Improving the accuracy of credit scoring models using an innovative Bayesian informative prior specification method

Abstract

A new Bayesian informative prior specification method (BAF method – Bayesian priors using ARIMA forecasts) is proposed to introduce additional information into credit risk modelling and improve model predictive performance. We use logistic regression to model the probability of default of mortgage loans comparing the Bayesian approach with various priors and the frequentist approach. But unlike previous literature, we treat coefficient estimates in the probability of default model as stochastic time series variables. We build ARIMA models to forecast the coefficient values in future time periods and use these ARIMA forecasts as Bayesian informative priors. We find that the Bayesian models using this prior specification method produce more accurate predictions for the probability of default as compared to frequentist models and Bayesian models with other priors.