Predicting forward default probabilities of firms: a discrete-time forward hazard model with firm-specific frailty

Abstract

Predicting the corporate default probability accurately is the core of credit risk management. There has been a relatively small amount of the literature on predicting a firm’s forward default risk. In particular, we wish to emphasize certain features of the panel data that are often overlooked in the analysis of default forecasting. First, the panel data are observed at discrete-time points with a large unit of time such as month, quarter, or year. Second, repeated survival status outcomes from the same firm are highly correlated. Thus, the continuous-time treatment or an independence assumption is often violated in practice. To avoid these potential drawbacks, we propose an extension of the discrete-time forward hazard model by assigning a frailty variable specifically to each firm. We use a real panel dataset to illustrate the proposed methodology. Using the dataset, our results first support the significance of including the firm-specific frailty variable in the extended model. Then, using an expanding rolling window approach, our results confirm that the extended model provides better and more robust out-of-sample performance than its alternative without frailty. Thus, accounting for firm-specific frailty can consistently yield more accurate predictions of firms’ forward default probabilities.

Keywords:

• Discrete-time forward hazard model
• Expanding rolling window approach
• Firm-specific frailty
• Forward default probability
• Maximum marginal likelihood
• Number of defaults

JEL Classifications:

• C53
• G20

Acknowledgements

The authors thank the reviewers for their valuable comments and suggestions that have greatly improved the presentation of this paper. The National Science and Technology Council of Taiwan provides support for this research.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by National Science and Technology Council [Grant Number NSTC 112-2410-H-259-030].