Advancing credit risk analysis

Researchers at the Business School and School of Mathematics developed a credit risk analysis model based on a rapidly evolving field of statistical methodology, known as “joint models for longitudinal and time-to-event data”. This is the first work that uses joint models in discrete time for credit scoring.

But what is “joint models for longitudinal and time-to-event data” and why does it matter?

These models address the problem of endogeneity by modelling both the time-to-event and the endogenous time-varying covariates, simultaneously. Endogeneity is present when the event's occurrence influences the explanatory variable. For example, a borrower's loan repayment, where we know that its path is related to the occurrence of the default. Such an innovative approach offers not only increased accuracy but also enables dynamic risk prediction, that is, to update predictions given new data.

The researchers applied the model to US mortgage loan data, with the largest sample size used in the literature on joint models. They concluded that this approach could increase discrimination performance compared to the discrete survival model. Moreover, when the variable whose values can be partially described by their past observations and handled by autoregressive terms, this performance can be further improved.

The computational resources for the completion of this study were provided by the Edinburgh Compute and Data Facility (Eddie) using the CmdStan interface with 4 CPU cores of 16GB of memory. Unfortunately, due to the high computational cost of this approach, it is yet to be adopted by financial institutions. Some approximations in the estimation can be a potential solution to this problem.

Medina-Olivares, V., Calabrese, R., Crook, J., & Lindgren, F. (2022). Joint models for longitudinal and discrete survival data in credit scoring. European Journal of Operational Research. https://doi.org/10.1016/j.ejor.2022.10.022

Why don't you explore featured projects demonstrating the use of similar resources and related training opportunities? Have a look at the carousels below.