Comparison of Financial Distress Prediction Models Using Financial Variables

This study delves into the critical realm of financial distress prediction and its implications for mitigating fraudulent behavior in corporate settings. The study's objectives encompass a comparative assessment of traditional statistical methods and cutting-edge machine learning algorithms to predict financial distress within companies listed on NYSE and NASDAQ. Leveraging a quantitative approach, the study uses data from 1577 companies for the year 2021, sourced from the FMP cloud database. Findings from hypothesis testing employing statistical techniques unveil an initial weak association between financial variables and the Beneish M Score, indicative of potential predictive power. However, the integration of diverse machine learning models highlights their potential to replicate this predictive capability. Notably, the Random Forest and XGBoost algorithms showcase substantial predictive power of predicting distressed and non-distressed firms with accuracy of 95.19% and 92.79%. These models also outperform all other machine learning models, distinguishing distressed from non-distressed instances with remarkable precision and recall rates.