透過您的圖書館登入
IP:54.196.106.106
  • 學位論文

整合支撐向量機模型(SVM)與市場基礎模型應用於台灣營建公司財務危機預測之研究

A study of Integrating Support-Vector-Machine (SVM)Model and Market-Based Model in Predicting Taiwan Construction Contractor Default

指導教授 : 曾惠斌
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


摘 要 近年來政府為了振興台灣經濟不斷釋放利多消息,如兩岸三通相關議題及擴大內需產業,營建產業的熱絡是幾十年難得一見的榮景,豪宅的興建、飯店式管理只租不賣,大小建案隨處可見,營建產業是高財務槓桿操作的行業最容易受到國內外經濟不景氣之影響,資金運作上容易產生調度問題進而發生財務危機,以致於資金往來之相關廠商也受到連鎖反應,就整體產業之競爭力亦受到限制。 公司發生財務危機一直是政府機構、金融機構、企業單位及投資者所關注的議題。應用於偵測公司發生財務危機預測模型從1966年由 Beaver提出後至今,陸陸續續國內外有相當多的學者致力於模型的建立與改良,並使用不同的估算技術來預估,以提高模型的預測能力。 有鑑於營建產業的產業特性及編列財務報表之一般會計處理原則與其他產業不盡相同,如高營運風險、高財務槓桿…等;目前的產業環境也急需財務危機預測之研究,故擬建立一套屬於台灣營建公司財務危機預測模型,以提高模型的準確度,藉此模型的應用能及早預測營建產業可能發生之財務危機公司,以提供營建產業經營者、管理者、金融機構、保險公司,投資大眾及營建相關產業…等;能更具體的瞭解及正確的辨別營建公司之財務危機。 根據研究顯示,各行業應用於支撐向量機模型(Support Vector Machine;SVM)或人工神經網路模型(ANN),使用樣本匹配的方法,預測企業財務危機,由於支撐向量機模型(SVM)或人工神經網路模型(ANN)沒有能力處理分類間的不平衡(default &non-default)造成樣本選擇的偏差。 本研究提出一種強化支撐向量機模型(Enforced Support Vector Machine-based model;ESVM),來預測營建產業財務危機,使用會計年度(firm-year)有效的資料,應用於強化支撐向量機模型(ESVM),以期望解決分類間不平衡的狀況;以傳統的Logistic迴歸會計模型提供一個標準,評估強化支撐向量機模型(ESVM)之財務危機預測能力。預測營建公司財務危機風險,所有相關的財務變數,使用逐步Logistic迴歸方法方法選擇變數後,將選定之變數放入各模型中進行比較。 本研究應用市場基礎模型(Merton)、會計基礎模型(Logistic、ESVM)、Hybrid模型1(結合市場基礎模型-Merton 及會計基礎模型-Logistic)、Hybrid模型2(結合市場基礎模型Merton及會計基礎模型ESVM),以預測台灣營建產業之財務危機,比較五種模型,以找出最佳財務危機預測模型。 研究以台灣經濟新社報資料庫(Taiwan Economic Journal;TEJ)取得上市與曾經上市之台灣營建公司之資訊。實證結果如下: 1.市場基礎模型:Merton模型(AUC:0.7197)。 2.會計基礎模型:20個變數Logistic模型(AUC:0.7425);ESVM模型 (AUC:0.7974),4個變數Logistic模型(AUC:0.8154);ESVM模型 (AUC:0.8258)。 3.Hybrid模型:21個變數Hybrid模型1(AUC:0.7459);Hybrid模型 2(AUC:0.8189),5個變數Hybrid模型1(AUC:0.8220);Hybrid 模型2(AUC:0.8252)。 綜合以上數據顯示出:臺灣市場資訊無法真實反應、效率低,選變數提升模型預測能力,Hybrid模型對市場及會計基礎模型具有整合提升預測能力,臺灣會計資訊較市場資訊可靠度、穩定度為優,財務危機預測以ESVM模型受選變數影響較少、穩定性、預測能力最佳。

並列摘要


Abstract Over the last few years, the Taiwan government has made great effort to revitalize Taiwan’s economy, for example, by negotiating direct trade, transportation and communications links with China, and by promoting the growth of those industries that are oriented towards the Taiwanese domestic market. Recently, the construction industry maintains high growth rate that have not been seen for over a decade. New construction projects, both large and small, are appearing throughout the country, including luxury homes and rental apartment buildings. As the construction industry relies heavily on financial leverage, it is particularly vulnerable to both domestic and external economic shocks. In addition, construction firms often face cash-flow problems, which can lead to serious financial difficulties and cause ripple effect on related parties. This situation may act as a constraint on the competitiveness of the whole construction industry. The issue of companies’ financial distress has always been concerned by government agencies, financial institutions, business enterprises, and investors. Since W.H. Beaver first proposed a model for predicting bankruptcy in 1966, a large number of scholars pursue this line of research by adopting a wide range of estimation techniques to develop a new predictive model or refine the previous predictive models for financial failure. The characteristics and the accounting principles of construction industry differ from those of other industries. Given the current economic climate, there is an urgent need for further research on the prediction of financial failure, particularly in the construction industry. Therefore, this study attempts to establish a financial failure forecasting model on public construction companies in Taiwan. By enhancing the accuracy of the forecasting model, it is anticipated that the established model can provide financial distress early warning of construction companies, thereby helping construction industry business owners and managers, financial institutions, insurance companies, investors, companies in related industries to accurately identify which construction industry firms are likely to be at risk of financial failure. Previous research has suggested that there are some problems with the use of the Support Vector Model (SVM) and the Artificial Neural Network (ANN) model in forecasting financial failure with sample matching. Because neither SVM nor ANN is capable of distinguishing between default and non-default, this will lead to bias in sample selection. This study employs an Enforced Support Vector Machine-based Model (ESVM) to forecast financial failure in the construction industry, and uses firm-year data to solve the problem of unbalanced samples. The traditional logistic regression model can provides as a benchmark to evaluate the ESVM model’s prediction ability. After selected by stepwise logistic regression method, the variables were imported into the models for comparison. Five predictive models are applied in this study as to identify the best model for predicting financial failure in the Taiwanese construction industry, including the market model (Merton), two accounting models (Logistic and ESVM), Hybrid Model 1 (an integrated model with the Merton market model and the Logistic accounting model), Hybrid Model2 (an integrated model with the Merton market model and the ESVM accounting model). The data of Taiwanese construction firms are collected from the Taiwan Economic Journal (TEJ) database. The empirical results of three types of models are shown as follows. 1.Market model: Merton model (AUC=0.7197) ; 2. Accounting models: Logistic model with 20 variables (AUC= 0.7425) , ESVM model with 20 variables (AUC=0.7974) ;Logistic model with 4 variables (AUC:0.8154), ESVM model with 4 variables (AUC=0.8258); 3. Hybrid model: Hybrid model 1 with 21 variables (AUC=0.7459), Hybrid model 2 with 21 variables (AUC=0.8189), Hybrid model 1 with 5 variables (AUC=0. 8220), Hybrid model 2 with 5 variables (AUC= 0.8252). The empirical result indicates the following implication. First, the information from Taiwanese stock market is hardly to reflect the real situation of the Taiwanese construction contractors, this implicates the Taiwanese stock market is still a low-efficiency market. Second, choosing variables can improve the models’ performance. Third, Hybrid model can enhance the predicting ability by integrating the market-based model and accounting-based model. Fourth, the accounting information in Taiwan is more reliable and stable than the information from the stock market. Fifth, the ESVM model can reduce the disturbance of selecting error thus it has higher stability and better performance in predicting construction contractor default.

參考文獻


33. 陳生祥,「運用資料探勘技術建構企業財務危機預警模式-結合財務與非財務資料」中原大學資訊管理學系,碩士論文,民國94年7月。
29. 黃景鋒,「公司信用風險之衡量─以財務比率與選擇權評價法為例」,淡江大學財務金融學系,碩士論文,民國94年6月。
38. 蔡鳳娟,「運用財務比率分析及公司治理指標於企業財務危機之研究」,淡江大學國際貿易學系國際企業學碩士在職專班,碩士論文,民國95年6月。
39. 蕭芳茗,「台灣上市櫃公司財務危機之預測-障礙選擇權評價法與Z-score模型」,淡江大學財務金融學系,碩士論文,民國95年6月。
37. 李明萱,「我國銀行金融預警模型之實證研究」,臺灣大學會計學研究所,碩士論文,民國95年1月。

被引用紀錄


李偉銘(2012)。以成長型階層式自映射網路模式及軌跡分析建構企業財務危機預測模式〔碩士論文,國立臺北科技大學〕。華藝線上圖書館。https://doi.org/10.6841%2fNTUT.2012.00520
陳柏誠(2015)。考量市場與會計基礎模型於營建產業財務危機預測之研究〔博士論文,國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342%2fNTU.2015.01184
張琮勛(2011)。破產預測模型之比較〔碩士論文,國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342%2fNTU.2011.10910
許維城(2011)。以支援向量迴歸機器學習方法預測實際波動度〔碩士論文,國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342%2fNTU.2011.02238

延伸閱讀