Analysis Of Financial Early Warning Model Based On Asymmetric Exponential Power Distribution

This paper, in binary response precondition, introduces AEPD (Asymmetric exponential power distribution), proposed by Zhu and Zinde-Walsh (2009) to build a more predictable financial early warning system for Taiwan companies. We can take mean value as zero and variance as one, then our model becomes SAEPD (standard asymmetric exponential power distribution). Besides, using more predictable financial accounting variable, including current ratio, debt to total assets ratio, return on asset after tax and before interest, total assets turnover and operating cash flow ratio, we can design a creative model. We use the principle defined in TEJ (Taiwan Economic Journal) as the difference between financial crisis companies and normal companies for positive analysis. We can obtain in-sample financial data by taking random sample and one-to-one matching method for financial crisis companies and normal companies from1999to2005and use maximum likelihood Estimation to estimate parameters in our new model in order to predict out-of-sample for all companies from2006to2008. Furthermore, we will compare our proposed model with past early warning model, Logit and Probit models.Based on the empirical results, for in-sample prediction, we find that the accuracy of SAEPD model is92.12%, higher than Logit (91.5%) and Probit (91.1%).Besides, the accuracy can be improved up to55.44%when defined financial crisis companies is predicted as financial crisis companies by SAEPD model. This is better than Logit (50.26%) and Probit (46.11%). For out-of-sample prediction, the distinction among SAEPD, Logit and Probit model is tiny. However, through the comparison of information criterion, we can know that SAEPD is more reliable than Logit and Probit model. Therefore, we propose combining binary response of SAEPD, Logit and Probit model to enhance the prediction accuracy.