上市上櫃公司使用風險值於財務預警之研究

Abstract

在過去有許多研究指出，台灣股票市場為一個有效率的市場，因此相信股價能增加過去以財務資訊來建立的預警模式的預測能力。故本研究試著加入風險值變數，來增加預警模式的預測能力。加入的五種風險值變數指標分別為變異數-共變異數法中的歷史移動平均法與指數加權移動平均、歷史模擬法、拔靴複製法、蒙地卡羅模擬法。 研究結果可歸納為以下幾點： （1）隨著預測時間的增加，模型的預測能力越差；在傳統財務變數方面來看，主要影響的變數在前一季為股東權益報酬率與每股盈餘；而在前兩季以後的模型中，負債比率都達1%的顯著水準以上，其他的顯著變數包括純益率、資本營業報酬率等獲利能力的變數，在危機發生前兩年的顯著變數為應收帳款週轉率變數，顯示公司發生危機前兩年，公司在週轉方面已經開始慢慢出現問題。 （2）危機發生前一季，在傳統財務預警模型中加入風險值變數後，所有模型的風險值變數都達1%的顯著水準以上，變異數-共變異法中的移動平均法、拔靴複製法與蒙地卡羅模擬法所計算的風險值變數較能增加整體模型的預測能力。 （3）在危機發生前兩季，加入風險值的預測模型，在保留樣本方面沒有顯著增加其預測能力，但在測試樣本中都有提高模型的預測能力；而在危機發生前三季、前一年、前兩年都沒有顯著增加模型的預測能力。

Since researches showed that Taiwan stock market is efficient, it is suspected that further study of the data of stock price index can increase the predictive ability of alert model. VaR is treated as additional variables to enhance the predictive ability of alert model in this study. The additional variables of VaR index are: Sample Moving Average, Exponentially Weighted Moving Average, Historical Simulation, Boostrap, and Monte Carlo Simulation. The results are summarized as follows: (1) With the increasing of the predictive period, the predictive ability of alert model is decreased. From the viewpoint of tradtional financial variable, the significant variables of the season before are the return of equity and earning per share. With two former seasons considered, the results of the alert model showed that variables of debt ratio all reached the 1% significant level. Two years before the distress, the most significant variable is the receivables turnover ratio. It is implied that the company may have the trouble of turnover. (2) Taking one season before the distress, the alert model with VaR variable among the models studied, reaches the 1% significant level. The VaR of Sample Moving Average method, Boostrap method and Monte Carlo Simulation method could comparatively induce more the predictive ability. (3) Two seasons before the distress, the alert model with VaR variable could not significantly increase the predictive ability in the retained sample. However, by using the testing sample, the predictive ability for all model are enhanced. As to three seasons, the one-year and the two years before the distress, none model showed increasing predictive ability.