An application to rate banks using a new variant of agglomerative clustering algorithm

The study aims to contribute to the field of credit ratings, by presenting models grounded on new variants of neighbourhood linkage method (NLM), an agglomerative hierarchical clustering technique. These models have been applied so as to analyse and predict the long-term bank credit ratings provided by an international rating agency. For this cause, the long-term ratings provided by an Indian arm of international rating agency have been considered. The dataset consists of 35 banks operating in India; this consists of 21 rated banks and 14 unrated banks. In order to validate the optimal number of cluster formations, the study uses a novel performance measure called 'modified structure strength'. Ultimately, based on the best performing NLM variant's cluster formations (of rated banks), unrated banks' potential ratings have been predicted. This model is agnostic to country or region and can be employed to forecast credit ratings of any bank across geography.