Evolutionary approaches for estimating a coupled markov chain model for credit portfolio risk management

Abstract

The analysis and valuation of structured credit products gained significant importance during the sub-prime mortgage crisis in 2007. Financial companies still hold many products for which the risk exposure is unknown. The Coupled Markov Chain approach can be used to model rating transitions and thereby default probabilities of companies. The likelihood of the model turns out to be a non-convex function of nthe parameters to be estimated. Therefore heuristics are applied to find the ML estimators. In this paper, we outline the model and its likelihood function, and present a Particle Swarm Optimization algorithm, as well as an Evolutionary Optimization algorithm to maximize this likelihood function. Numerical results concluDe the paper. © Springer-Verlag Berlin Heidelberg 2009.