Overlapping Observations in Credit Risk Models

Abstract

Parameters in logistic regression models for probability of default are typically estimated using the maximum likelihood method. The aim of this paper is to verify whether the use of overlapping observations improves precision or causes deterioration of estimation results in these models. Our Monte Carlo simulations demonstrate that the difference between parameter estimates using all overlapping observations in a sample and only non-overlapping observations in a reduced sample is not statistically significant, but the variance of parameter estimates is reduced when overlapping observations are used.

Keywords

• PD Model
• Credit Risk
• Overlapping Observations
• Logistic Regression