當量K線為基礎並應用樣本選取之神經網路來預測股票交易策略之研究

Technical Analysis Based On Equivalume Charting for Stock Exchange Using Neural Networks with Instance Selection

詹國忠(Gwo-Jong Jan)

指導教授 ： 謝俊宏

國立臺中科技大學/資訊學群/資訊科技與應用研究所/碩士(2009年)

摘要

在股票市場中，經長時間觀察交易量的變化，對於幫助我們了解股價漲跌之趨勢是很有參考價值的資訊。也正因為如此，在技術分析中的「當量K線圖」(Equivolume Charting)就是設計在K線圖中加入交易量的表示，讓投資者能藉由觀察K線圖更能清楚的觀察出量的變化，來瞭解股票資料的支撐區與阻力區。
類神經網路(Artificial Neural Networks)具有顯著的學習與預測能力，近幾年來，許多運用在股市的預測研究上，它都能提供有效之預測結果，幫助投資人買賣時機點的建議。本文就是應用並開發樣本選取的神經網路(Genetic Algorithm IS,GAIS)，做為在改善類神經網路之效率與輸出結果；並將該網路建立在調量移動平均線(Volume Adjusted Moving Average)與波動難易度(Ease of Movement)這兩個當量K線圖的技術分析指標上的預測模型，用來預測台灣50股票指數基金的漲跌結果，再以預測的結果來擬定投資策略，期能獲取超額的報酬。
本研究以2004年1月至2007年12月的歷史資料做為訓練資料，我們應用此模型，針對2008年1月至2008年12月之資料進行預測。實驗結果顯示，樣本選取的神經網路預測準確度平均達百分之七十以上，而且使用此模型應用於買賣的模擬交易策略，所輸出之結果，比單純同時運用調量移動平均線與波動難易度技術分析的方法，有較好的獲利表現，確實可提供投資人進行投資時之參考依據。

關鍵字

樣本選取 ； 技術分析 ； 調量移動平均線 ； 波動難易度

並列摘要

In the stock markets, by having long-termed observation on the change of trading volume, it provides valuable information for us in understanding the future trends of the stock price fluctuations. Because of this reason, the Equivolume Charting in the technical analysis is designed to add up the trade volume indication on “K Chart”; so that, the investors can easily notice the changing volume of the trades, and understand the stocks’ support and obstacle areas by observing the K charts.
Due to the Artificial Neural Networks own obvious learning and predicting ability, some researches in recent years have been applying it in predicting the future trend of the stock markets, which also advise investors getting the right timing in buying and selling. This article uses and develops the Neural Networks with Instance Selection (GAIS) to be the outcomes and efficiencies in improving the Artificial Neural Networks. Meanwhile, it also employs GAIS to establish the two technical analysis indicators, the Volume Adjusted Moving Average and the prediction model of Ease of Movement in Equivolume Charting, to predict the fluctuation results of ETF index of the 50 Taiwan Stocks. Then, investors normally will employ the prediction results on their investment strategies to expect obtain profitable outcomes.
This research uses the historical data between January 2004 and December 2007 as the training data. By using this model, we apply it to the prediction of the data between Jan 2008 to Dec 2008. The experimental results showed that the average prediction accuracy of the Neural Networks with Instance Selection is over 70%. In addition, if we implement the outputs of this model in the simulation of the trading strategies for real transactions, normally, we will get better and profitable results. It’s definitely a valuable reference material for investors when doing investments.

並列關鍵字

Instance selection ； Technical analysis ； Volume Adjusted Moving Average ； Ease of Movement

參考文獻

3.E. Fama, Efficient capital markets: II, Journal of Finance, P1575-1617, 1991.

4.Hansen J. V., McDonald J. B., Nelson R. D., “Time series prediction with genetic-algorithm designed neural networks: An empirical comparison with modern statistical models”, Computational Intelligence, P171-184, 1999.

5.Kai Fu, Wenhua Xu, ”Traning neural network with genetic algorithms of forecasting the stockprice index”, IEEE IOnernational Conference Intelligent processing Systems, P401-403, 1997.

6.Kyoung-jae Kim, “Artificial neural networks with evolutionary instance selection for financial forecasting”, Expert Systems with Applications, P519-526, 2006.

8.Monica Lam, “Neural network techniques for financial performance prediction:integrating fundamental and technical analysis”, Decision Support Systems, P567-581, 2004.