臺灣博碩士論文加值系統

在全球各地，無論英文、日文、中文，凡研究語音辨識的技術，首推使用隱藏式馬可夫模型（Hidden Markov Model,HMM），而一窺美國語音辨識技術專利的申請數可以發現自1988年至1998年語音辨識的技術專利到達高峰，之後日趨減少，由此可以顯示出此項辨識方法的技術趨於成熟，是以本論文將藉由此種以機率與統計做為基礎的隱藏式馬可夫模型理論，將其做創新的應用於股市單日交易預測上。
然而在此種藉由機率與統計原理做基礎的理論上，用來描述訊號所具有的特徵，此方法具有整體的隨機和局部的穩定性，且能容忍多變異性訊號，正符合了股市的虛中帶實，實中帶虛，看似無法則，然事實上就如目前使用的種種線型分析，如MACD、K線圖，又在在顯示其中隱藏一些可以預測的特徵，是以在尚無法洞聽股市變化的今日，本研究嘗試應用此模型來學習出股市的語法，就如同此模型應用在各國語言的語音辨識之上，而在學習出股市語法後，進而加以預測其下一個會出現的語句，進而達到具獲利性的決策模型。
而股票交易市場可謂是經濟的櫥窗，由於集中市場的交易，每日的交易量動輒數百億，股市熱絡時甚至可達上千億，然股市憂關技術面、心理面、個體經濟面、總體經濟面、資金面、政治面等多方面因素，對股市投入了不可預知的多維度的變異性，而對於大多數投資人而言，無不希望以積極的手法從中獲取超額的報酬，是以自證管會自民國八十三年元月五日起開放以信用交易資券相抵交割之方式進行當日股票賣之沖銷，而由於證管會規定股票當日沖銷有平盤以下不可放空之規定，是以而本研究將以臺灣加權股價指數作為實證對象，使其不會因此項規定而受限。經由本研究的實證，可得以下之結論。本研究之結果發現：隱藏式馬可夫模型的預測方法將優於投信投顧所使用的當日沖銷決策模式開盤八法及金融預測上具代表性的弱式效率市場假設決策模型，且經統計檢定後實驗證明其差異極為顯著。

Around the world, the Hidden Markov Models (HMM) are the most popular methods in the machine learning and statistics for modeling sequences, especially in speech recognition domain. According to the number of patent applications for speech recognition technology form 1988 to 1998, the trend shows that this method has become very mature. In this thesis, we will make a new use of the HMM and apply it on day trading stock forecast.
However, the HMM is based on probability and statistics theory. In a statistics framework, the HMM is a composition of two stochastic processes, a Hidden Markov chain, which accounts for temporal variability, and an observable process, which accounts for spectral variability. The combination contains uncertainly status just likes the stock walk trace. Therefore, the HMM and the stock walk trace have the same idea by coincidence. In this thesis, we will try to learn the stock syntax; just like how the HMM model was used in speech recognition in different languages, and the take the next step ahead in price prediction.
Additionally, the stock market is the reflection of the economy. The stock trace is impacted by many factors such as policy, psychology, microeconomics, economics, and capital, etc. There, in this thesis, the TAIFEX Taiwan index futures (TX) and day trade are used to avoid all the uncertainty factors.
After the all experiments, it is proven that the HMM is better than the benchmark method- Random Walk method and the Investment Trust & Consulting Association method- Modified Trading method. Moreover, the result is very conspicuous by the statistics testing of significance.

Contests
Chapter 1 Introduction 1
1.1 Motivation 1
1.2 Research scope of this work 5
1.3 Contribution 6
1.4 Thesis organization 7
Chapter 2 Background 8
2.1 Overview stock market theory 8
2.2 Overview stock pricing theory 13
2.3 Overview forecast time series theory 17
2.4 Overview Modified Trading method theory 20
Chapter 3 Hidden Markov Model 22
3.1 Overview Hidden Markov Model 22
3.2 HMM Parameter 24
3.3 Vector quantification 26
3.4 Building Hidden Markov Model 29
3.4.1 Forward procedure 30
3.4.2 Backward procedure 31
3.5 Estimate the HMM 32
3.6 Viterbi algorithm 35
Chapter 4 Implementation 38
4.1 Research scope 38
4.2 Simulation architecture 40
4.3 Simulation model 41
4.3.1Random Walk model 41
4.3.2 Modified Trading model 42
4.4 HMM model 44
4.4.1 HMM template training model 44
4.4.2 Codebook 46
4.4.3 HMM training 47
4.4.4 HMM forecasting architecture 49
Chapter 5 Experiment 51
5.1 Research scope 51
5.2 Random Walk experiment 51
5.2 Modified Trading experiment 52
5.3 HMM experiment 53
5.4 Statistic tests 55
5.5 Time series analysis of regression 58
5.6 Statistic Test conclusions 59
Chapter 6 Conclusion 61
6.1 Conclusion 61
6.2 Discussion 62
6.3 Future directions 63
Bibliography 64

Bibliography
[1].Anderson, W.N. Jr., Kleindorfer, G.B., Kleindorfer, P.R., Woodroofe, M.B. “Consistent estimates of the parameters of a linear system.” Ann.Math. Stat.40, 2064-2075, 1969
[2].Adlar Jeewook Kim, ,” Input/Output Hidden Markov Models for Modeling Stock Order Flows”, M.I.T. A.I. Lab Tech. Rep., No. 1370, Jan,2001
[3].A. Waibel, T. Hanazawa, G. Hinton, K. Shikano, and K. Lang,”Phoneme recognition using time-delay neural networks,” IEEE Trans. Acoust., Speech, Signal Processing, vol. 37,pp.328-339,1989
[4].A.Poritz, “Hidden Markov Models: A guided tour,” in Proc. Int. Conf. Acoust., Speech, Signal Processing, pp.7-13,1988
[5].Acero A., Stern R.M., “Environmental robustness in automatic speech recogtnition,” Proc ICASSp-90, pp 849-952, Albuq., NM,1990
[6].Admati, A.R. and P. Pfleidere,”The Theory of Intraday Patterns: Volume and Price Variability,” Review of Financial Studies, vol.1,pp.3-40,1988
[7].Bahl L.R., Jelinek F., “A maximum likelihood approach to continuous speech recognition,” IEEE Trans. Pattern Analysis Machine Intelligence,5, ,1983
[8].Bahl L.R., Brown P.F., De Souza P.V., Mercer R.L., “A tree-based statistical language model for natural language speech recognition,” IEEE Trans. On Acoustics Speech Signal Process, 37, No. 7.pp.1001-1008,1989
[9].Bollerslev, T., “On the Correlation Structure for the Generalised Autoregressive Conditional Heteroskedastic Process,” Journal of Time Series Analysis, 9, pp. 121-32. 1988
[10].Balasubramanian V.,”Equivalence and reduction of Hidden Markov Models”, M.I.T. A.I. Lab Tech. Rep., No. 1370, Jan,1993
[11]. Bo-Sung Kim, Bok-Gue Park, Jun-Dong Cho, Young-Hoon Chang, “Low Power Viterbi Search Architecture using Inverse Hidden Markov Model,”pp. 724 -732.,2000
[12].Claudio Becchetti and L. P. Ricotti, “Speech Recognition Theory and C++ Implementation,” John Wiley & Sons Ltd, 1999.
[13].Roberts, H., "Statistical versus Clinical Prediction of the Stock Market," unpublished manuscript, Center for Research in Security Prices, University of Chicago,May, 1967
[14].Robinson, P., "Time Series with Strong Dependence," in C. A. Sims (ed.), Advance sin Econometrics: Sixth World Congress, Vol. I, Cambridge University Press, pp. 47-95.,1994
[15].Turner, C. M. , R. Startz and C. R. Nelson, "A Markov model of heteroscedasticity, risk and learning in the stock market," Journal of Financial Economics, 25, pp. 3-22.,1989
[16].Steinhaus, S. "Comparison of mathematical programs for data analysis," University of Frankfurt, Germany, http://www.informatik.uni-frankfurt.de/ , 1999
[17].Starck, J.L., Murtagh, F., and A. Bijaoui Image and Data Analysis: The Multiscale Approach, Cambridge University Press,1998
[18]. Chen, S. H. and C. W. Tan "Estimating the Complexity Function of Financial Time Series," The Journal of Management and Economics, Oct.-Nov., Vol. 3.,1999
[19].Thomas Hellstrom, Kenneth, Kolmstrm,”Predictable Patterns in Stock Returns” Department of Mathematics and Physics, Malardalen University Press,1998
[20].Bagella M., Becchetti, L., Carpentieri, A., “The first shall be last. Do contraian strategy premia violate financial market efficency,” mimeo,1988
[21].D. Easley, N. Kiefer, and M. O’Hara. The information content of the trading process. Journal of Empirical Finance,vol., 4 pp.159-186, 1997.
[22].L. Rabiner.” A tutorial on hidden markov models and selected applications in speech recognition.,“ In Proceeding of IEEE, Vol. 77, No. 2, pages 257—286, 1989.
[23].L. R. Rabiner, “A tutorial on hidden Markov models and selected applications in speech recognition,” AT&T Bell Syst. Tech. Journal, vol. 62, no. 4, April,1983
[24].L.B. Jackson, Digital Filters and Signal Processing with MATALAB Exercises, Kluwer Academic Publishers, 1995.
[25].Y. Bengio. “Markovian models for sequential data,” Neural Computing Surveys, 2:129—162, 1999.
[26].Y. Bengio and P. Frasconi.”Input/output hmms for sequence processing,” IEEE Transactions on Neural Networks, 7(5) pp.1231-1249, 1996.
[27].J.G. Wilpon, et al., “Automatic Recognition of Keyword in Unconstrained Speech Using Hidden Markov Models,” IEEE ASSP Magazine, Vol. 38, pp. 1870-1878, 1990.
[28].J. R. Deller, “Discrete-Time Processing of Speech Signals,” Macmillan,1993.
[29]. J.G. Bergh, F.K. Soong, and L.R. Rabiner, “Incorporation of Temporal Structure Into a Vector-Quantization-Based Preprocessor for Speaker-Independent, Isolated-Word Recognition,” AT & T Technical Journal, Vol. 64, No. 5, May-June 1985.
[30]. G. Wilpon and L.R. Rabiner, “A Modified K-Means Clustering Algorithm for Use Isolated Word Recognition,” IEEE Trans. Acoustics, Speech, and Signal Processing, Vol. 33, No. 3, pp. 587-594, June 1985.
[31]. L¨utkepohl, H.: Introduction to multiple time series analysis. Berlin, Heidelberg, New York:,Springer 1991
[32]. Liptser, R.S., Shiryayev, A.N.: Statistics of random processes. Vols. I and II. Berlin, Heidelberg, New York: Springer 1977
[33].Kaas, Robert E and Adrian E. Raftery.“Bayes Factors,” Technical Report 254,Department of Statistics, University of Washington., 1995
[34].Lin, Tsungan and Bill G. Horne. “Learning Long-Term Dependencies in NARXRecurrent Neural Networks.” 1996 IEEE Conference on Neural Networks. pp. 1329-1338,1996
[35]. Neal, Radford M. “Monte Carlo implementation of Gaussian process models for Bayesian regression and classification,” Technical Report No. 9702, Dept. Of Statistics, University of Toronto. ftp://ftp.cs.toronto.edu/pub/radford, 1997
[36]. Valtchev, V., S. Kapadia, S. J. Young. “Recurrent Input Transformation for Hidden Markov Models.” Cambridge University Engineering Department Technical Report.,1995