透過您的圖書館登入
IP:44.204.99.5
  • 學位論文

使用集成模型進行營建公司財務違約預測之研究─以美國營建領域公司之資料為例

Default Prediction of Construction Firms Using Ensemble Model - Lesson Learned from US Construction Industry

指導教授 : 曾惠斌

摘要


鑒於金融危機後,大眾對於公司體質以更高的標準檢視,回顧過往文獻,大多數研究皆建立在整體之產業上,鮮少針對單一產業,更甚,針對營建產業進行單獨討論。然而,營建產業公司存在其特殊的財務特性 (Huang, 2011),是以本研究認為針對營建單一產業進行研究有其必要性。此外,隨著近年來人工智能與資料科學領域興起,我們得以使用更有效率的分類方式來進行預測模型的建置,是以本研究之目的為從諸多方法中,擇尋適合之方法進行建模,並以實際資料集作為佐證。 佐證資料集的選擇上,本研究承襲 Chen (2015) 之研究,以具備標竿性之美國營建產業公司數據集,輔以交叉驗證法拆分成訓練集以及測試集作為模型的訓練與檢測使用。於原始數據集中,一共包含了從1970至2006年,共1560筆觀測資料,其中之29筆為違約樣本,剩餘之1531筆為正常樣本。 本研究採用集成學習算法中的Adaptive Boosting (Fruend and Schapire, 1996) 演算法進行建模,並與採用Back-Propagation Neural Network (Rumelhart et al., 1986)算法之模型進行比較,說明此算法之成果與適用性。同時,參考Chen (2015) 提出之混和模型之概念與數據集不平衡處理對於模型成果之影響,於本研究中分別以Adaptive Boosting模型驗證,發現在此模型之下,採用單純會計變數以及會計變數混合市場變數之成效雖然差異不大,但是從該演算法過程中,卻可以觀察到市場變數實為權重較高的特徵之一。另外,經驗證說明,此模型之成果同樣會受到數據集不平衡之影響,並且在本研究中應用Synthetic Minority Over-sampling Technique (SMOTE) 手法處理之,得到與Chen (2015) 研究中相呼應的結果。

並列摘要


Since financial crisis in 2008, public started to evaluate a firm’s financial condition with strict demand. Reviewing past studies, most of which are based on overall industries, only few of them focus on particular industry. However, Huang (2011) pointed out that construction industry had its own financial characteristics. Hence, we do state that specific research on construction industry is necessary. In addition, as the development of artificial intelligence technology and data mining techniques, people are able to build up prediction model with efficiency. In this research, our purpose is to find an appropriate algorithm to create the prediction model which will then be proved by empirical experiment. This research inherited from Chen (2015), analyzing on US construction industry data set, which collected firm data from 1970 to 2006. Totally 1560 observations were collected, in which 29 observations are default and the remaining 1531 are normal. In this thesis, the prediction model is based on Adaptive Boosting algorithm (Fruend and Schapire, 1996), which is then compared to Back-Propagation Neural Network (Rumelhart et al., 1986) model. The result shows that Adaptive Boosting model can be well performed on default prediction. In addition, followed the concept of hybrid model and the affection of sample unbalance problem proposed by Chen (2015), this research conduct further experiment on both issue. The result reveals that hybrid model can only make slight improvement on our model, while market variable does really hold a high weight to the model. Also, sample unbalance problem can significantly affect the prediction accuracy, which we then use Synthetic Minority Over-sampling Technique (SMOTE) to deal with, and obtain the results responded to Chen (2015).

並列關鍵字

default prediction AdaBoost hybrid model SMOTE

參考文獻


1. Abidali, A. & Harris, F. (1995). “A methodology for predicting company failure in the construction industry.” Construction Management and Economics, 13(3), 189-196.
2. Altman, E. (1968). “Financial Ratios, Discriminant analysis and the prediction of corporate bankruptcy.” Journal of Finance, 23(4), 589-609.
3. Beaver, W. (1966). “Financial ratio as predictors of failure.” Journal of Accounting Research, 4, 71-111.
4. Black, F. & Scholes, M. (1973). “The Pricing of Options and Corporate Liabilities.” Journal of Political Economy, 81(3), 637-654.
5. Chava, S. & Jarrow, R. A. (2004). “Bankruptcy prediction with industry effects.” Review of Finance, 8(4), 537-569.

延伸閱讀