Bankruptcy Prediction: An Evolutionary Analysis of Models and Their Contextual Applications

Abstract

In today’s rapidly changing world, recent events like the COVID-19 pandemic have underscored the importance of effective risk management. Accurate bankruptcy predictions help anticipate and mitigate risks, as bankruptcies impact not only the failing firm itself, but also the society and economy on a larger scale. This thesis explores the evolution of bankruptcy prediction models from the 1960s to today and examines their applicability across different countries and economic contexts through a comprehensive literature review.

The thesis finds that bankruptcy prediction models have evolved from simple accounting-based approaches to hybrid models that integrate both accounting- and market-based information, enabling more comprehensive, real-time, and forward-looking predictions. Machine learning has further enhanced models by allowing them to manage more complex relationships between variables, reducing the limitations of parametric models, and improving accuracy. Incorporating qualitative variables into the models has refined predictions by considering a wider range of information when assessing a firm’s bankruptcy risk.

Models tend to perform best in economic environments similar to those in which they were developed, as predictors can vary significantly between open and closed economies or global and local industries. Crisis periods seem to reduce models’ prediction accuracy, but it remains unclear whether this decline occurs before, during, or after a crisis. Crises also alter the relevance of different predictive variables.

The thesis concludes that bankruptcy prediction models have evolved significantly since the 1960s, reflecting a broader attempt to capture the many causes and aspects of bankruptcy. However, simply adding more variables to a model does not automatically improve its accuracy, as the relevance of different variables depends on many contextual factors. The effects of country-specific factors and financial crisis on model performance remain complex, requiring future research.