摘要
股票价格预测作为金融预测领域中一项重要的研究方向,准确预测股票价格的涨跌可以帮助投资者盈利或及时止损.经研究发现,某些因素(如政策、社会突发事件等)会对同板块下的多只股票价格产生影响,导致同板块的多只股票在某个时间段内出现相似的走势,即板块效应.因此,同板块下多只股票的价格走势对于股票预测具有参考作用.针对这一现象,提出了一种基于板块效应的深度学习股价走势预测方法.首先,使用皮尔森(Pearson)相关系数和XGBoost算法对同板块下多只股票的收盘价进行分析,以筛选出与预测股票相关性高的多只股票,并使用自编码器对这些股票的收盘价进行降维,以提取股票的价格走势;其次,构建了一个基于卷积神经网络(convolutional neural networks, CNN)与长短期记忆(long short-term memory, LSTM)网络的混合深度学习预测模型,使用一维卷积神经网络提取输入数据的特征,使用LSTM网络对股票价格进行预测.该模型使用银行、医药、酒业、娱乐传媒4个板块的股票作为实验数据集.为了提高模型的预测效果,通过随机搜索对LSTM网络的神经元个数进行简单的分析,以选择较优的神经元个数.最后,通过实验分析,基于同板块数据集的深度学习预测模型具有良好的预测效果.
As an important research direction in the field of financial forecasting, accurate prediction of stock price rise and fall can help investors to make profits or stop losses in time. It has been found that certain factors(such as policies, social emergencies)can have an impact on the prices of multiple stocks in the same sector, resulting in similar movements of multiple stocks in the same sector in a certain period of time, i.e. the sector effect. Therefore, the price trends of multiple stocks under the same segment are useful for stock forecasting. To address this phenomenon, a deep learning stock price trend prediction method based on the plate effect is proposed. Firstly, the Pearson correlation coefficient and XGBoost algorithm are used to analyze the closing prices of many stocks in the same sector so as to screen out the stocks with high correlation with the predicted stocks. Then, the autoencoder is used to reduce the dimension of the closing prices of these stocks, so as to extract the price trend of the stocks. Secondly, a hybrid deep learning prediction model based on convolutional neural network and long short-term memory network is constructed. One-dimensional convolutional neural network is used to extract the features of input data, and LSTM network is used to predict stock prices. The model uses stocks in four sectors, namely, banking, pharmaceuticals, alcohol, and entertainment media, as the experimental data set. In order to improve the prediction effect of the model, the number of neurons of the LSTM network is simply analyzed by random search to select the better number of neurons. Finally, the experimental analysis shows that the deep learning prediction model based on the same board dataset has good prediction effect.
作者
李庆涛
林培光
王基厚
周佳倩
张燕
蹇木伟
Li Qingtao;Lin Peiguang;Wang Jihou;Zhou Jiaqian;Zhang Yan;Jian Muwei(School of Computer Science and Technology,Shandong University of Finance and Economics,Jinan 250014,China)
出处
《南京师范大学学报（工程技术版）》
CAS
2022年第1期30-38,共9页
Journal of Nanjing Normal University(Engineering and Technology Edition)
基金
国家自然科学基金项目(61802230)。
