- 學位論文
整合KMV模型與資料探勘技術建構企業信用評等模式以探討企業違約風險
Construct Credit Rating Model for Risk of Corporate Defaults Using KMV and Data Mining Techniques
摘要
近年來，越來越多的投資者透過財務資料來確定最佳的投資組合。因此資本市場所流通的公開財務資料對於信用評等系統在衡量信用風險中佔有重要地位。它可以幫助投資者直接了解銀行對借款人的信用水平，並直接預測借款人的違約概率，藉此快速評估投資風險。 信用評等模型的開發可以幫助借貸市場由於信息不對稱引起的逆向選擇和道德風險問題。本研究旨在採用隨機森林篩選與信用評等相關的財務特徵變數來建立信用評等預測模式，實驗對象為北美上市上櫃公司共6570筆財務資料，再使用資料探勘技術，包括隨機森林(RF)、決策樹（DT），類神經網路（ANN）和支持向量機（SVM）模型，以探討具有較好預測信用評等的方法。在研究變數中納入KMV模型中的違約距離增加預測能力，並使用四等級以及九等級的分類目標來探討各種資料探勘方法的預測能力。 由實驗結果可以得知，使用隨機森林篩選特徵變數之後再運用隨機森林建構預測模型為最佳的信用評等預測方法，四等級分類達到95.54%預測率，九等級則可達87.75%，說明隨機森林在篩選特徵以及分類預測上的卓越性。
並列摘要
In recent years, more and more investors can obtain financial information to determine the optimal portfolio. Credit soring systems in the internal rating approach occupies an important position to measure credit risk. It helps investors to realize bank's credit level of borrowers, and predicts borrowers’ probability of default (PD) directly. Credit rating model development help lending market due to information asymmetry caused by adverse selection and moral hazard problems. The Moody’s KMV (KMV) is employed as a tool to evaluate the market-based information of each corporation. The study intends to compare the different integration data mining methods, including Decision tree (DT), Back–propagation network (BPN) and Support vector machine (SVM) models, with KMV model in order to explore which method has better prediction in of credit rating. To verify the proposed method, using the hybrid model, which applies random forests (RF) to extract useful information for credit rating. The results show that KMV model does provide valuable information in credit rating predictions. Moreover, the proposed approach provides better classification results and generates meaningful rules for credit ratings.