臺灣博碩士論文加值系統

技術分析(Technical analysis)包括了客觀性的技術指標(Technical indicator)與主觀性的型態辨識(Chart Recognition)等辦法。不同的技術指標往往會給出不同的交易訊號；同樣，從不同時間尺度去觀察股票型態也通常會有不同的結論。為解決交易決策上的困擾，本文試圖通過監督式的機器學習(Supervised Machine Learning)演算法對標的歷史資料進行學習，分別構建技術指標與交易訊號、股票型態與交易訊號之間的量化聯繫，基於這種量化聯繫預測交易訊號。在這樣的演算法模型之下，我們構建了中低頻交易策略(Trading Strategy)，對大陸、臺灣、韓國、及美國四個不同市場的日資料進行了測試。發現以技術指標作為訓練模型的輸入資料時，在大陸和臺灣市場上有可觀的收益；以型態評分資料作為訓練模型輸入資料時，在大陸和韓國市場上有可觀收益。同時，也構建了日內5分鐘的中高頻交易策略，對大陸股指期貨，及主要商品期貨產品進行了測試，發現在波動較高的期貨產品中有較好的回測績效。

Technical analysis includes objectivity technical indicators and subjectivity chart recognition etc. Different technical indicators tend to delivery different trading signals; similarly, observing the stock patterns from different time scales also often have different conclusions. To solve the puzzle on the trading decisions, this article attempts to using the underlying historical datas to train supervised machine learning algorithms model, in order to construct the quantifying contact between technical indicators and trading signals, and between trading patterns and the signals, then predicting trading signals based on this quantifying contact. Under such an algorithm model, we constructed a low-frequency trading strategies to test the historial data of Japan, Taiwan, South Korea, and the United States. When the input data are Technical indicators, the Mainland and Taiwan markets have substantial revenue; when the input data are patterns, the Mainland and Korean markets have considerable benefits. It also builds a 5-minute intraday high frequency trading strategies on the Mainland stock index futures and main commodity futures. The test result presents high volatility futures products have satisfying performance.

第一章 簡介 1
第二章 技術指標與型態辨識 4
2.1 技術指標 4
2.2 型態辨識 (Pattern Recognition) 9
第三章 交易訊號 12
3.1 時間序列趨勢分析法 12
3.2 趨勢資料轉換 14
第四章 監督式的機器學習演算法及訓練 15
4.1 類神經網路 15
4.2 支持向量回歸 18
4.3 模型訓練 22
第五章 交易策略及其測試結果 24
5.1 資料來源 24
5.2 交易規則 24
5.3 實證分析 28
5.3.1 中低頻數據測試結果 28
5.3.2 中高頻數據測試結果 36
第六章 總結及未來工作 46
參考文獻（References） 48
附錄一：技術指標及其輸入型態 50

[1] 大衛·凱勒. 超越技術分析. 北京:中國經濟出版社, 2013. 02-111.
[2] 吳宗正. 投資技術分析. 臺灣:華泰文化事業股份有限公司, 2010. 02-436.
[3] Christopher M. Bishop. Pattern Recognition and Machine Learning. American:Springer, 2006. 325-352.
[4] Chun-Ji Chen. Chart Recognition by linear Scaling Method. Master Thesis. National Tsing-Hua University. 2016.
[5] Jheng-Long Wu and Pei-Chann Chang. A Trend-Based Segmentation Method and the Support Vector Regression for Financial Time Series Forcecasting, Mathematical Problems in Engineering. Volume 2012 (2012).
[6] J. Verbesselt, R. Hyndman, G. Newnham. Detecting Trend and Seasonal Chages in Satellite Image Time Series, Remote Sensing of Environment Volume 114, Issue 1, 15 January 2010, Pages 106–115.
[7] C. Yan, J. Fang, L. Wu. An Approach of Time Series Piecewise Linear Representation Based on Local Maximum Minimum and Extremum, Journal of Information & Computational Science 10:9 (2013) 2747-2756.
[8] R. DeMark Thomas. The New Science of Technical Analysis. 1999.
[9] V. N. Vapnik, The Nature of Statistical Learning Theory, Springer, New York, NY, USA, 1995.
[10] V. N. Vapnik, Statistical Learning Theory, Adaptive and Learning Systems for Signal Processing, Communications, and Control, JohnWiley & Sons, New York, NY, USA, 1998.
[11] Z. Liu. Chaotic time series analysis. Mathematical Problems in Engineering. Vol. 2010.
[12] B. J. Chen, M. W. Chang, and C. J. Lin. Load forecasting using support vector machines: a study on EUNITE Competition 2001. IEEE Transactions on Power Systems, vol. 19, no. 4, pp. 1821–1830, 2004.
[13] F. Girosi, M. Jones, and T. Poggio. Regularization theory and neural networks architectures. Neural Computation, vol. 7, pp. 219–269, 1995.