Dr. Nikolay Dobrinov, Senior Data Scientist, Deep Future Analytics
ABOUT THE PRESENTER
Dr. Nikolay Dobrinov has over 11 years of experience in credit risk analytics and specializes in forecasting and stress testing the performance of retail and commercial loan/line portfolios. Nikolay has developed numerous credit risk models, as well as performed independent reviews of such models, for both internal planning and regulatorypurposes (CCAR, DFAST, CECL) for banks and fintech companiesof all sizes.
How can we quantify model selection risk and what are it’s impacts on loss reserves and economic capital?
A large set of model specifications, that performs similarly in-sample, but may have distinct paths out of sample, may be available to choose from even within a single modeling framework. Dr. Dobrinov shows our latest research into various measures to quantify the model selection risk, its magnitude in credit risk modeling and its impact on loss reserves.The results highlight two sources of risk: a systematicdifference between the expectation value of the lifetime loss forecast and the medianof its simulated distribution that should be incorporated in the calculation of the loss reserves, and the 95% confidence interval from thedistribution which should contribute to the model risk component ofthe economic capital.