臺灣製造業危機預警模型之建構—Z-score、區別分析與Logistic模型之實證比較

The Construction of Financial Crisis Warning Model from Manufacturing Industry-The Empirical Comparison of Z-score,Multivariate Discriminate Analysis and Logistic Model

三雨辛(Yu-Shin San)

指導教授 ： 洪志興 ； 劉定焜

朝陽科技大學/管理學院/財務金融系/碩士(2011年)

摘要

近年多數研究財務危機預警的學者，多數以上市、上櫃個別產業或者以全產業為研究對象，在製造業方面的實證研究較為少見。有鑑於製造業對於我國的重要性，故本研究建構2001至2009年上市(櫃)製造業公司之balanced panel data，運用Altman (1968) 所發展的Z-score model，建構台灣製造業危機預警模型。樣本選取方面，除參考過去較多文獻採用之1:1與1:2模式進行危機與正常樣本之實證配對。同時，並考量1:3之配對比例 (Lee and Teng，2009；蔡明春等人，2009) 。為了比較不同預警模型的預測準確性，本研究進一步建構區別分析與Logistic model，除納入Z-score model之財務變數外，加入公司治理、總體經濟與過去文獻鮮少探討之風險變數，亦納入淨值與現金流量之考量。此外，本研究進一步將樣本予以分類，將樣本分為上市、上櫃與全樣本，由於台灣廠商以中小企業為主體，進一步將樣本區分為中小型與大型製造業公司。相較於過去文獻，本研究樣本具較長研究期間，並將製造業樣本予以分類。最後，進一步運用Z-score model、區別分析與Logistic model建構之危機預警機制，藉此推估不同分類及不同預警模型，何者具有較佳之預測能力，選出適合預測製造業財務危機的機制。
實證結果顯示，在樣本配對比率上，Z-score model與其他二種模型之最佳正確分類率結果不同。Z-score model以1:1的配對方式為最高；而區別分析與Logistic model皆為1:3的配對比率為最佳；在不同分類樣本之分類正確率而言，發現皆以預測中小型製造業樣本時，各模型之正確分類率為最佳。
Z-score model之分類正確率低於區別分析法與Logistic model，不論在各分類樣本與配對比率上，Z-score model之最高分類準確性 (55.25%) 皆低於其他二種模型之最低分類準確性 (70.8%；71.1%)，表示Z-score model並不適用臺灣之財務資料。若以Z-score model之年資料與季資料比較，年資料之準確性較高，與余義賢 (2008) 以季資料使用Z-score model之估計結果相符，顯示使用季資料建構Z-score model於預測上，呈現不穩定的狀態。
整合各模型之不同，本研究發現以區別分析法預測財務危機之準確性最高，可達85.3%，而Logistic model建立之危機預警模型最高準確率可達到89.7%。由此可知，不論在分類樣本與配對比率上，皆以Logistic model之正確分類率為佳。整體而言，正確分類率由高至低分析為Logistic model、區別分析與Z-score model。

關鍵字

區別分析 ； Logistic 模型 ； 財務危機預警模型 ； 製造業 ； Z-score 模型

並列摘要

Recently, most of the researchers study in finance crisis forecasting bankruptcy that select companies of the individual industry or the entire industry in TSEC and OTC. The empirical studies of the manufacturing industry are rarely. But the manufacturing industry is very important in our country, so this research constructs the panel data of manufacturing industry companies in Taiwan from 2001 to 2009, and we use the Z-score model developed by Altman (1968). We follow former literatures to use 1:1 and 1:2 patterns on the financial distress and non-financial distress, and consider simultaneous pair of proportion 1:3 construction (see Lee and Teng, 2009; Ming-Chun Tsai et al., 2009). This research constructs multivariate discriminate analysis (MDA) and Logistic model for different financial crisis precaution model. The model besides integrates finance of variable into the Z-score model, corporate governance, macroeconomic factors and a rare discussed risk of variable in the past literatures, we also take into account the net asset value (NPV) and the cash flow.
In addition, this research classifies the sample into full sample and sub-sample of TSEC and OTC. As results of main industry are small and medium-sized enterprises (SMEs) in Taiwan, we divide the sample into SMEs and large manufacturing industries. Compare with the past studies, this paper has long period and classifies the manufacturing industry sample. Finally, we apply Z-score model, MDA and Logistic model for financial distress prediction. Under the estimation of the different classification and the different financial distressprediction model, we try to find out the superior forecasting mechanism for finance crisis of manufacturing industry.
The results show that the best accurate rate of Z-score model are different from other two kind of models had different. Z-score model with 1:1 pair has the highest accurate rate, and MDA and Logistic model with 1:3 pair are best in accurate rate. The classification of accuracy in the different classified sample, we find that the sub-sample of SMEs has the highest accurate rate for all. The classification of accuracy of Z-score model is lower than the MDA and Logistic model. No matter in each classified sample and pair accurate rate, Z-score model’s highest classified accuracy (55.25%) is lowest than other two kind of models (70.8%; 71.1%). The result shows that Z-score model is not suitable for financial distress prediction model in Taiwan. Compare the sample of year and the season form Z-score model, accurate rate of the yearly data is higher than seasonal data which consisted with Yu (2008).
Integration of individual financial distress prediction model, this study finds that the accuracy rate of MDA with the highest rate of 85.3%, and Logistic model with the highest rate of 89.7%. We can conclude that no matter in the classification or pair, the Logistic model is the financial distress prediction model among all. As a whole, the accurate rates from high to low are Logistic model, MDA and Z-score model.

並列關鍵字

Z-score model ； Logistic model ； manufacturing industry ； financial distress prediction model ； multivariate discriminate analysis

參考文獻

8.何文榮、李復禮 (2006)。公司治理與財務危機關係之研究－以上市公司為例。華人前瞻研究。2(2)，1-25。

10.林豐騰 (2009)。企業財務危機預測－整合財務指標、公司治理因素及智慧資本構面模型。績效與策略研究。6(2)，59-72。

15.陳乙文、黃鈴羢 (2005)。影響財務危機預警模型因子之研究。建國科大學報：管理類。24(2)，153-168。

19.陳業寧、王衍智與許鴻英(2004)。台灣企業財務危機之預測：信用評分法與選擇權評價法孰優。風險管理學報。6(2)，155-179。

23.黃劭彥、李超雄、洪光宏、吳東憲 (2006)。以經營效率觀點建立台灣資訊電子業財務危機預警模型。文大商管學報。11 (2)，1-20。