頁籤選單縮合
題名 | 變數篩選法對財務危機預警模型正確率影響之研究=A Study on Accuracy of Financial Distress Prediction: Reflections on Variable Selection Methods |
---|---|
作者 | 吳智鴻; 周德勝; 李家琳; 林筱芸; 鄭姿玟; 蔡嘉川; 林孝怡; | 書刊名 | 管理研究學報 |
卷期 | 5:1 2005.01[民94.01] |
頁次 | 頁77-115 |
分類號 | 494.7 |
關鍵詞 | 財務危機預警模式; 變數篩選檢定方法; Financial distress pre-warning model; Variable selection methods; |
語文 | 中文(Chinese) |
中文摘要 | 本研究先利用在過去文獻中,學者所採用的五種篩選變數檢定方法(因素分析法、t檢定、單變量ANOVA、逐步排除法及Mann-Whiteney-Wilcoxon法),遴選出顯著數後,再以財務危機研究中常用於建構財務危機預警模式之四種統計方法(區別分析法、Logit、Probit與類神經網路法),藉以分析上述變數篩選法是否會影響其財務危機預警模式預測正確率。以找出何種變數篩選法能得節較佳的正確率,並利用此最佳模型進行九十年及九十一年的樣本外預測。 研究樣本為臺灣上市公司,採1:3的配對方式,選取民國87年至89年間有發生危機的公司22家及66家正常公司,並以危機發生時點為基準,往前選取三年及往後兩年的財務比率資料進行分析。研究結果顯示:1.在未用數篩選法下所得到模式的預測效果略優於其他透過五種變數篩選法後所建構之財務危機預警模式;2.四種財務危機預警模型之預測力,以類神經網路法的預測效果最佳,而預測效果效差則為區別分析法,且具有最高的型Ⅱ誤差;3.以危機前一年的資料所建構的財危機構式具有最高的預測正確率;4.在樣本外預測中,九十年度是以Logit較佳,九十一年度,以類神經網路法之預測力較佳。 |
英文摘要 | Most of past studies often concentrated on how to build a financial distress pre-warning model with the best prediction accuracy, but rarely focused on the prediction accuracy affected by various variable selection methods. Hence, this study attempts to analyze the impact of model predicting ability among five variable selection methods (factor analysis, t text, ANOVA, stepwise Discriminant analysis, Mann-Whiteney-Wilcoxon text) used in past research. First, the financial ratio variables were selected via various variable selection methods, and then as the independent variables to different financial distress models (Discriminant analysis, Logit regression, Profit regression, and Neural Network) which were wildly used in the past decades, to classify the financially sound firms and the financially sound firms and the financially distressed firms. Based on the prediction of models, this study is able to offer the direction and suggestions of investment for corporate mangers, officers in government, and investors. In order to increase the accuracy of model prediction, the sample size of firms were set to a 1 to 3 ratio (i.e., one financially distressed firm to three financially sound firms) and the independent variables were gathered from information of financial ratios in TEJ database. All fanatical ratios data of twenty-two distressed firms and sixty-six financially sound firms included the time period from 1998 to 2000. The time period of training data and validating data started three years before the point of financial failure and ended two years after the point of financial crisis, respectively. The findings of this study are as following: 1. The financial distress model with the best predication accuracy was developed without using variable selection methods. 2. The Mann-Whiteney-Wilcoxon test raises the accuracy of model while the data distribution is non-normal. 3. Neural Network has a superior ability to predict the financial crisis firms than other models and variable selection methods done. In contrast, the Discriminant analysis has the worst predicting accuracy and the highest Type II error. 4. The model has the best prediction ability which was constructed on the information of financial ratios which were recorded in one year prior to the point of financial failed. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。