Predicting Business Failure

Being able to make an objective assessment of a firm’s probability of getting into distress and eventually failing is of great importance to a large number of economic agents. Because of its importance to this wide array of practitioners, it is not surprising that business failure or bankruptcy prediction is one the most extensively researched topics in the corporate finance and financial accounting literature. Although some earlier examples of academic studies and practical applications exist, the modern era of failure prediction is generally considered to have started in the 1960s, with relatively straightforward analyses of bankruptcy likelihood for small samples of stock exchange quoted companies in the United States. Ever since, researchers have attempted to improve the performance and the validity of prediction models in various ways. One important source of improvement has been the application of more sophisticated statistical methods, from univariate tests and discriminant analysis in the 1960s and 1970s, over logit and probit regressions in the 1980s and 1990s, to the more recent hazard models and artificial intelligence techniques. Moreover, a better understanding of the potential causes of distress and failure has led to the inclusion of many new predictors, ranging from more complex accounting ratios and market variables to macroeconomic and industry-related measures and non-financial information (e.g., company ownership or managerial characteristics). Finally, due to ever-better data availability, many studies have also widened the scope of prediction in terms of the type of companies that are studied (e.g., models for private companies or SMEs), in terms of geography (in both developed and emerging markets), and in terms of the definition of failure (i.e., not only formal bankruptcy but also reorganization procedures, voluntary liquidations, and company sales). The continuous search for improvement has led to a vast and rich literature consisting of hundreds of published prediction studies. It goes without saying that this chapter can never hope to give a complete discussion of this body of work, which means that it should be read as a primer to the main insights and developments in the field and as a starting point in exploring the literature. We predominantly include studies that have been highly influential and papers published in high-quality business and economics journals. Our discussion is non-technical in nature, meaning that we focus on the intuition behind the different methodologies and approaches and refer to the original sources for the mathematical and statistical details.status: publishe