公司財務結構與違約機率之分析

Abstract

對於研究公司財務結構以及違約機率方面，有很多學者做過類似的研究。例如，Altman (1968)及Ohlson (1980)等學者都有作過。但是大部份的學者都是只有用公司單期的資料來做分析，Shumway (2001)將這些模型定義為靜態模型。有別於這些模型，他認為利用公司多期的資料來分析，可以達到更好的效果。故他提出了離散型倖存模型。該模型的優點為能夠有效的使用公司的所有歷史資料來分析，所以可以預測取樣公司每一時間點發生財務危機的機率。因此，Shumway證明動態的離散型倖存模型的預測能力，比靜態模型要來的佳。 本研究使用三個模型來做分析，分別為Logit，MDA以及離散型倖存模型。Shumway (2001)中證明離散型倖存模型為多期的Logit模型，本研究將利用此種特性來進行分析。在本研究中，我們採用56項財務比率進行逐步迴歸，來做進一步的挑選，挑選最適合台灣市場的財務比率變數。 另外，本研究所採取的樣本為上市上櫃公司﹝排除金融產業﹞。我們欲找出這種類型的樣本的分別的最適解釋財務比率以及模型，以期對於台灣地區公司的財務預測有所貢獻。

There are many researchers publish this kind of studies, like Altman (1968), Ohlson (1980), and so on. Altman (1968) is the first one who used multiple discriminant analysis (MDA) method to analysis bankruptcy prediction. There are five significant financial ratios in Z-score model. Besides, Olson (1980) also used the logit model to analysis this subject and Zmijewski (1984) used the probit model. But most researchers only used single period data to analysis. Shumway (2001) defined these models as static models. Compares to theses model, he thought that it might be more powerful by using multi-period data. So he proposed discrete-time survival model. The advantage of this model is that one can use all the historical data to analysis the bankruptcy probability effectively. Thus we can predict the bankruptcy probability every point. So, Shumway proved that discrete-time survival model is better than other static models. We use three models in this study, such as logit model, MDA, and discrete-time survival model. Shumway (2001) proved that discrete-time survival model is multi-period logit model. We will use this characteristic to analysis this model. In this study, we pick fifty-six variables and use stepwise to choose the best explainable variables in theses three models. And the variables picked must be the best ones which fit Taiwan. Otherwise, in this study, our sample firms are which list in TSE and GTM (exclude financial industry). We try to find the best predictable models which fit these two kinds of data. And we expect to have some contribution to predict bankruptcy probability.