External versus internal credit ratings: what are the implications for rating stability and accuracy?

-

The paper explores the trade-off between rating accuracy and stability - both theoretically and empirically. The most important and novel theoretical result is that there is not necessarily a tradeoff between rating accuracy and stability. That is, regression analysis on simulated data shows that volatile ratings are not always more accurate. Empirically, all of our findings are novel as we have been the first academics to use internal credit rating data from the Global Credit Data Consortium. Consistent with our theoretical findings, we do not find evidence of an accuracy-stability tradeoff in the data.

The objective of this project is to present a detailed and comprehensive analysis of banks' internal credit ratings. These ratings will be then compared against external ratings produced by the top external credit rating agencies (CRAs) and implications w ith respect to rating stability and accuracy will be assessed. The main dataset underlying the analysis is provided by the Pan-European Credit Data Consortium (PECDC), which covers aggregate industry ratings for a large sample of its member banks. The consortium hosts the world's largest database on bank's default probabilities and over the period from 2003 to 2012, it contains information on approximately 40,000 corporate defaults. We are the first researchers to have secured access to this unique dat abase. The dataset compiled by PECDC contains information on multi-period forward looking default probabilities (i.e. through-the-cycle estimates). This information is unique and it allows us to compare banks' estimates of default probabilities, observe d default frequencies and accuracy ratios to ratings on similar assets produced by the major CRAs. To our best knowledge, no other study has attempted to compare the trade-off between rating stability and accuracy for a large sample of internal and extern al credit ratings. The second part of the paper will investigate the determinants of banks' internal credit ratings. While this exercise will employ existing models of determinants of ratings, it will contribute to the literature due to its focus on inte rnal credit ratings. This difference is important because external credit ratings are often inflated as the issuer pays for the credit rating. This is not the case for our internal data and we will therefore contribute by providing unbiased estimates of t he impact of selected firm characteristics on credit ratings.