Machine Learning or Econometrics for Credit Scoring : Let's Get the Best of Both Worlds

In the context of credit scoring, ensemble methods based on decision trees, suchas the random forest method, provide better classi cation performance than standardlogistic regression models. However, logistic regression remains the benchmark in thecredit risk industry mainly because the lack of interpretability of ensemble methodsis incompatible with the requirements of nancial regulators. In this paper, we pro-pose to obtain the best of both worlds by introducing a high-performance and inter-pretable credit scoring method called penalised logistic tree regression (PLTR), whichuses information from decision trees to improve the performance of logistic regression.Formally, rules extracted from various short-depth decision trees built with pairs ofpredictive variables are used as predictors in a penalised logistic regression model.PLTR allows us to capture non-linear e ects that can arise in credit scoring datawhile preserving the intrinsic interpretability of the logistic regression model. MonteCarlo simulations and empirical applications using four real credit default datasetsshow that PLTR predicts credit risk signi cantly more accurately than logistic regres-sion and compares competitively to the random forest method