University of Groningen Faculty of Economics and Business
Master’s thesis International Economics and Business
The Impact of Sovereign Credit Rating Changes on Government Bond yields
Name Student: Jussi Leinonen Student ID number: s1931369
1
ABSTRACT
During recent years, the informational value of sovereign credit rating changes has been questioned and challenged by several academics. Therefore, this thesis analyzes the changes in government bond yields around the time of different rating events, with the aim of determining whether rating changes actually provide additional information about governments’ credit-worthiness. This is done by utilizing panel regressions and event studies with a dataset consisting of daily government bond yields of 19 European countries. Rating events are found to contain important information. This information content is found to be most important for rating downgrades. There is also some evidence that crisis period strengthens the reaction.
2
1 Introduction
After the downgrade of U.S sovereign debt on August 5, 2011, Nobel laureate professor Paul Krugman argued that “there is no reason to take Friday’s downgrade of America seriously” as, according to Krugman, credit rating agencies are not up to their task when it comes to rating sovereign debt. Furthermore, regarding downgrades of problem countries, such as Greece and Portugal, Krugman argues that “rating agencies were just following the markets, which had already turned on these problem debtors”. (The New York Times, 2011)
Krugman is not alone with his criticism towards rating agencies. According to de Grauwe (2010: p. 1), credit raters “systematically fail to see crises coming. And after the crisis erupts, they systematically overreact, thereby intensifying it”. The late downgrade of Dubai’s bonds can be seen as an example of this, as Dubai’s bonds were downgraded only after the repayments were postponed and after “we had all read about it in the Financial Times”.
Therefore, as the informational value of sovereign credit rating changes has been questioned by several academics, this thesis investigates whether this indeed is the case by analyzing changes in government bond yields around the time of the rating changes. Through this analysis the aim is to determine whether rating changes actually provide additional information about governments’ credit-worthiness and thereby whether they have an impact on governments’ cost of borrowing. In other words, the ultimate goal of this thesis is to determine whether policy makers actually should care about the sovereign credit ratings. This question is answered by using a dataset consisting of daily government bond yields of 19 European countries from May 15, 1993 until May 14, 2013.
3 price adjustments, price movements around the time of the credit rating announcement illustrate market’s adjustment to new information. In addition, during times of financial distress, markets become more sensitive to new information thereby increasing volatility, thus making it worthwhile to investigate whether bond yield reactions to rating changes differ between times of stability and distress. This thesis makes use of methods by Afonso, Furceri and Gomes (2012) and enhances their model by exploring additional hypotheses. Furthermore, extending the dataset into May, 2013 enables a proper estimation of whether crisis influences to the impacts of credit rating events.
2 Literature Review
Sovereign credit ratings proxy the likelihood that a borrower, in this case a country, will default on its obligations. Ratings are important to national governments since many investors prefer rated securities over unrated ones, even if the unrated one would contain similar credit risk. Furthermore, sovereign ratings are also important because changes of these ratings often have an impact on credit ratings of e.g. local municipalities and private firms since they are unlikely to have higher rating than their home country. (Cantor & Packer, 1996)
There is a negative relationship between bond returns and credit ratings. Yields tend to rise as ratings worsen, reflecting the increase in the default risk premium. (Cantor, & Packer, 1996) Thus, there is an incentive for private and public sector to manage their finances in a way that do not worsen the ratings since the alternative would result in an increase in the cost of borrowing. Therefore, changes in sovereign ratings could help to impose market-based financial discipline on the public and private sector. (Larrain, Reisen, & von Maltzan, 1997)
4 market government bond yield spreads. With a dataset consisting of 16 emerging market countries `between 1990-2000, they find that rating changes also influence stock returns and that changes have spill-over effects especially on the neighboring countries.
Using emerging market data from 2001-2008, Ismailescu and Kazemi (2010) investigate rating announcements on sovereign credit default swap (CDS) spreads that reflect markets’ opinion on sovereign credit risk. They find that only positive announcements have an impact, while negative rating announcements have no impact on sovereign CDS markets.
One of the few papers investigating credit rating announcements in developed countries is the paper by Afonso et al. (2012) which studies the reaction of government bond yields and CDS spreads on rating announcements. With a dataset consisting of 24 EU countries from 1995 to 2010, they find that particularly negative announcements have an impact on bond yields. They also find that the reaction of CDS spreads to negative rating events has increased since the financial crisis escalated after the bankruptcy of Lehman Brothers.
2.1 Credit rating agencies
Credit rating agencies provide information about credit-worthiness of different institutions, securities and sovereigns. They do this in two ways. First, they offer independent evaluations of issuers’ abilities to meet their financial commitments in the form of credit rating. This credit rating is a letter combination which reflects the issuers’ financial situation. Second, credit rating agencies offer “monitoring services” which encourage issuers to take actions in order to avoid future rating downgrades. Thus, the overall purpose of credit rating agencies is to reduce information costs, maximize the number of potential borrowers and support market liquidity. (de Haan & Amtenbrink, 2011; IMF, 2010)
5 The three biggest credit rating agencies are the U.S. based Standard & Poor’s (S&P), Moody’s and Fitch. This thesis uses data from these three rating agencies simply because of data availability and their public recognition. Table 1 describes each rating symbol reflecting the issuers’ financial position from each of the three rating agencies included in the dataset.
Rating agency Number equivalent
Interpretation S&P Moody's Fitch
Investment grade
Highest quality AAA Aaa AAA 21
High quality AA+ Aa1 AA+ 20
AA Aa2 AA 19
AA- Aa3 AA- 18
Strong payment capacity A+ A1 A+ 17
A A2 A 16
A- A3 A- 15
Adequate payment capacity BBB+ Baa1 BBB+ 14
BBB Baa2 BBB 13
BBB- Baa3 BBB- 12
Speculative grade
Likely to fulfill obligations, ongoing uncertainty
BB+ Ba1 BB+ 11
BB Ba2 BB 10
BB- Ba3 BB- 9
High credit risk B+ B1 B+ 8
B B2 B 7
B- B3 B- 6
Very high credit risk CCC+ Caa1 CCC+ 5
CCC Caa2 CCC 4
CCC- Caa3 CCC- 3
Near default with possibility of recovery
CC Ca CC 2
C C C 1
Default D RD
D Source: Fitch (2013), Moody's (2013) and Standard
& Poor's (2013)
Table 1, Credit ratings and number equivalents
6 AAA from Standard & Poor’s illustrates issuers “extremely strong capacity to meet financial commitments” (Standard & Poor’s, 2013). Similar definitions are used by Moody’s and Fitch as well. (IMF 2010) The sub-categories investment grade and speculative grade are used to broadly divide securities between safe (investment grade) and risky (speculative grade) securities. The threshold between investment grade and speculative grade is important as many investors are not willing, or are even forbidden, to purchase assets below investment grade. Jaramillo and Terada (2011) investigate credit ratings and corresponding sovereign bond yield spreads and find that obtaining investment grade reduces the spreads by 36 %. Upgrades within the investment grade are found to reduce spreads only between 5 – 10 %. The rating changes themselves can take form of one or several notches at each time. Most common rating change is one notch, say from AAA to AA+ (or from 21 to 20), but especially in recent years, changes of several notches have occurred. For example on July 13, 2011 Fitch Ratings downgraded Greek 10-year government bond 4 notches from B+ to CCC (from 8 to 4).
Credit rating agencies also give indications about future rating changes with which they encourage issuers to take actions in order to avoid future downgrades. For example negative “review” or “watch” procedures indicate that rating downgrade is likely within next 90 days. Negative “outlook” in turn indicates potential for a downgrade within the next one or two years. Because of limited number of observations, watchlistings and outlook announcements are grouped together and labeled as outlooks in this thesis. Furthermore, since better credit rating indicates higher credit-worthiness and thus lower risk, sovereign bond yields of countries that have a high rating are lower than countries with low rating. Thus, sovereign yields tend to rise as ratings decline. (Cantor, & Packer, 1996)
2.2 Hypotheses
7 yields. Thus, credit rating upgrades and positive outlook announcements are expected to lower the government bond yields.
H1: A government credit rating upgrade or a positive outlook announcement decreases
government bond yield.
On the other hand, a credit rating downgrade or a negative outlook announcement signal that issuer’s credit-worthiness has deteriorated, thus implying a higher risk of default. Since higher risk is associated with higher bond yields, this credit rating downgrade and negative outlook announcement is expected to increase government bond yields.
H2: Credit rating downgrade or a negative outlook announcement increases government
bond yield.
Actual rating changes and outlook announcements contain different information. Whereas a rating change reflects country’s improved or decreased credit-worthiness, an outlook announcement is only an implication of future possible rating change, leaving some room governments to influence the outcome. Thus, the informational value provided by actual rating change announcement is bigger and because of this it is expected to have a bigger impact on bond yields.
H3: Credit rating changes have bigger impact on bond yields than outlook announcements.
Papers by Larrain et al. (1997) and Afonso et al. (2012) find that negative rating events have larger impact in absolute terms than positive rating events. Furthermore, Afonso et al. (2012) suggest that this stronger impact by negative announcements is further strengthened by the financial crisis as the negative market mood during crisis times causes them to react to negative news even stronger. Based on these findings, negative rating events are expected to have larger impact on bond yields in absolute terms than positive rating events and this impact is expected to strengthen during crisis times. Crisis times will be defined later with the help of Google Trends by looking the search interest of different formations of words “sovereign”, “debt” and “crisis”.
8
H5: Crisis times strengthen the impact of negative events.
Since outlook announcements indicate the possibility of future rating changes, the rating changes following these announcements are likely to be less surprising than rating changes that were not preceded by a corresponding outlook announcement. Therefore, credit rating changes that were not preceded by corresponding outlook announcement can be expected to have bigger impact on bond yields than rating changes that were preceded by outlook announcement.
H6: Rating changes that are preceded by corresponding outlook announcement have lower
impact on sovereign bond yields than rating changes that were not preceded.
Rating upgrade into the investment grade increases trading of the security which in turn increases liquidity of security and reduces its risk. On the other hand, downgrade to the speculative grade reduces trading since many investors are unwilling to hold securities with speculative grade credit rating. Jaramillo et al. (2011) find that upgrade into the investment grade reduces bond yield spread significantly more than upgrade within investment grade. Based on these, upgrades into the investment grade and downgrades to the speculative grade are expected to have larger impact on bond yields than similar upgrades or downgrades within each two grades.
H7: Upgrade (downgrade) into investment (speculative) grade is expected to have larger
impact on bond yields than corresponding rating changes within each grade.
3 Methods
9 important because rating changes can be anticipated. Thus, this thesis employs these two different and separate statistical measurement tools; a panel regression and an event study.
3.1 Panel regression
Following Kaminsky et al. (2001), the panel regression is conducted using pooled ordinary least squares method (pooled OLS). This pooled OLS model studies how sovereign yields react to rating announcements. The basic model separates the effects of rating changes and outlook announcements since, as mentioned earlier, the informational value between those two events is likely to be different. Outlooks are announcements of possible future rating changes while rating announcements change the actual ratings. Thus, the baseline pooled OLS model used is the following:
(1) ∆ = + + + +
Where, ∆ = daily change in 10-year government bond yield of country i on day t
= Constant
= One-day lag of the 10-year government bond yield of country i
= The magnitude of sovereign credit rating change (notches) for country i on day t. Following Kaminsky et al. (2002), rating upgrades (downgrades) have positive (negative) sign.
= Outlook announcement for country i on day t. Variable takes value 1 (-1) if there is a positive (negative) outlook and zero otherwise.
= Error term with zero mean and variance .
10 countries we will use cluster robust standard errors in order to allow variances to differ across countries. (Wooldridge, 2003)
However, the pooled OLS is not the only possibility when conducting panel regressions. Other options include fixed effects (FE) and random effects. Considering the choice of the method the literature is diverse. For example, while Afonso et al. (2012) use fixed effects, Kaminsky et al. (2001) use pooled OLS. Among other factors, such as the focus of interest, the choice of the method is data dependent.
In order to confirm the validity of pooled OLS method, we conduct several tests. First, we test whether random effects exist. This is done using Lagrange Multiplier (LM) test, which tests whether random effects exist i.e. whether error variance differs across countries or time. (Breusch & Pagan, 1980) The test statistic rejects existence of random effects and thus suggests that the pooled OLS model is preferred over RE (see appendix 1). Second, we test the existence of fixed effects using simple F-test. The presence of fixed effects in the model leads intercepts to differ across countries or time, meaning that, for example, country specific effects would influence the bond yield reactions to rating events. Even though intuitively this seems very plausible since especially during the sovereign debt crisis bond yields of some countries rose sharply, the F-test suggests that fixed effects do not exist. This further confirms the validity of the pooled OLS as the correct estimation method.
The variables in equation (1) study different hypotheses. First, a statistically significant with a negative sign confirms hypothesis 1 and 2, indicating a negative relationship with credit ratings and government bond yields. Second, if the coefficient is statistically significant and smaller in magnitude than , hypothesis 3 is confirmed. In that case rating changes indeed have bigger impact on bond yields than outlook announcements.
11 (2) ∆ = + + + + + +
Where, = takes value 1 if upgrade occurs, 0 otherwise
= takes value 1 if downgrade occurs, 0 otherwise
= takes value 1 if positive outlook occurs, 0 otherwise
= takes value 1 if negative outlook occurs, 0 otherwise
Hypothesis 4, which expects stronger reaction for negative events than for positive events, is confirmed in case and are smaller in magnitude compared to and respectively.
Finally, in order to answer hypotheses 5–7, model in (2) is further modified by adding an interaction term according to each hypothesis. In order to investigate whether recent sovereign debt crisis has resulted in stronger reactions to the rating announcements, a dummy variable !"!" is added into the model. It takes value 1 if downgrade happened during the crisis period and zero otherwise. Crisis period is defined with the help of Google Trends by looking the search interest of different formations of words “sovereign”, “debt” and “crisis”. This proxy for crisis period illustrates the wide media coverage which may have resulted as stronger reactions in government bond yields. Different formations show a consistent picture that the crisis became “a hot topic” in November 2009. Thus, November 1, 2009 is used as the dividing date between “normal” times and crisis period. Because some papers, such as Afonso et al. (2012), use the Lehman Brothers bankruptcy on September 15, 2008 as a dividing date between times of stability and distress, this date is used as robustness check for the validity of the crisis period.
(3.1) ∆ = + + + !"!" + ( !"!" ∗ ) +
(3.2) ∆ = + + + !"!" + ( !"!" ∗ ) +
12 is negative for rating upgrades and positive for rating downgrades and larger in magnitude than corresponding obtained in (2).
Finally, Hypotheses 6 and 7 are answered in the same way by using interaction terms.
(4.1) ∆ = + + + & + (&∗ ) +
(4.2) ∆ = + + + &+ (&∗ ) +
Where, & = takes value 1, if rating change was preceded by corresponding outlook announcement, 0 otherwise
alternatively,
(5.1) ∆ = + + + '+ ('∗ ) +
(5.2) ∆ = + + + '+ ('∗ ) +
Where, ' = takes value 1, if rating change is an upgrade (downgrade) into investment (speculative) grade
Hypothesis 6 expected rating changes that are preceded by corresponding outlook announcements to have lower impact on bond yields than rating changes were announced without such indication. In addition, hypothesis 7 expected rating changes that change country’s rating grade to have larger impact to bond yields than rating changes within each grade. Concerning hypothesis 6, if + in (4) is smaller in magnitude than estimated in equation (2), it is confirmed. On the other hand, concerning hypothesis 7 if + is larger in magnitude in (5) than estimated in equation (2), hypothesis 7 is confirmed.
3.2 Event studies
13 percentage change government bond yield) around the time of the announcement. In order to analyze this, the event study methodology is applied. This method aims to determine whether an event, here a sovereign credit rating announcement, causes an abnormal movement in the security price, here government bond yield. The event study method measures the impact of a specific event on the value of a specific security. The usefulness of this method lies in the rationale that the effects of an event will be reflected immediately in the price of a security that in some way is related to the event. This characteristic has led to wide use of event studies for example in finance to examine the stock price reactions to different firm related events. In other words, event study methodology is used to investigate whether an event creates abnormal returns. (MacKinlay, 1997; Benninga, 2008)
In this case the event study measures the impact of sovereign credit rating announcement on the yield of a government bond. It is very likely that credit rating announcement has impact on sovereign bond yields since credit rating should reflect the default risk of each country. Thus, the event study here links rating events to abnormal differences in yields between model generated and actual yields. This means investigating whether the government bond yields differ from “normal times” after each rating event. (Afonso et al., 2012)
Evaluation of the event's impact requires a measure of the abnormal yield. The abnormal yield is the actual ex post return of the security over the event window minus the normal return of the security over the event window (MacKinlay, 1997). The normal return is the expected return that would have realized had the event not taken place. Following MacKinlay (1997), the relation between abnormal and normal expected return is the following:
(4) = ∆− )(∆|&)
Where, = Abnormal return for country i at day t
∆ = Change in10-year government bond yield (adjusted) of country i on day t
)(|&) = Expected normal return
14 constant mean return model which assumes that mean return of a security remains constant over time, meaning that & is constant. The second option is to assume that there is linear relationship between the security i return and market or some reference security return. In this case & is the market return or the return of the reference security. The linear market model represents a potential improvement over the constant mean return model because it removes the portion of the return that is related to variation in the market's return and thus reducing the variance of the abnormal return. However, this is only if of the regression is high. Furthermore, the use of linear model requires the security returns to be jointly normal. (MacKinlay, 1997)
Because of higher explanatory potential, the linear market return model is chosen for the estimation of the expected normal returns. According to MacKinlay (1997), the normal expected returns can be estimated by using following ordinary least squares (OLS) regression:
(5) ∆ = + +∆,,+
Where, ∆ = Change in10-year government bond yield (adjusted) of country i over period t
∆,, = Change in market bond yield over period t
, + = Parameters of the linear model
= error term, with )( = 0), var() = / 0
Here a problem arises. Normal returns are usually calculated over an estimation period in which no events take place. However, this would eliminate too many observations as especially during the crisis, the rating announcements are be clustered around short time period. MacKinlay (1997) recognizes this problem and suggests using a market-adjusted return model which can be considered as restricted market model with equal to 0 and + equal to 1.
15 on changes in yield spreads, thus assuming the change in bond yield spread as abnormal return. However, this approach might overstate the reactions for countries with very low yield spread over Germany, since small changes in very narrow spreads are very high in percentage terms. Because of this, the focus here is on the changes in actual yields. Therefore, the abnormal return is the daily change in bond yield of country i minus the daily change in bond yield of Germany.
Furthermore, the abnormal returns must be aggregated across countries in order to draw overall inferences (MacKinlay, 1997). Thus, following Ritter (1991) who provides an example of the use of market-adjusted model and the notation by MacKinlay (1997) the average abnormal returns are calculated as follows:
(6) 1 =
2∑ (∆− (4+ +∆567,) 2
8 )
Where, 1 = Average Abnormal return at day t
∆ = Daily change in 10-year government bond yield (adjusted) of country i at day t
∆567, = Daily change in 10-year German government bond yield at day t
, + = Parameters of the linear model, with = 0 and + = 1
: = number of events for day t
After successfully calculating the abnormal returns, they need to be standardized. Standardization enables assessment of the statistical significance of the abnormal returns. Because rating events occur on a different date to different countries, we can assume that abnormal returns are cross-sectionally independent. (Kolari & Pynnönen, 2010) Thus, following Ritter (1991) the standardized abnormal returns (SAR) for each rating event are calculated as follows:
(7) "1 = 1 ∗ ;:/
16 The statistical significance of the test statistic SAR tells whether the credit rating changes result in abnormal government bond yields. In other words, it shows whether credit rating announcement provides enough new information about issuer’s credit-worthiness to have an impact on the issuer’s cost of borrowing i.e. whether the impact on bond yield is large enough to make the actual yield differ from a yield in a case in which no rating announcement occurred. All abnormal returns are assumed to be cross-sectionally independent, meaning that
a country’s abnormal bond yield is not dependent from that of another country. Furthermore, SARs are assumed to have a normal distribution.
In order to analyze the reaction over multiple days, abnormal returns need to be aggregated also over time. Following MacKinlay (1997) the cumulative abnormal returns are calculated as follows:
(8) 1 (=, = ) = ∑>>8? @1
Following Ritter (1991) These CARs are further standardized:
(9) "1 (=, = ) =AB72(>@,>?)∗; C@,C? DC@,C?
Where, >@,>? = ∑>>8>? @
SCARs are estimated over three-day period [-1, 1] in which the credit rating event is considered to occur at day 0. This event window is decidedly short in order to avoid spill-over effects from other countries as for example Afonso et al. (2012) find some evidence that rating announcement in event countries affect sovereign bond yields in other countries. Thus, keeping the event window short we make sure that no rating events in other countries occur at the same time.
17 than surrounding outlook announcements, hypothesis 3 is confirmed and if they are bigger in magnitude for negative events than for positive events hypothesis 4 is in turn confirmed. If SARs and SCARs obtained from rating events during crisis times are bigger in magnitude than those obtained prior the crisis, the hypothesis 5 is confirmed. Furthermore, hypothesis 6 is confirmed if SARs and SCARs of the rating changes that were preceded by corresponding outlook announcements are small in magnitude than rating changes that were not. Finally, if SARs and SCARs generated by upgrades (downgrades) into investment (speculative) grade are, bigger in magnitude than those generated by rating changes within each rating grade, hypothesis 7 is confirmed.
As described earlier, this thesis heavily utilizes some of the methodology and findings by Afonso et al. (2012). However, with respect to event studies this thesis contributes to the existing literature, not only by investigating additional hypotheses, but also by using a different approach in calculating abnormal returns which is argued to be subject to less bias. Furthermore, the crisis period is defined with a different approach that is based on a more subjective definition of a crisis period. Instead of using a date on which crisis may well have escalated in economic terms, this thesis divides times crisis and stability according to public’s perception of it, proxied by the search interest of crisis-related word combinations. Furthermore, while the dataset used by Afonso et al. (2012) lasts only until October 10, 2012, the dataset here is extended until May 14, 2013 which allows better estimation of the crisis effects.
4 Data
18 Table 2 presents the collected rating events. In total there are 169 credit rating changes of which 67 are upgrades and 102 rating downgrades. In more detail, the dataset consists of 127 outlook announcements, with 37 being positive and 90 being negative. 123 rating changes were preceded by corresponding outlook announcement, meaning that only 42 rating changes were announced without prior indication. Downgrades (89) are preceded more often by corresponding outlook announcements than upgrades (34). There are only 10 cases in which country’s rating grade has been changed. All those cases are negative meaning that there are no events in which country’s rating has been upgraded from speculative to investment grade. Not surprisingly, during the European sovereign debt crisis downgrades are much more common than upgrades. Since November 1, 2009 there are only 3 upgrades (all of which Czech Republic) and 78 downgrades.
Rating Change Outlook Rating Change_exp Gtrade Change Crisis
Positive 67 37 34 0 3
Negative 102 90 89 10 78
Total 169 127 123 10 81
Table 2, Rating events
19
5 Results
This chapter presents the results. First, table 2 presents descriptives concerning the dependent variable, daily percentage change in 10-year government bond yields. The numbers show that the mean changes are negative, however very close to 0, with highest daily increase being 34.15 % (Hungary) and biggest daily decrease being 59.62 % (Greece).
ΔYield Obs Mean Std.Dev. Variance Min Max
Austria 5216 -0.0222% 0.01096 0.00012 -6.8777% 12.1196% Belgium 5216 -0.0190% 0.01046 0.00011 -6.1855% 7.2559% Czech rep. 3400 -0.0350% 0.01142 0.00013 -9.2200% 10.8209% Denmark 5216 -0.0214% 0.01410 0.00020 -12.2324% 30.1489% Finland 5216 -0.0277% 0.01244 0.00015 -9.5485% 14.5656% France 5216 -0.0198% 0.01125 0.00013 -7.7127% 7.1540% Germany 5216 -0.0222% 0.01304 0.00017 -12.4091% 9.8693% Greece 3682 0.0330% 0.01820 0.00033 -59.6233% 18.3583% Ireland 5216 -0.0094% 0.01102 0.00012 -17.6554% 9.7075% Italy 5216 -0.0157% 0.01096 0.00012 -12.8163% 9.5067% Netherlands 5216 -0.0197% 0.01144 0.00013 -9.0909% 8.4933% Norway 5216 -0.0173% 0.01133 0.00013 -6.7848% 11.6632% Portugal 5171 -0.0052% 0.01248 0.00016 -23.2961% 15.4509% Spain 5216 -0.0121% 0.01090 0.00012 -14.4849% 6.6551% Sweden 5216 -0.0217% 0.01399 0.00020 -12.4166% 27.4138% Switzerland 5216 -0.0222% 0.01948 0.00038 -19.8355% 22.4719% United Kingdom 5216 -0.0191% 0.01327 0.00018 -9.1803% 18.6246% Hungary 3726 -0.0056% 0.01479 0.00022 -11.2226% 34.1519% Poland 3226 -0.0311% 0.01101 0.00012 -9.3858% 10.2711% Total 92229 -0.0168% 0.01293 0.00017 -59.6233% 34.1519%
Table 3, Descriptive statistics for daily changes in government bond yields
Closer examination of the dataset reveals that the far majority of the daily changes is under 10% and that there are only a few changes higher 20%. We inspect the dataset to see whether these outliers coincide with rating announcement and find none. Thus, reactions to rating announcements are not likely to be biased because of outliers.
20 Following the near bankruptcy of Greek government, the maximum values for Greek government bond yields are much higher compared to other countries whose maximum yields are all under 14%. From September 2011 until September 2012 Greek government bond yield stayed over 20%. As a robustness check we will exclude rating events that happened during this time in order to check whether they bias the results.
Yield Obs Mean Std.Dev. Variance Min Max
Austria 5216 4.62 1.37112 1.87997 1.45 7.88 Belgium 5216 4.83 1.39914 1.95760 1.93 8.68 Czech rep. 3400 4.35 1.18249 1.39829 1.52 7.69 Denmark 5216 4.75 1.78230 3.17659 0.97 9.32 Finland 5216 4.83 1.96184 3.84882 1.34 11.38 France 5216 4.61 1.39719 1.95213 1.67 8.41 Germany 5216 4.38 1.45940 2.12983 1.15 7.81 Greece 3682 7.69 6.79112 46.11927 3.21 48.60 Ireland 5216 5.53 1.72140 2.96322 3.04 13.90 Italy 5216 5.83 2.47670 6.13406 3.21 13.79 Netherlands 5216 4.50 1.36346 1.85902 1.49 7.85 Norway 5216 5.06 1.52857 2.33654 1.66 9.25 Portugal 5171 6.23 2.69600 7.26840 3.00 16.21 Spain 5216 5.64 2.22886 4.96780 3.03 12.48 Sweden 5216 5.00 2.33184 5.43746 1.14 12.05 Switzerland 5216 2.85 1.15167 1.32635 0.38 5.62 United Kingdom 5216 5.08 1.71839 2.95287 1.38 9.05 Hungary 3726 7.79 1.17136 1.37208 4.99 11.91 Poland 3226 6.20 1.56772 2.45775 3.07 12.31 Total 92229 5.17 2.44796 5.99250 0.38 48.60
Table 4, Descriptive statistics for government bond yields
21 Against expectations positive outlook announcements seem to increase bond yields. However, the cumulative change in yield spread is very modest and especially during the five day period [-2, 2] surrounding the announcement, the change is close to zero. In line with the expectations negative outlook announcements increase government yields. Furthermore, this increase seems to be lower compared to the increase surrounding rating downgrades. In total, the illustration in figure 1 suggests that, as expected, rating changes have bigger impact on bond yields than outlook announcements.
Figure 1, Bond yield spreads over Germany surrounding rating announcement (full sample)
5.1 Panel regression
22 (negative) outlook announcements. Changes in the bond yields are expressed in decimal points.
The results from model (1) show that the coefficient for RC is negative and statistically significant suggesting that, as expected, rating upgrades lower the sovereign bond yields while rating downgrades increase the yields. Rating change of one notch is estimated to change the bond yield by 0.18% on day 0. The coefficient for OL is also negative in line with the expectations, but not significant. Furthermore, the results suggest rating changes to have larger impact on bond yields compared to outlook announcements. In total these findings give partial support to hypotheses 1–3. They are supported related to rating changes, while outlook announcements are found not to have significant impact on bond yields.
(1) (2) (3) (constant) 0.0000 (0.56) 0.0000 (0.55) -0.0000 (-0.04) Yield (t-1) -0.0000*** (-3.09) -0.0000*** (-3.17) -0.0000 (-0.70) RC -0.0018*** (-3.16) OL -0.0004 (-0.23) UPGRADE -0.0022*** (-3.45) -0.0023*** (-4.95) DOWNGRADE 0.0033** (2.12) 0.0037 (1.65) OUTLOOK_POSITIVE -0.0000 (-0.03) 0.0001 (0.08) OUTLOOK_NEGATIVE 0.0006 (0.24) 0.0045* (1.92) UPGRADE (t-1) -0.003** (-2.37) DOWNGRADE (t-1) 0.0155*** (4.09) OUTLOOK_POSITIVE (t-1) -0.0012 (-0.68) OUTLOOK_NEGATIVE (t-1) 0.0037* (1.83) N 92229 92229 73768 R2 0.0002 0.0001 0.0018 F 9.29*** 12.20*** 22.89*** SE 0.0129 0.0129 0.0128
Dependent variable: Daily percentage change in sovereign bond yield
*** Significant at 0.01 level ** Significant at 0.05 level * Significant at 0.1 level
Table 5, The impact of credit rating changes and outlook announcements on bond yields1
1 Analysis is repeated excluding Greek government credit rating from period September 1, 2011 – September
23 Model (2) in table 5 separates the effects between positive and negative events by including several dummy variables which take value 1 if rating event in question occurs and 0 otherwise. The coefficient for credit rating upgrades is negative and statistically significant, giving further support to hypothesis 1, which expected upgrades to have reducing impact on bond yields. A credit rating upgrade is estimated to decrease bond yield by 0.22%. Similarly, hypothesis 2, expecting rating downgrades to increase government bond yields, is further confirmed. The coefficient for rating downgrades is positive and statistically significant, estimating that rating downgrades increase bond yields by 0.33% on day 0. Coefficients related to negative events are again larger compared to positive events, giving further support related to hypothesis 4. Furthermore, in line with the baseline model, the coefficients for outlook announcements have their expected signs but are not statistically significant. This finding provides further evidence that outlook announcements do not have impact on government bond yields.
Since credit rating announcement might occur any time of the day, even when the market has already closed, the market reaction to the announcement does not necessarily take place during the same day. For this reason, model (3) in table 5 adds lags of the independent variables from (2) into the model. Coefficients for the lagged dummy variable further confirm earlier findings. Important finding from (3) is that the coefficients from lagged values are estimated to be bigger in magnitude compared to originals. Here, credit rating upgrade is estimated to decrease government bond yield by 0.3%, thus increasing the impact by 0.1 %-points compared to the original. Rating downgrade in turn is estimated to increase bond yield by 1.55%, increasing the impact by 1.22 %-points. The impact of positive outlook announcement remains still insignificant here. However, the coefficient for negative outlook announcement turns significant at the 10% level, giving some evidence that outlook announcements have impact on government bond yields. In total, these findings show again stronger impact for negative rating events.
24
Next in table 6 we analyze the results related to hypotheses 5 and 6. Hypothesis 5 predicts rating changes during financial distress (such as economic crisis) to have larger impact on bond yields compared to rating changes that take place during financial stability. Hypothesis 6 in turn predicts that rating changes which are preceded by corresponding outlook announcements have smaller impact on bond yields compared to bond yields that take place without any prior indication from the credit rating agencies.
Model (4) studies the crisis effects. Variables UPGRADE*CRISIS and DOWNGRADE*CRISIS are interaction terms which are constructed by multiplying dummy variable UPGRADE which takes value 1 if upgrade occurs and 0 otherwise, with dummy variable CRISIS which takes value 1 if the date of the observation is later than November 1, 2009 and 0 otherwise. (4) (5) (6) (constant) 0.0000356 (0.58) 0.0000 (0.47) 0.0000 (0.46) Yield (t-1) -0.0000*** (-3.16) -0.0000*** (-2.95) -0.0000*** (-2.91) UPGRADE -0.0023*** (-3.42) -0.0016 (-0.92) DOWNGRADE 0.0032 (1.39) -0.0003 (-0.14) CRISIS -0.0027 (-0.79) UPGRADE*CRISIS 0.003 (0.66) DOWNGRADE*CRISIS 0.0029 (0.63) EXPECTED 0.0039** (2.11) -0.0028 (-2.76) UPGRADE*EXPECTED -0.0051 (-1.42) DOWNGRADE*EXPECTED 0.007* (2.08) N 92229 92229 92229 R2 0.0002 0.0002 0.0002 F 12.48*** 25.68*** 17.94*** SE 0.0129 0.0129 0.0129
Dependent variable: Daily percentage change in sovereign bond yield
*** Significant at 0.01 level ** Significant at 0.05 level * Significant at 0.1 level
Table 6, The impact of expected rating changes and rating changes during economic crisis on
government bond yields
25 needs to be added to coefficient of the original variable. Thus, for example for rating upgrades during the crisis the coefficient to be interpreted is + . (Brambor, Clark & Golder, 2005) Model (4) shows that the estimate for downgrades during crisis is 0.32+0.29=0.61%, which suggests that crisis to increase the reaction resulting from rating downgrade. For rating upgrades, crisis is actually found to lower the reaction. However, since only 3 rating upgrades occur during crisis, this result is trivial. Furthermore, as the interaction variable for downgrade is not significant, the support for hypothesis 5 is very weak at best.
Hypothesis 6 is studied with model (5) for rating upgrades and with model (6) for rating downgrades. Against expectations, in both cases the expected rating changes seem to have stronger impact compared to rating changes that were not expected. However, this effect is only significant for rating downgrades for which expected downgrades increase bond yield by 0.67%. As a comparison, when expectations were not accounted for, rating downgrade was estimated to increase bond yield by 0.33%. For upgrades the reaction from expected rating changes is -0.67%, illustrating a stronger reaction also for expected upgrades. In total, these findings are surprising and provide evidence against hypothesis 6.
Hypothesis 7 predicted that credit rating upgrades (downgrades) into investment (speculative) grade would experience higher bond yield reactions. However, the dataset does not allow reliable estimation of this hypothesis as there are only 10 cases in which rating grade is changed. Furthermore, all these events are negative. We run (2) with a dummy variable which takes value 1 when rating grade is changed and 0 otherwise. The result shows unexpected negative sign with reaction estimated 0.23%. However, this estimate is not statistically significant (t-stat -0.30) and thus no further inference can be made from this.
5.2 Event studies
26 Table 7 presents the results related to hypotheses 1–4. Average abnormal returns (AR) and cumulative abnormal returns and their corresponding significance level are calculated using equations 4–9. The results show that rating upgrades seem to be more anticipated than rating downgrades since the magnitude of bond yield changes is higher with upgrades prior to the announcement than with downgrades. However, in accordance with the results from pooled OLS, abnormal returns related to negative events are bigger in magnitude compared to positive events. Furthermore, all events have expected signs and abnormal returns from rating changes are bigger than abnormal returns from outlook announcements. However, these results provide only limited support for the hypotheses since almost all abnormal returns are not statistically significant. Only day 0 returns for rating downgrades are significant on 10% level.
t UPGRADE DOWNGRADE OUTLOOK POSITIVE OUTLOOK NEGATIVE
AR -1 -0.0086 (-0.99)t 0.0000 (0.01) -0.0015 (-0.73) 0.0023 (1.04) 0 -0.0006 (-0.59) 0.0063* (1.98) 0.0018 (1.05) 0.0017 (0.52) 1 -0.0033 (-1.61) 0.0047 (0.71) 0.0004 (0.31) 0.0011 (0.46) CAR [-1, 1] -0.0125 (-1.06) 0.011 (0.86) 0.0007 (0.14) 0.0051 (0.65) [-1, 0] -0.0092 (-0.95) 0.0063 (1.02) 0.0003 (0.09) 0.004 (0.73) [0, 1] -0.0039 (-1.28) 0.011 (1.12) 0.0022 (0.73) 0.0028 (0.50) N 67 102 37 90 t t-statistic in parenthesis *** Significant at 0.01 level ** Significant at 0.05 level * Significant at 0.1 level
Table 7, ARs and CARs surrounding rating changes and outlook announcements
The results related to hypothesis 5 are presented in table 8 in which the dividing date between crisis and stability is again November 1, 2009. Here, we have to focus only on rating downgrades since only 3 rating upgrades occur during the crisis. Even though majority of the downgrades occurs during crisis, there are 24 downgrades in the dataset that take place before the crisis.
27 t Upgrade_Crisis Upgrade_Pre-Crisis Downgrade_Crisis Downgrade_Pre-Crisis
AR -1 -0.1923 (-0.99)t 0.0001 (0.09) 0.0002 (0.05) -0.0005 (-0.24) 0 -0.0175 (-1.20) -0.0004 (-0.41) 0.0066* (1.62) 0.0054* (1.81) 1 -0.0572 (-1.84) -0.0009 (-0.95) 0.005 (0.58) 0.0035* (1.69) CAR [-1, 1] -0.2669 (-1.11) -0.0012 (-0.44) 0.0118 (0.71) 0.0084 (1.16) [-1, 0] -0.2097 (-1.01) -0.0003 (-0.17) 0.0068 (0.85) 0.0048 (0.94) [0, 1] -0.0746 (-1.64) -0.0013 (-0.68) 0.0116 (0.91) 0.0089* (1.76) N 3 65 78 24 t t-statistic in parenthesis *** Significant at 0.01 level ** Significant at 0.05 level * Significant at 0.1 level
Table 8, ARs and CARs during and prior crisis
Finally, hypotheses 6 and 7 are examined in table 9. In line with pooled OLS, the results here suggest that rating changes that are preceded by corresponding outlook announcements experience larger reaction than rating changes that were announced without prior indication. ARs presented in table 9 are mostly bigger in magnitude for both upgrades and downgrades, compared to ARs in table 7. Especially after the rating change announcements, expected changes experience higher abnormal returns. With respect to rating grade changes, the results show some, but only very weak support. Abnormal returns following downgrades from investment grade into speculative grade are positive and larger compared to all downgrades. Especially, the abnormal returns at day t+1 are significant and large suggesting that these changes may have occurred after the markets had already closed. However, as mentioned earlier, the dataset does not allow here for full interpretation since there are only 10 events in which the rating grade changes. Thus, support for hypothesis 7 is merely suggestive and weak.
t UPGRADE EXPRECTED DOWNGRADE EXPECTED GRADE CHANGE
AR -1 -0.0008 (-0.75)t 0.0001 (0.02) 0.0117 (0.74) 0 -0.0031* (-1.89) 0.007** (2.05) -0.0051 (-0.50) 1 -0.0031 (-1.45) 0.0114*** (3.12) 0.0553** (3.06) CAR [-1, 1] -0.007 (-1.44) 0.0184* (1.81) 0.0619 (1.41) [-1, 0] -0.0039 (-1.44) 0.0071 (1.08) 0.0066 (0.26) [0, 1] -0.0062 (-1.64) 0.0184** (2.60) 0.0502 (1.78) N 34 89 10 t t-statistic in parenthesis *** Significant at 0.01 level ** Significant at 0.05 level * Significant at 0.1 level
28
5.3 Robustness checks
In order to confirm the robustness of the results, several additional tests are performed. First, we check whether the relationship between bond yield changes and rating announcements is non-linear. Previous chapter assumes this relationship to be linear; however it is possible that the reactions of high yield bonds differ from those of low yield bonds. Bond’s high yield reflects issuer’s decreased credit worthiness, thus it is likely that high yield bonds react differently to rating announcements than low yield bonds. In order to analyze this, we divide sample according to yield levels. The results of this analysis are presented in table 10. First, model (1) repeats the results from full sample. Then, model (2) contains only observations that have government bond yield higher than 4%. Finally, models (3) and (4) divide sample further for observations that have yields higher than 6% and 8%, respectively.
The results show that the expected signs are confirmed to all yield levels. Furthermore, there is some evidence that the reaction changes along with yield levels. While for upgrades, higher yield levels seem to result in smaller reactions, for downgrades higher yield levels lead to stronger reactions. With the full sample, bond yield change from credit rating upgrade is estimated -0.22%. However, when considering only yields over 8% this downward reaction in bond yield shrinks to -0.14%. The interpretation here has to be made with some caution because of the low number of upgrade events for this sample and because the coefficient for UPGRADE is statistically insignificant.
(1) (2) (3) (4) (constant) 0.0000 (0.55) 0.0007*** (2.99) 0.0009*** (3.00) 0.0016*** (3.56) Yield (t-1) -0.0000*** (-3.17) -0.0001** (-2.66) -0.0001** (-2.44) -0.0001** (-2.45) UPGRADE -0.0022*** (-3.45) -0.0023*** (-3.01) -0.0021** (-2.64) -0.0014 (-1.57) DOWNGRADE 0.0033** (2.12) 0.0038** (2.20) 0.0043** (2.27) 0.0038 (1.02) N 92229 65340 22912 8530 Full Y > 4 Y > 6 Y > 8
Dependent variable: Daily percentage change in sovereign bond yield
*** Significant at 0.01 level ** Significant at 0.05 level * Significant at 0.1 level
29 In turn, when considering rating downgrades, the reaction becomes stronger for high yield bonds. With full sample, a credit rating downgrade increases bond yields by 0.33%. For yields higher than 6% this reaction increases to 0.43%, thus growing by 0.1%-points. For sample of bond yields higher than 8%, the magnitude of the reaction decreases again, however for this sample DOWNGRADE turns insignificant.
These results present some evidence that the relationship between rating changes and bond yield changes is non-linear and varies according to the yield level. This is not surprising finding when considering how bond yields reflect the issuer’s credit worthiness. With higher bond yields, market participants’ increased concerns strengthen the reaction resulting from rating downgrades. In turn, rating upgrades may have then less impact on yields because doubts related to credit-worthiness are not cleared by an occasional rating upgrade.
This thesis uses unique method in defining crisis period by looking at search interests of different combination of words related to sovereign debt crisis from Google Trends. Some papers, such as Afonso et al. (2012) take the Lehman Brothers bankruptcy from September 15, 2008 as the day marking the beginning of the crisis. In order to confirm the validity of the results we repeat the analysis for crisis effects by using September 15, 2008 as the dividing date between stability and crisis. Since period September 15, 2008 – November 1, 2009 does not include any credit rating upgrades; the results remain unchanged concerning UPGRADE. However, the same time period saw 12 rating downgrades, increasing the amount of crisis downgrades to 90 and decreasing the amount of pre-crisis downgrades to 12. This naturally makes it difficult to draw sound conclusions because the number of pre-crisis downgrades is low.
In table 6 the bond yield change resulting from downgrades during crisis was estimated as 0.32%+0.29% = 0.61%, thus during crisis the upward change from rating downgrade is estimated to increase by 0.28%-points2. The interaction coefficient, however, is not statistically significant, providing only suggestive evidence that crisis period increases reactions to rating changes. We run the regression again with a difference that now crisis dummy takes value 1 for dates later than September 15, 2008 and 0 otherwise. This alternative crisis-specification estimates downgrade during crisis to increase bond yields by
30 0.55-0.22 = 0.33%3, which equals to the estimate in table 5. This alternative specification for crisis period thus suggests crisis times not to strengthen the reactions for rating downgrades. However, the statistically insignificant results do not allow full interpretations. Rather, based on available data, findings suggest that crisis reactions seem to be sensitive to the definition of crisis period, taking away support for hypothesis 5. Alternatively, findings suggest that assuming the crisis in Europe to have started already at the time of Lehman Brothers bankruptcy is incorrect as that date does not lead to any differences between in reactions to credit rating changes.
Results in table 4 suggested that lagging independent variables results in stronger yield changes. For this reason, we repeat analysis by using cumulative two-day change in bond yields between t=0 and t+1 as a dependent variable. Results for this analysis are presented in table 11. Using the cumulative change as dependent variable confirms all the expected signs and further suggests that the effects from rating changes are extended beyond t=0. While one day reaction for rating upgrade was a decrease in bond yield by -0.22%, the two-day reaction estimates a statistically significant decrease of -0.51%.
(1) (2) (constant) 0.0002 (1.20) Yield (t-1) -0.0001*** (-3.70) -0.0001*** (-3.31) RC -0.0074** (-2.86) OUTLOOK -0.0017 (-0.66) UPGRADE -0.0051*** (-4.39) DOWNGRADE 0.0098 (1.43) OUTLOOK_POSITIVE -0.0007 (-0.34) OUTLOOK_NEGATIVE 0.0021 (0.63) N 92229 92229 R2 0.0009 0.0005 F 6.63*** 22.08*** SE 0.0191 0.0191
Dependent variable: percentage change in sovereign bond yield between t=0 and t+1
*** Significant at 0.01 level ** Significant at 0.05 level * Significant at 0.1 level
Table 11, Bond yield changes between t=0 and t+1
For rating downgrades the increase in the magnitude of the reaction is even larger. In table 4 the estimated one-day reaction in bond yield was an increase of 0.33%. According to table
31 11, the two-day reaction to downgrade is 0.98%, meaning an increase of 0.65%-points. However, some caution is needed here since the coefficient for DOWNGRADE is not statistically significant. Similar increased reactions, although not statistically significant, are also reported for outlook announcements.
Nickell (1981) introduces a bias related to dynamic panel data models which have some form of lagged value of the dependent variable as independent control variable. This problem arises because of correlation between the independent variable and error term and is likely to cause severe bias with panel data model with small number of time periods (small T) and a large number of cross-sectional observations (high N). However, the panel here is the opposite with daily data providing extremely high number time periods (more than 5,000). Roodman (2006) concludes that with large T, this so called Nickell bias becomes insignificant and fixed effects estimator is robust. Since we have conducted our panel regression with pooled OLS, it is possible that the results suffer from the Nickell bias. In order to confirm the robustness of the results in this sense, we run equations (1) and (2) using fixed effects. (1) (2) (constant) 0.0002** (2.44) 0.0002** (2.42) Yield (t-1) -0.0001*** (-4.87) -0.0001*** (-4.92) RC -0.0018*** (-3.09) OUTLOOK -0.0004 (-0.19) UPGRADE -0.0023*** (-3.74) DOWNGRADE 0.0032* (2.04) OUTLOOK_POSITIVE -0.0001 (-0.06) OUTLOOK_NEGATIVE 0.0005 (0.18) N 92229 92229 R2 0.0001 0.0001 F 14.04*** 20.03***
Dependent variable: Daily percentage change in sovereign bond yield
*** Significant at 0.01 level ** Significant at 0.05 level * Significant at 0.1 level
Table 12, Estimates from using fixed effects
32 FE are around 0.01%-points, suggesting that the estimates are also robust against Nickell bias.
6 Conclusions
During recent years, the informational value of sovereign credit rating changes has been questioned and challenged by several academics. Credit rating agencies have been blamed for not being up to their tasks and credit ratings have been argued to provide little new information about sovereign issuer’s credit-worthiness. Therefore, as the informational value of sovereign credit rating changes has been questioned by several academics, this thesis analyzes changes in government bond yields around the time of the rating changes, with the aim of determining whether rating changes actually provide additional information about governments’ credit-worthiness and thereby whether they have an impact on governments’ cost of borrowing.
The recent financial crisis and especially the European sovereign debt crisis have increased the default risk among the developed countries and brought the rating announcements to the center of attention. This thesis makes use of these unprecedented events in the world economy, investigating how sovereign bond yields of several European countries respond to credit rating changes by utilizing panel regressions and event studies with a dataset consisting of daily government bond yields of 19 European countries from May 15, 1993 until May 14, 2013.
33 as new information of deteriorated credit-worthiness which in turn has a positive impact on government bond yields. Again, for corresponding outlook announcements, the results lack of statistical significance.
Negative rating events are found to have stronger impact on bond yields than positive events. This finding, in line with Brooks et al. (2004) and Afonso et al. (2012), suggests that markets put more weight on negative announcements as they signal the increased likelihood of issuer’s insolvency. Information about improved credit worthiness does not cause as much stir within the markets because it may be regarded as returning to “normal” situation. The fifth hypothesis expected a period of economic crisis to further strengthen the impact of rating changes. We hypothesize, with the help of Google Trends, crisis period to start November 1, 2009. We find some evidence, in line with Afonso et al, (2012), from event studies as well as from panel regressions, that the crisis period indeed increases the reaction resulting from credit rating downgrades but not from upgrades. However, the robustness check suggests that this finding might be sensitive to definition of crisis period since the use of the Lehman Brothers bankruptcy as the starting date for the crisis period, produces reaction equal to results obtained without a crisis variable. Thus, the evidence for hypothesis 5 is not robust.
Furthermore, we hypothesized that credit rating changes which are preceded by corresponding outlook announcements, would have lower impact on bond yields than rating changes that were announced without such prior indication. Surprisingly the results suggest the opposite. Rating changes, especially rating downgrades that are preceded by corresponding outlook have stronger impact compared to those that were not. Previous literature does not consider this, thus leaving us alone with this puzzling finding. One explanation for this may be that outlook announcements create excitement or nervousness among the markets that builds up and does not get released until the actual rating change is announced. This reaction then is shown as a stronger reaction compared to situation in which credit rating is changed without any prior outlook announcement.
34 that, in line with Jaramillo et al. (2011), rating changes resulting to changes in rating grades would have stronger impact on bond yields compared to rating changes within rating grade. Furthermore, since the dataset does not include any events in which sovereign credit rating has been upgraded from speculative grade into investment grade, these findings can only be applied to credit rating downgrades.
In total, the evidence thus suggested that despite the criticism, sovereign credit rating changes do provide new information to the markets. This information content is considered to be more important rating downgrades than for rating upgrades. Outlook announcement are found to provide only little new information, however they seem to play a role by magnifying the reaction resulting from the possible following rating changes. Furthermore, there is some evidence that crisis period may strengthen the reaction to rating downgrades adding to the previous literature which suggests that during crisis security prices and bond yields are more information sensitive. This evidence boils down to one suggestion for the policymakers; they still should care about their governments’ respective credit ratings.
6.1 Limitations
Findings in this thesis are subject to limitations. First, the dataset related to some specific rating events is very limited and thus does not allow full interpretations. This is the case especially when comparing the rating change reactions between times of crisis and stability as crisis naturally contains more rating downgrades and vice versa. Furthermore, the limited number of observations related to rating grade changes does not allow us to evaluate the effects resulting from this, intuitively very important, rating event.
35 Germany as abnormal returns. Flight-to-quality effect is likely to increase spread during the crisis, thus magnifying the abnormal returns.
36
References
Afonso, A., Furceri, D., & Comes, P. 2012. Sovereign credit ratings and financial markets linkages: Application to European data. Journal of International Money and Finance, 31: 606–638.
Benninga, S. 2008. Financial Modeling. 3rd ed. Cambridge: The MIT Press.
Brambor, T., Clark, W. R., & Golder, M. 2005. Understanding Interaction Models: Improving Empirical Analyses. Political Analysis, 13(1): 1–20.
Breusch T. S., & Pagan A. R. 1980. The Lagrange Multiplier Test and its Applications to Model Specification in Econometrics. The Review of Economic Studies, 47(1): 239–253.
Brooks, R., Faff, R.W., Hillier, D., & Hillier, J. 2004. The national market impact of sovereign rating changes. Journal of Banking & Finance, 28(1): 233–250.
Brown, S. & Warner, J. B. 1985. Using Daily Stock Returns: The Case of Event Studies. Journal of Financial Economics, 14(1): 3–31.
Cantor, R. & Packer, F. 1996. Determinants and impact of sovereign credit ratings. FRBNY Economic Policy Review, 2(2): 37–54.
Carter Hill, R., Griffits, W. E., & Lim, G. A. 2011. Principles of Econometrics. 4th ed. USA: John Wiley & Sons, Inc.
De Grauwe, P. 2010. Crisis in the eurozone and how to deal with it. CEPS Policy Brief, February 2010.
37 Fitch. 2013. Definitions of Ratings and Other Forms of Opinion. http://www.fitchratings.com/web_content/ratings/fitch_ratings_definitions_and _scales.pdf, Accessed March 20, 2013.
IMF. 2010. Credit Rating Systems. Global Financial Stability Report. October 2010.
Ismailescu, I. & Kazemi, H. 2010. The reaction of emerging market credit default swap spreads to sovereign credit rating changes. Journal of Banking and Finance, 34(12): 2861–2873.
Jaramillo, L. & Terada, C. M. 2011. Sovereign Credit Ratings and Spreads in Emerging Markets: Does Investment Grade Matter? IMF Working Paper 11/44.
Kamisky, G. & Schmuckler, S. L. 2002. Emerging market instability: Do sovereign ratings affect country risk and stock returns? The World Bank Economic Review, 16(2): 171–195.
Kolari, J. & Pynnönen, S. 2010. Event Study Testing with Cross-sectional Correlation of Abnormal Returns. The Review of Financial Studies, September: 1–30.
Larrain, G., Reisen, H. & von Maltzan, J. 1997. Emerging market risk and sovereign credit ratings. OECD Development Centre Working Paper no. 124.
MacKinlay, A. C. 1997. Event studies in economics and finance. Journal of Economic Literature, 35(1): 13–39.
Moody’s. 2013. Rating Symbols and Definitions, February 2013. http://www.moodys.com/researchdocumentcontentpage.aspx?docid=PBC_7900 4. Accessed March 20, 2013.
38 Reinhart, C. M. 2010. This time different chartbook: country histories of on debt, default,
and financial crises. NBER Working Paper 15815.
Reisen, H. & von Maltzan, J. 1999. Boom and bust and sovereign ratings. OECD Development Centre Working Paper no. 148.
Ritter, J. R. 1991. Long-Run Performance of Initial Public Offerings. The Journal of Finance, 45(1): 3–27.
Roodman, D. 2006. How to do Xtabond2: An Introduction to Difference and System GMM in Stata. Center for Global Development Working Paper No. 103.
Standard & Poor’s, 2013. Credit Ratings Definitions & FAQs. http://www.standardandpoors.com/ratings/definitions-and-faqs/en/us. Accessed March 20, 2013.
The New York Times. 2011. Credibility, Chutzpah and Debt.
http://www.nytimes.com/2011/08/08/opinion/credibility-chutzpah-and-debt.html?_r=0, accessed February 20, 2013.
