Similarity failure proximity score: A network-based metric for bankruptcy prediction

• Introduce the Similarity Failure Proximity Score (SFPS) network metric • SFPS combines graph embeddings with clustering to encode similarity structure • SFPS outperforms traditional benchmarks for company bankruptcy prediction • SHAP identifies SFPS as the strongest predictor among all features This study proposes the Similarity Failure Proximity Score (SFPS), a network-based metric that improves bankruptcy risk identification by capturing a company’s governance-linked proximity to bankrupt companies. SFPS translates relational structure in board-interlock networks into company-level failure proximity scores, reflecting structural similarity to bankruptcy-prone companies in the governance network. Using Slovak company data from 2016 to 2023 and annual corporate governance networks, we show that SFPS consistently enhances out-of-sample bankruptcy discrimination across seven machine learning classifiers, outperforming benchmarks based on financial ratios, non-financial variables, classical network measures, and uniform proximity proxies. The economic value of SFPS is operational: it provides a rankable early-warning signal for banks, regulators, and credit analysts. By targeting screening, monitoring, and intervention toward companies structurally close to bankrupt companies that standard accounting indicators may miss, SFPS can help reduce monitoring costs and expected credit losses. This is particularly relevant in low-transparency, governance-intensive environments where financial reporting can be delayed or incomplete and where distress may propagate through shared decision-makers. Results are robust across alternative similarity specifications and across director-centric and company-only network constructions. Overall, SFPS contributes to network-informed risk models by exposing latent vulnerabilities that complement Basel-aligned and financial-ratio-based approaches in credit screening and systemic oversight.