + εi,t with: ∆CDSi, t = the change in a credit default swap for country ‘i’ at time period ‘t’. β0 = Intercept β1 = Slope
∆Ratingi, t = the change in a sovereign credit rating for country ‘i’ at time period ‘t’ εi,t = error term

* (2) ∆Si, t = β0 +β1 * ∆Ratingi, t + εi,t with: ∆Si,t = the change in the sovereign bond spread for country ‘i’ at time period ‘t’ β0 = Intercept β1 = Slope
∆Ratingi, t = the change in a sovereign credit rating for country ‘i’ at time period ‘t’ εi,t = error term

By using the econometric technique of the Ordinary Least Squares these population models can be estimated:
- (3) ∆CDSi, t = β0 +B1 * ∆Ratingi,t + ei,t
- (4) ∆Si, t = β0 +B1 * ∆Ratingi,t + ei,t

Classical assumptions
Now that the models are introduced they are tested for if the classical assumptions are met in order for the OLS estimators of parameters β, to be the best available. This is important, in order to have an unbiased and consistent estimation of the effect of changes in credit ratings on the bond and CDS spread.
The full overview can be seen in the Appendix, Section VII: Testing classical assumptions. In short this section shows that equation (3) for all three areas (Non-European/European/EU) contains heteroskedasticity which is corrected for in the following analyses by running robust regressions. Furthermore equation (4) for EU countries could be biased since the Ramses RESET test indicates that a misspecification is present, which is why the regression results in this case could be biased, which should be kept in mind when interpreting the empirical results.

T-test
In the next two subsections model (3) for the ∆CDS spread and model (4) for the ∆bond spread will be tested, using a two-sided t-test, with the following hypotheses:
H₀ : B1 = 0
H₁ : B1 ≠ 0 α = 0.05
The zero-hypothesis states that there is no effect of credit rating changes on the CDS spread for equation (3) or the bond spread for equation (4). The alternative hypothesis states that the zero-hypothesis is not true, using a significance-level of five percent.

CDS spreads surrounding sovereign credit rating events
The regressions with a two-sided t-test are shown below for: Europe, the EU and non-European countries using equation (3):
∆CDSi, t = β0 +B1 * ∆Ratingi,t + ei

For European countries:

For EU countries:

For Non-European countries:

The results show that for all three areas there is a significant effect of the change in a credit rating on the CDS spread surrounding the credit rating event within a two-day window (t-1, t+1) . Thus the null-hypothesis can be rejected, with a significance level of 0.05.
Put in other words, ‘B1’ is not equal to zero with a 95% confidence interval. The slope and the intercept for the three areas for equation (3) differ from one another, each of which will be discussed in turn. * For the European countries the estimated model is the following:
∆CDSi, t = 4.09 + 9.10 * ∆Ratingi,t + ei,t
This implies that a downgrade of the rating by 1 increases the CDS spread by (4.09 + 9.10 * (1)), within the two-day window. This means an increase of 13.19 basis points of the CDS spread. An upgrade by 1 decreases the CDS spread by (4.09 + 9.10 * (-1)). This means a decrease of 5.01 basis points.

* For the EU countries the estimated model is the following:
∆CDSi, t = 3.72 + 4.92 * ∆Ratingi,t + ei,t
This implies that a downgrade of the rating by 1 increases the CDS spread by (3.72 + 4.92 * (1)) within the two-day window. This means an increase of 8.64 basis points of the CDS spread. An upgrade by 1 decreases the CDS spread (3.72 + 4.92 *(-1)). This means a decrease of 1.2 basis points.

* For the Non-European countries the estimated model is the following:
∆CDSi, t = 9.10 + 10.85 * ∆Ratingi,t + ei,t
This implies that a downgrade of the rating by 1 increases the CDS spread by (9.10 + 10.85 * (1)) within the two-day window. This means an increase of 19.95 basis points of the CDS spread. An upgrade by 1 decreases the CDS spread by (9.10 + 10.85 * (-1)). This means a decrease of 1.75 basis points

The results show that a downgrade by 1 in a credit rating has the biggest effect on the CDS spread for the Non-European countries, followed by European countries and lastly for EU countries for the period 1990-2011, since 19.95 > 13.9 > 8.64. Whereas an upgrade by 1 in a credit rating has the biggest effect on the CDS spread for EU countries, followed by Non-European countries and lastly for European countries for the period 1990-2011, since 5.01 > 1.75 > 1.2. Lastly a downgrade for all three areas has a bigger impact on the CDS spread than a subsequent upgrade.

Bond spreads surrounding sovereign credit rating events
In order to check if the same result hold for the bond spread, as for the CDS spread, surrounding a sovereign credit rating event, the regression is executed using equation (4): ∆Si, t = β0 +B1 * ∆Ratingi,t + ei,t.. Hereby the window for equation (4) is: one day before the credit rating announcement up until one day after the rating announcement. Because of lack of data the only area which is examined is the EU.

The regression for equation (4) is as follows for EU-countries:

The regression shows that for the change in a sovereign credit rating, with a two-day window surrounding the credit rating event, there is no significant effect on the bond spread for the EU, with α = 0.05. Because 0.863 > 0.05, the null hypothesis cannot be rejected, which indicates that it cannot be proven that B1 is different from zero.

However, when the window is increased from a two-day window to a three-day window surrounding the credit rating change (t-1, t+2), the results are different as the following regression shows:

In this case there is a significant effect of the change in credit rating on the bond spread, surrounding the event within a three-day window, for the EU, with a significance-level of 0.05, since 0.005 < 0.05. * For the EU the estimated model is the following:
∆Si, t = 0.037 + 0.039 * ∆Ratingi,t + ei,t
This implies that a downgrade of a credit rating by 1 increases the bond spread by (0.037 + 0.039 * (1)). This means an increase of 0.076 in the bond spread, alternatively +7.6 basis points. An upgrade of a credit rating by 1 decreases the bond spread by (0.037 +0.039 * (-1)). Which is a decrease of 0.002, alternatively -0.2 basis points.

Concluding remarks
The CDS spread changes significantly after a rating announcement within a two-day window for: Europe, EU-countries and non-European countries. Furthermore the bond spread for EU-countries is significantly affected by a change in a sovereign credit rating within a three-day window but not within a two-day window. This means that the three-day window appears to be a better fit for explaining the bond spread difference during a credit rating event. Thus in the CDS market investors appear to react faster to changes in sovereign credit ratings than investors in the sovereign bond market. These results are consistent with empirical findings from Blanco et al. (2005), discussed earlier in the theory section, who in short stated that CDS spreads adjust faster than bond spreads to new information because they appear to be more liquid than bonds.
An overview of the effect of a rating change, on the CDS or bond spread is shown in the table below: Countries | Effect of an upgrade of the rating by 1 on the CDS spread | Effect of a downgrade of the rating by 1 on the CDS spread | Effect of an upgrade of the rating by 1 on the bond spread | Effect of a downgrade of the rating by 1 on the bond spread | European Union | -1.2 | 8.64 | -0.2 | +7.6 | European | -5.01 | 13.9 | - | - | Non-European | -1.75 | 19.95 | - | - |

The table shows that the effect of a rating change on the bond spread is smaller than for the CDS spread regarding the EU. Lastly, the table indicates that a downgrade by 1 has a bigger impact on the CDS and bond spread than an upgrade by 1.

3.2 Effect of business cycles on investors’ behavior surrounding rating events
The third research question is: Do investors react differently on changes in sovereign credit ratings, dependent on the economic business cycle that a country is in?
In order to answer this question the reaction of investors is measured in two ways: either by changes in bond spreads or changes in CDS spreads, surrounding a credit event. Hereby the window used for the CDS spreads is from one day before the rating announcement up until one day after the rating announcement. The window used for the bond spreads is from one day before the rating announcement up until two days after the rating announcement, since the last subsection suggested that bond spreads adjust slower than CDS spreads.
To determine the economic business cycle a sovereign country is in, three approaches are used: either by looking at quarterly changes in GDP, looking at the change in net bank lending tightness or by comparing the pre-crisis period with the period of crisis. Each of these methods will be discussed in turn.

Effect of previous changes in quarterly GDP on bond spreads and CDS spreads
In this subsection a business cycle is postulated as the quarterly change in GDP in constant prices.
Thereby a dummy is constructed first for the economic growth or downfall a quarter preceding a rating announcement. If there is an increase in the GDP of an EU-country, a quarter previous to the rating announcement, the value of the dummy is one. If there is a decrease of GDP for EU-countries, a quarter previous to the rating announcement, the value of the dummy is zero.

The change in GDP in constant prices can be written as follows: * ∆GDPEU, t-1= GDP EU, t-1 – GDP EU, t-2
With:
∆GDPEU, t-1 = change in quarterly GDP in constant prices for a EU country, one quarter previous to a rating announcement
GDP EU, t-1 = Quarterly GDP in constant prices for a EU country, one quarter previous to a rating announcement
GDP EU, t-2 = Quarterly GDP in constant prices for a EU country, two quarter previous to a rating announcement
Whereby the GDP dummy is defined as follows:
∆GDPEU, t > 0, growth_dummy = 1 (Economic Growth)
∆GDPEU, t < 0, growth_dummy = 0 (Economic Downfall)

In order to check whether an economic growth a quarter preceding a rating event has a different impact on CDS and bond spreads than an economic downfall, a quarter preceding a rating event, the following interaction terms are constructed: * changegood = number_diff * growth_dummy * changebad = number_diff * (1-growth_dummy)
Hereby ‘changegood’ represents an interaction term between economic growth, a quarter preceding the rating announcement, and a rating change. And ‘changebad’ represents an interaction term between economic downfall, a quarter preceding the rating announcement, and a rating change.

Using the interaction terms it now can be tested if a preceding quarterly economic growth or downfall has a different effect on the government bond spread, within the three-day window: from one day before the rating announcement up to two day after the rating announcement, for EU-countries.
The regression is as follows:

The regression results show that neither a country’s preceding quarterly economic growth nor downfall affects the sovereign bond spread surrounding a credit event within a three-day window, with a confidence interval of 95 percent, since both P-values: 0.405 and 0.215 are bigger than 0.05.

Again the interaction terms: ‘changegood’ and ‘changebad’ are used, but now in order to test if they have a different effect on CDS spread for EU-countries surrounding a sovereign credit rating event.
The regression is as follows:

The regression results show that a sovereign country’s preceding quarterly economic downfall before a credit rating announcement takes place significantly affects the CDS spread surrounding a sovereign credit rating event within a two-day window (t-1, t+1). This however is not the case for periods of economic growth, since the P-value is larger than 0.05 with 0.89 respectively.

Below an F-test is performed to check if there is a significant difference between the interaction terms ‘changegood’ and ‘changebad’:

The results show that with a significance level of 0.05, the null-hypothesis cannot be rejected, meaning that there is no significant difference between ‘changegood’ and ‘changebad’ It should however be noted that with a significance level of 0.1 the zero hypothesis can be rejected, meaning that there would be a difference between ‘changegood’ and ‘changebad’. But then it is more likely that a Type I error is made, which states that a true null-hypothesis is rejected when it is in fact true.

The interaction terms ‘changegood’ and ‘changebad’ didn’t differentiate between the size of the preceding increase in GDP or preceding decrease in GDP which investors might take into account when estimating default risk during rating changes.
To test if changes in CDS spreads are caused by preceding changes in GDP and/or changes in credit ratings, the following regression is executed:

To test if changes in bond spreads are caused by preceding changes in GDP and changes in credit ratings, the following regression is executed:

Both regressions show that there is no effect of a preceding change in GDP (in absolute values) on the bond and CDS spreads using a significance level of 0.05. But the earlier results, using the interaction terms ‘changegood’ and ‘changebad’, showed that in the CDS market investors do respond to a preceding economic downfall but not to an economic growth. Thus it can be concluded that investors within the CDS market use information about a decrease in a country’s level of GDP, a quarter preceding a rating announcement, but do not take into account the relative size of the decrease of preceding GDP when determining default risk.

Another possibility is that investors might not take into account a country’s preceding absolute change in GDP for estimating default risk, but rather rely on their expectations of absolute changes within the quarterly GDP. The proxy that will be used for the expected absolute change in GDP within a given quarter will be the actual change in GDP within that quarter.

To test if changes in the CDS spread are caused by absolute changes in quarterly GDP and/or credit rating changes, the following regression is executed:

The regression shows that the expected quarterly GDP plays a significant role surrounding a credit rating event in explaining the CDS spread. But now the rating announcement doesn’t significantly affect the spread anymore, since 0.062 is bigger than the significance level of 0.05. But one should be weary when interpreting these results because rating changes also depend on a country’s GDP, whereby it cannot be said with certainty that investors only use their expectations about GDP for estimating default risk within the CDS market, but could use the ratings instead which already reflect this change in GDP. Thus multicollinearity might exist, which causes difficulty in identifying the separate effects of rating changes and changes in GDP on spreads. To see if this is the case the correlation is shown below between rating changes and changes in GDP:

It shows that credit rating changes and changes in GDP are not perfectly correlated, meaning no perfect multicollinearity. However an additional approach is used in order to test for imperfect multicollinearity, namely the Variance Inflation Factor (VIF). This test can indicate if the imperfect multicollinearity is severe or not. (Studenmund, 2006) Although there is no table of formal critical VIF values, a common rule of thumb is that if the VIF is bigger than five, the multicollinearity is severe and if there is no multicollinearity the VIF is equal to one. The VIF test for the CDS spread regarding the two explanatory variables: changes in GDP and changes in rating, is shown below: It shows that the VIF is 1.09, which is far lower than 5, which indicates that hardly any imperfect multicollinearity exists between changes in GDP and the credit ratings.

To test if changes in the bond spread are caused by anticipated changes in quarterly GDP and/or credit rating changes the following regression is executed:

It shows that as well as the rating change as the quarterly GDP significantly affect the bond spread with a significance level of 0.05. Thus it can be concluded that investors within the bond market use information about an anticipated change in a country’s level of GDP and also use changes within credit ratings for determining the default risk of EU countries.

Concluding remarks
The results of the previous subsection, regarding changes in GDP on the CDS and bond spread, are summarized below:

For bond spreads: * No effect was found of a preceding quarterly economic growth or downfall, taking into account the relative size of the credit rating change, on the bond spread surrounding a credit rating event. * No effect was found of a preceding quarterly change in GDP (in absolute values) on the bond spread surrounding a credit rating event. * An effect was found of the expected quarterly GDP on the bond spread surrounding a credit rating event.

For CDS spreads: * An effect was found of a preceding quarterly economic downfall, taking into account the size of the credit rating change, on the CDS spread surrounding a sovereign credit rating event. However this effect was not found in the case of a preceding quarterly economic growth. * No effect was found of a preceding quarterly change in GDP (in absolute values) on the CDS spread surrounding a credit rating event * An effect was found of the expected quarterly GDP on the CDS spread surrounding a credit rating event. But in this case rating changes did not significantly affect the CDS spread anymore.

Thus it appears that in the bond market investors use both changes in credit ratings and expected changes in quarterly GDP for estimating changes in default risk of a sovereign country.
Whereas in the CDS market investors use information about a country’s preceding quarterly decrease in GDP as well as the expected quarterly GDP for estimating changes in default risk of a sovereign country.

Effect of changes in bank lending tightness
It is interesting to look at the changes in bank lending tightness within the euro area to see if it might be a good indicator for a banks’ risk aversion in period of economic crisis or growth. It seems plausible that in a period of crisis, banks are less willingly to lend credit, because they might have less access to money and a face greater risk since borrowers are less likely to be able to repay their debt. This is supported by research of Ivashina and Scharfstein (2008), who argue that new lending declined substantially during the financial crisis across all types of loans. Some of this decline could have reflected a drop in demand as firms scaled back expansion plans during a recession. However, they show that there may also have been a supply effect: banks with less access to deposit financing and at greater risk of credit-line drawdowns reduced their lending more than other banks.

This notion is examined by looking at the effect of quarterly changes in the bank lending tightness index (as discussed in Section II), for the euro area, on the CDS and bond spread surrounding a change in a sovereign credit rating.
The regression is as follows for the CDS spread:

This regression shows that a preceding net change in the index of bank lending strictness has a significant effect on the CDS spread surrounding a credit rating event, within a two-day window (t-1, t+1), because 0.025 < 0.05.

If the net change in the bank lending tightness index increases by 1, the CDS spread, within a two-day window surrounding a credit rating event, increases with (61.17 + (4.32 * Number_diff) + (-20.32 * (1))) = 40.85 basis points, ceteris paribus the other explanatory variable. If the net change in the bank lending tightness index decreases by 1, the CDS spread, within a two-day window surround a credit rating event, increases with (61.17 + (4.32 * Number_diff) + (-20.32 * (-1))) = 81.17 basis points, ceteris paribus the other explanatory variable.

Next it is tested if a net change in bank lending strictness also affects the bond spread surrounding a credit rating event. The regression for the bond spread is as follows: This regression shows that there is no significant effect of the net change in the index of bank lending strictness on the bond spread surrounding a credit rating event within a three-day window (t-1, t+2), because the P-value of 0.064 is larger than 0.05.

Concluding remarks
The previous subsection showed that a net change of bank lending strictness has an effect on the CDS spread surrounding the credit rating announcement but does not affect the bond spread. These results hold true for the euro area (EU17), which was the only area taken into consideration because the bank survey was only conducted within this area.

Pre-crisis period versus the crisis period
Another way to roughly postulate business cycles is to define the period from: 1990 up until the 14th of September, 2008, as the period before the economic recession and the period from: the 15th of September, 2008, up until 2011, as the period during the economic recession. (Ivashina & Scharfstein, 2008) These periods are taken, since at the 15th of September, 2008, the American bank Lehman Brothers filed for bankruptcy, which further offset the economic crisis that threw economies around the world in recession. The seeds of the economic crisis were already sown in late 2007, however after the collapse of Lehman Brothers the crisis became full blown and that is why these previous discussed periods are used for analyses.

To see if business cycles do affect the bond and CDS spreads when a sovereign credit rating change takes place, the change in spreads should be larger (or smaller) in the period of economic recession than in a period of economic growth.

This will be tested for the two equations, already explained in a previous subsection:
- (3) ∆CDSi, t = β0 +B1 * ∆Ratingi,t + ei
- (4) ∆Si, t = β0 +B1 * ∆Ratingi,t + ei

First equation (3) is used in order to test whether in the pre-crisis period, sovereign credit rating changes affect the CDS spread differently than in the crisis period for three areas (EU/Europe/Non-European).

∆CDS spread for EU-countries
The regression for the effect of changes in credit ratings for EU-countries on the CDS spread, before the 15th of September, 2008, is the following:

The regression shows that there is a significant effect of the changes in a sovereign credit rating on the CDS spread. A downgrade by 1 in the credit rating, results in an increase of the CDS spread by (0.29 + 1.74 * (1)). This means an increase in the CDS spread by 2.03 basis points surrounding a credit rating event within a two-day window (t-1, t+1). An upgrade by 1 in the credit rating, results in a decrease of the CDS spread by (0.29 + 1.74 * (-1)). This means a decrease in the CDS spread by 1.45 basis points surrounding a credit rating event within a two-day window (t-1, t+1).

The regression for the effect of changes in credit ratings for EU-countries on the CDS spread, after the 15th of September, 2008, is the following:

The regression shows that there is a significant effect of the changes in a sovereign credit rating on the CDS spread. A downgrade by 1 in the credit rating, results in an increase of the CDS spread by
(5.19 + 4.82 * (1)). This means an increase in the CDS spread by 10.1 basis points surrounding a credit rating event within a two-day window (t-1, t+1). An upgrade by 1 in the credit rating, results in a decrease of the CDS spread by (5.19 + 4.82 * (-1)). This means a increase in the CDS spread by 0.37 basis points surrounding a credit rating event within a two-day window (t-1, t+1).

To compare the pre-crisis with the crisis period for EU-countries the results are shown in the table below. It shows that there is a much larger change in CDS spread during credit rating changes in period of crisis compared to the pre-crisis period. Furthermore a downgrade in the crisis period has a much larger effect on the CDS spread than a subsequent upgrade compared to the pre-crisis period. It should be noted that result is odd that an upgrade by 1 for an EU country in period of crisis still increases the CDS spread instead of decreasing it. Period | Effect of a rating downgrade by 1 on the CDS spread in basis points | Effect of a rating upgrade by 1 on the CDS spread in basis points | Pre-crisis period EU | +2.03 | -1.45 | Crisis period EU | +10.1 | +0.37 | Difference in spread | +8.07 | +1.82 |

∆CDS spread for European countries
The regression for the effect of changes in credit ratings for European countries on the CDS spread before the 15th of September, 2008, is the following: The regression shows that there is a significant effect of the changes in a sovereign credit rating on the CDS spread. A downgrade by 1 in the credit rating results in an increase of the CDS spread by (-0.43 + 2.05 * (1)). This means an increase in the CDS spread by 1.62 basis points surrounding a credit rating event within a two-day window (t-1, t+1). An upgrade by 1 in the credit rating results in a decrease of the CDS spread by (-0.43 + 2.05 * (-1)). This means a decrease in the CDS spread by 2.48 basis points surrounding a credit rating event within a two-day window (t-1, t+1).

The regression for the effect of changes in credit ratings for European countries on the CDS spread after the 7th of September, 2008, is the following:

The regression shows that there is a significant effect of the changes in a sovereign credit rating on the bond spread. A downgrade by 1 in the credit rating results in an increase of the CDS spread by (3.01+ 10.96 * (1)). This means an increase in the CDS spread by 13.97 basis points surrounding a credit rating event within a two-day window (t-1, t+1). An upgrade by 1 in the credit rating results in a decrease of the CDS spread by (3.01+ 10.96 * (-1)). This means a decrease in the CDS spread by 7.95 basis points surrounding a credit rating event within a two-day window (t-1, t+1).

To compare the pre-crisis with the crisis period for European countries the results are shown in the table below. It shows a similar pattern as before with the EU-countries, namely that in the period of crisis the CDS spread is much more strongly affected by rating changes than in the pre-crisis period and that downgrades have an even larger effect than subsequent upgrades in the period of crisis. Period | Effect of a rating downgrade by 1 on the CDS spread in basis points | Effect of a rating upgrade by 1 on the CDS spread in basis points | Pre-crisis period Europe | +1.62 | -2.48 | Crisis period Europe | +13.97 | -7.95 | Difference in spread | +12.35 | -5.47 |

∆CDS spread for non-European countries
The regression for the effect of changes in credit ratings for non-European countries on the CDS spread before the 15th of September, 2008, is the following: The regression shows that there is no significant effect of the changes in a sovereign credit rating on the CDS spread, with a significance level of 0.05.
The regression for the effect of changes in credit ratings for non-European countries on the CDS spread, after the 15th of September, 2008, is the following: The regression shows that now there is a significant effect of the changes in a sovereign credit rating on the CDS spread during the period of economic crisis, since 0.032 < 0.05. A downgrade by 1 increases the bond spread in the period of crisis by (5.35 + 9.86 *(1)). This means an increase of 15.21 basis points. An upgrade by 1 decreases the bond spread in period of crisis by (5.35 + 9.86 *(-1)). This means a decrease of 4.51.

Because changes in credit ratings for non-EU countries in the period before the crisis (1990 up to the 6th of September, 2008) didn’t significantly affect the CDS spread, no comparison can be made regarding to the CDS spread during rating changes in period of crisis.

∆Bond spread for EU countries
The regression for the effect of changes in credit ratings for non-European countries on the bond spread before the 15th of September, 2008, is the following:

The regression shows that there appears to be no significant effect of changes in sovereign credit ratings on the bond spread in the period preceding the crisis using a significance level of 0.05.

The regression for the effect of changes in credit ratings for non-European countries on the bond spread, after the 7th of September, 2008, is the following:

The regression shows that there appears to be no significant effect of changes in sovereign credit ratings on the bond spread in the period of crisis using a significance level of 0.05.

From subsection 3.1 it was concluded that changes in credit ratings do affect bond spreads for the period 1990-2011 as a total. But this subsection shows that when looking at the pre-crisis period and the crisis period separately there appears to be no significant effect of changes in credit ratings on bond spreads, using a significance level of 0.05. This makes it inconclusive to distinguish whether business cycles affect European bond spreads surrounding a change in a sovereign credit rating by looking at pre-crisis versus the crisis period.

Conclusion
This paper examined the following research questions:
- Do investors react to changes in sovereign credit ratings?
- Why do investors react to changes in sovereign credit ratings? - Do investors react differently on changes in sovereign credit ratings, dependent on the economic business cycle that a country is in?

Under the very strong assumptions that only the change in a sovereign credit rating explains changes in credit default swap (CDS) spreads or bond spreads, the answer to the first research question is that investors do react to changes in sovereign credit ratings. The results show that there is a significant effect of credit rating changes on CDS spreads within a two-day window, from one day before the rating announcement up to one day after the rating announcement for European countries, EU-countries and non-European countries using the period 1990 up until 2011. Furthermore the bond spread for EU-countries is significantly affected by a change in a sovereign credit rating within a three-day window, but not within a two-day window. Thus in the CDS market investors appear to react faster to changes in sovereign credit ratings than investors in the sovereign bond market, but this may be due to differences in liquidity risk.

To answer the second research question, investors within the sovereign bond or CDS market should react to changes in sovereign credit ratings because this is new public information about a change in the default risk of a sovereign country. Hereby the reaction can be measured as the change in bond spread or CDS spread surrounding a credit rating event. Next to the default risk also the recovery rate, liquidity risk and risk aversion could explain deviations in the bond and CDS spread. However investor’s might not only use credit rating agencies as a sole information source for estimating default risk, but also other factors might be at play here. These other factors could be: home bias, availability bias, anticipation of rating drift by credit rating agencies, business cycles or other information made public about changes in a country’s default risk.

The answer to the third research question is that some of the empirical evidence found supports the notion that business cycles affect investors’ behavior surrounding credit rating events, although it is not entirely conclusive. The empirical analyses show that in the bond market investors use both changes in credit ratings and expected changes in quarterly GDP for estimating changes in default risk of a sovereign country. Whereas in the CDS market investors use information about a country’s expected changes in quarterly GDP, the preceding quarterly change in GDP if it is a decrease and changes in bank lending strictness for estimating changes in default risk. Remarkable is that sovereign credit rating changes do not appear to be a statistically significant factor anymore for investors to consider within the CDS market, when including the former mentioned explanatory variables into the regression analysis. This paper also found evidence that in the pre-crisis period for the European and EU countries, before the 15th of September, 2008, the CDS spreads react much more moderately on changes in credit ratings then after the 15th of September, 2008, in the period of crisis. Also rating downgrades have an even larger impact on CDS spreads than rating upgrades in the period of crisis compared to the pre-crisis period. Thus investors show irrationalities because they tend to be influenced more by periods of economic crisis then by periods of economic growth.

My research used a rather limited number of explanatory variables for the econometric analyses. Therefore some recommendations for future research would be to include: behavioral biases, liquidity risk, risk aversion and fundamentals all together in models that estimate the effect of changes in credit ratings on investors’ behavior (measured as CDS spreads and bond spreads). Furthermore the scope of this research could be increased for the number of regions included in the analyses in order to distinguish if investors’ behavior differs among those areas.

Reference list
Alpert, M., and Raiffa, H. (1982). A progress report on the training of probability assessors, in:
Kahneman, D., Slovic, P. & Tversky A., eds., Judgment Under Uncertainty: Heuristics and Biases.
Cambridge University Press, Cambridge. pp. 294–305.

Alquist, R. (2008). How Important Is Liquidity Risk for Sovereign Bond Risk Premia? Evidence from the London Stock Exchange. Working Paper. Bank of Canada. Retrieved on: 27, October, 2011, on http://www.bankofcanada.ca/wp-content/uploads/2010/02/wp08-47.pdf

Afonso, A., Gomes, P., Rother, P. (2007). What hides behind sovereign debt ratings?. European Central Bank. Working Paper Series, 711. Retrieved on: 12, September, 2011, on http://www.ecb.int/pub/pdf/scpwps/ecbwp711.pdf

Afonso, A., Furceri, D & Games, P. (2011). Sovereign credit ratings and financial markets linkages: application to European data. European Central Bank.

Baker, H. K. & Mansi, S. A. (2002). Assessing Credit Rating Agencies by Bond Issuers and Institutional Investors. Journal of Business Finance & Accounting, 29(9) & (10), Nov/Dec. 2002.

Barberis, N., Thaler, R. (2003). A survey of behavioral finance. In: Constantinides, G.M., Harris, M. & R. M. Stulz (ed.), Handbook of the Economics of Finance, edition 1, volume 1, chapter 18, pages 1053-1128 Elsevier.

Berndt, A. et al. (2005). Measuring Default Risk Premia from Default Swap Rates and EDFs. Bank for International Settlements.

Blanco, R., Brennan, S & Marsh, I (2005). An Emprical Analysis of the Dynamic Relationship between Investment Grade Bonds and Default Swaps. Journal of Finance 60, 2255–2282.

Bongaerts, D. Jong, F. & Driesen, J. (2010). Derivative Pricing with Liquidity Risk: Theory and
Evidence from the Credit Default Swap Market. Retrieved on: 26, October, 2011, on http://www.afajof.org/afa/forthcoming/6795p.pdf

Brealey, R. A., Meyers, S. C., Marcus, A. J. (2009). Fundamentals of corporate finance (6th edition). New York: Mcgraw-Hill.

Calice, G., Chen, J., Jullian Williams (2011). Liquidity Interactions in Credit Markets: An Analysis of the Eurozone Sovereign Debt Crisis. Retrieved on: 27, October, 2011, on http://www.efmaefm.org/0EFMAMEETINGS/EFMA%20ANNUAL%20MEETINGS/2011-Braga/papers/0417.pdf

Cantor, R. & Packer, F. (1995). Sovereign Credit Ratings. Federal Reserve bank of New York. Current Issues in economics and finance. Vol 1. Number 3. Retrieved on: 14, September, 2011, on http://www.newyorkfed.org/research/current_issues/ci1-3.pdf

Cantor, R. & Packer, F. (1996). Determinants and impact of sovereign credit ratings.
Economic Policy Review, Federal Reserve Bank of New York, Vol. 2, October.

Cavallo, E. A., Powell, A., Rigobón, R. (2008). Do Credit Rating Agencies Add Value? Evidence from the Sovereign Rating Business Institutions. Inter-American Development Bank.

Chari, V., Kehoe, P. (1997), ‘Hot money’, Research Department, Federal Reserve Bank of
Minneapolis.

Coudert, V., Gex, M. (2010). Credit default swap and bond markets: which leads the other?. Banque de France. Financial Stability Review. No. 14 – Derivatives – Financial innovation and stability.

Dell’Ariccia, G., Schabel, I. & Zettelmeyer, J. (2006). How Do Official Bailouts Affect the Risk of Investing in Emerging Markets?. Journal of Money, Credit, and Banking 38(7): 1689-
1714.

Edwards, S (1984), ‘LDC foreign borrowing and default risk: an empirical investigation,
1976-80’, American Economic Review, Vol. 74, No. 4, pages 726-34.

European Central Bank (2011a). The Euro Area Bank Lending Survey. Retrieved on: 27, October, 2011, on http://www.ecb.int/stats/pdf/blssurvey_201110.pdf?30040c6af1d568d586fa6ebba9766aef

European Central Bank (2011b). Map of the Euro Area 1999-2011. Retrieved on: 11, November, 2011 on http://www.ecb.int/euro/intro/html/map.en.html

Eurostat (2011a). Government bond yields - 10 years' maturity. Retrieved on: 27, October, 2011, on http://epp.eurostat.ec.europa.eu/cache/ITY_SDDS/en/irt_lt_mcby_esms.htm Eurostat (2011b). Harmonised indices of consumer prices (HICP). Retrieved on: 30th, October, 2011, on http://epp.eurostat.ec.europa.eu/cache/ITY_SDDS/en/prc_hicp_esms.htm Ferrucci, G. (2003). Empirical determinants of emerging market economies’ sovereign bond spreads. Bank of England.

Financial Times (2011). Greek rating now worst in the world. Retrieved on: 10, November, 2011, on http://www.ft.com/intl/cms/s/0/fc261efe-95de-11e0-ba20-00144feab49a.html

Fischhoff, B., Slovic, P. & Lichtenstein, S. (1977), Knowing with certainty: the appropriateness of extreme confidence, Journal of Experimental Pyschology: Human Perception and Performance 3:
552−564.

Hull, J., Predescu, M., White, A. (2004). The relationship between credit default swap spreads, bond yields, and credit rating announcements. University of Toronto.

Ivashina, V. & Scharfstein, D. (2008). Bank lending during the financial crisis of 2008. Journal of Financial Economics. Retrieved on: 31th, October, 2011 on: http://www.people.hbs.edu/dscharfstein/ivashina-scharfstein%20jfe%202010.pdf

Moody’s Investors Service, 2011. Retrieved on: 11, November, 2011, on http://www.moodysanalytics.com/~/media/Microsites/ERM/2011/through-cycle-EDF/CDS-Implied%20EDF%20measures%20for%20Sovereigns%20brochure.ashx

Olléon-Assouan, E. (2004). Techniques used on the credit derivatives market: credit default swaps, Banque de France, Financial Stability Review, 4, June

Packer, F. (2003). Mind the gap: domestic versus foreign currency sovereign ratings. BIS Quarterly Review. Retrieved on: 14, September, 2011, on http://www.bis.org/publ/qtrpdf/r_qt0309f.pdf

Schwarz, K. (2010). Mind the Gap: Disentangling Credit and Liquidity in Risk Spreads. Retrieved on 27, October, 2011 on: http://finance.wharton.upenn.edu/~kschwarz/Spreads.pdf

Studenmund, A.H., (2006). Using Econometrics 5th Edition: A Practical Guide. Boston: Pearson Education, inc.

U.S. Department of the Treasury (2011). Resource Center: Daily Treasury Yield Curve Rates. Retrieved on 27, October, 2011 on http://www.treasury.gov/resource-center/data-chart-center/interest-rates/Pages/TextView.aspx?data=yield

Xinhua News Agency (2011). Fitch cuts Italy and Spain by 2 notches. Retrieved on: 10, November, 2011, on http://news.xinhuanet.com/english2010/business/2011-10/08/c_131178204.htm

Zhu, H. (2004). An empirical comparison of credit spreads between the bond market and the credit default swap market. Bank for International Settlements.

Appendix
Section I: Rating symbols & definitions Rating Moody’s | Explanation of the ratings: | Aaa | Issuers or issues rated AAA.n demonstrate the strongest creditworthiness relative to other domestic issuers | Aa1 | Issuers or issues rated AA.n demonstrate very strong creditworthiness relative to other domestic issuers | Aa2 | | Aa3 | | A1 | Issuers or issues rated AA.n present above-average creditworthiness relative to other domestic issuers | A2 | | A3 | | Baa1 | Issuers or issues rated AA.n present average creditworthiness relative to other domestic issuers | Baa2 | | Baa3 | | Ba1 | Issuers or issues rated AA.n present below-average creditworthiness relative to other domestic issuers | Ba2 | | Ba3 | | B1 | Issuers or issues rated AA.n present weak creditworthiness relative to other domestic issuers | B2 | | B3 | | Caa1 | Issuers or issues rated AA.n present very weak creditworthiness relative to other domestic issuers | Caa2 | | Caa3 | | Ca | Issuers or issues rated AA.n present extremely weak creditworthiness relative to other domestic issuers | C | Issuers or issues rated AA.n present the weakest creditworthiness relative to other domestic issuers |
(Source: Moodys.com)

Rating Standard&Poor’s: | Explanation of the ratings: | AAA | Extremely strong capacity to meet financial commitments. Highest Rating. | AA+ | Very strong capacity to meet financial commitments. | AA | | AA- | | A+ | Strong capacity to meet financial commitments, but somewhat susceptible to adverse economic conditions and changes in circumstances. | A | | A- | | BBB+ | Adequate capacity to meet financial commitments, but more subject to adverse economic conditions. | BBB | | BBB- | Considered lowest investment grade by market participants. | BB+ | Considered highest speculative grade by market participants. | BB | Less vulnerable in the near-term but faces major ongoing uncertainties to adverse business, financial and economic conditions. | BB- | | B+ | More vulnerable to adverse business, financial and economic conditions but currently has the capacity to meet financial commitments. | B | | B- | | CCC+ | Currently vulnerable and dependent on favorable business, financial and economic conditions to meet financial commitments. | CCC | | CCC- | | CC | Currently highly vulnerable. | C | Currently highly vulnerable obligations and other defined circumstances. | D | Payment default on financial commitments. |
(Source: Standard&Poors.com)

Rating Fitch: | Explanation of the ratings: | AAA | Highest credit quality. ‘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 foreseeableevents. | AA+ | Very high credit quality.‘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. | AA | | AA- | | A+ | High credit quality. ‘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. | A | | A- | | BBB+ | Good credit quality.‘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. | BBB | | BBB- | | BB+ | Speculative.‘BB’ ratings indicate an elevated vulnerability to default risk, particularly in the event of adverse changes in business oreconomic conditions over time. | BB | | BB- | | B+ | Highly speculative.‘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. | B | | B- | | CCC+ | Substantial credit risk.Default is a real possibility. | CCC | | CCC- | | CC | Very high levels of credit risk.Default of some kind appears probable. | C | Exceptionally high levels of credit risk.Default appears imminent or inevitable. | D | ‘D’ ratings indicate an issuer that in Fitch Ratings’ opinion has entered into bankruptcy filings, administration, receivership, liquidation or other formal winding-up procedure, or which has otherwise ceased business. |
(Source: FitchRatings.com)

Section II: Tables
Table 2.1:

(Source: S&P.com)

Table 2.2: Linear rating transformation RatingMoody’s | NumberMoody’s | Default Probability | Rating_S&P | NumberS&P | DefaultProbability | Aaa | 1 | 0.02 | AAA | 1 | 0.02 | Aa1 | 2 | 0.032 | AA+ | 2 | 0.033 | Aa2 | 3 | 0.04 | AA | 3 | 0.042 | Aa3 | 4 | 0.056 | AA- | 4 | 0.059 | A1 | 5 | 0.08 | A+ | 5 | 0.084 | A2 | 6 | 0.114 | A | 6 | 0.119 | A3 | 7 | 0.144 | A- | 7 | 0.154 | Baa1 | 8 | 0.182 | BBB+ | 8 | 0.2 | Baa2 | 9 | 0.23 | BBB | 9 | 0.259 | Baa3 | 10 | 0.307 | BBB- | 10 | 0.367 | Ba1 | 11 | 0.408 | BB+ | 11 | 0.518 | Ba2 | 12 | 0.544 | BB | 12 | 0.733 | Ba3 | 13 | 0.848 | BB- | 13 | 1.215 | B1 | 14 | 1.323 | B+ | 14 | 2.014 | B2 | 15 | 2.064 | B | 15 | 3.338 | B3 | 16 | 4.168 | B- | 16 | 5.384 | Caa1 | 17 | 8.418 | CCC+ | 17 | 8.682 | Caa2 | 18 | 17 | CCC | 18 | 14 | Caa3 | 19 | 17.946 | CCC- | 19 | 14.936 | Ca | 20 | 20 | CC | 20 | 17 | C | 21 | 20 | C | 21 | 18.25 | | | | D | 22 | 20 |

Table 2.3: Defining the countries residing within geographical areas Euro area (EU17) | European Union countries (EU) | European countries | Austria | Austria | Albania | Belgium | Belgium | Andorra | Cyprus | Bulgaria | Armenia | Estonia | Cyprus | Austria | Finland | Czech Republic | Azerbaijan | France | Denmark | Belarus | Germany | Estonia | Belgium | Greece | Finland | Bosnia and Herzegovina | Ireland | France | Bulgaria | Italy | Germany | Croatia | Luxembourg | Greece | Cyprus | Malta | Hungary | Czech Republic | Netherlands | Ireland | Denmark | Portugal | Italy | Estonia | Slovakia | Latvia | Finland | Slovenia | Lithuania | France | Spain | Luxembourg | Georgia | | Malta | Germany | | Netherlands | Greece | | Poland | Hungary | | Portugal | Iceland | | Romania | Ireland | | Slovakia | Italy | | Slovenia | Latvia | | Spain | Liechtenstein | | Sweden | Lithuania | | United Kingdom | Luxembourg | | | Macedonia | | | Malta | | | Moldova | | | Montenegro | | | Netherlands | | | Norway | | | Poland | | | Portugal | | | Romania | | | Russia | | | San Marino | | | Serbia | | | Slovakia | | | Slovenia | | | Spain | | | Sweden | | | Switzerland | | | Turkey | | | Ukraine | | | United Kingdom |
Section III: Figures

Figure 3.1: flows of a bond with a CDS (without default)

Principal
CDS premium
CDS
Interest payment
Sovereign bond
Third Party
Investor

Sovereign Country Principal
CDS premium
CDS
Interest payment
Sovereign bond
Third Party
Investor

Sovereign Country

Figure 3: flows of a bond with a CDS (after the default has occurred)
Principal
Third Party
Investor

Sovereign Country Bond
(market value)
Principal
Third Party
Investor

Sovereign Country Bond
(market value)

Section IV: Extended theory

Bonds
(Brealey et al., 2009) Governments borrow money by selling (sovereign) bonds to investors. The governments are in this case the issuers of the bonds and the holders of the bonds are the investors.

Initially the cash flows are as follows:
Investor
Investor
Sovereign Country
Sovereign Country Face value Sovereign bond

Each bond has a face value (also called the principal or par value) and a coupon. The face value is the payment the investor gets at the maturity of the bond. The coupon is the interest payments paid to the bondholder. For example the face value of the bond is $1000, the coupon rate is 5 percent and the bond matures in 2 years. This means that for two consecutive years the bondholder receives an interest payment of 5 percent of the face value, namely $50 and at the end of the two years the investor also receives the face value of $1000.

The price of the bond depends, besides the risk, on the interest rate that investors could receive by investing elsewhere. For example, they could also decide to place their money in the bank and receive interest instead of buying a government bond. If the bank offered an interest of two percent a year and this was a risk-free investment, the opportunity cost (foregone earnings) when investing in government bonds is 2 percent a year. Thus investors would require, when holding government bonds, at least 2 percent return a year on their investment.

In case of our earlier example the price of a two year bond, with a coupon of 5 percent and a face value of $1000, can be calculated as follows:
$50/ (1.02) + $50/ (1.022) + $1000/ (1.022) = $1058.25

Thus the price of $1058.25 is more than the face value. This is a general result. When the market interest rate exceeds the coupon rate, bonds sell for less than face value. When the market interest rate is below the coupon rate, bonds sell for more than face value.
If a bond sells at $1085.25, it’s coupon rate is 5% with a maturity of 2 years we can also calculate what the interest rate should be if the bond is correctly priced. This is also called the yield to maturity (ytm).
In this case we need to calculate what the yield to maturity (ytm) is:
$1058.25 = $50/(1+ytm) + $50/ (1+ytm2) + $1000/ (1+ytm2)
Solve for ‘ytm' gives 0.02, which is 2 percent.

(Brealey et al., 2009) The yield to maturity is defined as the discount rate that makes the present value of the bond’s payments equal to its price. In other words, the yield to maturity measures the average rate of return to an investor who purchases the bond and holds it until maturity, accounting for coupon income as well as the difference between purchase price and face value.

(Brealey et al., 2009) When a bond issuer, for example a government, may default on its obligations is called default risk (or credit risk). (Moody’s Investor Service, 2011) Hereby meant by obligations are: paying in full and on time. Governments need to compensate for this default risk by promising a higher rate of interest on their bonds.

Section V: Graphs
Graph 5.1
(Source: Eurostat)

Graph 5.2
(Source: U.S. Department of the Treasury)

Graph 5.3
Graph 5.4
Graph 5.5
Graph 5.6
Graph 5.7
Graph 5.8
The sample used to estimate the average and median difference of CDS rates as shown in figures 5.3 up until 5.8, are countries which had available CDS rates and a rating upgrade or downgrade between the periods: 1990-2011 The number of observations of the sample and the countries used are shown below: Number of observations: | | 68 | local currency upgrade - Non-European countries | 26 | local currency downgrade - Non-European countries | 95 | foreign currency upgrade - Non-European countries | 20 | foreign currency downgrade - Non-European countries | 48 | local currency upgrade – European countries | 105 | local currency downgrade – European countries | 56 | foreign currency upgrade – European countries | 99 | foreign currency downgrade – European countries | 28 | local currency upgrade – EU countries | 81 | local currency downgrade – EU countries | 38 | foreign currency upgrade – EU countries | 79 | foreign currency downgrade – EU countries |

Countries used for Graph 5.3 and Graph 5.4 LC upgrade - Europe | LC downgrade - Europe | FC upgrade - Europe | FC downgrade - Europe | Belgium | Bulgaria | Belgium | Bulgaria | Bulgaria | Croatia | Bulgaria | Cyprus | Cyprus | Cyprus | Cyprus | Estonia | Czech Republic | Czech Republic | Czech Republic | Greece | Estonia | Estonia | Estonia | Hungary | Latvia | Greece | Iceland | Iceland | Lithuania | Hungary | Latvia | Ireland | Malta | Iceland | Lithuania | Italy | Poland | Ireland | Malta | Latvia | Romania | Italy | Poland | Lithuania | Russia | Latvia | Romania | Portugal | Serbia | Lithuania | Russia | Romania | Slovakia | Malta | Serbia | Russia | Slovenia | Poland | Slovakia | Spain | Spain | Portugal | Slovenia | Ukraine | Turkey | Romania | Spain | | Ukraine | Russia | Turkey | | | Spain | Ukraine | | | Ukraine | | |

Countries used for Graph 5.5 and Graph 5.6 LC upgrade - EU | LC downgrade - EU | FC upgrade – EU | FC downgrade -EU | Belgium | Bulgaria | Belgium | Bulgaria | Bulgaria | Cyprus | Bulgaria | Cyprus | Cyprus | Czech Republic | Cyprus | Estonia | Czech Republic | Estonia | Czech Republic | Greece | Estonia | Greece | Estonia | Hungary | Latvia | Hungary | Latvia | Ireland | Lithuania | Ireland | Lithuania | Italy | Malta | Italy | Malta | Latvia | Poland | Latvia | Poland | Lithuania | Romania | Lithuania | Romania | Portugal | Slovakia | Malta | Slovakia | Romania | Slovenia | Poland | Slovenia | Spain | Spain | Portugal | Spain | | | Romania | | | | Spain | | |

Countries used for Graph 5.7 and Graph 5.8 LC upgrades - Non-European countries | LC downgrades - Non-European countries | FC upgrade - Non-European countries | FC downgrade - Non-European countries | Bolivia | Bahrain | Bahrain | Bahrain | Brazil | Colombia | Bolivia | Ecuador | Chile | Ecuador | Brazil | El Salvador | China | Egypt | Chile | Japan | Colombia | El Salvador | China | Kazakhstan | Ecuador | Japan | Colombia | Lebanon | Hong Kong | Kazakhstan | Costa Rica | Thailand | Indonesia | Lebanon | Ecuador | Tunisia | Israel | Malaysia | Hong Kong | Venezuela | Japan | South Africa | Indonesia | Vietnam | Kazakhstan | Thailand | Israel | | Korea | Tunisia | Japan | | Lebanon | Venezuela | Kazakhstan | | Panama | Vietnam | Korea | | Peru | | Lebanon | | Philippines | | Morocco | | Saudi Arabia | | Panama | | South Africa | | Peru | | Thailand | | Philippines | | Uruguay | | Saudi Arabia | | Venezuela | | South Africa | | Vietnam | | Thailand | | | | Uruguay | | | | Venezuela | | | | Vietnam | |

Graph 5.9

Graph 5.10

The sample used to estimate the average and median difference of bond spreads as shown in figures 5.9 and 5.10, are EU countries which had available bond spreads and a rating upgrade or downgrade between the periods: 1990-2011 The number of observations of the sample and the countries used are shown below:

Number of observations: | | | | | 120 | foreign currency upgrade EU countries | 91 | foreign currency downgrade EU countries | 61 | local currency upgrade EU countries | | 103 | local currency downgrade EU countries |

Countries used for Graph 5.9 and Graph 5.10 FC_upgrade EU | FC_downgrade EU | LC_upgrade EU | LC_downgrade EU | Lithuania | Lithuania | Slovakia | Lithuania | Slovakia | Latvia | Bulgaria | Latvia | Latvia | Hungary | Lithuania | Hungary | Czech Republic | Greece | Spain | Greece | Hungary | Italy | Hungary | Ireland | Greece | Ireland | Latvia | Portugal | Romania | Portugal | Romania | Poland | Bulgaria | Finland | Italy | Cyprus | Slovenia | Sweden | Poland | Spain | Poland | Cyprus | Malta | Czech Republic | Spain | Spain | Czech Republic | Bulgaria | Portugal | Bulgaria | Greece | Malta | Italy | Romania | Sweden | Italy | Ireland | Belgium | Cyprus | Belgium | Finland | | Finland | Finland | Malta | | Portugal | Romania | Denmark | | Belgium | | Sweden | | Ireland | | Cyprus | | Slovenia | | Belgium | | | |

Graph 5.11

Section VI: Data
Table 6.1: Dataset variables & description

Variable: | Description | country_id | Specific number assigned to each country | country | Country name | statadate | The date of the credit rating announcement | rating | Sovereign credit rating | number_diff | The linear transformation of the change in a sovereign credit rating. Whereby a positive number difference indicates a downgrade and a negative number difference an upgrade | dp_diff | The transformation of a change in a sovereign credit into a change in default probability. Whereby a positive default probability difference indicates a downgrade and a negative default probability difference an upgrade. | lcr_dummy | This dummy indicates if the rating change is for a local currency (LC) sovereign bond. Whereby 1 indicates that is a LC bond and 0 otherwise | fcr_dummy | This dummy indicates if the rating change is for a foreign currency (FC) sovereign bond. Whereby 1 indicates that it is a FC bond and 0 if otherwise. | moodys_dummy | This dummy indicates if a rating change is from Moody’s. 1 indicates a rating change from Moody’s and 0 if otherwise. | fitch_dummy | This dummy indicates if a rating change is from Fitch. 1 indicates a rating change from Fitch and 0 if otherwise. | sp_dummy | This dummy indicates if a rating change is from S&P. 1 indicates a rating change from S&P and 0 if otherwise. | eu_dummy | This dummy indicates if a country belongs to one of the 27 member states of the European Union. 1 indicates an EU country and 0 indicates that it is a non-European country | europe_dummy | This dummy indicates if a country belongs to Europe. 1 indicates an European country and 0 indicates that it is a non-European country | eu17_dummy | A dummy which is equal to 1 when a country belongs to the euro area and is zero when a country does not belong to the euro Area | cds(t-1) | The credit default swap spread of a country one day before the rating announcement | cds(t+1) | The credit default swap spread one day after the rating announcement | cds_diff((t+1)-(t-1)) | The difference between the credit default swap spread one day after the rating announcement minus the credit swap spread one day before the announcement | dailybondy_(t-1) | The daily bond yield one day before a rating announcement | dailybondy_(t+1) | The daily bond yield one day after a rating announcement | dailybondy((t+1)-(t-1)) | The difference between the bond yield one day after a rating announcement minus the bond yield one day before the announcement | banklendingconditions | Index based on a quarterly survey for banks of EU-countries on the credit lending tightness/ease. A positive net percentage indicates that a larger proportion of banks have tightened credit standards, whereas a negative net percentage indicates that a larger proportion of banks have eased credit standards. | germanbondy(t-1) | German government bond yield one day before a rating announcement | germanbondy(t+1) | German government bond yield one day after a rating announcement. | bondspread(t-1) | The bond yield for a given country one day before the rating announcement minus the German bond yield one day before the rating announcement. Also known as the bond spread | bondspread(t+1) | The bond yield for a given country one day after the rating announcement minus the German bond yield one day after the rating announcement. Also known as the bond spread | bondspread_diff((t+1)-(t-1)) | The difference between the bond spread one day after a rating announcement and one day before a rating announcement | growth_dummy | Dummy for economic growth. If 1, this means an economic growth, measured as an increase in GDP per capita in constant prices. If 0, this means an economic downfall, measured as a decrease in GDP per capita in constant prices | changegood | Interaction term of growth_dummy and number_diff. If smaller than zero, there is economic growth a quarter before a rating change. If 0, there is no economic growth a quarter before a rating change. | changebad | Interaction term of growth_dummy and number_diff. If bigger than zero, there is economic downfall a quarter before a rating change. If 0, there is no economic downfall a quarter before a rating change. | gdpchange | The quarterly change in GDP in constant prices in the quarter a rating change takes place | gdpchangelag | The quarterly change in GDP in constant prices, a quarter preceding a rating change |

Table 6.2 Dataset variables with some main characteristics Variable | Obs | Mean | Std. Dev. | Min | Max | country_id | 1882 | 77.96759 | 41.52065 | 3 | 147 | country | 0 | | | | | statadate | 1882 | 16188.49 | 1708.994 | 11000 | 18826 | rating | 0 | | | | | number_diff | 1882 | .021254 | 1.548292 | -11 | 14 | dp_diff | 1882 | .1265983 | 2.427411 | -13.662 | 13.778 | lcr_dummy | 1882 | .4399575 | .4965137 | 0 | 1 | fcr_dummy | 1882 | .5600425 | .4965137 | 0 | 1 | moodys_dummy | 1882 | .2848034 | .4514407 | 0 | 1 | fitch_dummy | 1882 | .3198725 | .4665509 | 0 | 1 | sp_dummy | 1882 | .3953241 | .4890501 | 0 | 1 | eu_dummy | 1882 | .2624867 | .4401027 | 0 | 1 | europe_dummy | 1882 | .3900106 | .487882 | 0 | 1 | eu17_dummy | 1882 | .1451356 | .3523311 | 0 | 1 | cds(t-1) | 556 | 327.0245 | 569.3985 | 0 | 6065.598 | cds(t+1) | 557 | 332.5538 | 573.5635 | 0 | 5899.398 | cds_difft1t1 | 556 | 6.081307 | 52.4318 | -257.32 | 368.6 | dailybondy(t-1) | 494 | 4.921984 | 3.740649 | 0 | 20 | dailybondy(t+1) | 494 | 4.886194 | 3.776922 | 0 | 20 | dailybondy(t+2) | 494 | 4.943583 | 3.801703 | 0 | 20 | dailybondy(t+1)-(t-1) | 369 | .0210298 | .2019727 | -1.26 | 1.16 | banklendingstrictness | 1139 | 2.811861 | .2576548 | 2.27 | 3.18 | germanbondy(t-1) | 1882 | 4.217821 | 1.067594 | 0 | 9.19 | germanbondy(t+1) | 1882 | 4.212752 | 1.079134 | 0 | 9.11 | germanbondy(t+2) | 1882 | 4.204883 | 1.095496 | 0 | 9.11 | bondspread(t-1) | 494 | .744838 | 4.002216 | -9.11 | 13.83 | bondspread(t+1) | 494 | .7247571 | 4.014578 | -9.01 | 13.98 | bondspread(t+1)-(t-1) | 494 | -.020081 | .4750894 | -5.04 | 1.16 | germanbondy(t+2) | 1882 | 4.204883 | 1.095496 | 0 | 9.11 | Bondspread(t+2) | 494 | .7819838 | 4.084765 | -9.01 | 14.55 | Bondspread((t+2)-(t-1)) | 494 | .0371458 | .4747667 | -4.68 | 4.42 | growth_dummy | 442 | .7375566 | .4404609 | 0 | 1 | changegood | 442 | -.2918552 | 1.144196 | -3 | 5 | changebad | 442 | .3371041 | .7388959 | -2 | 4 |

Section VII: Testing classical assumptions
In this section it is tested if the classical assumptions are met for the following models:
(3) ∆CDSi, t = β0 +B1 * ∆Ratingi,t + ei,t
(4) ∆Si, t = β0 +B1 * ∆Ratingi,t + ei,t

Hereby for equation (3), the following areas are considered, namely: Europe, EU-countries and non-European countries. For equation (4) only the EU is considered due to lack of data for other regions. For equation (3) a window of two days surrounding the change in a credit rating is taken (t-1, t+1) and for equation (4) a three-day window is taken (t-1, t+2).
The classical assumptions are shown in the table below and will be discussed in turn. (Studenmund, 2006) The importance of the first four classical assumptions is that if these are met, the OLS method gives unbiased and consistent estimators of parameters β. If the last three classical assumptions are also met, this means that the OLS gives unbiased and consistent estimators of the variance of parameters β. Classical assumption: | Description: | 1 | The regression model is linear in the parameters β, is correctly specified, and has an additive error term | 2 | There is no perfect multicollinearity between the explanatory variables | 3 | The error term is uncorrelated with all explanatory variables | 4 | The population-average of error term has an expected value of zero | 5 | The error term has a constant variance | 6 | There is no serial correlation in the error term | 7 | The error term has a normal distribution |
(Source: Studenmund, 2006)
1. The regression model is linear in the parameters β, is correctly specified, and has an additive error term
As well as equations: (3) and (4) are assumed linear in the parameters β and both equations have an additive stochastic error term. To test if both equations are correctly specified, the Ramsey Regression Specification Error Test (RESET) is used. (Studenmund, 2006) This is a general test that determines the likelihood of an omitted variable or some other specification error. This test only signals when a major specification error might exist. It doesn’t say exactly what the error is.
Hereby the hypotheses which are tested for are:
H0 = There are no omitted variables
H1 = There are omitted variables

The Ramsey RESET for equation (3): ∆CDSi, t = β0 +B1 * ∆Ratingi,t + ei,t, for Non-European/European and EU countries is shown below.
Non-European countries:

European countries:

EU countries:

The Ramsey RESET shows with a P-value of 0.13, 0.85 and 0.58, for: Non-European, European and EU countries respectively, that there is no major specification error regarding equation (3).

The Ramsey RESET for equation (4): ∆Si, t = β0 +B1 * ∆Ratingi,t + ei,t , For EU countries is shown below.

The Ramsey RESET shows with a P-value of 0.0370, which is smaller than the significance level of 0.05, that there might be a major specification error in equation (4), however it does not specify the details of that error.

2. There is no perfect multicollinearity between the explanatory variables
Because there is only one explanatory variable, namely the change in a credit rating, this cannot be perfectly correlated with other explanatory variables.

3. The error term is uncorrelated with all explanatory variables
It is assumed that the observed values of the explanatory variables are determined independently of the values of the error term. Explanatory variables are considered to be determined outside the context of the regression equation in question.
This is tested for equation (3) for three areas and for equation (4) for the EU:

(3) ∆CDS Europe, t = β0 +B1 * ∆Ratingi,t + ei

(3) ∆CDSEU, t = β0 +B1 * ∆Ratingi,t + ei

(3) ∆CDSnon-European, t = β0 +B1 * ∆Ratingi,t + ei

(4) ∆SEU, t = β0 +B1 * ∆Ratingi,t + ei

For both equation (4) and (3), it shows in every case that the residual and the explanatory variable: change in a credit rating, are uncorrelated.

4. The population-average of error term has an expected value of zero
This assumption cannot be tested, since the population-average of the error term is unobservable. However we can test if the residuals for the models (3) and (4) have an expected value of zero.

The tables above show that the mean of the residuals is equal to zero for equation (3) and (4).

5. The error term has a constant variance
In order to check if the error term has a constant variance, equation (3) and (4) are tested on heteroskedasticity. If there is heteroskedasticity, this means that the error term has no constant variance. Put in other words: it is assumed that the observations of the error term are not drawn continually from the same identical distribution. It can be tested using the Breusch-Pagan test, assuming that the models are linear in the parameters β, which is already stated in an earlier classical assumption.
The hypotheses are the following:
H0 = The error term has a constant variance (homoskedasticity)
H1 = The error term does not have a constant variance (heteroskedasticity). Put differently, the variance in the error term increases (or decreases) as the expected value of Y increases.

Below are the Breusch-Pagan tests for heteroskedasticity for equation (3) and (4):

(3) ∆CDSi, Europe = β0 +B1 * ∆Ratingi,t + ei

(3) ∆CDSi,EU = β0 +B1 * ∆Ratingi,t + ei

(3) ∆CDSi,non-European = β0 +B1 * ∆Ratingi,t + ei

(4) ∆Si, EU = β0 +B1 * ∆Ratingi,t + ei

The results show that for equation (3), for each area: Europe, EU and non-European countries, that there is heteroskedasticity, with a 95% confidence interval, since the F-test with the accompanied P-value is lower than 0.05. Therefore the null hypothesis is rejected, there is no constant variance. Furthermore for equation (4) there is no heteroskedasticity, with a 95% confidence interval, since the F-test with the accompanied P-value is higher than 0.05. Therefore the null hypothesis cannot be rejected in the latter case.
Because we found heteroskedasticity in the first three cases for equation (3), this needs to be corrected for. This can be done by a robust estimation technique. See subsection 3.1 whereby the regressions of equation (3) are corrected for by using the robust command.

6. There is no serial correlation in the error term
If a systematic correlation exists between one observation of the error term and another, then the error term is said to be serially correlated. This could be tested using the Breusch-Godfrey test, using the following hypotheses:
H0 = there is no serial autocorrelation in the error term
H1 = there is serial autocorrelation in the error term

However, equations (3) and (4) are not tested for serial correlation in the error term, because these equations are not suitable.

7. The error term has a normal distribution
It is assumed that the error term has a normal distribution. This is assumed because the sample sizes used for the analyses are relatively large, which suggest that the error term should lean towards a normal distribution. This can be examined visually by examining histograms of the residuals below:

Residuals for Europe, for equation ∆CDSi, Europe = β0 +B1 * ∆Ratingi,t + ei:

Residuals for EU countries, for equation ∆CDSi,EU = β0 +B1 * ∆Ratingi,t + ei:

Residuals for non-European countries, for equation ∆CDSi,non-European = β0 +B1 * ∆Ratingi,t + ei:

Residuals for EU-countries, for equation ∆Si, EU = β0 +B1 * ∆Ratingi,t + ei:

The histograms for all four equations with their residuals respectively appear to be stacked around the center, with only a few outliers. The residuals lean towards a normal distribution.

--------------------------------------------
[ 1 ]. (Blanco et al., 2004) The economic effect of a CDS is similar to that of an insurance contract. The legal distinction comes from the fact that it is not necessary to hold an insured asset (e.g., the underlying bond or loan) in order to claim "compensation" under a CDS. Speculators can take long (short) positions in credit risk by selling (buying) protection without needing to trade the cash instrument.
[ 2 ]. Germany still has the highest credit rating given by Moody’s, S&P and Fitch Ratings, which means it has the lowest possible default risk. Whereas the U.S. received a downgrade in the credit rating, in August, 2011, from S&P from AAA to AA+, while Moody’s and Fitch Ratings still give the U.S the highest possible rating. This means that the U.S. Treasury bonds may no longer be a good proxy as of August, 2011, for measuring risk aversion