The Prediction of Financial Distress with Macroeconomics and Industrial Variables –The Empirical Findings from the United States Listed Firms

• Main Authors: Lin, Kuei-Te, 林魁德
• Other Authors: Liang, Te-Hsin
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/90866735659881909701

碩士 === 輔仁大學 === 應用統計學研究所 === 98 === The 2007 financial crisis was triggered by a liquidity risk in the banking system of the United States (US). This crisis has resulted in the collapse of large financial institutions, the bailout of banks by national governments, and downturns in stock markets and economies around the world. These events have highlighted the importance of corporate default prediction machine for bankers, obligors, and investors. The main purpose of this research is to construct an alarm system for short-term financial distress by applying the financial ratio and sensitivity variables of macroeconomics. The samples in this study were collected from 2005 to 2009 data of NYSE-listed companies in the US. The companies that failed during the period of financial crisis because of the bankruptcy (Chapter 11) were selected. In the first stage, this study used independent sample t-test to select significant predictive variables, as well as to deal with multicollinearity problems among those selected variables in creating derivative variables. In second stage, the data set was split into two data sets, a training data set and a testing data set, according to the ratio of 80:20. In third stage, using logistic regression, this study built a whole model with the training data set, and built three kinds of ensemble models in order to calculate the average, maximum, and vote of the predictive probabilities, or results from 101 models with 80% sampling from the training data set (i.e., re-sampling method). In the final stage, this study compared the predictive abilities of the four models by testing the data set. Empirical analyses show that the both of financial ratio and financial ratio with modified macroeconomic variables of ensemble models are more powerful in bankruptcy prediction than “the whole model.” Among three ensemble models, the “voting ensemble model” has comparatively superior performance on cumulative lift, cumulative response rate, AR, FPR, TNR, and PR. The “maximum ensemble model” was observed to have comparatively inferior performance on misclassification rate, FNR, and TPR. In addition, the models that consider modified macroeconomic factors were observed to significantly improve the predictive ability of bankruptcy prediction models.