(A) comparative study of three methods for prediction bankyuptcy

Abstract

The purposes of this paper are to compare the three classification procedures - DA, RPA and ACLS - within the context of bankruptcy and to find the conditions under which one of them dominates the others. RPA was developed to alleviate the methodological or statistical problems of traditional classification procedures. ACLS was developed to solve the bottleneck problems of the direct knowledge acquisition method in developing expert systems. So far, the above two procedures have been compared with discriminant analysis (DA) in the current literature which reported that they performed better than DA. However the two procedures were not compared each other. It is found that first, the essential differences between RPA and ACLS are assumption about prior probabilities and methods to estimate misclassification costs, second, if the estimates of the prior probabilities are relatively accurate RPA performs better than DA and ACLS but if the estimates are very biased ACLS dominates RPA. The conclusions from our findings are first, because we can estimate the relatively accurate range of prior probability of bankruptcy, RPA is recommendable when we are to predict bankruptcy of a firm, second, if decision situation is too volatile and unpredictable to estimate prior probabilities and misclassification costs then it is desirable to using ACLS in that ACLS shows average performance of RPA.