Analysis of credit-rating migrations with genetic algorithms

Abstract

Modelling dependent credit-rating migrations of assets classified into M credit classes and S industries, M × S + 2M×S parameters have to be estimated. For a realistic choice of M and S, this number is huge and it greatly exceeds the number of available observations. To avoid brute-force calculations, we suggest sequential and parallel genetic algorithms. Considering a practically important combination of M = 7 and S = 6, the approach is tested on Standard and Poor's data.