earticle

영어

In the default prediction problem, the cost from the failure of forecasting defaults is much bigger than that of forecasting non-defaults. The cost asymmetry is deeper in the corporate default prediction than the retail as corporate loan portfolios are not granular. However, the two types of costs are treated equally in general as default prediction models are usually estimated to minimize prediction errors or maximize statistical performance. This practice might not fulfill the goal of risk management to minimize economic losses. To mitigate this issue, this study apply cost-sensitive learning approach to default prediction, which minimizes economic costs instead of statistical errors. We define economic costs and test them for various levels of the cost asymmetry by employing Logistic regression, XGBoost, and LightGBM. As a result of empirical experiments with Taiwanese and Polish corporate default data, we first find that the proposed cost-sensitive models are superior to the cost-insensitive counterparts in terms of economic cost, mostly regardless of the cost asymmetry scenarios. Secondly, nevertheless, the decreases in the statistical performance are relatively small – economic costs decrease 24.6% at the expense of the decrease in AUC of 4.6% on average. This suggests that financial firms can adopt the proposed default prediction models without violating the regulatory requirement on model quality. Lastly, we find that the features of high prediction power in the cost-sensitive and insensitive models are different, which has an important implication for credit monitoring.