Classical and quantum computing methods for estimating loan-level risk distributions

Abstract

Understanding the risk distribution for a single loan due to modelling and macroeconomic uncertainty could enable true loan-level pricing. To implement such analysis, rapid real-time methods are required. This article develops a Monte Carlo method for estimating the loss distribution on classical computers. For this analysis, multihorizon survival models incorporating the competing risks of default and pay-off were created on Freddie Mac data. The Monte Carlo simulation results fit well to a Lognormal distribution. Knowing in advance that a lognormal distribution is a reliable fit means that only tens of Monte Carlo simulations are needed to estimate tail risk in a single account forecast. We also explored the feasibility of using quantum computers that may be available in the near future to perform the calculations. Leveraging previous quantum algorithms for value-at-risk estimation, we show that a simplified version of the problem has potential for significant speed enhancement, but the full competing risks approach will require additional quantum algorithm development to be feasible.