Title: A study of credit risk model robustness through a crisis
Authors: Anthony Bellotti - University of Nottingham Ningbo China (China) [presenting]
Abstract: The COVID-19 crisis has led many organizations to worry about how their predictive models will perform during the changing social and economic environment. Some financial institutions have reverted to simpler models or human judgement, avoiding more complex models that they consider less robust to change. In light of the uncertainty, others have applied conservative decision making such as lowing credit limits. In an attempt to quantify the effects of dramatic change in population caused by crisis on credit risk models, we conduct an empirical study based on models built before and during the last financial crisis in 2009, using the Freddie Mac loan-level data for mortgage originations from 2004 to 2016. We compare credit risk models using logistic regression with more complex models, such as random forests and artificial neural networks, to determine how they perform during and after the financial crisis.