Ensemble methods for bankruptcy resolution prediction: a new approach

Abstract

When a company goes bankrupt, it generates an extremely important uncertainty for all stakeholders as to whether the company will be reorganized or liquidated. This study aims to provide a successful methodology to predict whether a bankrupt SME will reorganize or liquidate. This could prevent significant economic and social losses and would contribute to reduce the number of SMEs that are helped to reorganize when they have little chance of success or that are liquidated when they could be viable. This useful and valid methodology applies algorithms (e.g., k-nearest neighbors) and techniques of ensemble learning and performance evaluation algorithms for the first time, considering the reviewed literature. By applying this methodology, it is possible to achieve a performance far superior to that known in the literature, specifically with an average accuracy of 94 percent using a data set with only financial variables of 1683 Spanish SMEs in the period 2011–2019