Relevance of loan characteristics in probability of default prediction for commercial mortgage loans

Abstract

The current papers examines the sensitivity of loan characteristics on mortgage default probability for UK commercial mortgages. Commercial real estate (CRE) mortgages are major asset holdings for commercial banks, life insurance companies and thrift institutions. The slumping market for real estate threatened to drag down regional banks and other smaller financial institutions in the 08/09 financial crisis and led to the collapse of some financial institutions. Despite the prominence of CRE mortgages, modeling and analysing credit risk of CRE mortgages has been lagging behind those of non-CRE commercial loans. Modelling the probability of default for commercial real estate mortgages is more complicated than that for non-commercial real estate loans. Many distressed loans passed traditional underwriting standards suggesting that, in addition to LTV and DSCR ratios, other characteristics should be taken into consideration such as the inclusion property characteristics. The accuracy of default prediction is tested comparing two traditional statistical methods a) logistic regression (logit) and b) multiple discriminant analysis (MDA) using a unique dataset of defaulted commercial loan portfolios from 60 financial institutions lending in the UK between 2005 – 2017. Overall, both models show that the inclusion of property characteristics such as geography and asset type have been significant factors in determining default probability and improve model accuracy, while LTV shows no clear significance.