臺灣博碩士論文加值系統

本研究主要探討14檔中國股票型基金買賣的時間點，利用近年常見的機器學習技術來輔助投資決策，將決策區分為買入、保持、賣出三種分類，並以公開資訊的數據及技術分析中的指數移動平均收斂發散法（MACD）中的最佳報酬買賣點作為訓練資料，再建立非線性支援向量機（SVM）及倒傳遞類神經網路（BPN）模型，使用模型進行2014年至2016年每日交易預測，並將3年之間交易的報酬進行分析比較。研究結果發現：
一、指數移動平均收斂發散法（MACD）最佳策略交易，全部中國股票型基金平均報酬率有47.85%，相對於上海綜合指數平均報酬47.13%、上證B股指數34.84%、深證B股指數30.17%，投資報酬與中國股市相類似。
二、在經歷2015年中國股票市場震盪下，倒傳遞類神經網路（BPN）模型所獲得的平均報酬86.59 %高於支援向量機（SVM）平均報酬率63.45%，在統計檢定的驗證下，倒傳遞類神經網路（BPN）優於支援向量機（SVM）模型。
三、支援向量機（SVM）及倒傳遞類神經網路（BPN）模型3年累計報酬，經過統計檢定的驗證，皆優於指數移動平均收斂發散法（MACD），在統計檢定的驗證下，二種模型報酬率皆有顯著差異。

This research project aims to assist the investment decision of buying or selling for 14 China equity funds by using the machine learning method. We divided the investment decision into three categories: buying、selling and holding. By using the best investigate point of Moving Average Convergence-Divergence (MACD) method in the fund open data as the training data, we built the non-linear Support Vector Machine (SVM) model and Back Propagation Network (BPN) model to execute the daily trading forecast from 2014 to 2016. The results of average rate of return of investment (ROI) of three models were compared and analyzed. The findings of this research are presented as follows:
1.The average rate of ROI of MACD model, the best market trading strategy, of all China stock funds is 47.85%. Comparing to Shanghai Composite Index 47.13%, Shanghai B-share index 34.84% and the Shenzhen B-share index 30.17%, they are almost the same.
2.Under the China stock market’s bumpy ride in 2015, the average rate of ROI of BPN model is 86.59% which is higher than the SVM model’s 63.45%. BPN model is better than SVM model. The significant difference is showed under the verification of statistical test.
3.The average rates of ROI of both SVM and BPN models are all better than MACD model. The significant differences are also showed under the verification of statistical test.

摘要.............................................................I
ABSTRACT.............................................................II
誌謝.............................................................III
目錄.............................................................IV
圖目錄.............................................................VII
表目錄.............................................................VIII
第1章 緒論.............................................................1
1.1 研究背景與動機.............................................................1
1.2 研究目的.............................................................4
1.3 研究範圍.............................................................5
1.4 研究流程與架構.............................................................5
1.5 研究假設.............................................................7
第2章 文獻探討.............................................................8
2.1 技術分析.............................................................8
2.1.1 基金績效指標.............................................................8
2.1.2 指數移動平均.............................................................11
2.1.3 相對強弱指標.............................................................13
2.1.4 移動平均收斂發散法.............................................................13
2.2 機器學習模型.............................................................14
2.2.1 支援向量機.............................................................15
2.2.2 類神經網路.............................................................18
2.3 文獻整理.............................................................23
第3章 研究方法.............................................................25
3.1 研究步驟.............................................................25
3.2 研究資料.............................................................26
3.2.1 研究期間.............................................................26
3.2.2 研究對象.............................................................26
3.2.3 資料來源.............................................................27
3.3 資料處理.............................................................27
3.3.1 缺失值處理.............................................................27
3.3.2 正規化處理.............................................................28
3.3.3 輸入特徵.............................................................29
3.4 尋找最佳報酬.............................................................31
3.4.1 對照模型建置.............................................................31
3.4.2 最佳報酬對照.............................................................32
第4章 預測模型之建構.............................................................35
4.1 建構流程.............................................................35
4.2 特徵篩選.............................................................36
4.3 支援向量機模型.............................................................38
4.4 倒傳遞類神經網路模型.............................................................40
4.5 ｋ折交叉檢驗.............................................................41
4.6 最佳化資料調整.............................................................42
第5章 驗證結果與分析.............................................................44
5.1 支援向量機結果分析.............................................................44
5.1.1 預測結果.............................................................44
5.1.2 統計檢定.............................................................46
5.2 倒傳遞類神經網路模型.............................................................48
5.2.1 預測結果.............................................................48
5.2.2 統計檢定.............................................................50
5.3 綜合比較.............................................................52
5.3.1 各模型累積報酬比較.............................................................52
5.3.2 統計檢定.............................................................54
第6章 結論與建議.............................................................57
6.1 研究結論.............................................................57
6.2 研究限制.............................................................58
6.3 研究建議.............................................................58
參考文獻.............................................................60

一、中文部分
[1] David Olson, Yong Shi著、鄭滄祥(2008)譯，「資料探勘」，高立圖書有限公司
[2] Jiawei Han, Micheline Kamber,Jian Pei著、郝沛毅,李御璽,黃嘉彥(2014)譯，「資料探勘」，高立圖書有限公司
[3] Joel Grus著、藍子軒(2016)譯，「Data Science from Scratch 中文版」，碁峰資訊股份有限公司
[4] Michael Bowles著、賴屹民(2016)譯，「機器學習使用Python進行預測分析的基本技術」，碁峰資訊股份有限公司
[5] Sebastian Raschka著、劉立民,吳建華(2016)譯，「Python機器學習」，博碩文化股份有限公司
[6] 李顯儀(2016)，「基金管理」，全華圖書
[7] 林宜賢(2007)，「應用平滑支撐向量迴歸與類神經網路於共同基金績效之預測」，國立台灣科技大學資訊管理學系碩士論文
[8] 林傑宸(2011)，「基金管理-資產管理入門寶典」，智勝文化事業
[9] 葉怡成(2009)，類神經網路模式應用與實作，儒林圖書有限公司
[10] 張恆勖(2011)，「應用平滑支撐向量迴歸於中國大陸 QDII基金之投資策略績效評比」，國立台灣科技大學資訊管理學系碩士論文
[11] 陳志龍(2006)，「運用類神經網路與技術指標預測股票型基金漲跌及交易時機之研究－以臺灣50指數股票型基金為例」，朝陽科技大學資訊管理學系碩士論文
[12] 陳彥江(2011)，「移動平均線定價交易模型在台灣指數股票型基金之應用―以寶來金融指數股票型基金為例」,輔仁大學統計資訊學系碩士論文
[13] 陳美雪(2015)，「移動平均線與台灣指數股票型基金之探討」，東海大學財務金融學系碩士論文
[14] 鄭捷(2016)，「今天不學機器學習明天就被機器取代-從Python入手+演算法」，佳魁資訊
[15] 蘇禧(2011)，「證券投資基金交易策略基於湍流粒子群優化和移動平均收斂」，國立台灣科技大學資訊工程學系碩士論文

二、英文部分
[1] Fan, A. and Palaniswami, M. (2001), ”Stock Selection Using Support Vector Machines”, Proceedings of the International Joint Conference on Neural Network, Vol.3, pp. 1793-1798.
[2] Hunag, W., Nakanori Y. and Wang, S.-Y. (2004), “Forecasting Stock Market Movement Direction with Support Vector Machine”, Computer and Operation Research Vol.32, pp.2513-2522.
[3] R.Kohavi et al. (1995), “A Study of Cross-validation and Bootstrap for Accuracy Estimation and Model Selection” In IJCAI, Vol. 14, pp. 1137-1145.
[4] Rumelhart, David E.; Hinton, Geoffrey E.; Williams, Ronald J. (1986), “Learning Representations by Back-propagating Errors” Nature 323 (6088), pp. 533-536.
[5] Shashua, Amnon(2009), “Introduction to Machine Learning” Class Notes 67577
[6] Vapnik, V. and Cortes, C. (1995) “Support Vector Networks,” Machine Learning, Vol.20, pp.273-295.
[7] Vellido, A., Lisboa, P.J.G. (1999), ”Neural Networks in Business：A Survey of Applications (1992-1998)”, Expert System with Applications, Vol.51, pp.51-70.
[8] Vapnik, V. and Lerner A. (1963). “Pattern recognition using generalized portrait method.” Automation and Remote Control Vol.24, pp.774-780.
[9] Vert, Jean-Philippe, Koji Tsuda, and Bernhard Schölkopf (2004). "A primer on kernel methods." Kernel Methods in Computational Biology.
[10] Yin-Wen Chang, Cho-Jui Hsieh, Kai-Wei Chang, Michael Ringgaard and Chih-Jen Lin (2010). "Training and testing low-degree polynomial data mappings via linear SVM". J. Machine Learning Research 11, pp. 1471-1490.

三、網頁部分
[1] Investment Company Institute https://www.ici.org/
[2] Scikit-Learn Machine Learning in Python http://scikit-learn.org/
[3] Stock-AI開放財經資料庫 https://stock-ai.com/
[4] Technical Analysis Library http://www.ta-lib.org/
[5] 中華民國證券投資信託暨顧問商業同業公會 http://www.sitca.org.tw/
[6] 行政院主計處 https://www.dgbas.gov.tw/
[7] 維基百科 https://www.wikipedia.org/