透過您的圖書館登入
IP:3.15.237.255
  • 學位論文

依產業別建立台灣上市上櫃公司信用風險模型-支援向量機應用

Credit Risk Modelling using Support Vector Machines (SVM): Evidence from the Taiwan Market

指導教授 : ARRAY(0xbfde464)

摘要


本文主旨在於利用財務等相關資訊建立衡量企業信用風險模型,主要研究期間與對象為2000年到2008年底,台灣上市上櫃公司中因財務危機被轉為全額交割、停止交易或終止上市上櫃之公司以及用以比較之正常公司為對象。因於各產業間有其不同的財務特性。因此,本研究將樣本數較多的產業分開,分別篩選具有重要指標的變數,並且建立模型。 本研究所使用的模型有區別分析、Logistic、倒傳遞類神經網路、以及支援向量機。採用兩階段模型建構方式,先利用區別分析和Logistic模型篩選出各產業重要的指標,以做為支援向量機和倒傳遞類神經網路的輸入變數,再比較各種方式的分類結果。本研究實證發現: 1.各產業間皆有其不同的財務特性,且篩選出的重要財務指標並不一致。利用各產業所建立的模型分析得知,預測能力較使用所有產業的預測效果為佳。 2.同一產業間區別分析以及 Logistic所篩選出來的變數也不太相同。其預測能力在不同的產業下也各有優劣。並無特定模型預測績效一定為佳的現象。 3.支援向量機使用所有變數所建立出來的模型其預測能力都較倒傳遞類神經網路差,但是經過變數篩選過後,其預測能力及穩定性皆有提升。而類倒傳遞神經網路經過篩選變數之後也有收斂速度較快的優點。

並列摘要


The purpose of this study is to develop a credit risk model for Taiwanese firms. The sample period is from 2000 through 2008. The sample includes the listed firms on stock market and the over-the-counter (OTC) in Taiwan, that had been either listed throughout the sample period or classified into three categories, full-cash delivery, eliminated stock trading,or delisted stocks at some point during the sample period." Due to the fact that each industry in Taiwan has a very different financial structure than the other industries, we develop industry-specific credit risk models for Taiwanese firms. We adopt a two-step model development. In the first step, we use discriminate analysis and logistic analysis to select variables that best describe each industry in Taiwan. In the second step, we then apply the variables obtained in the first step in back-propagation network and support vector machine, and compare the bankruptcy prediction performance of each model. Our empirical tests show: 1. Each industry in Taiwan has a very different financial structure, and the variables suggested by each model for each industry are also different. Moreover, industry-specific models performs better than the model based on all industries in predicting bankruptcy. 2. The variables selected by discriminate analysis for each industry are different than what logistic analysis suggests, and the prediction performance of logistic analysis is not always better than discriminate analysis. 3. Back-propagation network has better prediction performance than support vector machine in predicting bankruptcy for most industries in Taiwan. Using the selected variables, support vector machine, however, is shown to outperform back-propagation network for some industries.

參考文獻


2. 李三榮(2002),「Basel II」,台灣金融財務季刊,第3輯第2期,第105-124 頁。
5. 林俊男(2005),「 台灣財務危機公司預警模型之探討︰景氣衰退期與成長期之比較」,中原大學企業管理研究所碩士論文。
6. 胡志宏(2003),「新版巴塞爾資本協定對我國金融業信用風險管理之衝擊」,元智大學管理研究所碩士論文。
11.莊尚穆(2003),「應用Multinomial Logit迴歸之投資風險預警系統」,大同大學資訊經營學系碩士論文。
14.劉易昌(2004),「支援向量機於財務預測上之應用」,靜宜大學資訊管理研究所碩士論文。

被引用紀錄


楊淨萍(2010)。亞洲單一貨幣化與貨幣危機預警模型 —外匯壓力指數關聯性研究〔碩士論文,中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu201000463

延伸閱讀