應用Multinomial Logit迴歸之投資風險預警系統

The Application of Multinomial Logit Regression In Warning System of Investment Risks

莊尚穆(Shang-Mu Chuang)

指導教授 ： 洪育忠

大同大學/經營設計學院/資訊經營學系所/碩士(2004年)

摘要

企業的經營與社會經濟脈動息息相關，一旦企業遭遇經營危機而面臨倒閉，勢必造成社會不安，並讓投資人的心血付出慘痛的代價。例如過去亞洲金融風暴對台灣經濟造成的衝擊，而後陸續有企業傳出跳票與倒閉事件，使得有許多企業因為不景氣而瀕臨財務危機的情形，進而導致投資人的權益遭受到極大的損害。我們可知投資市場上充斥著許多投資陷阱，一旦決策錯誤便造成重大損失，因此本研究之投資風險預警模式的主要目的就是希望試圖以統計方法來建構一套財務風險預警系統，儘可能在公司發生財務危機之前，先偵測出公司可能的經營危機。
本研究樣本選取以中華信用評等公司所進行信評的相關銀行為對象，除了剔除未達信用評等四年及公司財報不完整之相關證券金融保險公司，並且參考台灣證券交易所之資料，最後篩選出具有完整資料之公司20家，作為最後分析之樣本。而研究期間為民國89年至91年的樣本資料當作實驗組，來建立整體的預測模型；而以92年的樣本資料為對照組，來驗證模型的準確率。在研究期間內的樣本數共有20家。運用15個財務比率，先進行多元逐步迴歸分析，由15個財務比率中萃取出3個較具影響力的指標，再進行區別分析、Multinomial Logit迴歸分析等統計計算工作，以求出模型的正確率。並比較其區別分析、Multinomial Logit迴歸分析之鑑別率，並利用較佳之評估企業財務績效的迴歸模型設計出於PDA上之決策輔助系統。而從本研究過程中，可得到下列結論:
一、經由多元逐步迴歸分析結果，將原本十五個財務指標變數，逐步萃取出流動準備比率、利息收入占年平均放款餘額比率和股東權益報酬率等三個可以預測公司評等的顯著財務變數。
二、 構建區別分析模型的自變數為流動準備比率、利息收入占年平均放款餘額比率和股東權益報酬率，其Fisher線性區別函數:
　公司評等佳（財務結構良好）之區別函數：
Y1= -16.96+0.181X4+4.123X8-0.133X11
公司評等普通（財務結構穩定）之區別函數：
Y2= -20.688+0.134X4+4.852X8-0.197X11
公司評等不佳（財務結構不穩定）之區別函數：
Y3= -21.723+0.09759X4+5.062X8-0.215X11
X4=流動準備比率
X8=利息收入占年平均放款餘額比率
X11=股東權益報酬率
代入實驗組樣本與對照組的樣本數值，可得到60.0%與65%的正確率。
三、 構建Multinomial Logit迴歸模型的自變數為流動準備比率、利息收入占年平均放款餘額比率、股東權益報酬率，將原始樣本與對照組的樣本數值代入預測公司的財務結構之Multinomial Logit迴歸實證模型，可得到68.3%與75.0%的正確率。
四、就模型預測正確率而言，區別分析在原始樣本與對照組樣本的正確率各為60.0%與65.0%，Multinomial Logit迴歸分析在原始樣本與對照組樣本的正確率各為68.3%與75.0%，可得Multinomial Logit迴歸分析的正確率遠高於區別分析。
五、系統的建置是以歷史資料來作為Multinomial Logit迴歸分析的樣本，經過Multinomial Logit迴歸分析，得到最佳模式來作為系統的架構。並採用eMbedded Visual Tools 3.0中的eMbedded Visual Basic 3.0於Windows CE作業系統上做程式開發的工作。使用系統時，一般投資人可輸入流動準備比率、利息收入占年平均放款餘額比率、股東權益報酬率等三個簡單財務比率，經過系統的評估，即可預測公司未來營運狀況的好壞。

關鍵字

財務危機 ； 投資風險預警

並列摘要

Enterprise operation and social economic development are closely linked, once enterprise encountered with crisis and going to close down, not only the society will be disturbed and investors will have great loss. For example, Asia financial storm did impact Taiwan economic, and the follow-up business crises and bankruptcy events revealed that many enterprises were on the edge of financial crises due to recession and result in investors extremely loss in rights and interests. We know the investment markets are full of traps, and a wrong decision could lead to serious loss; therefore, main purpose of the investment risk prediction model in this research is hoping to establish one financial crisis prediction system by statistic method, and detect potential business operation crisis before the occurrence of financial crisis.
Samples of this research are the banks that are rated by Taiwan Ratings; in addition to eliminate the companies that do not have four years ratings and with incomplete company financial statements, and also refer to Taiwan Stock Exchange Corporation’s (TSEC) data to select 20 companies with complete information as final analysis samples. Furthermore, the experimental group’s samples are obtained from the research period of 2000 to 2002, and use to establish the total prediction model; and the control group’s samples are gathered from 2003, and use to verify model’s accuracy. There are 20 samples for research. Apply 15 financial ratios to make multiple stepwise regression analysis and extract 3 indicators with more influence to make statistical computing, such as discriminate analysis and multinomial logit regression, to get model’s accuracy. Moreover, compare the discriminant analysis and multinominal logit regression’s discriminate ratio, and use better Regression Model of business financial performance to design the decision-making assistant system of PDA. Conclusions formed from the research process are as below:
First, through multiple stepwise regressions analysis to extract three significant financial variables that could predict company ratings from the original fifteen financial indicator variables, and these three variables are liquidity reserve ratio, ratio of interest revenue to average loan and ratio of return on stockholders’ equity.
Second, the independent variables, which construct discriminant analysis model, are liquidity reserve ratio, ratio of interest revenue to average loan and ratio of return on stockholders’ equity, and its Fisher’s linear discriminate function is:
　 Discriminate function of company with good rating (good financial structure):
Y1= -16.96+0.181X4+4.123X8-0.133X11
Discriminate function of company with good average rating (Stable financial structure):
Y2= -20.688+0.134X4+4.852X8-0.197X11
Discriminate function of company with good bad rating (Unstable financial structure):
Y3= -21.723+0.09759X4+5.062X8-0.215X11
X4= liquidity reserve ratio
X8= ratio of interest revenue to average loan
X11= ratio of return on stockholders’ equity
Accuracy ratios of 60.0% and 65.0% could be obtained after substation experimental group and control group’s sample values.
Third, the independent variables, which construct Multinomial Logit Regression Model, are liquidity reserve ratio, ratio of interest revenue to average loan and ratio of return on stockholders’ equity; and the accuracy ratios of 68.3% and 75.0% could be obtained after substitution original sample and control group’s sample values to predict company financial structure’s Multinomial Logit Regression empirical model.
Forth, in terms of model prediction’s accuracy ratio, discriminant analysis’ accuracy ratios in original sample and control group are 60.0% and 65.0% spectively; and Multinomial Logit Regression’s accuracy rations in original sample and control group sample are 68.3% and 75.0%, respectively; the obtained accuracy ratios of the Multinomial Logit Regression are much higher than discriminant analysis’.
Fifth, the establishment of system is to use historical data as Multinomial Logit Regression’s sample, and through Multinomial Logit Regression to get optimized model to be system’s framework. Moreover, adopts eMbedded Visual Basic 3.0 of eMbedded Visual Tools 3.0 to develop program on Windows CE. While use the system, general investors could input liquidity reserve ratio, ratio of interest revenue to average loan and ratio of return on stockholders’ equity, and the system will evaluate these three simple financial ratios to predict company’s future operation performance.

並列關鍵字

Financial Crisis ； Investment Risk Prediction

參考文獻

[1]. Beaver, W.H, “Financial Ratios as Predictors of Failure”, Empirical Research in Accounting: Selected studies, 1966, Supplement to Journal of Accounting Research, pp.71-111, 1966.

[2]. Altman, E.I.,” Financial Ratios, Discriminant Analysis and The Predication of Corporate Bankruptcy”, Journal of Finance, Vol.23, No.4, pp.589-609. , Sept. 1968.

[4]. Eisenbeis, R.A., “Pitfalls in the Application of Discriminant Analysis in Business, Finance and Economics”, Journal of Finance, pp.875, 1977.

[5]. Deakin, E.B., “A Discriminant Analysis of Predictors of Business Failure”, Journal of Accounting Research, Vol 10(1), pp.167-179. , Spring 1972.

[6]. Blum, M., “Fail Company Discriminant Analysis”, Journal of Accounting Research, pp.1-25, 1974.