A self tuning model for risk estimation

Credit scoring models often use linear or logistic regression to investigate the relation between observed characteristics and credit ratings. The basic relation is, however, a form of Bayes’ theorem. This paper proposes a model in which estimation techniques from hidden Markov models are adapted to evaluate the parameters of a risk profile. The risk being estimated might be financial, as in credit scoring, or alternatively whether an observed member of a population might represent some terrorist threat.