Financial distress: An evaluation of the predictive power of accrual and cash flow information using ordinal multi-state prediction models

The primary stream of research testing the importance of cash flow information has concerned the ability of accounting information to predict financial distress. Most prior financial distress studies have primarily used a dichotomous bankrupt/nonbankrupt response for financial distress. These dichotomous distress studies suffered from many limitations. The primary objective of this study was to better evaluate the ability of cash flows to predict financial distress by correcting for many of these limitations through the use of ordinal multi-state prediction models. Another purpose of this study was to test the feasibility of using ordinal multi-state models to predict financial distress and the appropriateness of the multi-state scale as stated in this study.; Separate models for lag periods from one to three years prior to financial distress were constructed to test the predictive ability of cash flows and accrual ratios. These models (cash flows, accrual, and mixed models) were constructed using ordinal logistic regression (OLR), thus taking into consideration the ordinal response scale of financial distress. The ordinal response variable in this study was financial distress with the following four states of distress: (1) financially healthy, (2) dividend reduction/elimination, (3) debt accommodation and/or loan/interest default, and (4) bankruptcy. The predictor variables were the relevant cash flows and accrual ratios. Rank probability scores and classification accuracy were both used to test the predictive strength of the models.; The results of this study were consistent with the theoretical model developed in this study and the opinions expressed by the FASB. Cash flows are useful in predicting financial distress when combined with accrual data; however, cash flows are not more useful by themselves than accrual ratios in predicting financial distress.; Proportional odds tests also indicated that the ordinal scale used in this study may not fit the data very well for the accrual ratios tested. Further analyses indicated that the scaling problem occurred because the bankrupt firms, overall, were not as financially distressed as the loan interest/principal default and/or debt accommodation firms. This finding calls into question the use of a simple dichotomous bankrupt/nonbankrupt response as a proxy for financial distress.