Explainable Artificial Intelligence for Predicting Corporate Financial Distress Using Dynamic Cash Flow and Working Capital Indicators

Abstract

Corporate financial distress rarely emerges as a single-period accounting event; it develops through interacting liquidity pressures that progressively weaken a firm’s capacity to finance operations and meet contractual obligations. This study proposes an explainable artificial intelligence framework that detects this deterioration from the temporal structure of cash flow and working capital behavior. Rather than relying primarily on static financial ratios, the framework constructs dynamic indicators capturing operating-cash-flow persistence, cash-flow volatility, working-capital absorption, receivables acceleration, inventory accumulation, payable compression, cash-conversion-cycle deterioration, liquidity-buffer erosion, and deviations from firm-specific historical baselines. Machine-learning models estimate forward distress probabilities across defined prediction horizons, while temporal validation and cost-sensitive classification address changing financial conditions and asymmetric consequences of missed distress events. Explainability mechanisms decompose predicted risk into firm-level financial drivers and reveal whether deterioration originates from weakening cash generation, inefficient working-capital conversion, short-term financing pressure, or interacting liquidity constraints. By integrating prediction, temporal deterioration analysis, and interpretable risk attribution, the framework transforms financial-distress modelling from retrospective classification into an explainable early-warning system for identifying emerging liquidity vulnerability and supporting proactive corporate, lending, investment, and restructuring decisions.