A method to incorporate transition risk stress testing into probability of default (PD) models for retail portfolios

Climate risk is one of the type of risks in a bank's portfolio which is not fully recognized, and its impact on the future overall risk changes is hidden due to lack of sufficient knowledge at the moment. One of the most common data comes from Network for Greening the Financial System (NGFS) scenarios related to climate change (physical risk) and climate policy and technology trends (transition risk). In the paper we focused on the transition risk scenarios and their impact on the economy and in particular on credit risk. Our main goal was to check the tendency in the probability of default (PD) default prediction in relation to climate risk potential future scenarios. We used data related to credit risk observed in Southern Europe banks for mortgage products for the years 2003-2019. Based on PD models we predicted the changes in the PD parameter over many years ahead by considering the set of scenarios collected in NGFS data. We selected the two scenarios 'carbon tax revenue from the residential and commercial sector' and 'electricity price at the final level in the transportation sector' for building the final models. From the PD logit model and linear predictors for the PD model we found that the main determinants predicting PD correlating with NGFS scenarios are LTV, customer income, unemployment rate, and crude oil prices. The quality of univariate models is above average, and the quality of the PD model is on an average level. The proposed models can be used in banking as stress tests in climate risk managemen