The Vasicek Credit Risk Model : A Machine Learning Approach

This paper explores the ability of the Machine Learning (ML) techniques to calibrate models that replicate the outputs of the Vasicek credit risk model. This model measures the loss distribution of a portfolio made up of loans that can be exposed to multiple systemic factors and it is widely used in the financial sector and by regulators. Under some assumptions this model provides a closed-form expression of the loss distribution but, in the general case, it requires computationally demanding Monte Carlo simulations to estimate this distribution. The ML approach only requires an initial calibration process and our results show that, for different portfolios, we can replicate outputs with a high degree of accuracy. Using just two variables, the confidence level and a Gaussian copula based loss distribution estimate, the tree based models provide quick and accurate estimates of the real loss distribution. This is the case for granular and also for concentrated portfolios