基于LSTM神经网络的沪深300指数预测模型研究
|
摘要
将LSTM用于沪深300指数的股价预测中,并在通用变量开盘价、收盘价、最高价、最低价的基础上新加入了日成交量与日成交额,以此来预测第二日的最高价,获得了比较好的预测效果,并与SVR模型和Adaboost模型预测作对比,LSTM获得的测试集RMSE要更低.接着,用SVR、Adaboost和LSTM进行岭回归集成,即,先用训练集对这三种模型进行训练,然后用训练数据进行测试,将它们的测试结果作为自变量,以相应的真实第二日最高价作因变量,进行岭回归,再对测试集数据做出预测,得到测试集的RMSE进一步降低;再者,查看回归方程发现SVR系数为负,与因变量呈负相关关系,进一步选取Adaboost和LSTM两种模型在训练集上的预测结果做自变量,相应的真实第二日最高价作因变量,再次进行岭回归,得到测试集的RMSE再次降低,进一步验证了回归集成算法的有效性,可以为广大投资者做买卖决策时提供重要的参考价值.
This article, LSTM is used to predict the stock price of the CSI 300 index, in which, daily volume and daily turnover are added as the variables on the basis of the most frequently used variables: opening, closing, highest and lowest price. The highest price of the second day is forecasted and good prediction effect is obtained. Compared to the SVR and Adaboost, LSTM gets a lower test set RMSE. Then, the three models of SVR, Adaboost and LSTM are ensembled by ridge regression, namely, the three models are trained first with the training data set, then they are tested also by using training data set and three test results are got. Further, the three test results are used to be independent variables and the corresponding real second highest prices are used to be dependent variables, so that, a ridge regression model is solved. The test data set is input into the regression ensemble algorithm,and the corresponding second highest prices are predicted, we find that the RMSE of test data set is further reduced. Additionally, we find that the regression coefficient of SVR in the above ridge regression model is negative, and this means that it presents negative correlation with the dependent variable. So, we only ensemble Adaboost and LSTM by ridge regression,the training and the testing processes are the same as the above regression ensembling, and the RMSE of test data set is reduced once again. The effectiveness of the regression ensemble algorithm is further verified, and this provides important reference value for investors when they make buying and selling decisions.
This article, LSTM is used to predict the stock price of the CSI 300 index, in which, daily volume and daily turnover are added as the variables on the basis of the most frequently used variables: opening, closing, highest and lowest price. The highest price of the second day is forecasted and good prediction effect is obtained. Compared to the SVR and Adaboost, LSTM gets a lower test set RMSE. Then, the three models of SVR, Adaboost and LSTM are ensembled by ridge regression, namely, the three models are trained first with the training data set, then they are tested also by using training data set and three test results are got. Further, the three test results are used to be independent variables and the corresponding real second highest prices are used to be dependent variables, so that, a ridge regression model is solved. The test data set is input into the regression ensemble algorithm,and the corresponding second highest prices are predicted, we find that the RMSE of test data set is further reduced. Additionally, we find that the regression coefficient of SVR in the above ridge regression model is negative, and this means that it presents negative correlation with the dependent variable. So, we only ensemble Adaboost and LSTM by ridge regression,the training and the testing processes are the same as the above regression ensembling, and the RMSE of test data set is reduced once again. The effectiveness of the regression ensemble algorithm is further verified, and this provides important reference value for investors when they make buying and selling decisions.
引文
[1] Rounaghi M M, Zadeh F N. Investigation of market efficiency and financial stability between S&P500 and London stock exchange:monthly and yearly forecasting of time series stock returns using ARMA model[J]. Physica A Statistical Mechanics&Its Applications, 2016, 456:10-21.
[2] Lin Z. Modelling and forecasting the stock market volatility of SSE composite index using GARCH models[J]. Future Generation Computer Systems, 2018, 79:960-972.
[3] Herwartz H. Stock return prediction under GARCH-An empirical assessment[J]. International Journal of Forecasting, 2017, 33(3):569-580.
[4] White H. Economic prediction using neural networks:the case of IBM daily stock returns[C]. 1988,2(2):451-458.
[5] Ozbayoglu M. Neural based techical analysis in stock market forecasting[J]. Intelligent Engineering Systems through Artificial Neural Networks, 2008, 18:261-265.
[6] Hammad A A A, Ali S M A, Hall E L. Forecasting the Jordanian stock prices using artificial neural networks[M]. Intelligent Engineering Systems Through Artificial Neural Networks. 2007:502-505.
[7] Pang X, Zhou Y, Wang P, et al. An innovative neural network approach for stock market prediction[J]. Journal of Supercomputing, 2018(1):1-21.
[8]林楠.基于BP神经网络和GARCH模型的中国银行股票价格预测实证分析[D].兰州:兰州大学,2014.
[9] Graves A. Supervised Sequence Labelling[M]. Springer Berlin Heidelberg, 2012:5-13.
[10] Oz C, Ming C L. American sign language word recognition with a sensory glove using artificial neural networks[J]. Engineering Applications of Artificial Intelligence. 2011, 24(7):1204-1213.
[11] Hochreiter S, Schmidhuber J. Long Short-term Memory[J]. Neural Computation, 1997, 12(9):1735-1780.
[12]孙瑞奇.基于LSTM神经网络的美股股指价格趋势预测模型的研究[D].北京:首都经济贸易大学,2015.
[2] Lin Z. Modelling and forecasting the stock market volatility of SSE composite index using GARCH models[J]. Future Generation Computer Systems, 2018, 79:960-972.
[3] Herwartz H. Stock return prediction under GARCH-An empirical assessment[J]. International Journal of Forecasting, 2017, 33(3):569-580.
[4] White H. Economic prediction using neural networks:the case of IBM daily stock returns[C]. 1988,2(2):451-458.
[5] Ozbayoglu M. Neural based techical analysis in stock market forecasting[J]. Intelligent Engineering Systems through Artificial Neural Networks, 2008, 18:261-265.
[6] Hammad A A A, Ali S M A, Hall E L. Forecasting the Jordanian stock prices using artificial neural networks[M]. Intelligent Engineering Systems Through Artificial Neural Networks. 2007:502-505.
[7] Pang X, Zhou Y, Wang P, et al. An innovative neural network approach for stock market prediction[J]. Journal of Supercomputing, 2018(1):1-21.
[8]林楠.基于BP神经网络和GARCH模型的中国银行股票价格预测实证分析[D].兰州:兰州大学,2014.
[9] Graves A. Supervised Sequence Labelling[M]. Springer Berlin Heidelberg, 2012:5-13.
[10] Oz C, Ming C L. American sign language word recognition with a sensory glove using artificial neural networks[J]. Engineering Applications of Artificial Intelligence. 2011, 24(7):1204-1213.
[11] Hochreiter S, Schmidhuber J. Long Short-term Memory[J]. Neural Computation, 1997, 12(9):1735-1780.
[12]孙瑞奇.基于LSTM神经网络的美股股指价格趋势预测模型的研究[D].北京:首都经济贸易大学,2015.