Assessing financial distress of SMEs through event propagation: An adaptive interpretable graph contrastive learning model

Abstract

Accurate assessment of financial distress of SMEs is critical as it has strong implications for various stakeholders to understand the firm's financial health. Recent studies start to leverage network data and suggest the effect of event propagation for predicting financial distress. Yet such methods face methodological challenges in determining and explaining event propagation due to heterogeneous entities and events. In this research, we propose to extend graph contrastive learning and interpretable machine learning in the context of a firm network formed by distinct entities (e.g., firms and persons) and events (i.e., positive and negative), and employ the propagation influence of events in firm networks for financial distress assessment of SMEs. To this end, we design a novel artifact, i.e., adaptive interpretable heterogeneous graph contrastive learning, by drawing on homophily and social learning theories. Our experimental results demonstrate the effectiveness of the proposed artifacts and suggest the differing effects of positive vs. negative events on the financial distress of SMEs. This research contributes to the IS and explainable graph AI literature by improving the assessment and interpretability of network-based financial distress of SMEs.