Finance-Insurance
Time: 14.00 pm
Date:13th of January 2025
Room 3049

Arnaud GERMAIN (Univ. Catholique de Louvain) “Cluster aggregating: application to Early-Warning System for Non-Performing Clients”

Abstract : We introduce a new ensemble learning strategy called clagging (for cluster aggregating), which consists in combining models fitted on different clusters. First, we divide the training set into k clusters with k=1,..,K. Next, we fit a model on each of those 1+2+…+K clusters. Finally, the output for a given test point is obtained by combining the predictions using the distance of the test point to the clusters’ centroids. We perform an extensive horse race study, considering both regression and classification tasks. Our results suggest that clagging outperforms bagging, where a bootstrapped sample is traditionally created by drawing observations with replacement until the size of the bootstrapped samples coincides with the size of the original training set. Clagging can also improve the performance compared to a standard fit on the whole training set.

In addition, we apply clagging in the context of default prediction in finance. In its “Guidance to banks on non-performing loans”, the European Central Bank requires banks to implement an Early Warning System to identify potential non-performing clients at a very early stage. Relying on a unique dataset provided by a systemic European bank, we show that clagging boost the out-of-sample performance compared to a case where we fit a single prediction model on the whole dataset and a case where we rely on domain knowledge to determine the clusters.

Organizers:  Jean-David FERMANIAN