标题: 应用倒传递类神经网路于开盘行为研究分析以台湾加权指数为例
Applying Back Propagation Neural Network for Discovery Behavior of Opening Patterns of Taiwan Stock Market
作者: 黄万成
Huang, Wan-Cheng
陈安斌
Chen, An-Pin
管理学院资讯管理学程
关键字: 开盘行为;倒传递类神经网路;台湾股价加权指数;Behavior of Opening Patterns;Backpropagation Neural Network;Taiwan Stock Price Index
公开日期: 2009
摘要: 由于资讯的迅速传递,台湾股市又容易受到突发事件的影响,有不少投资人的财产经过一夜股市的变动而大幅缩水,导致股市投资者追求低风险的投资机会,以降低不可掌握的风险则显得更为迫切,而面对隔夜的风险通常使得投资人长期的获利于一夕之间即回吐,因此长期性投资人则需要一套避险的模式来保全自己的财产。
本研究应用人工智慧领域中的倒传递类神经网路,对台湾加权指数之开盘后十五分钟的历史资料,以时间09:05、09:10、09:15收盘价之涨跌型态做分群,产生八个模型的群组,再将各群中的资料输入至倒传递类神经网路,预测相对于当日内台湾加权指数收盘价的涨幅程度,并以台湾期货指数为投资交易对象来实验。实验结果证实,实验模型经由分群后结合倒传递类神经网路,来预测准确率显着优于对照组只有使用倒传递类神经网路及随机漫步的模型,而且模型中以M2(涨、涨、跌)、M7(跌、跌、涨)的投资交易准确度及获利力为最佳之绩效模型。因此,以上实验藉由分群技术辅助更能掌握环境的变化去作动态学习,进而提供投资人更明确的买卖交易资讯,以辅助决策者做正确抉择的依据。
Because the rapid change of information transmission, Taiwan stock market is also easy to influence by unexpected events. Many investors’ property changes in the stock market overnight sometimes shrink substantially. Therefore, it is urgent that the stock market investors seeking low risk investment opportunities to reduce the unpredictable risk. In order to prevent the dramatic overnight losses, the long-term investors need to build a hedging model to save their own property.
In this study, we apply the theory of artificial intelligence in the field of back-propagation neural network to clustering the historical data of the behavior of opening patterns after 15 minutes in Taiwan weighted index price by the time 09:05, 09:10, 09:15 of the closing price. Produce eight types of groups, then each group of data entry to the back-propagation neural networks to predict relative to the same day's closing price of the Taiwan stock price index, and tests the investors in Taiwan's futures index as trading partners.The experimental result confirmed that after the experimental model through the combination of clustering propagation neural network to predict the exact rate was significantly better than the control group which only using back propagation neural network and the random walk model. In addition, the model with M2 (up, up, down), M7 (down, down, up) of the investment transaction accuracy and the profitability are the best profit performance model. Therefore, these experiments assisted by clustering has better grasp of the changes in the environment to make dynamic learning. Thus provide investors with more specific transactions information to assist decision-makers to make the right choice.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT079664526
http://hdl.handle.net/11536/43729
显示于类别:Thesis