Journal of Credit Risk

Risk.net

Estimating credit risk parameters using ensemble learning methods: an empirical study on loss given default

Han Sheng Sun and Zi Jin

  • This study investigates two well-established ensemble learning methods: Stochastic Gradient Boosting and Random Forest, and proposed two new ensembles.
  • An ensemble approach can bring significant increase in the model discriminatory power comparing to a single decision tree. 
  • The methods can be directly applied to predicting the exposure at default and probability of default with some simple modifications. 
  • The proposed approaches introduce a novel modelling framework to estimate and validate credit risk parameters based on the internal data of different portfolios.

ABSTRACT

In credit risk modeling, banks and insurance companies routinely use a single model for estimating key risk parameters. Combining several models to make a final prediction is not often considered. Using an ensemble or a collection of models rather than a single model can improve the accuracy and robustness of prediction results. In this study, we investigate two well-established ensemble learning methods (stochastic gradient boosting and random forest) and propose two new ensembles (ensemble by partial least squares and bag-boosting) in the application of predicting the loss given default. We demonstrate that an ensemble approach significantly increases the discriminatory power of the model compared with a single decision tree. In addition, the ensemble learning methods can be applied directly to predicting the exposure at default and probability of default with some simple modifications. The proposed approaches introduce a novel modeling framework that banks and other financial institutions can use to estimate and validate credit risk parameters based on the internal data of different portfolios. Moreover, the proposed approaches can be readily extended to general portfolio risk modeling in the areas of regulatory capital and economic capital management, loss forecasting, stress testing and pre-provision net revenue projections.

 

To continue reading...

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an indvidual account here: