Have Financial Statements Become Less Informative? Evidence from the Ability of Financial Ratios to Predict Bankruptcy

Abstract

Using a hazard model, we examine secular changes in the ability of financial statement data to predict bankruptcy over a forty-year period. We identify three trends in financial reporting that could influence predictive ability with respect to bankruptcy: the increase in FASB standards, many of which have a fair value emphasis, the perceived increase in discretionary financial reporting behavior, and the increase in the magnitude of unrecognized intangible assets and other unrecognized assets and obligations. A parsimonious three variable model provides significant explanatory power throughout the time period. In dividing the entire time period into two sub-periods, we find a slight deterioration in predictive power in the second time period. We also examine a predictive model using explanatory variables based on market values. We argue that if the market-based variables fully capture the information in prices about bankruptcy probability, this model is expected to dominate the financial statement based model and the difference can be interpreted as the incremental role of non financial statement information in bankruptcy prediction. We find that a combined model that uses both types of variables outperforms a financial statement only model but does not completely subsume these variables. In particular, the leverage variable remains significant even in the combined model. The evidence indicates there is an increase in the incremental predictive power of non-financial statement information in the second sub-period, where the financial statement data show a slight deterioration. The predictive power of the combined model is essentially the same in both time periods. This is consistent with the non-financial statement data compensating for a loss in the predictive power of the financial ratios. However, these differences are small and the striking feature of the results is the robustness of the predictive models over a long period of time.