Financial Distress Prediction Models Based on Traditional Accounting Ratios

Abstract

This research presents a novel methodological framework for financial distress prediction
that challenges the prevailing paradigm of complex, multi-source predictive modeling. While
contemporary literature increasingly favors models incorporating market data, textual analysis, and macroeconomic indicators, this paper argues for a deliberate and sophisticated
re-engagement with traditional accounting ratios. Our originality stems not from the rediscovery of these ratios, but from a fundamentally new approach to their synthesis and
interpretation. We introduce the concept of ’Ratio Synergy Networks’ (RSNs), a graphtheoretic methodology that models the non-linear, interdependent relationships between
classic liquidity, profitability, leverage, and activity ratios. This approach moves beyond
treating ratios as independent variables in a regression model, instead capturing how the
predictive signal of one ratio is contingent upon the values of others, mimicking the holistic
assessment performed by expert analysts. Furthermore, we develop a temporal ’Ratio Trajectory Clustering’ (RTC) algorithm that identifies archetypal paths to distress, classifying
firms not just by their static financial position but by the dynamic deterioration pattern
of their ratios over a multi-year horizon. Applying this dual-framework to a comprehensive
dataset of U.S. public firms from 1990 to 2023, we demonstrate that a model built exclusively on carefully processed traditional ratios can achieve predictive accuracy that matches
or exceeds state-of-the-art hybrid models, while offering superior interpretability and robustness in economic downturns. The findings suggest that the diminishing returns of adding
novel data sources may be greater than previously assumed, and that significant latent predictive power remains untapped within the conventional accounting statement, accessible
only through more advanced relational and temporal analytics. This research contributes
a counter-intuitive yet empirically robust perspective to the financial forecasting literature,
advocating for depth over breadth in predictive feature engineering.