How significant is the informational content of credit rating announcements

(1)

Amsterdam Business School

Master Thesis

How Significant is the Informational

Content of Credit Rating

Announcements

Author: Jihad Mekroud 10825185 Supervisor: Dr. Stefan Arping

A thesis submitted in fulfilment of the requirements for the degree of Master of Science

at the

Faculty of Economics and Business

(2)

Declaration of Authorship

I, Jihad Mekroud, declare that this thesis titled, How Significant is the Informational Content of Credit Rating Announcements and the work presented in it are my own. I confirm that:

This work was done wholly or mainly while in candidature for a research degree at this University.

Where any part of this thesis has previously been submitted for a degree or any other qualification at this University or any other institution, this has been clearly stated.

Where I have consulted the published work of others, this is always clearly at-tributed.

Where I have quoted from the work of others, the source is always given. With the exception of such quotations, this thesis is entirely my own work.

I have acknowledged all main sources of help.

Where the thesis is based on work done by myself jointly with others, I have made clear exactly what was done by others and what I have contributed myself.

Signed: Jihad Mekroud

Date: 15. December 2015

(3)

ii

Amsterdam Business School

Abstract

Faculty of Economics and Business Master of Science

How Significant is the Informational Content of Credit Rating Announcements

by Jihad Mekroud

Capital market efficiency requires that prices fully reflect all available informa-tion. This study analysis the significance of the informational content of credit rating changes and further how stock liquidity is related to the results. The findings show that capital markets for stocks are not entirely efficient. Market participants anticipate rating changes, mostly for downgrade event. In terms of stock liquidity, smaller firms are more prone to rating changes. Hereby, the outcome show that bid-ask spread is more sensitive for smaller firms than for larger. Furthermore, the results validate that speculative grade firms react stronger than investment grade firms. Collectively, the findings revealed room for more research, in particular about stock return performance before and after and down- or upgrade event. Hereby, the results indicated that before a downgrade event stock returns are declining, but start to increase rapidly just after the event, same counts for upgrade events but the other way around.

(4)

List of Figures

1 Estimation and Event Window Length . . . 14

2 Overview of to tested Event Windows - Short-Term . . . 14

3 Overview of to tested Event Windows - Long-Term . . . 15

4 Reaction of Abnormal Returns . . . 19

(7)

List of Tables

3.1 List of Variables . . . 17

4.1 Downgrade Announcements - Abnormal Returns . . . 21

4.2 Upgrade Announcements - Abnormal Returns . . . 23

4.3 Long-Term Reaction . . . 24

4.4 Mean CAR of Investment Grade and Speculative Grade . . . 25

A.1 Standard & Poor’s Rating . . . 36

A.2 Distributions of Credit Rating Announcements per Year . . . 37

A.3 Mean CAR for 2005 and 2006 . . . 37

A.4 Mean CAR for 2007 and 2008 . . . 38

A.5 Mean CAR for 2009 and 2010 . . . 38

A.6 Mean CAR for 2011 and 2012 . . . 39

A.7 Mean CAR for 2013 and 2014-15 . . . 39

A.8 Energy and Financials Industry . . . 40

A.9 Health Care and Industrials Industry . . . 41

A.10 Information Technology and Materials Industry . . . 42

A.11 Telecom Service and Utilities Industry . . . 43

A.12 Multivariate Regression on Total Downgrade Events . . . 44

A.13 Multivariate Regression on Total Downgrade Events . . . 44

A.14 Multivariate Regression on Downgrade Events - Financial Firms . . . 45

A.15 Multivariate Regression on Downgrade Events - Financial Firms . . . 45

A.16 Multivariate Regression on Downgrade Events - Non-Financial Firms . . . 46

A.17 Multivariate Regression on Downgrade Events - Non-Financial Firms . . . 46

A.18 Multivariate Regression on Downgrade Events - Large Firms . . . 47

A.19 Multivariate Regression on Downgrade Events - Large Firms . . . 47

A.20 Multivariate Regression on Downgrade Events - Small Firms . . . 48

A.21 Multivariate Regression on Downgrade Events - Small Firms . . . 48

A.22 Multivariate Regression on Total Upgrade Events . . . 49

A.23 Multivariate Regression on Total Upgrade Events . . . 49

A.24 Multivariate Regression on Upgrade Events - Financial Firms . . . 50

A.25 Multivariate Regression on Upgrade Events - Financial Firms . . . 50

A.26 Multivariate Regression on Upgrade Events - Non-Financial Firms . . . . 51

A.27 Multivariate Regression on Upgrade Events - Non-Financial Firms . . . . 51

A.28 Multivariate Regression on Upgrade Events - Large Firms . . . 52

A.29 Multivariate Regression on Upgrade Events - Large Firms . . . 52

A.30 Multivariate Regression on Upgrade Events - Small Firms . . . 53

A.31 Multivariate Regression on Upgrade Events - Small Firms . . . 53

(8)

Chapter 1

Introduction

Rating agencies have an important role in the economic system since on the one hand the regulator may rely on ratings as information and on the other hand ratings are helping investors at investment decisions. In fact, they generate credit ratings for a broad number of industries. They rate debt issued by companies and governments, so called corporate and government bonds whereas this research study only concentrates on credit ratings of companies. The rating itself is an important measure for market participants in order to value a company’s credibility and in respect of borrowing costs. Standard & Poor‘s (2008), for instance, defines an issuers credit rating as ”(...) a current opinion of an obligor‘s overall financial capacity (its creditworthiness) to pay its financial obligations”.

The impact of credit rating announcements on stock prices usually should not have any impact on stock prices since the informational value of credit rating announce-ments is already priced into the stock price. There is no new information offered to the market by a credit rating announcement . It is more about collecting already existing information (Wakemann 1990).

In belief of the efficient market hypothesis market prices fully reflect all available information (e.g Fama 1970). However, recent studies in this field showed ambiguous findings. Different empirical researches in the earlier years on this topic (e.g. Weinstein 1977, Pinches and Singleton 1978) showed that there is no significant reaction before, at and after the announcement date on the stock prices. Thus, rating announcements might have no relevant information for the market.

Nevertheless, a rating change may still reflect non-public information and thus may contain important information to market participants (Ingram 1983). In particular, nowadays where liquidity is a significant and economically important factor for asset

(9)

Chapter 1. Introduction 2

pricing, less liquid stocks might be more prone to changes of credit rating announcements than more liquid stocks.

For the empirical research this study is going to examine how strong the informa-tional content of credit rating announcements is. In order to examine the informainforma-tional content, it is necessary to consider stock returns. In particular, this research study will value abnormal returns around the event which will be the credit rating announcement. These returns will be cumulated over a specific event window and then will be tested for statistical significance. Further, there will be a cross sectional test which will in-clude firm characteristics. The economic effect of the event differs by firm. Therefore, statistical properties of cross sectional regressions will be examined. This sample will include US stocks from January 2005 to January 2015. The firm characteristics that are used for the cross sectional regression analyses will be amongst others financial metrics measures such as leverage, return on assets and the firm size. In particular, this study will conduct some new regression analyses that haven’t been done before in other stud-ies. This means, the respective regression analyses will be conducted on financial and non-financial firms as well as on large and small firms.

(10)

Chapter 2

Theory and Hypothesis

2.1 Credit Ratings

2.1.1 Function of Credit Rating Agencies

As already mentioned briefly in the introduction, credit rating agencies primarily are supposed to predict the default for all kinds of debt securities and debt issuers (IOSCO 2004). They are not specifically concentrated on the probability of default but more on the valuation of the riskiness of different securities like debt. Since they assess debt quality and are specialized in capital markets transactions, credit rating agencies act as information intermediaries that try to overcome informational asymmetries between both market sides. Also, many investors are kind of limited in their investment decisions in such a way that they are only allowed to invest with a top rating category (investment grade) (Maher and Sen 1997).

Furthermore, credit rating agencies might possess non-public information from different borrowers or issuers as part of the rating process (Langohr 2009). Given that credit rating agencies may provide new information to financial markets, they seem highly relevant also from the international policy perspective, especially when it comes to issues about the international financial system. They have the ability to intensify or alternatively weaken financial crises. The Basel II Accord indicates that banks for instance are allowed to provide a credit for a percentage of their own equity. Moreover, this percentage depends on the credit risk involved, which may depend on the credit rating (Basel Committee on Banking Supervision 2004). In the past there have been many studies regarding the function of credit rating agencies in order to get a better understanding of the credit rating industry and to make policy recommendations to reform it (Basel Committee 2000). The results of these studies indicate that it might

(11)

Chapter 2. Theory and Hypothesis 4

be dangerous for instance to link regulatory risk weights to credit ratings (Amato and Furfine 2003). Thus, it might be more useful to rely on the economics and information provision since without good information provisions, rating might not be useful for regulation and risk management.

2.1.2 Credit Rating Process

Usually a company receives a credit rating by contacting the rating agency for the purpose to assign a rating to a new debt issue. In order to understand why this is crucial for the issuer, a closer look at the relationship between issuer and investor might be useful. Basically, there exists a simple principal-agent relationship when it comes to a bond issuance. The issuer, in this case the agent, may have information that is not available to the investor, in this case the principal. Because the issuer may face higher costs if revealing this information towards the investors, he is better off hiding this information. The investor will not trust the reliability of such information and thus will ask for a higher risk premium, which increases the cost of the transaction. This means basically that the investor will ask for a higher interest rate. In general, this would also affect issuers with low credit risk as they may also face difficulties issuing bonds. And exactly for this situation credit rating agencies exist. They reduce information asymmetry by revealing hidden information and providing investors with a screening instrument. Eventually, this lowers the risk premium required by investors and thus the cost of the transaction for the issuer (IOSCO Report on the activities of credit rating agencies 2003).

The rating process varies widely and depends on the methodologies used within the process. Some credit rating agencies base their rating judgement on qualitative and quantitative indicators and then report this to the rating committee. Whereas other credit rating agencies rely only on quantitative models and thus the rating will be rather a result of statistical technical analyses. However, there is no single approach which can be seen as the most appropriate one. Thus, it depends on the specific case to determine which approach is the right one. Generally, the largest credit rating agencies tend to apply similar rating approaches and use similar rating instruments. Larger international rating agencies have a rating committee which is able to initiate, withdraw and change a rating. A rating committee usually consists of lead analysts, supervisor or managing directors and junior analysts. Rating decisions will be then set by a majority vote of the committee (IOSCO Report on the activities of credit rating agencies 2003).

The company that is requesting a credit rating, needs to submit a bunch of docu-mentations such as annual reports for the past years, quarterly reports and prospectus

(12)

to some recent debt issuance. The lead analysts review all the documents and also ex-amine company characteristics as well as the bond issue. Based on their results, the analyst team prepares a rating report which they present to the rating committee. The committee reviews the report and discusses with the analyst team. Following this, they set the final rating decision (Huanga, Hsinchun Chena et al. 2003).

Once the rating decision was set, the issuer then will be informed about the rating decision and will be offered a draft of the rating press release and/ or report. The issuer can review the rating decision and can request a reconsideration of the decision. In case the issuer presents new information material, there might be a chance of reconsideration. Eventually, the rating release/ report will be publicly available in order to make sure that no non-public information is being hidden by the issuer (IOSCO Report on the activities of credit rating agencies 2003).

2.2 Literature Review

There have been several research studies in the past, which discussed the impact of rating announcements on capital markets. But still, there are ambiguous findings. In respect of the contribution to the research topic, these different results might help to understand what empirical design and approach was used and why these studies ended up with ambiguous findings.

As one of the first studies on this topic Katz (1974) examined the impact of credit rating changes on bond yields. In particular, he tried to figure out to what extent rating reclassifications are anticipated by market participants. He developed a regression model in order to forecast the expected yield to maturity after reclassification and compared afterwards the actual with the expected one to see whether there were any kinds of anticipations. As a result, there were no anticipations at all. Furthermore, he observed a significant negative reaction after a negative rating announcement but somehow no significant reaction for a positive rating announcement (Katz and Steven 1974)

Steiner and Heinke (2001) came to almost similar results. In their research they examined daily excess eurobond returns associated with rating change announcements. As an interesting outcome the results for negative rating announcements where signifi-cant for bond price reactions whereas no announcement effects were noticed for positive announcements. Furthermore, they found significant movements up to 100 trading days prior to the rating change (Steiner and Heinke 2001).

Two other important empirical studies in this field were done by Weinstein (1977) and Hull et al. (2004). Weinstein, as one of the representative of the efficient market

(13)

hypothesis, examined the reaction of credit rating announcements on bond prices. In contrast to the results presented by Katz, Weinstein in fact found little reaction on bond prices prior to the announcement and thus contradicts the findings of Katz. In particular, he found some evidence between 18 and 7 months before the rating announcement but no evidence 6 months prior to the announcement and little change 6 months after the rating announcement (Weinstein 1977). Also Hettenhouse and Sartoris (1976) concluded that bond prices react to other information released prior to the rating change and thus capital markets anticipate rating changes (Hettenhouse and Sartoris 1976).

A highly significant study, which confirms the informational content of rating announcement is the study done by Holthausen et al. (1992). Instead of using monthly returns they decided to use daily data for the first time. They argued that the advantage of using daily data instead of monthly data is that one may decrease the probability that the price reaction is due to other information published during the month. They calculated excess bond returns as raw bond returns minus the return on a risk free rate. An additional extensive sample was used. The study belongs to one of the few, which was able to show reactions after positive rating announcements (Hand, Holthausen and Leftwich 1992).

Another paper from Avramov, Chordia, Jostova et al. (2008) examined the credit risk puzzle which indicates that low credit risk firms realize higher returns than high credit risk firms. Normally, when believing into the fundamental principle of financial economics one would expect that high credit risk firms face higher expected returns as investors pay a premium for bearing risk. In particular, the paper shows that the credit risk effect reacts at the most for worst-rated stocks around downgrades. It seems that low rated firms are facing considerable negative returns due to strong institutional selling pressure. Further, this paper argues that the credit risk effect is caused by mispricing which is a product of illiquidity, due to insufficient analyst coverage, and short selling constraints. This paper is contributing to this research study in such a way as it is more comprehensible in explaining why credit rating downgrade announcements (might) cause significant stock price reactions compared to credit rating upgrades (Avramov, Chordia, Jostova et al. 2008).

When assessing the stronger credit risk effect for low credit firms, it is reasonable to consider the results from the paper ”The Long-Run Stock Returns Following Bond Ratings Changes” by Dichev and Piotroski 1989. Basically, they examined the long-run effect of abnormal returns following credit rating changes. In particular they calculated abnormal returns as both cumulative abnormal returns (CAR) and buy-and-hold re-turns (BHAR), after controlling for firm size and book-to-market ratio. They found no significant abnormal returns for stocks with credit rating upgrade. Moreover, they found

(14)

significant abnormal returns for stocks following downgrades. They conclude that poor returns of downgrade firms are more identifiable for smaller and low credit quality firms. Smaller firms tend to have lower bond ratings (e.g. Keenan, Carty, and Shtogrin 1998). Another important aspect is that downgrades are strong predictors of future deterio-rations in earnings (Dichev and Piotroski 1998). In respect of this research study, the regression analysis that will be conducted will include control variables such as firm size and book-to-market ratio when calculating abnormal returns, similar to several recent studies (e.g. Fama and French 1992).

In order to understand how liquidity might affect returns and further to understand why different effects are observable among up- and downgrades, the research paper ”Asset pricing with liquidity risk” written by Acharya and Pedersen (2005) gives useful insights. In particular, this paper presents a theoretical model which helps to explain how asset prices are affected by liquidity risk. As some recent literature about effect of rating announcements on asset prices found statistical significant abnormal returns mostly for downgrades, the paper from Acharya and Pedersen might give more color on this unilateral reaction. They used liquidity adjusted CAPM in order to show that security’s return depends amongst others on its expected liquidity and the covariance between security’s return/liquidity and markets’ return/liquidity. Further, the paper argues, that the expected return of a security is increasing with its expected illiquidity. Eventually, the model shows that positive shocks to illiquidity, are associated with a low contemporaneous return.

All in all, it is still not clear enough how changes of positive rating announcements changes impact prices. While some studies indicate abnormal effects, others don’t show any reaction which might be due to a small number of positive rating announcements. Jorian and Zhang (2007) found empirical evidence that different results might be due to the empirical design (Jorion, Phillippe and Zhang 2007). Thus the methodology approach used for the cross section may be different among the studies.

(15)

2.3 Hypothesis

As mentioned in the literature, there have been some studies already done in this research field. Nevertheless, it is still not clear whether specific firm characteristics affect the rating change and therefore a affect the stock price. In particular, this study exam-ines liquidity parameters such as bid-ask spread and whether liquidity plays a crucial role. More liquidity in stocks should result in a lower level of information asymmetry, hence statistically significant reactions should barely be observable for more liquid stock compared to less liquid stocks.

H1: Credit Rating Announcements do not have any informational value for the market.

H2: Significant reaction are rather expected from illiquid stocks than from liquid stocks

H3: Stocks react differently on changes in credit ratings, dependent on specific firm

(16)

Chapter 3

Research Methodology

In general, research methodology is used for the purpose to show how data was generated and how it was collected. Furthermore, this chapter shows how the results were attained. A thorough explanation about where the data comes from is helpful for the reader in order to be able to assess the validity and reliability of the study conducted. It makes clearer, why a specific method or procedure was appropriate for the study.

Generally, two different kinds of research techniques might be applied: scientific and historical approach. Both approaches try to obtain quality information on a specific research topic by using eight steps which include:

• formation of research topic

• development of hypothesis

• conceptual definitions

• data gathering

• data cleaning and analysis

• data testing

• research conclusion

It needs to be mentioned that if the results do not support the hypothesis, it does not necessarily mean that the research failed. In other words, the rejection of a specific research topic may lead to a reconsideration and thus, optimisation of a particular hypothesis. Consequently, this process is part of the very foundation of the research itself, i.e. improving the knowledge about a specific topic. However, conducting a

(17)

Chapter 3. Research Methodology 10

scientific analysis will always lead to biased results which depends on the perspective of the researcher. It cannot be totally objective since the perspective of the researcher will always have some influence in the choice of independent variables and interpretation of results. Results can support a hypothesis, but they can never prove it (Bauer et al. 2000).

Finally, the research design used for this work involves both philosophical assump-tions that direct the collection and data analysis which includes both the qualitative and quantitative approaches in the research study. This is based on the central premise that the combination of both qualitative and quantitative approaches provides a better un-derstanding of the research problem compared to a single approach. Moreover, with the emergence of more strategies and tools for combining the different types of data, it has become more possible to cross multiple disciplinary boundaries (Creswell and Clark 2011).

3.1 The Event Study Methodology

The key objective in this event study is to assess the response of stock prices to such an event that is connected with the release of key information to the stock market, in particular credit rating announcements. As mentioned earlier in the literature review there is a semi-strong hypothesis which indicates that the stock price completely reflects all available public information in the market (Watts and Zimmerman 1986). In case the event, the credit rating change announcement, has any informational content, the stock price would imply any statistically significant abnormal return connected to this event. The methodology of this event study is concentrated on the effect of credit rating change on stock prices. It allows the measurement of stock market reactions to the release of information about credit ratings.

3.2 Data

An important part for this research study is collecting data regarding rating an-nouncements. In particular, it is about the specific date when a rating agency changed the credit rating of a firm for US companies. The information concerning the required variables was received by the WRDS database. Different upgrade and downgrade data was collected from several sets of companies. Hereby, large numbers of companies were selected for the analysis with the purpose to ensure that the findings were spread enough to provide an actual impression of the trends in the actual scenario. Furthermore, this

(18)

study implements statistical instruments such as descriptive statistics and regression analyses to take empirical tests on hypotheses. In addition, SPSS and STATA have been used as the statistical analysis software.

In order to get a large sample distribution, the data chosen included observations between January 2005 and January 2015 as it was approximated to be just enough to provide the trend which was necessary in making such conclusions. The sample focused on rating announcement made by Standard & Poor’s since Standard & Poor’s belongs to one of the biggest rating agencies (ESMA 2014). The required data is delivered by the Compustat database of Wharton Research Data Services. In order to calculate the estimation of normal returns of each of the selected firms during the research period, the use of the S & P 500 index seemed appropriate. The chosen index served as benchmark as the including information is similar to what is included in the dataset.

Since this research study aims to consider differences between large and small firms, the average natural logarithm of total assets will be used as the main criteria of classifying whether a firm is large or small. A firm that exceeds the average natural logarithm of total assets is categorized as large and vice versa. The information regarding the total assets of the firms were found on the WRDS database. The same counts for all the other financial variables used for the multivariate regressions. In order to value the impact of a specific sector, a further classification was applied which separated the financial and non-financials firms. The necessity to split the firms in two groups emerged due to the fact, that during the financial crisis the financial sector got more negatively affected in terms of rating downgrades and strong profit losses (Bartram and Bodnar 2009). By controlling for this event, the results might be more reliable as it is somehow possible to control for exogenous effects that are correlated to the credit rating events.

The data from Compustat included credit ratings for each company of each month for the specified period (68.977 observations). Then duplicates had to be dropped to keep only the first observation of the credit rating for each company (1.526 left). After dropping the duplicates, announcement dates had to be adjusted since Compustat indeed offers the credit rating for each company but usually to the end of each month. Thomson One was used to get the exact rating date. Thomson One is a strong research tool that offers a broad range of financial data. The exact date for each rating event had to be looked up for each observation in the sample. Finally, the dataset was categorized between downgrades and upgrades so that in the end 424 upgrade and 431 downgrade observations were left for the study period. The data chosen was very selective as it was necessary to make sure that only samples were selected which contained historical stock prices. Furthermore, the rating changes were instrument in the determination of

(19)

the relationship, if any, between stock returns and the other factors such the firm size, leverage and the ratio between assets and debt of the firms.

Lastly, a cross-sectional analysis will be conducted to examine whether specific company or industry characteristic may have a different effect on the results. The data is going to be obtained from Datastream and cross checked with the CRSP database.

3.3 Methodology Approach

To simplify the research and to ensure that closer tabs were kept on the proceeding, the research decided to employ the four steps event study which helped in achieving the research schedule and providing the required results. The event study is conducted through the EVENTUS package. This research study is following 4 Steps:

• Identifying the right event data

• Specify the benchmark model of return for the purpose of inspecting the behaviour of the firm’s stock returns which is one of the most fundamental variables under observation by this research

• Calculate abnormal returns around the event date

• Test abnormal returns for significance

The returns that are above the expected returns or simply the returns that are above the normally anticipated levels are said to be abnormal. For calculating the announcement returns the market model was used, where the market is the Standard & Poor’s 500 Composite Index. The market model is given in accordance with the following parameters:

E(Rit) = αi+ βiRM t+ uit (1)

Where Rit is the log return of the stock i at time t and RM the log market return

of index M in time t. The calculation of the stock prices as well as the calculation of the index was achieved through the use of the logarithmic function below. Note that Rit and Pt are the returns to the normal expectation, the security under consideration,

the instantaneous time during which the study was conducted and the closing price on the day of investigation, respectively. The function Pt−1 on the other hand is taken to

represent the closing price during the day before the day under investigation.

Rit= ln

Pt

Pt−1

(20)

After calculating the expected return of each stock for each day, the abnormal re-turn can be calculated as the observed rere-turn over the event window minus the expected return. This statement can be expressed using the equation illustrated below. The fac-tors ARit, Ritand αi+ βiRM t are the abnormal return, the actual returns obtained and

the benchmark returns, respectively.

ARit= Rit− (αi+ βiRM t) (3)

To calculate these abnormal returns, α and β had to be estimated for each firm using OLS and the historic firm and market returns. Given the historically estimated value of α and β the expected return of the stock could be calculated. Finally, the abnormal return is calculated as the difference between the actual return on the stock and the expected return.

In order to implement an appropriate Market Model Event Study, several decisions are necessary including the frequency over which returns are measured, the length of the estimation period and the event window for calculating the abnormal returns. For the latter this research applies daily stock returns as brought up earlier. Nowadays, daily stock returns have become standard, in particular when conducting short-term event studies. When it comes to high-frequency event studies, one might even use 15, 30 or 60 minutes returns (Belinda Mucklow 1994). However, for long-term studies it is recommended to use less frequent, even quarterly returns (Bremer, Buchanan and English 2011).

An event study usually differs between estimation window and event window. The event window is not part of the estimation window and does not overlap. It is set after the estimation window. Specifically this means that the trading days before the event day do not count as part of the estimation window but as event window. The estimation period during which the experiment was to be conducted is a crucial instrumental factor in researches. Moreover, there is no common rule concerning the length. In this research it was decided to select the estimation with respect to many previous researchers which is 255 days (Mackinlay 1997). The estimation period ended 30 days before the event date. The 30 days period prior to the event date was assigned for the event window period. The following table should offer a better understanding about the estimation and event window:

(21)

Figure 1: Estimation and Event Window Length

Furthermore, this research study conducted a comparison between several event windows while there was also a calculation of the abnormal returns characterizing each of the days constituting the event windows under consideration. Daily stock returns were collected for the calculation of the abnormal returns which characterized the period. The event period consisted of 30 day prior and 30 days after the event date and thus a total of 60 calculations representing all the abnormal returns during this period. The following table should offer an overview about the single event windows that were part of the research study. In particular, the mean cumulative abnormal returns (CAR) were observed for each window and then used for cross-sectional analyses, which will be discussed later on.

Figure 2: Overview of to tested Event Windows - Short-Term

Apart from the short-term analysis, this research will also examine a long-term analysis in order to assess how significant possible abnormal returns do react one year

(22)

and a half year before the event. This might offer more insight into how strong the in-formational content of credit rating changes are and whether market participant already anticipate a rating change long before it actually takes place.

Figure 3: Overview of to tested Event Windows - Long-Term

So far, once the abnormal returns were calculated for each event, they getting summed up in order to calculate the CAR. According the efficient market hypothesis, the CAR should be close to zero as all available information are already incorporated into the stock price (e.g. Fama 1970).

CARi(T 1,T 2)= T 2

X

t=T 1

ARi,t (4)

Statistical tests of abnormal returns are usually based on the cross average of each measure. Thus, to get the cross sectional average, one takes the CAR and averages across N observations to get a single figure representing the mean CAR over all events.

meanCAR(T 1,T 2) = 1 N N X i=1 CARi(T 1,T 2) (5)

Through the averaging technique it was possible to replicate the impacts of the event on a particular occasion especially considering the fact that the entire factor de-scribing the abnormal returns was centrally located within a given event during this research. In particular, this helped with considering only information that did have any relation hence impact on the event investigation. The mean CAR can be used in order to measure the impact and significance of company and industry characteristics on credit rating changes.

(23)

Multivariate regression analyses will be used in order to consider firm character-istics as independent variables and the mean CAR for the different event windows as dependent variable. Further, these analyses will help to identify variation that occurred in the abnormal returns throughout the entire event window periods. This research study will contain some variables in the regression to conduct a couple of cross-sectional analyses in order to examine whether specific firm characteristics have an impact and to solve for omitted variable bias.

meanCAReventwindow= α0+ α1(BAS) + α2(Lev) + α3(F irmSize) + α4(ROA)+

α5(BV S) + α6(DP S) + α7(N umberof Analyst) + uit

(6)

Where BAS is the bid-ask spread to see to what extent liquidity plays a role in terms of impact through credit rating changes since bid-ask spread is a good measure for liquidity. The variable Lev will be calculated as the average leverage of the company when the announcement was made. Further, choosing this variable enables to test the power and the strength of the firm with respect to financial matters. Usually one can say the lower the debt to asset ratio the less risky a firm and vice versa (Bodenhorn 2003). Thus, a firm that is characterized by high leverage, should be expected to face higher reactions regarding changes in prices and credit compared to firms with lower leverage.

Another variable employed here is the Firm Size variable which is calculated as the natural logarithm of total assets of the company for that year when the announcement was made. As mentioned before this variable is important as it acts as a representative of the theory. According to the theory, small firms are expected to experience larger abnormal returns compared to large firms (e.g. Jones 2012 see also Bernard and Thomas 1990 and Fama 1998). Further, larger firms are more closely followed (e.g. more an-alysts). Thus, events should be more predictable, all others equal. This would imply, that a rating announcement should not offerthat much new information to the market compared to smaller firms. Barely significant abnormal return should be observable (Khotari, Warner 2006). First, a regression will be conducted to see how firm size gen-erally impacts the outcome. Then the dataset will be separated into large and small firms. This will be determined by taking the average firm size of all firms. Those firms, which are below the average firm size, will be seen as small firms and vice versa.

The variable ROA defines the return on assets and should indicate how profitable a company is relative to its assets. Better rated companies usually should have positive relation to the return on assets measure. Similar explanation counts for the variable DPS

(24)

which is the dividend per share. Firms with strong fundamentals would have sufficient earnings in order to payout a dividend which would also indicate that they should have a better rating.

Arbel and Strebel (1982) found that unattended firms somehow experience infor-mation asymmetry. To control for, this study will use the variable ”number of analysts” which shows basically by how many analysts a firm is covered.

These multivariate regression analyses will be conducted first on the total dataset. Then the dataset will be separated between financials and non-financials. As explained earlier, Bartram and Bodnar found out that during the crisis financial firms suffered more than non-financials (Bartram and Bodnar 2009). The Global Industry Classification Standard code (GICS) from Compustat helped in order to identify financial firms (Code 40 for the financial industry). Further, the dataset will be separated between small and large firms.

The following table should give an overview about the variables used for the cross sectional analyses.

Table 3.1: List of Variables

List of Variables

Variables Symbol Description

Firm Specific Variables

Liquidity BAS Bid-Ask-Spread

Firm Size Firm Size Natural Log of Total Asset

Leverage Lev Long-Term Debt divided by Total Asset

Return on Asset ROA Net Income divided by Total Asset

Dividend per Share DPS Dividend per Share

(25)

Chapter 4

Analysis Results

Chapter 4 discusses all the results of the conducted event study. The event win-dows which were used for the studies were mentioned earlier. First, a short- and then a long-term analysis is conducted for negative and positive credit rating announcements.

4.1 Abnormal Returns Reaction

Before analyzing the statistical significance of a change in credit ratings and thereby the informational power of credit rating announcements, this section should give an overview about how abnormal returns reacted around the event date. Hereby the methodology approach discussed in section 3.3 will be applied. After calculating the abnormal return over an event window of -7 and +7 days around the event date, the re-sults were plugged into a graph to visualize the abnormal return reaction of credit rating announcements. But it needs to be emphasized that to this point no statistical analysis where conducted, in order to assess whether the abnormal returns are significant or not. This will be done in the following sections.

(26)

Chapter 4. Analysis Results 19

Figure 4: Reaction of Abnormal Returns

Comparing the two graphs in figure 4, there is obviously a higher and also more volatile reaction for downgrades than for upgrades. The reaction for upgrades is not that strong over the event window, whereas the reaction of downgrades before the event date seems to be indeed strong, with -0.31% two days and -0.70% one day before and -0.42% at the event date. The reaction for upgrades was not that strong at the event date with +0.13%. Furthermore, the downgrade reaction seem to pursue an odd development as the reaction turns into positive abnormal returns with +0.37% on day 5 after the event date. It seems that the stock market overreacted before and during the event date which subsequently lead to a high negative return. A deep cut in stock prices was obviously exploited by other market participants who started to buy the underpriced stock. Thus, positive reaction just after the event date was observable.

Apart from this interpretation, in terms of informational content these results indicate that credit rating downgrade announcements obviously revealed some informa-tion in beforehand which let the stock returns tumble. This can be further explained by the behaviour of market participants. In recent research studies it was examined how market participants react on good and bad news for different states of the econ-omy. The main results revealed that market participants overreact to shocks that might move stock prices. In 1985 Werner De Bondt and Richard Thaler examined market overreaction and published their results in the Journal of Finance with the title ”Does the Market Overreact?”. They examined US stock returns over a period of three years. They splitted the stocks in two groups: best performing and worst performing stocks. Surprisingly, the worst performing stock portfolio did outperform the market index, whereas the best performing stock portfolio underperformed. The reason for this is sim-ply that market participants essentially overreacted for both portfolios. When market

(27)

participants overreacted to bad news, this lead to a decline in stock prices until after a certain time market participants realized that their pessimism was not justified. So they concluded that the stocks were underpriced and the stock began to rebound. For the best performing stock portfolio the opposite was true. Market participants argue that the stock price was not justified (De Bondt, Thaler 1985).

4.2 Market Reaction to Downgrade Announcements

4.2.1 Short-Term Reaction

The first event study examines the market reaction around a negative credit rating change. In particular, the event study should indicate whether any significant effects were arising around the event window. The results of the event study are presented in the table below whereas this table reports Abnormal Returns over the short-term window of (-7/+7) around the announcement day.

Overall, the results show that the stock market is reacting negative just before the downgrade of a credit rating. This might be a sign that market participants are antici-pating the credit rating change, since the results in Table 4.1 got some highly significant evidence. Two days before the announcement the abnormal return is statistically signif-icant with -0.31%. Right before the announcement day the abnormal return decreases further to -0.70% and becomes statistically significant at the 1% level (t-statistic -4.499). At the announcement day itself and the day after the results indicates negative abnormal returns which are both statistically significant at the 1% level.

(28)

Table 4.1: Downgrade Announcements - Abnormal Returns

Market Model, Standard & Poor’s 500 Composite Index Abnormal Returns

Day AR(%) t-statistics

t-7 -0.29% *** 3.592 t-6 -0.08% 0.304 t-5 0.19% 1.543 t-4 0.01% -0.522 t-3 -0.22% * -1.765 t-2 -0.31% ** -2.108 t-1 -0.70% *** -4.499 t(Event Day) -0.42% *** -3.927 t+1 -0.64% *** -5.526 t+2 0.02% 0.234 t+3 0.01% * 1.903 t+4 -0.03% -0.091 t+5 0.37% *** 2.953 t+6 0.08% 1.112 t+7 0.17% * 1.956 Number of Events 428 * significance at 10% level ** significance at 5% level *** significance at 1% level

These empirical results are in line with the results found earlier in the literature. Amongst others, Grier and Katz (1976) and Griffin and Sanvicente (1982) found an empirical significant evidence. The latter explored stock price reactions to rating changes and tested the cumulative residuals significance between the event and the controlled samples. Their findings indicate that the cumulative abnormal returns within the short-term window around the announcement day were indeed statistically significant for downgrading stocks.

The fact, that downgrading lead to negative abnormal returns just before the announcement day, indicates the informational value of credit rating announcements. Furthermore, it might show that rating agencies posses private information that is not

(29)

accessible to market participants. However, to confirm this statistically significant evi-dence this research study will later on differ between industries and conduct regression by including firm characteristics as independent variables.

4.3 Market Reaction to Upgrade Announcements

4.3.1 Short-Term Reaction

The next event study examined the short-term effect of abnormal returns facing a positive credit rating announcement. The time window that was used for that event was similar to the event window of downgrading events, [-7/+7] around the announcement day.

Surprisingly and completely different from the previous results for the downgrading events, there are obviously barely any statistically significant values. Overall, positive absolute returns are observed over the [-7/+7]-day window. However, these returns are not statistically significant.

Furthermore, these findings are similar to the findings of Steiner and Heinke (2001). As an interesting outcome the results for negative rating announcements where signifi-cant whereas no statistically signifisignifi-cant effects were noticed for positive rating announce-ments. Moreover, they found significant movements up to 100 trading days prior to the rating change.

(30)

Table 4.2: Upgrade Announcements - Abnormal Returns

Market Model, Standard & Poor’s 500 Composite Index Abnormal Returns

Day AR(%) t-statistics

t-7 -0.01% -0.310 t-6 -0.17% ** -2.413 t-5 0.06% 0.854 t-4 0.00% -0.032 t-3 0.08% 0.126 t-2 -0.02% -0.029 t-1 0.01% 0.096 t(Event Day) 0.13% 1.354 t+1 -0.20% *** -2.648 t+2 -0.05% 0.332 t+3 0.15% 1.148 t+4 0.00% 0.126 t+5 -0.03% -0.354 t+6 0.02% 0.399 t+7 -0.02% -0.163 Number of Events 452 * significance at 10% level ** significance at 5% level *** significance at 1% level

4.4 Long-Term Reaction

This section discusses the results which were extracted by examining the long-term effect of credit rating announcements. In particular, this research examined the reaction of stock prices for one year, 255 trading days, and half a year, 127 trading days, before and after the credit rating announcement. As Steiner and Heinke (2001) concluded, some bond prices react to other information released prior to the rating change. In particular they found some statistical significant price movements 100 trading days prior to the rating change.

(31)

Table 4.3: Long-Term Reaction

Long-term Analysis

Upgrade Downgrade

Mean CAR Mean CAR

Event Window (t-stat) (t-stat)

(-255 , -1) 1.09% -8.80% *** (-0.293) (-9.133) (-127 , -1) 0.78% -6.68% *** (-0.322) (-8.953) (+1 , +127) -4.97% *** 15.86% *** (-4.498) (18.105) (+1 , +255) -8.24% *** 27.77% *** (-5.009) (24.824) Observations 420 427 * significance at 10% level ** significance at 5% level *** significance at 1% level

By examining these results, apparently downgrades show highly statistically sig-nificant reactions before and after the rating event whereas CARs for upgrade events appear to be statistically significant after the rating event. Interestingly, the CAR seems to be negative before the downgrade event for both event windows and suddenly turn positive after the event. To the contrary, CARs seem to turn negative after the upgrade events. An explanation for this appearance was discussed in 4.1.

4.5 Effect between Investment Grade & Speculative Grade

This section examines how reaction of credit rating change differs between invest-ment grade and speculative grade stocks. One would expect that stocks with speculative grades would react stronger to new information than investment grade stocks. The rea-son for that is based on the idea that speculative stocks are supposed to be less traded than more secure stocks with investment grade. Further, for trading speculative stocks market participants would ask for a much higher risk premium as their risk of failure is higher. This should turn into higher returns due to higher volatility.

(32)

Table 4.4 shows the results of the reaction between investment grade and specu-lative grade stocks for different event windows. The observations within the two cate-gories, up- and downgrade, were separated again in two groups, investment grade and speculative grade. Table A.1 in the appendix gives an overview about the single rating categories. A firm with at least a credit rating of BBB- is classified as investment grade. Any firm with a rating below BBB- is classified as speculative grade. In total, there are 540 investment grade observations, where 293 and 247 counted for upgrades and downgrades, respectively, and 314 speculative grade observations, where 131 and 183 counted for upgrade and downgrades, respectively.

Table 4.4: Mean CAR of Investment Grade and Speculative Grade

Investment Grade Speculative Grade

Upgrade Downgrade Upgrade Downgrade

Mean CAR Mean CAR Mean CAR Mean CAR

Event Window (t-stat) (t-stat) (t-stat) (t-stat)

(-30 , -11) 0.64% 0.00% -2.37% * -0.74% ** (1.426) (-0.233) (-1.866) (-2.142) (-10 , -2) 0.34% -0.76% 0.41% -0.9% ** (0.938) (-1.408) (0.481) (-2.529) (-1 , 0) 0.07% -1.18% *** -0.57% -1.04% *** (0.527) (-4.403) (-0.719) (-4.095) (0 , +1) 0.11% -0.68% *** -0.30% -1.66% *** (0.833) (-3.076) (-0.161) (-6.979) (+2 , +10) -0.12% 0.98% *** -1.31% ** 2.39% *** (-0.517) (2.883) (-2.123) (3.043) (+11 , +30) -0.57% ** 0.86% * -1.07% 5.59% *** (-2.074) (1.941) (-1.064) (5.698) Observations 293 247 131 183 * significance at 10% level ** significance at 5% level *** significance at 1% level

Comparing the two different rating segments, they seem to have similar signifi-cant results on up- and downgrade events. But a thorough analysis indicates that stocks with speculative grades, as expected, show higher returns compared to investment grade stocks. A highly significant and interesting reaction is noticeable for speculative grade stocks regarding downgrades. In particular, for the event windows (+2 , +10) and (+11

(33)

, +30) abnormal returns of +2.39% at a statistical significance of 1% level and +5.59% at a statistical significance of 1% level are observable, respectively. Comparing these re-actions with the rere-actions of investment grade stocks, where for the event windows (+2 , +10) and (+11 , +30) statistical significance abnormal returns of +0.98% and +0.86% were observable, respectively, the assumption stated can be confirmed. Furthermore, the results show statistical significant negative CARs for the downgrade category up to one day after the event date for both the investment and speculative grade stocks. In contrast to downgrades, barely statistical significant reactions can be observed for the upgrades column, neither for investment nor speculative grades stocks. In general, it is only observable, that for both categories the CAR turns into a negative figure after the event date, even though they are not statistical significant. Speaking in terms of liquid-ity, the results from Acharya and Pedersen (2005) could be affirmed by the observable reaction. Speculative stocks and in general stocks which were facing a downgrade event experience a liquidity shock in the downturn. During the downturn investors seem to worry about a security’s performance and tradability both in market downturns and when liquidity dries up. As Acharya and Pedersen examined in their paper, expected illiquidity increases expected return. Since speculative stocks are less traded than invest-ment grade stocks, speculative stocks are essentially less liquid than investinvest-ment grade stocks. Thus, under consideration of Acharya and Pedersen, these results would fit into the understanding of financial markets.

4.6 Market Reaction separated in Industries

The following section elaborates the effect across different industries. These analy-ses seemed reasonable as the effect across industries might differ. In order to identify the different industries, this research simply used the Global Industry Classification Stan-dard values for each company. In appendix A.8, A.9, A.10 and A.11 the different results for the different Industries can be found. The results for the Telecom Service industry are negligible as the observations are not sufficient to give any statistical significant insights.

In general, the data shows that the financial industry experienced most of the credit rating changes over the last 10 years. This is most reasonable attributable to the financial crisis as the financial industry suffered most during the crisis. The banks had a crucial role as they were selling so called Collateral Debt Obligations which were struc-tured financial products on mortgage loans. Eventually these products became worthless and banks had to write down huge losses, which lead some banks to bankruptcy (e.g. Lehman Brothers). Analyzing the effect on the financial industry more thoroughly, it

(34)

is quite interesting to see how strong positively the financial firms reacted after the downgrade rating event within the (+1 , +255) event window, which stated a statistical significant positive CAR of 41.46% and hence the highest CAR compared to all other in-dustries. Furthermore, it is observable that the financial industry experienced statistical significant negative CARs before the downgrade event which was mostly stronger com-pared to other industries. In particular, the day before the event, event window (-1 , 0), indicates a highly statistical significant negative CAR with -2.69% (t-statistics -6.915). Surprisingly, one year (255 trading days) before the downgrade event, still a statistical significant positive CAR is observable. The CAR turns statistical significant negatively half a year (127 trading days) before the rating downgrade with -3.44% (t-statistics -6.556).

Comparing other industries, such as industrials, materials and information tech-nology, the results indicate similar to the financial industry negative statistical significant CARs half a year (127 trading days) before the downgrade event. Examining the effect by rating upgrades the results show barely statistical significant results. In fact, the energy, health care and information technology sectors solely show statistical significant negative CARs on a long-term basis, for one year (255 trading days) and half a year (127 trading days) after a rating upgrade event. When speaking in terms of informational content, the results show that in some industries mostly downgrade events had statis-tical significant effects on the stock returns. In particular, before the downgrade event negative returns were mostly observable. On the one hand this could imply that market participants were already anticipating credit rating downgrades. On the other hand this could also imply that due to the shrinking stock prices and therefore negative returns rating agencies decided to change credit ratings. This will be verified by considering the results from the other sections.

4.7 Market Reaction separated in Years

This section analysis the CAR performance for each year over the last ten years for different event windows. This study should give more insights about which years were statistically most significant as some results might be biased by other events happened at the same time such as the financial crisis.

Starting with the results from 2005 and 2006 it is observable that there was barely a statistical significant CAR, even though a number of observations were essentially high with an average of around 42 observations. There are some figures for the downgrade event for the event window (0 , +1) and (+11 , +30) in 2006 which seemed statistically significant. For 2007 some statistically significant figures were observable in particular

(35)

for downgrade events for the event windows (-1 , 0) and (0 , +1). For the upgrade events a statistically significant CAR for the event window (-30 , -11) with a CAR performance of 1.8% and a t-statistics of 1.750 is observable. The statistical results for this year do not seem to be reliable as they do not match to other results within this research study.

Now considering 2008, the first one notes is that observations are significantly higher for downgrades than it was in the years before. This might be explained by the financial crisis. Many firms across different industries were affected negatively by the financial crisis as many firms and not only banks were somehow connected, also known as the contagion risk. Thus, rating agency downgraded banks and also other firms during the crisis as they noticed that they were connected too strong to higher risk as supposed. No statistically significant reaction was observable for the upgrade events, except for the event window (+2 , +10) with a negative CAR of -2.01% which is negligible as other event windows showed no significant reactions. Considering the downgrade events, the results show that highly statistical significant CARs were observable already before the event. In particular, for the event windows (-30 , -11) and (-10 , -2) a negative CAR of -4.99% and -2.25%, respectively, was noticeable for the observations. During this year, as mentioned earlier, many firms’ performances were affected by the financial crisis and rating agencies might have reacted as response to the shrinking stock prices. After the downgrade events in 2008 interestingly enough, positive CARs were observable within the short-term window, in particular a highly statistical significant CAR of 4.88% and 2.92% for the event window (+2 , +10) and (+11 , +30) were observable, respectively. It is interesting as the stock returns obviously turns negative until the credit rating downgrade occurs which then turns the stock return positive afterwards. It seems, as credit rating downgrades kind of offers market participants such information that they no longer believe the stock is overvalued and they start to buy the stock instead of selling it, which eventually increases the return.

For the years 2009 and 2010, only 2009 shows some statistically significant CARs related to credit rating downgrades. The number of observations for this year is similar to 2008 and particularly high with 85 observations for downgrade events. A similar path to 2008 is noticeable which indicates positive CARs after the downgrade event, positive CARs of +2.00% and 8.41% for the event window (+2 , +10) and (+11 , +30), respectively. In contrast to 2009, 2010 shows barely statistically significant reactions or at least not such a significant reaction which could give some insights.

In 2011 and 2012, one can observe that here again obviously some events were taking place that might have affected firms’ performance across all industries. For 2011, no statistical significant reaction is observable whereas the contrary is true for 2012. That year was characterized with the European economic crisis. For the event windows

(36)

(-10 , -2) and (-1 , 0) highly statistically significant CARs of -6.55% and -3.59% were observable, respectively. This happened right before the downgrade event during the European economic crises in 2012. Also here, during the post event window (+11 , +30) one can observe a highly statistically significant positive CAR of 4.59%, which is relatively similar to the CARs noticed during 2008 and 2009. For the remaining years, only 2014-15 showed statistically significant CARs right before (-1 , 0) and after (0 , +1) the downgrade event with a negative CAR of -2.24% and 2.95%, respectively.

All in all, the results shown here in this section indicate that especially downgrade events might have been affected by other events that were happening at the same time.

4.8 Cross-Sectional Analysis

The following section deals with cross-sectional tests analyses, which should pro-vide a more complete picture of event-related tests. In order to get an understanding about the necessity of a cross-sectional analysis, a quick explanation will be provided. A cross-sectional test basically examines how stock return reacts due to credit rating changes with respect to firm’s characteristics. In regards to this research one would essentially regress the mean CARs for the different event windows against the firm char-acteristics. Usually, abnormal returns vary cross-sectionally which is due to the different economic effect which differs from firm to firm. For instance, when considering the inde-pendent variable ”Number of Analyst”, one should expect that events are more expected or predictable for firms which are being closely covered, in other words which are being covered by more analysts. If it turns out that unexpected information was affecting stock returns, this might have various consequences. For instance, standard estimates of cross-sectional coefficients might be biased (Eckbo, Maksimovic and Williams 1990).

4.8.1 Downgrade Events

Now, starting with the regression analyses. For the multivariate regressions con-ducted for this research study, the robust command was used in order to mitigate the effect of outliers and heteroscedasticity. With robust standard errors, the regression method ignores outliers and follows the true trend of the majority of the data.

First, the focus is on the overall downgrade events, in particular table A.12 and A.13. During the event window (-30 , -11) the ”bid-ask spread” (BAS) variable seems statistically highly significant at a 1% level. An increase of one unit in the BAS seems to reduce the CAR for this event window by -0.397. BAS is a measure of liquidity and

(37)

this result might indicate that if BAS increases, hence illiquidity, stock return decreases. Interestingly, considering the event window (+2 , +10) the BAS turns positive. In terms of financial markets interpretation it makes more sense that the value is positive as a liquid stock should reflect information faster compared to an illiquid stock. A more liquid stock indicates that it is more traded hence information will be reflected in the price way faster. Illiquid stocks have higher spreads since market participants disagree more about their respective fundamental value, which might be due to a lack of information. Thus, the more information is available, the tighter the spreads, the higher the liquidity (Ang, Shtauber, Tetlock 2013). Furthermore, it is observable that the ”Number of Analyst” variable is statistically highly significant for the post-event windows (+1 , +255) and (+1 , +127) with a negative value of -0.0250 and -0.0145 for each additional unit of ”Number of Analyst”, respectively. This basically means that the more a firm is covered by Analysts the lower the CAR. This makes sense as the closer a firm is covered, the more information is available and the faster new information is elaborated into the stock price. Walking further through the table A.12 and A.13 one can see in event window (-10 , -2) a statistically significant value for the ”Return on Asset” (ROA) variable. This makes sense as a higher return on asset should be attractive for investors. This should increase stock return.

The regression results for financial firms can be obtained in table A.14 and A.15. The first thing that needs to be mentioned is that the R2, which indicates how much of the variance is explained by the variables, is quite high compared to the previous overall table A.12 and A.13. In fact, the results are slightly similar for the BAS variable, meaning that before the rating event it is significantly negative and it turns significantly positive after the rating event. The ROA variable shows a statistically significant positive value before the downgrade event but it turns negative after the downgrade event, e.g. the event window (+2 , +10). The relationship exists because companies with high ROA have more liquid stock, causing the value of actual return on stock to increase hence lowering CAR. For the event window (-255 , -1) the variable ”Dividend per Share” (DPS) is highly significant for explaining the CAR for financial firms in that event window. This basically means that firms that pay high dividends face lower CAR. When share demand increases, the CAR reduces and higher reduction is experienced in companies with higher dividend and lower in companies with lower dividend per share. Considering the ”Number of Analyst” variable, it seems that a similar statistical significant negative value for the long-term post-event window (+1 , +255) is observable.

For the non-financial firms in table A.16 and A.17 similar reactions are observable regarding the ROA, Number of Analyst, DPS and BAS. Differences occur in terms of the R2 figure, which is significantly lower. Further, DPS variable seems to be highly

(38)

statistical significant with a slightly negative value in the long-term before the event takes place, in particular for the event windows (-255 , -1) and (-127 , -1).

The results in table A.18 A.19 A.20 A.21 are separated into two groups, namely small and large firms. Comparing these two groups, the major differences that are observable are amongst others that CAR of smaller firms seem to have stronger and a more sensitive relationship to BAS compared to larger firms. That makes sense as larger firms usually have a smaller BAS because of their higher trading volume than smaller firms. Also the ROA for smaller firms seem to be more statistically significant compared to larger firms. In fact, smaller firms tend to have a higher ROA compared to larger firms (Lafrance 2012). Furthermore, the R2 for larger firms is higher than for smaller firms. This indicates that more of the variation can be explained for larger firms than for smaller firms.

4.8.2 Upgrade Events

The following subsection analysis the CAR regression results of upgrade events. First, this section analyses the overall effect which is presented in table A.22 and A.23. The only variable that appears statistically significant and similar to the downgrade events here is the ”Number of Analyst” variable for the long-term post event windows (+1 , +127) and (+1 , +255). This variable shows similar negative values which indicates a decrease in CAR by an increase in number of analysts.

The next results within the next tables A.24, A.25, A.26 and A.27 are between financial and non-financial firms. Similar to the upgrade results, the financials seem to have a relatively high R2. Comparing the financials with non-financials it seems that before the upgrade event for the event window (-30 , -11) BAS is statistically highly significant for financials compared to non-financials. But the change (-0.0003) on CAR by one unit of BAS is too low as this would barely have an effect. Therefore, it is negligible for the moment. On the long-term post-event window (+1 , +127) BAS turns statistically significant positive for financials, similar to downgrade events, and indicates an increase of 0.6868 in CAR for one unit of BAS. On the other hand non-financials for this specific event window show statistical significant results for the variables ”Firmsize” and ”Number of Analyst” for the post-event windows (+1 , +255) and (+1 , +127).

The last tables A.28, A.29, A.30 and A.31 present the results of the regression analysis between large and small firms. The major differences between large and small firms are specifically in the long-term post-event window (+1 , +255) where DPS and BAS are highly statistically significant for large firms whereas for small firms only the ”Number of Analyst” seem to be statistically significant for this post-event window. In

(39)

general, BAS seem to play a more crucial role for the upgrade events, more specific for larger firms compared to smaller firms.

Overall, separating the downgrade and upgrade events in different groups and conducting multivariate regression analyses helped this research to give more insight of how the individual CARs are affected by firm characteristics. Not all of the variables included in the regression were significant. However, some variables show essential sig-nificant reactions. Most interesting is the effect between large and small firms, as the results given provide a foundation to further work on.

How significant is the informational content of credit rating announcements

Amsterdam Business School

Master Thesis