Reverse-engineering banks’ financial strength ratings using logical data analysis

We study the problem of evaluating the creditworthiness of banks using statistical, as well as combinatorics-, optimization-, and logic-based methodologies. We reverse-engineer the Fitch credit risk ratings of banks using ordered logistic regression and Logical Analysis of Data (LAD). It is shown that LAD provides the most accurate rating model. The obtained ratings are successfully cross-validated, and the derived model is used to identify the financial variables most important for bank ratings. The study also shows that the LAD rating approach is (i) objective, (ii) transparent, and (iii) generalizable. It can be used to build internal rating systems that (iv) have varying levels of granularity, allowing their use in the banks’ operations related to the credit granting decision process (pre-approval, determination of pricing policies), and (v) are Basel 2 compliant, allowing their use in the financing and provisional decisions pertaining to the determination of the amount of regulatory capital. The impact on the reduction of the credit and operational risk, the extended scope, and the monetary value of such a risk rating system are discussed