An Explainable Early-Warning System for Financial Distress: Board Governance Capacity as a Predictive Signal

The identification of corporate financial distress using accounting information is adversely affected by high inflation because the information quality of accounting ratios decreases significantly during rapid changes in price levels. Therefore, decision makers require non-monetary warning signals that can provide relevant information without being affected by inflationary noise. In this study, we develop an explainable early-warning system by combining a formative Board Governance Capacity Index (BGCI) with eight accounting ratios to evaluate non-financial companies listed on Borsa Istanbul. We collected company data for the period 2014–2023, comprising 1,642 firm-year observations from 190 firms. Financial distress is defined as the presence of qualified or adverse audit opinions or negative book equity. The BGCI consists of five publicly reported dimensions: board size, board independence, percentage of female directors, percentage of foreign directors, and external audit quality. We evaluate the performance of six machine learning classifiers using firm-grouped cross-validation, isotonic probability calibration, out-of-time evaluation, and TreeSHAP for feature attribution. Incorporating BGCI improves out-of-sample performance for all six classifiers, with the AUC of the benchmark logistic regression model increasing by 0.067 points, from 0.644 to 0.711, and the calibrated Brier score decreasing by 0.054. Component ablation analysis confirms that BGCI performs better than individual governance attributes. BGCI has the highest mean out-of-sample Shapley value among the nine predictors (0.696) and increases the pooled out-of-time AUC from 0.671 to 0.719. Fixed-effects logistic regression estimates controlling for firm size, sector, and year effects indicate that a one-unit increase in BGCI is associated with a 4.3-percentage-point reduction in the predicted probability of financial distress. Partitioning predictions by macroeconomic regime indicates that the additional AUC gain provided by BGCI increases significantly from 0.016 during periods of relatively low inflation to 0.118 during periods of relatively high inflation, thereby improving the accuracy of early-warning signals when conventional accounting-based signals become less reliable.