آناليز مميزي آميخته درپيش بيني ورشكستگي

Mixture Disciminant Analysis For Bankruptcy Prediction

Linear Discriminate Analysis (LDA) is a valuable tool for two or multigroup classification. Assuming in LDA classes have multivariate normal distribution with common covariance matrix. The case of normal classes rarely holds. Therefore, for the purpose of effective classification in case of heterogeneity of classes it is natural to assume that the classes consist of subclasses with normal but unobserved distribution, classification in this case is called Mixture Discriminate Analysis (MDA), and estimation of the parameters is possible only through EM algorithm. Linear decision boundaries do not differentiate classes completely. In MDA, decision boundaries are nonlinear even with the assumption of equal covariance matrix among classes. This study with financial ratios of Altman model tries to predict financial failure in Tehran stock market, for this purpose sample consist 100 survivor firms and 44 bankrupt firms from 1378 to 1384. Using this sample estimated parameters for LDA and compared MDA. It has been demonstrated MDA is more effective for predicting financial failure of various firms and MDA has been 92.34 percent correct classification for all firms.