股票投資是一項重要的投資活動，現今有許多研究專注在股票市場，建立股票評價系統是其中之一。在傳統股票評價系統方面，先藉由專家問卷方法透過分析網絡程序法(Analytical Network Process)計算出股票指標之權重，再利用線性模型以加權的方式判斷出具有投資潛力的股票。此股票評價系統可有效的判斷出股票指標的重要性及選擇出具有投資潛力的股票,以協助投資者獲得最好的投資回報率。但傳統系統存在著兩個重要的問題，首先，是人為的主觀因素，股票專家們來自不同的領域，因此在股票指標上所給予的權重就會不一致，再者股票指標與股票績效之間是呈現非線性的關係，此兩問題影響股票評價系統的績效，所以傳統的股票評價系統預測準確度並不佳。
為解決傳統股票評價系統之問題，本研究使用人工智慧的方式發展出準線性迴歸模型以建立股票評價系統，主要概念是利用歷史的股票資料運用機器學習法代替專家問卷法，以避免人為的主觀因素並利用非線性模型提高整體系統精度。準線性迴歸模型包含線性部份和非線性部份，為了實現準線性迴歸模型，在本研究中設計一個階層式演算法，在線性部份計算出股票指標之權重，而非線性部份計算輸入與輸出之間的非線性關係，以提高整體模型的預測準確度並選擇出具有投資潛力的股票，因此本研究結果顯示(1)本研究所提出的股票評價系統解決了傳統股票評價系統所存在問題。(2)可同時提供兩種重要的投資訊息以幫助股票投資者判斷出重要的股票指標並選擇出具有高報酬潛力的股票。

Stock investment is one of important investment activities in the world. There are many researches focusing on stock market. Building stock evaluation system is an important one among them. A traditional stock evaluation system generally consists of three parts which are feature selection, correlation coefficient analysis and prediction model. It provides indicator evaluation index and stock evaluation index for investors to make a decision of the stock investment. The analytical network process (ANP) method with expert questionnaire and linear weighted method are commonly used method to build the stock evaluation system in the traditional methods.
However, there are two major problems on a traditional stock evaluation system. First is that the expert experiences were used to determine the importance stock indicators in the expert questionnaire. But each expert has subjective judgment of stock indicators who might give different importance levels of stock indicators. Second is that the linear weighted method is utilized to judge the importance stocks. The relation between stock indicators and stock performance usually are highly nonlinear. These may affect the performance of the stock evaluation system built.
It therefore, is highly motivated to develop a modeling scheme by using machine learning method to overcome those problems. We propose a quasi-linear regression (quasi-LR) model, consisting of linear part and nonlinear part, to develop an improved stock evaluation system. Hierarchical algorithm is developed to identify the quasi-LR model in such a way that the linear part describes the importance of stock indicators and the nonlinear part improves the accuracy of the stock performances forecast. Thus our stock evaluation system can help investors to judge the importance of stock indicators and select the stocks with higher profit. It solved the problems of traditional stock evaluation system and also increases the accuracy of the stock performances forecast.
In the quasi-LR model, linear regression method is used on linear part. The empirical risk minimization (ERM) concept and structural risk minimization (SRM) concept are used on nonlinear part. The former is back propagation neural networks (BPN) method and latter is support vector machine (SVM) method. We name them as quasi-LR-BPN and quasi-LR-SVM. The experiment results show that our proposed stock evaluation system outperforms the traditional stock evaluation system. On the other hand the “Market β” is the most important indicator among the eight indicators in our system and the traditional system. Furthermore, the quasi-LR-SVM does generate better accuracy rate then quasi-LR-BPN. Thus the SRM concept is better than the ERM concept on the stock evaluation system.

摘　要 I
Abstract II
Acknowledgements III
Contents IV
Table of Contents VI
Figure of Contents VII
Chapter 1 INTRODUCTION 1
1.1 Background and motivation 1
1.1.1 Stock investment 1
1.1.2 Introducing stock evaluation system 2
1.1.3 Problems in stock evaluation system 6
1.2 The purpose of our research 9
1.3 Overview of the thesis 11
Chapter 2 LITERATURE REVIEW 12
2.1 Traditional stock evaluation system 12
2.2 The nonlinear methods of the stock market. 16
2.2.1 Basic Idea of BPN 16
2.2.2 Basic Idea of SVM 19
Chapter 3 THE PROPOSED STOCK EVALUATION SYSTEM 21
3.1 Flowchart of the proposed system 21
3.2 Quasi-Linear Regression Model 25
3.2.1 Building quasi-Linear Regression model by BPN method 26
3.2.2 Hierarchical algorithm of quasi-LR-BPN 28
3.2.3 Building quasi-Linear Regression model by SVM method 31
3.2.4 Hierarchical algorithm of quasi-LR-SVM 32
Chapter 4 EXPERIMENTS 34
4.1 Experimental data set 34
4.2 Experimental results 36
4.2.1 Selection output of system 36
4.2.2 Quasi - Linear Regression Model 40
Chapter 5 CONCLUSIONS AND FUTURE WORK 44
5.1 Conclusions 44
5.2 Future Work 45
Reference 47

[1]Hall, J. W ,“Adaptive selection of US stocks with neural nets,” in Trading on the edge: neural, genetic, and fuzzy systems for chaotic financial markets, G. J. Deboeck, New York, wiley, 1994, pp. 45-65.
[2]Tsai, C.-F. and Wang, S.-P,” Stock Price Forecasting by Hybrid Machine Learning Techniques,” Proc. of Int. MultiConference of Engineers and Computer Scientists 2009, Vol 1, 2009.
[3]Olson, Dennis. and. Mossman, Charles, “Neural network forecasts of Canadian stock returns using accounting ratios,” Int. Journal of Forecasting, Vol. 19(3), 2003, pp. 453-465.
[4]S. I.Ao, “Analysis of the interaction of Asian Pacific indices and forecasting opening prices byhybrid VAR and neural network procedures,” Vienna, in Proc.Int. Conf.on Computational Intelligence for Modelling, Control and Automation 2003, 2003.
[5]Cheng-Lung Huang and Cheng-Yi Tsai, “A Hybrid SOFM-SVR with a Filter-Based Feature Selection for Stock Market Forecasting” Expert Systems with Applications, Vol. 36, Issue 2, March 2009, pp. 1529–1539
[6]Shih-Wei Lina, Tsung-Yuan Tsenga, Shuo-Yan Choub and Shih-Chieh Chen, “A simulated-annealing-based approach for simultaneous parameter optimization and feature selection of back-propagation networks,” Expert Systems with Applications, Vol. 34, Issue 2, 2008, pp. 1491-1499.
[7]Liad Wagman,” Stock Portfolio Evaluation: An Application of Genetic-Programming-Based Technical Analysis,” Genetic Algorithms and Genetic Programming at Stanford 2003, 2003, pp.213-220.
[8]W.Yan, M.Sewell and C. D. Clack, “Learning to Optimize Profits Beats Predicting Returns Comparing Techniques for Financial Portfolio Optimisation,” in Proc. of the Genetic and Evolutionary Computation Conference, 2008, pp1681-1688.
[9]Yung-Keun Kwon, Byung Ro Moon, “ A Hybrid Neurogenetic Approach for Stock Forecasting,” IEEE Transactions on Neural Networks, Vol. 18(3), 2007, pp.851-864.
[10]F. Reilly, and K. Brown, “Investment Analysis and Portfolio Management,” 7th Edition, 2004, ISBN:0324171730.
[11]J. Richard Dietrich, Steven J. Kachelmeier, Don N. Kleinmuntz and Thomas J. Linsmeier, “Market efficiency, bounded rationality and supplemental business reporting disclosures,” Journal of Accounting Research, Vol. 39(2), 2001, pp. 243-268.
[12]Fernando Fernández-Rodrígueza, Christian González-Martela and Simón Sosvilla-Rivero,” On the profitability of technical trading rules based on artificial neural networks: Evidence from the Madrid stock market,” Economics Letters, Vol. 69, 2000, pp. 89-94.
[13]B Egeli, M Ozturan, B Badur,”Stock market prediction using artificial neural networks,” Proc. Int. Conf. on Business, 2003, pp. 1-8.
[14]Versace, M., Bhatt, R., Hinds, O. and Shiffer, M, “Predicting the exchange traded fund DIA with a combination of genetic algorithms and neural networks,” Expert Systems with Applications, Vol. 27, 2004, pp. 417-425.
[15]D. Senol and M. Ozturan,” Stock price Direction Prediction Using Artificial Neural Network Approach: The Case of Turkey,” Journal of Artificial Intelligence, Vol. 1(2), 2008, pp. 70-77.
[16]Ebrahim Abbasi, and Amir Abouec, “Stock Price Forecast by Using Neuro-Fuzzy Inference System,” Int. Journal of Business, Economics, Finance and Management Sciences, Vol. 1:3, 2009.
[17]K Lai, L Yu, S Wang, C Zhou,” A Double-Stage Genetic Optimization Algorithm for Portfolio Selection,” Lecture Notes in Computer Science, Vol.4234, 2006, pp. 928-937.
[18]Lijuan Cao and Francis E.H Tay,”Financial forecasting using support vector machines,” Neural Computing & Applications, Vol. 10, 2001, pp.184-192.
[19]Kyoung-jae Kim, “Financial time series forecasting using Support Vector Machines,” Neurocomputing, Vol. 55, 2003, pp.307-319.
[20]Lee, Wen-Shiung, Tzeng, Gwo-Hshiung, Guan, Jyh-Liang, Chien, Kuo-Ting, Huang, Juan-Ming, “Combined MCDM techniques for exploring stock selection based on Gordon model,” Expert Systems with Applications, Vol. 36, 2009, pp.6421-6430.
[21]Lean Yu, Huanhuan Chen, Shouyang Wang, and Kin Keung Lai, “Evolving Least Squares Support Vector Machines for Stock Market Trend Mining,” IEEE Transactions on Evolutionary Computation, Vol. 13(1), 2009.
[22]An-Sing Chen, Mark T. Leung, “Regression neural network for error correction in foreign exchange forecasting and trading,” Computers & Operations Research, Vol. 31(7), 2004, pp. 1049-1068.
[23]Wun-Hua Chen, Jen-Ying Shih, Soushan Wu, “Comparison of support vector machines and back-propagation neural networks in forecasting the six major Asian stock markets,” Int. Journal of Electronic Finance, Vol. 1(1), 2006, pp. 49-67.
[24]Jiawei Han, Micheline Kamber, “Data Mining Concepts and Techniques 2ed”. 2006, ISBN: 1558609016.
[25]Matsunaga, A. and Ogawa, K, “Scatter correction in multinuclide data acquisition by means of a neural network,” Proc. of IEEE on nuclear science symposium, Vol. 2, 1999, pp. 948-952.
[26]Castillo, P. A., Merelo, J. J., Prieto, A., Rivas, V., and Romero, G,“G-Prop: global optimization of multilayer perceptrons using Gas,” Neurocomputing. Vol. 35, 2000, pp. 149-163.
[27]Ghosh, R. and Verma, B, “Ahierarchical method for finding optimal architecture and weights using evolutionary least square based learning,” Int. Journal of Neural System, Vol. 13, 2003, pp. 13-24.
[28]Wang, T. Y. and Huang, C. Y, “Applying optimized BPN to a chaotic time series problem,” Expert Systems with Applications, Vol. 32, 2007, pp. 193-200.
[29]Vapnik, V. N, “The Nature of Statististical Learning Theory” New York: Springer-Verlag. 1995.
[30]J. Hu and K. Hirasawa, “A method for Applying Neural Networks to Control of Nonlinear System,”Neural Information Processing Research and Developmen,. 2004, pp.351-369.
[31]古月 敬之, “ニューラルネットワーク計算知能-第二章 線形特性を有するニューラルネットワーク,” 渡辺桂吾編著, 2006, pp. 27-49.
[32]H. T. Toivonen, S. Totterman, and B. Akesson, “Identification of state-dependent parameter models with support vector regression,” Int. Journal of Control, Vol. 80(9), September 2007, pp.1454-1470.
[33]Tong-Seng Quah and Bobby Srinivasan, “Improving returns on stock investment through neural network selection,” Expert Systems with Applications, Vol. 17, 1999, pp.295-301.
[34]Bela G. Liptak and Béla G. Lipták, “Instrument Engineers' Handbook: Process control and optimization”2006, ISBN: 0849310814.