信用評等資訊與知識轉化流程整合性分析模式---關鍵因素粹取、視覺化呈現及建構自動信用評等模式知識地圖
No Thumbnail Available
Date
2009/08-2010/07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
隨著美國金融市場次級房貸風暴不斷蔓延擴散，投資人與金融市場主管機關不免對 促成風暴的推手之一信用評等機構的評等決策存在著高度不信任，原因在於這些信用評 等公司的評等訊息幾乎無法事前或適時反應問題受評企業與金融商品的信用風險，往往 事後或事態嚴重時才進行信用評等的調整，難免有事後諸葛的責難，更甚而對其所提供 的簡易評等符號訊息是否正確公允打上問號。此外，近年來在新版巴賽爾協定規範下， 金融機構為利計算風險性資產，以達到最低資本適足率的要求，對於發展內部信用評等 模型十分重視。因此，本三年期研究計畫之主要目的在於藉由關鍵因素粹取、視覺化呈 現及建構自動信用評等模式知識地圖方式，提出一個信用評等資訊與知識轉化流程整合 性分析模式。本計畫第一年將蒐集與整理信用評等相關的公開發行資料，涵蓋信用評等 訊息、發行面、財務資料、營運資料、交易市場資料等，透過關鍵因素粹取技術，例如 因素分析等方法，找出影響信用評等的關鍵因素，並利用自組織映射方法將各個因素的 特徵平面圖與各個受評標的群體的代表性特徵呈現出來，以協助信用評等分析人員進行 受評標的與其他同性質受評標的之比較分析，以及透過簡易視覺化方式處理大量且多元 化的資訊；第二年將應用多種機器學習（包括支持向量機、成長階層自組織映射等）與 多變量統計技術（包括多元區別分析、多元尺度分析等）建構各種自動化信用評等模式， 並比較分析各模式的決策品質；第三年將此一系統化方式所產生的信用評等知識地圖應 用範疇擴及至各種類的受評標的，包括金融業與一般產業發行人、發行債務、資產證券 化商品等，以更進一步呈現整個評等產業的知識內涵。
Due to the impact of subprime mortgage financial crisis spreading from the U.S. to other areas, investors and the regulator of financial markets highly distrust the rating decision of rating agencies which are now considered as the promoters of this crisis. The reason is that the rating information generated from these agencies cannot reflect the true credit risk of issuers or issues immediately, and they often adjust ratings after some negative events. Thus, the rating information becomes a lag indicator and investors put question marks on these simple rating symbols for their inaccuracies. In addition, under the Basel II Accord, financial institutions place importance on developing internal credit rating models for measuring risk-weighted assets to attain the minimal requirement on regulatory capital ratios. Therefore, the main objective of this three-year project is to propose an integrated process model to transform information to knowledge for credit ratings through extracting key features, visualizing relevant information, and representing knowledge maps of credit ratings. In the first year, we will collect the data regarding credit ratings and preprocess them, select key features through some feature selection approaches (such as factor analysis) and represent the feature plane maps of the key features to help rating analysts compare each rating object with the other similar objects and make the rating decision. Moreover, they can process large amounts of data through visualization approach proposed by this project. In the second year, we will apply some important machine learning methods (such as support vector machines, growing-hierarchical self-organizing maps, etc.) and multivariate statistical methods (including discriminant analysis, multidimensional scaling, etc.) to construct automatic credit ratings models, and compare the decision quality of these models. In the last year, the integrated process model will be applied to more rating objects, including issuers’ ratings, issue ratings and ratings of mortgage-backed securities (MBS) and collateralized debt obligations (CDO).
Due to the impact of subprime mortgage financial crisis spreading from the U.S. to other areas, investors and the regulator of financial markets highly distrust the rating decision of rating agencies which are now considered as the promoters of this crisis. The reason is that the rating information generated from these agencies cannot reflect the true credit risk of issuers or issues immediately, and they often adjust ratings after some negative events. Thus, the rating information becomes a lag indicator and investors put question marks on these simple rating symbols for their inaccuracies. In addition, under the Basel II Accord, financial institutions place importance on developing internal credit rating models for measuring risk-weighted assets to attain the minimal requirement on regulatory capital ratios. Therefore, the main objective of this three-year project is to propose an integrated process model to transform information to knowledge for credit ratings through extracting key features, visualizing relevant information, and representing knowledge maps of credit ratings. In the first year, we will collect the data regarding credit ratings and preprocess them, select key features through some feature selection approaches (such as factor analysis) and represent the feature plane maps of the key features to help rating analysts compare each rating object with the other similar objects and make the rating decision. Moreover, they can process large amounts of data through visualization approach proposed by this project. In the second year, we will apply some important machine learning methods (such as support vector machines, growing-hierarchical self-organizing maps, etc.) and multivariate statistical methods (including discriminant analysis, multidimensional scaling, etc.) to construct automatic credit ratings models, and compare the decision quality of these models. In the last year, the integrated process model will be applied to more rating objects, including issuers’ ratings, issue ratings and ratings of mortgage-backed securities (MBS) and collateralized debt obligations (CDO).