Bank loans recovery rate in commercial banks:A case study of non-financial corporations

The empirical literature on credit risk is mainly based on modelling the probability of default, omitting the modelling of the loss given default. This paper is aimed to predict recovery rates on the rarely applied nonparametric method of Bayesian Model Averaging and Quantile Regression, developed on the basis of individual prudential monthly panel data in the 2007–2018. The models were created on financial and behavioural data that present the history of the credit relationship of the enterprise with financial institutions. Two approaches are presented in the paper: Point in Time (PIT) and Through-the-Cycle (TTC). A comparison of the Quantile Regression which get a comprehensive view on the entire probability distribution of losses with alternatives reveals advantages when evaluating downturn and expected credit losses. A correct estimation of LGD parameter affects the appropriate amounts of held reserves, which is crucial for the proper functioning of the bank and not exposing itself to the risk of insolvency if such losses occur.