運用台股收盤價以類神經網路建立預測股價模型

Impress on Taiwan stocks closing price to establish a predictive stock price model with a neural network

詹祐驊(Yu-Hua Chan)

指導教授 ： 李勇昇

健行科技大學/電資學院/資訊工程系碩士班/碩士(2019年)

摘要

近年來科技日新月異，計算機運算演進速度大幅提升，促使資料儲存成本降低、累積速度增快，許多資料無論是否有價值都能夠被儲存，透過機器學習方法，讓電腦學習大量的資訊，對於未來的資訊做出預測，造就機器學習在各領域間奠定了舉足輕重的地位，創造出大數據分析一詞，躍升成為各行業熱門的應用工具。
由於證券市場股價變化難測，影響股價的因素多而複雜，因此如何準確預測股票價格走勢一直是投資者極欲研究的題材。因此本研究使用機器學習的方法，利用大量的統計資料進行分析，從大量的資料中挖掘出有用的規則，並利用其規則判斷精確結果。本研究針對台積電公司股票(2330)、中信金(2891)、富邦金(2881)、中鋼(2002)、國泰金(2882) 、統一企業(1216) 、華碩(2357) 、嘉澤(3533)股票收集股票的成交股數、成交筆數、成交金額及開盤價…等47種股票交易資訊，其中還包含了RSI、K、D、MACD、OSC、MTM、W%R、BIAS、OBV、PYS各項股票的技術指標，其目的是希望利用類神經網路(Neural Network)製作模型，預測股價未來的走向，根據最終的研究成果，在判斷隔日收盤價的平均誤差為台積電1.91元、富邦金0.18元、國泰金0.1元，都具有不錯的預測結果，此預測之結果可提供一般投資者、研究者，或公司決策者之參考。

關鍵字

機器學習 ； 類神經網路 ； 多層感知機 ； MLP ； 股票分析

並列摘要

In recent years, technology has been changing with each passing day, and the speed of computer operation has increased dramatically. This has led to a reduction in the cost of data storage and has sped up the accumulating process. Many materials can be stored regardless of their value. Through machine learning methods, computers can absorb tons of information to make predictions for future information. Machine learning now plays a major role in various fields, created the term “Big Data Analytics”, and has become a popular application tool in various industries.
Because the stock market price is unpredictable, the factors that affect the stock price are very complicated. Therefore, how to accurately predict the stock price trend has been the subject of intense research. Therefore, this study uses machine learning methods, using a large volume of statistical data to analyze, mining useful rules from big data, and using its rules to determine accurate results. This study is aimed at 47 stock trading information such as the number of shares, number of transactions, transaction amount and opening price etc. traded in stocks such as TSMC (2330), CTBC Financial Holding Co.Ltd. (2891), Fubon Financial Holding Co., Ltd. (2881), China Steel Corporation (2002), Cathay Financial Holding Co., Ltd. (2882) , Uni-President Enterprises Corp. (1216), ASUSTeK Computer Inc. (2357), LOTES Co., Ltd. (3533.TW), which includes the technical indicators of RSI, K, D, MACD, OSC, MTM, W%R, BIAS, OBV, PYS stocks, the purpose is to use the neural network to make models to predict the future direction of the stock price. According to the final research results, the average deviation of the closing price of the next day is 1.91 NT dollar for TSMC, 0.1045 NT dollar for Cathay Financial Holding Co. Ltd, and 0.18 NT dollar for Fubon Financial Holding Co., Ltd. Both came with good prediction results. The results of this forecast can be used as a reference for general investors, researchers, or company decision makers.

並列關鍵字

machine learning ； neural network ； multi-layer perceptron ； MLP ； stock analysis

參考文獻

1. 林碁域，股票價格趨勢預測之研究，國立高雄大學，亞太工商管理學系碩士在職專班碩士論文，2016。

2. 賀安平，從新聞文章預測股票走勢：使用SVM與LDA演算法，國立高雄應用科技大學，資訊管理研究所碩士班碩士論文，2016。

3. 蔡燿全、陳住銘、楊棠堯、王鐘億，類神經網路應用於股票投資策略之研究，私立逢甲大學，中華管理評論，1999年7月 Vol.2, No.5, pp.25~48。

4 Tariq Rashid , Make Your Own Neural Network CreateSpace Independent Publishing Platform , Seattle, 2016.

5. Farhadi, F. Learning Activation Functions in Deep Neural Networks. Ph.D. Thesis, École Polytechnique de Montréal, Montréal, QC, Canada, 2017.