Abstract:
|
It has been of interest to improve the default prediction models for risk management and credit portfolio pricing, especially after the financial crisis during 2007-2009. In this paper, we propose methods to predict default probabilities for companies and the number of defaults in the market based on large-scale time-to-event data and selected covariate information. We propose a competing risks model to incorporate exits of companies due to default and other reasons. To account for the variability in default risk over time, we model the firm-specific and macroeconomic covariate dynamics using a time series model. The residuals of time series are modeled by a dynamic factor model to capture the correlation among covariate processes. To estimate parameters in the covariate model, we derive the expectation maximization algorithm that allows for missing values in explicit forms. We also address the identification problems in the factor model by adding necessary constraints. For default predictions, we derive the point predictions as well as prediction intervals that synthetically take uncertainties in the parameter estimation and the future covariate process into account.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.