3. Research methodology
3.1 Research question and data collection methods
The research design in this dissertation aims to investigate: Constraints on the ability of credit rating agencies to anticipate economic crises.
The question was triggered by authors such as Rousseau (2006) who suggested a connection between incentives and higher solicited ratings and thus the failure of credit rating agencies to warn investors of imminent crises, so this paper initially set out to determine the reasons. It was later found that Gan (2004) and Poon and Firth (2005) more than ably answered the question of credit rating agencies and solicited versus unsolicited ratings. Both authors agreed that credit rating agencies do give higher ratings to solicited securities, but the reasons for this have little to do with incentives. Rather the reasons are twofold. First, unsolicited ratings are lower because credit rating agencies are more cautious when rating unsolicited securities because of lack of information. Second, securities with better private information tend to self-select into the solicited group of securities and hence the higher solicited ratings.
While Poon and Firth (2005) and Mukhopadhyay (2006:209) give a reason why unsolicited ratings are lower than solicited ratings, they still do not properly explain why credit rating agencies have failed to warn investors of imminent crises. In other words, if incentives are not the reason why credit rating agencies have failed to warn investors of imminent crises, what then is the reason?
37 This paper will take the views of the above authors (Gan, 2004; Mukhophyay, 2006;
Fairchild, et al., 2009; Poon and Firth, 2005) as correct and build on them. The reason the views of the above authors are taken as correct is that the only evidence that authors such as Rousseau (2006) give for incentives unduly influencing credit ratings is higher solicited ratings, and the authors above give a reason for higher solicited ratings and the reason is not incentives.
To provide alternative reasons (besides incentives) as to why credit rating agencies have failed to predict crises, a data set with particular characteristics (expanded on later) needs to be examined. In order to determine what these characteristics are, the data collection methods used by Poon and Firth (2005) as well as Gan (2004) will need to be discussed. This is as the data collection methods and reasons in these papers give clues as to which characteristics the data needs to have in order to answer the research question.
The data collection method of Gan (2004) was to compile a list of securities from the SDC Platinum Database for the period 1 January 1994 to 31 December 1998. The list of securities included information about each security such as: the first date the security was rated; the initial rating; the securities‟ SEC number (the paper is a US paper); and the initial pricing of the security.
After the data were compiled, Gan (2004) used the SEC number to determine whether the security was solicited or unsolicited. The SEC requires all public companies to register certain documents with them. Included in these documents is what they call the S-1, S-2, S-3 or S-4 SEC registration document. In one of these S-documents of a company, would be what they term an item 14 entry. This item 14 entry lists expenses that the company has incurred. These include items such as accounting fees, SEC registration fees, legal fees and the actual rating agency fees a company has paid.
Once all this information was collected Gan (2004) ran an ordered probit model to see whether rating agencies behaved differently towards solicited ratings versus unsolicited ratings. The study found that rating agencies did give lower unsolicited
38 ratings even though there was not much difference between the performance of solicited and unsolicited securities. The author also found that when public information about a company is held constant, then those companies with better private information will self-select into the solicited rating group because they know that by revealing their superior private information they will get a better rating. Note that public information is used in unsolicited ratings, while both private and public information is used in solicited ratings. The conclusion by Gan (2004) then should be interpreted with care; what it says is that solicited ratings are higher than unsolicited ratings, but only because companies with good private information happen to request solicited ratings.
Poon and Firth (2005) examined a different period. The authors examined Fitch ratings of banks as of 1 January 2002. The Poon and Firth (2005) paper also used a probit model to determine their result, which was that unsolicited bank ratings are lower, but that weaker banks tended to be those that received unsolicited ratings.
The ratings data used in the Poon and Firth (2005) paper was representative because Fitch‟s market share of bank ratings in Africa, Asia, Central Europe, Eastern Europe and Latin America is almost twice that of Moody‟s and Standard and Poor‟s.
In Australia, North America and Western Europe, Fitch‟s market share of bank ratings is equal to that of Standard and Poor‟s and Moody‟s. Also the authors used Bankscope data to control for the financial performance of the banks they studied.
The important aspect of the papers discussed above is that they give an idea of the characteristics of the data that is required in order to examine the rating behaviour of credit rating agencies. The first requirement is that there should be a way to divide the securities studied into various groups in order to compare them. In the case of the above papers the two groupings were solicited and unsolicited ratings. In this paper the different securities should be divided into different sectors and time periods in order to examine the rating agency‟s behaviour when rating the different securities at different times.
A second requirement is that the securities examined must have some sort of rating history available. For instance the data collected by Gan (2004) had information such as a security‟s first date of rating as well as what that first rating was. This
39 paper uses defaulted securities and the securities have some rating history such as the date of first rating and what the rating was as well as the last rating before default and what that rating was.
A third requirement is that a method must be there to observe the financial performance of these securities. To observe the financial performance of the securities studied Gan (2004) used the pricing information of the securities examined and Poon and Firth (2005) used data from the Bankscope financial database to get financial ratios for the banks they were examining. The reason the papers by Gan (2004) and Poon and Firth (2005) needed the financial information was to control for financial performance (fundamentals) when comparing whether solicited ratings are rated better than unsolicited ratings. In the case of this paper all the securities examined have defaulted and would therefore have been bad investment decisions, so in a way financial performance has been controlled for. In other words, there are no securities that have performed well and others that have performed badly in the dataset - they have all performed badly.