Promontory Financial, New York
In this issue we present two full-length research papers and two technical reports. We are pleased to include two papers co-authored by very prominent contributors in the field: Edward Altman and Frank Fabozzi. The first paper, "Corporate bond defaults are consistent with conditional independence", is by Kramer. The paper reconsiders the results in Das, Duffie, Kapadia and Saita (2007), which cast doubt on the empirical validity of the conditional independence assumption in the widely used doubly stochastic model of credit risk. The authors show that the result of Das et al is not robust to alternative econometric specification of the default intensity. The results in this paper suggest that the conditional independence assumption may be "rescued" by more careful choice of conditioning variables. This is an important contribution as well as a relief to practitioners, since the conditional independence assumption has come to be a mainstay in most current credit risk modeling practice.
The second paper, “Tests of the performance of structural models in bankruptcy prediction”, is by Fabozzi, Chen, Hu and Pan. The paper constructs a data set over a 20-year period from 1983 to 2002 for testing the ability of structural credit risk models to predict the event of bankruptcy, going on to develop a framework for testing several models against one another in a reasonably well-controlled setup. The paper assesses the relative performance of six structural credit risk models in predicting bankruptcy over various horizons. These six models are well-known structural models used in the literature: the Merton model, the Black–Cox model, the Leland–Toft model, the Longstaff–Schwartz model, the flat barrier model and the Geske model. The key predictive factor is equity returns and, while interest rates are not important, flexibility in determining the default barrier as differentiated by the various models is important. The comparison of the predictive performance of the various models is based on setting up a common characteristic: namely, the distance to default. The relative performance is judged using receiver operating characteristics curves rather than absolute predictions of default.
The next two papers in the issue are technical reports.A technical report describes a particular practical technique and enumerates situations in which it works well and others in which it does not. Such reports provide extremely useful information to practitioners in terms of saved time and duplication of effort. The contents of technical reports complement rigorous conceptual and model developments presented in the research papers and provide a lot of value to practitioners.
The first technical report, “A statistical modeling approach to building an expert credit risk rating system”, is by Waagepetersen. In this paper, the author outlines a method for developing a scoring model when historical data is unreliable or data is limited using an approach that mimics the decision behavior of experts. The paper presents statistical methods for characterizing the effect of financial measures of creditworthiness on the ratings given by experts. That is, experts rate credit risk based on some given financial variables, and the statistical model then captures the relationship between the financial measures and the credit rating. The statistical methods that are used and the conceptual developments given in the paper are straightforward and do not constitute significant advancement of a modeling technique, but the paper does provide practitioners faced with a lack of historical default data an illustration of how to go about the particular application of replicating expert ratings in such circumstances.
The second technical report, “The value of non-financial information in small and medium-sized enterprise risk management”, is by Altman, Sabato and Wilson. In this paper the authors estimate default probabilities from a very large data set of small and medium-sized enterprises (SME obligors) in the UK using the standard logistics specification. The particular objective of the paper is to ascertain the increase in the power of default prediction given by incorporating non-financial and nonaccounting variables into a default model applied to SMEs. The authors find that data relating to legal action by creditors to recover unpaid debts, company filing histories, comprehensive audit report/opinion data and firm-specific characteristics make a significant contribution to increasing the default prediction power of risk models built specifically for SMEs. The paper applies econometric models published in existing literature and it does not present significant new insights into the area of conceptual credit risk measurement. However, there is a paucity of published literature on default and loss-given-default modeling specifically forSMElending, in contrast to corporate lending. While SMEs constitute the majority of bank obligors, full historical financial information is not readily accessible for these private unrated companies, nor is the information available in a form that is consistent across companies. For practitioners engaged in modeling the credit risk of SME obligors, the case study presented in this paper provides valuable suggestions for improving their modeling through the incorporation of non-financial variables.