Improving SME default prediction through context-enriched classification models

In today’s rapidly evolving market, it has become imperative for companies to rely on efficient and effective corporate default forecasting systems to stay competitive. For a comprehensive understanding of market variations, it is crucial to consider both economic-financial information and contextual factors. This research presents an explainable perspective of predicting corporate defaults for small and medium-sized enterprises (SMEs), demonstrating the importance of incorporating contextual information to improve risk assessment. The study analyzes a dataset of Italian SMEs from the construction sector, considers the importance of sector-specific indicators, and evaluates various models using standard interpretable classification algorithms. The classifications are then compared with and without the inclusion of contextual data, and their performances are analyzed. The results of this study confirm that the integration of contextual data improves the performance of classification models, with improvements observed in metrics such as accuracy, AUROC, and specificity. Moreover, contextually enriched data contributes to a more even distribution of gains across variables considered, improving model robustness and limiting overfitting risk. This work contributes to enriching the existing body of literature on SME default prediction, particularly by emphasizing the role of contextual information in improving model explainability and performance.