雷欽隆臺灣大學：電機工程學研究所詹耀仁Chan, Yao-JenYao-JenChan2007-11-262018-07-062007-11-262018-07-062004http://ntur.lib.ntu.edu.tw//handle/246246/53099近年來，由於人工智慧的技術日漸成熟，有釵h研究者嘗試以類神經網路建立股票交易決策支援系統。但過去的研究並未將股票型態的對未來價格的影響列入考慮，而事實上股票型態是技術分析領域中一個很重要的部分。因此本研究提出一個能夠量化頭肩型態的方法，並且將量化的結果以及十八種技術指標作為類神經網路的輸入變數，以期可使我們的決策支援系統能夠同時考量股票型態及技術指標的影響，對股票未來走勢做綜合性的判斷。 本研究的實驗對象為台灣六家上市公司及電子加權指數與大盤指數，研究期間為民國八十八年至民國九十二年。本研究的實驗結果平均準確率高於百分之六十，若進一步將研究期間集中於出現頭肩型態的範圍內，則準確率可提高至約百分之八十。因此我們認為利用量化後的股票型態來預測股市的未來發展是合理且可行的。我們相信若在系統中加入更多的股票型態，將能使準確率更進一步提昇。Since the technique of artificial intelligence has been getting maturer in recent years, many researchers have been trying to build stock trading decision support systems based on neural networks. However, the influence of stock patterns has not been considered in previous researches and we know that is an important part in the filed of technical analysis. Thus, in this research, we propose a new method which could quantify head and shoulders patterns and we form the inputs of neural networks with the quantified results and eighteen types of technical indicators. This could let our system has the ability to consider the influences of stock patterns and technical indicators simultaneously. The sample data in this research are six quoted companies and two indices in Taiwan stock market. Experiment period is from 1999 to 2003. The average accuracy is greater than 60%. If we focus on the period which head and shoulders patterns appear, the accuracy is greater than 75%. Thus, we conclude that it is effective to predict stock markets by quantified patterns. We believe that the accuracy could be further improved by introducing more quantified patterns.Chapter 1 Introduction 1 1.1 Backgrounds and Objectives 1 1.2 Organization 3 Chapter 2 Preliminaries 4 2.1 Introduction to Stock Markets 4 2.1.1 Analyses of Stock Markets 4 2.1.2 Efficiency Market Hypothesis 7 2.1.3 Technical Indicators 9 2.2 Basic Concepts of Neural Networks 18 2.2.1 Models of a Neuron 18 2.2.2 The Back Propagation Algorithm 20 2.2.3 Concurrent Inputs and Sequential Inputs 24 Chapter 3 Previous Researches 27 3.1 The Efficiency of Taiwan Stock Market 27 3.2 Stock Trading Decision Support Systems Based on Neural Networks 30 3.3 Summary 33 Chapter 4 Implementation 34 4.1 Quantify Head and Shoulders Patterns 34 4.2 Selection of Input Variables 38 4.2.1 Input Variables 38 4.2.2 Genetic Algorithms 39 4.2.3 The Stepwise Additional Method 44 4.3 Desired Outputs 45 4.4 System Architecture 47 Chapter 5 Experiment Results 49 5.1 Experiment Periods and System Parameters 49 5.2 Compare Genetic Algorithms with the Stepwise Addition Method 51 5.3 Compare Concurrent Inputs with Sequential Inputs 54 5.4 The Efficacy of Quantified Head and Shoulders Patterns 55 5.5 An Interesting Phenomenon 57 Chapter 6 Conclusions and Future Works 61 References 63391759 bytesapplication/pdfen-US技術指標遺傳演算法人工智慧股票型態頭肩型態技術分析類神經網路genetic algorithmsneural networksartificial intelligencetechnical analysistechnical indicatorsstock patternsthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53099/1/ntu-93-R91921023-1.pdf