Discussion of Financial Ratios As Predictors of Failure

Abstract

Although a discussant's role in essence is of a critical nature, I would like to start by discussing the strong features of the paper before turning to some areas where critical questions may be useful. First of all, I am very glad to see the empirical approach taken by Professor Beaver for testing the ability of ratios to predict. As he indicated, this is just one of many possibilities for studying the ability of accounting data to predict. In the area of methodology, there are some very strong points to note. For instance, Professor Beaver took substantial care in an attempt to avoid bias in the selection of the two samples: the failed firms and the nonfailed firms. I also like very much his progression in the analysis, starting from the comparison of means, where many people would be tempted to stop, going on to an analysis of predictive power by way of a simple type of discriminant analysis, and then finally approaching the area of likelihood ratios and Bayesian inference. I would also like to stress the importance of the calibrating samples which Professor Beaver used in his analysis. As you know, if you have a body of sample data, you can analyze these and come up with a criterion that is in some sense a good criterion according to the sample data. However, the indication of the predictive power of the criterion based on the sample data generally tends to be biased upward, because the criterion will usually work better for the data from which the criterion was developed than for another random sample from the population. Hence, it is desirable to use calibrating samples where, in effect, half of the data is used in order to develop the criterion and the other half is used to test the predictive power of the criterion. The importance of calibrating samples can be appreciated by examining page 85 of Beaver's paper, where the predictive power (actually its complement, the percentage of firms that are misclassified) is given for a variety of ratios and years before