透過您的圖書館登入
IP:3.129.70.157
  • 學位論文

智慧型資料分析於金融市場之應用

Applications of Intelligent Data Analysis in Financial Markets: Selected Essays

指導教授 : 陳文華

摘要


在現代社會中,人類在金融市場中的交易活動日益頻繁,因此衍生出大量的金融資料,這些資料的型態包括財務報告、公司年報、公開說明書、財經新聞、分析師的報告、金融交易資訊等,人們通常亦因此而出現資訊超載的現象。故若這些資料未經適當且有效地分析與呈現,便無法成為對金融市場攸關者有用的資訊或知識,這些攸關者包括股東、債權人、內外部監控者、財務分析師及管理者。因此,本論文將著重探討利用現代擅長於處理大量文字及數值型態資料的智慧型資料分析方法在金融市場之各項應用議題。 本論文的主要目的包括:(1)歸納整理智慧型資料分析在金融市場之應用的相關研究,描繪出此議題之全貌及研究趨勢,並指出未來可行的研究方向;(2) 探討新興分類器--支持向量機器(support vector machines, 簡稱SVM)—應用在台灣信用評等模式的可行性。(此為一屬於分類型態的決策問題);(3) 應用一些備受肯定的文字採掘工具,包括詞庫斷詞法、文件向量模式及成長階層自組織映射演算法(Growing Hierarchical Self-Organizing Map, GHSOM),以建構台灣證券暨期貨市場的中文法令地圖,並評估其成效。(此為一屬於分群型態的決策問題);(4) 利用支持向量迴歸(support vector regression, 簡稱SVR)及倒傳遞類神經網路(backpropagation neural networks, 簡稱BP)預測六個主要亞洲股市的指數,其中有些市場具有淺碟型的股市特徵。(此為一屬於預測型態的決策問題)。 首先,本論文分別從應用及方法的觀點調查並歸納整理相關文獻,這些文獻係從一些具有代表性的國際學術期刊及研討會論文集搜錄而來,並同時可在此研究領域的四大線上資料庫(ABI/INFORM、SDOS-Elsevier、IEEE Xplore及ACM Digital Library)檢索取得。此外,我們指出目前所觀察到的研究趨勢及此領域未來可行的研究方向。除了相關文獻的討探外,在第三至第五章中,我們試著利用一些智慧型資料分析工具從一堆數值型的資料及文字型的資料中挖掘知識,以解決金融市場中的三個決策問題。在第三章中,我們應用SVM建構台灣的發行人信用評等模式,並與BP模式進行比較分析,研究結果顯示SVM分類模式表現優於BP模式,其分類正確率為84.62%,高於前人研究之其他多個分類的模式。在第四章中,我們應用GHSOM呈現中文的台灣證券暨期貨市場法令地圖,此地圖架構是依據文件內容的相關性所構成,在地圖架構中,我們新增了一個標題選擇的模組及利用彩色網頁來呈現地圖架構,為一個可讀性高且易於操作使用的系統介面。在第五章中,我們採用兩個人工智慧的方法,包括SVR及BP,以預測亞洲區六個主要證券市場的指數,包括日經指數、澳洲普通股指數、香港恆生指數、新加坡海峽時報指數、台灣發行量加權股價指數及南韓綜合指數。研究結果顯示SVM在澳洲普通股指數、香港恆生指數、台灣發行量加權股價指數及南韓綜合指數獲得較高的預測績效,而BP在日經指數及新加坡海峽時報指數獲得較高的預測績效。

並列摘要


The overwhelming amount of data stored in financial markets has led to the phenomenon of information overload. It includes financial reports, annual reports, prospectuses, financial news, analysts’ reports, and trading information. Without efficient analysis and presentation, however, the data is not useful to stakeholders, i.e., stockholders, creditors, auditors, financial analysts, and managers. Thus, in this dissertation, we study this important research issue, focusing especially on applying intelligent data analysis in financial markets. The goals of this dissertation are as follows: (1) Present a comprehensive picture and some future research directions of intelligent data analysis in financial markets; (2) Explore the applicability of support vector machines (SVM), a novel classifier, to Taiwan’s credit rating models. (A classification problem); (3) Apply some text mining tools, including the lexicon-based term extraction approach, document vector models, and growing hierarchical self-organizing maps (GHSOM) to the construction of legal maps for Taiwan’s securities and futures markets, and then execute an evaluation of those maps. (A clustering problem); and (4) Use support vector regressions (SVR) and backpropagation neural networks (BP) to forecast six indices of the major Asian stock markets, some of which have thin-dish like characteristics (a prediction problem). Our experiment results show that the SVM classification model performs better than the BP model for Taiwan’s issuer credit rating system. The accuracy rate (84.62%) is also higher than previous research. We use GHSOM to present a content-based, highly readable, and easy-to-use map hierarchy for Chinese language legal documents in the securities and futures markets. We also propose a fully enhanced topic selection module and a web-based user interface. To evaluate the legal maps, we apply three methods: a validity test, a questionnaire and a task experiment. The results of the validity test show that clustered legal documents organized by GHSOM are not only content-related, but also useful for investors who want to study and compare the same thread of information in different documents. However, responses to the questionnaire about the legal map were only a little positive. In task experiment, the results show that using our system the non-professional users only needed twice as much time as the lawyer to accomplish the task. The results of financial time series forecasting in six major Asian stock markets show that SVRs are superior in predicting the All Ordinaries, the Hang Seng, the Taiwan Weighted, and the KOSPI indexes, while BP models are superior in predicting the Nikkei 225 and the Straits Times Indexes.

參考文獻


[1]. 陳文華、施人英、吳壽山,2004,「探討文字採掘技術在管理者知識地圖之應用」,中山管理評論,12(6),35-64.
[2]. 陳文華、徐聖訓、施人英、吳壽山,2003,「應用主題地圖於知識整理」,圖書資訊學刊,1(1),37-58。
[3]. Abu-Mostafa, Y.S., Atiya, A.F. (2001). Introduction to the special issue on neural networks in financial engineering. IEEE Transactions on Neural Networks, 12(4), 653-655.
[4]. Agrawal, R., Srikant, R. (1994). Fast algorithms for mining association rules in large databases.
[6]. Altman, E. (1968). Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of Finance, 23(3), 589-609.

被引用紀錄


林啟義(2000)。台灣食品、用品業零售通路信用評等模式之探討〔碩士論文,元智大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0009-0112200611334278

延伸閱讀