Title: Copula-Heckit: Application of modified models with selectivity for loss amount prediction in case of bank failures
Authors: Henry Penikas - Bank of Russia (Russia) [presenting]
Abstract: A range of papers deals with the probability of bank default modeling. A separate stack of papers discusses bank failure case studies. However, some works do not limit themselves to the probability of failure prediction. They attempt to predict the loss amount in case of a bank failure. The Heckit model is used. It presumes a correlation of errors from the selection equation and the principal one. However, such a Pearson correlation coefficient does not capture the rich bivariate dependence patterns that might occur in the real-world. Copulas, including Archimedean ones, do allow for this. The objective is to demonstrate the advantage of modifying Heckit model to account for copulated errors, i.e. to use copula-Heckit model. We start with the simulated data and proceed with the empirical on. It is Russian bank license withdrawal data for 2013-2020. We show how copula-Heckit model outperforms the conventional one.