Hybrid bankruptcy forecasting for Indian firms: Integrating financial ratios, macroeconomic indicators, and random forest

Abstract

Bankruptcy forecasting in emerging markets is complicated by macroeconomic and regulatory volatility. This study evaluates whether a hybrid model that integrates firm financial ratios, macro indicators, and a Random Forest classifier outperforms traditional ratio-only approaches for Indian firms. Each bankrupt company is analyzed over a five-year window preceding its actual failure date, resulting in ten bankrupt firms paired with ten matched healthy peers. Using these firm-specific five-year pre-bank-ruptcy panels, we estimate logistic regression and Random Forest models with strati-fied 5-fold cross-validation and derive a parsimonious four-factor risk score. Relative to ratio-only baselines, the hybrid design improves accuracy from 0.76→0.80 (logit) and 0.82→0.86 (Random Forest), and lifts the Area Under the ROC Curve (AUC) from 0.70→0.78, indicating that the model correctly ranks a bankrupt firm as riskier than a healthy firm 78% of the time. Debt-to-Equity, Current Ratio, Net Profit Margin, and GDP Growth dominate feature importance, and rising risk scores typi-cally cross ~0.40 two to three years before failure. Robustness checks, including alternative class-balance weights, sector dummies, and rolling-window estimation, yield comparable gains and stable feature rankings. The resulting bankruptcy Early-Warning System (EWS) is transparent, portfolio-scalable, and easily embedded into bank risk dashboards. The evidence shows that multidimen-sional hybrid models provide earlier and more reliable warnings than ratio-based for-mulas, offering practical value to lenders, investors, and regulators in volatile settings.