基于注意力机制的LSTM股价趋势预测研究

针对中国股票市场,提出了一种基于注意力机制的LSTM股价趋势预测模型。选取42只中国上证50从2009年到2017年的股票数据为实验对象,根据股票市场普遍认可的经验规则,分别对每个技术指标进行量化处理得到股票涨跌的趋势数据,并和交易数据混合作为预测模型的输入,然后使用基于注意力机制的LSTM模型提取股价趋势特征进行预测。实验结果表明:引入股票离散型趋势数据到预测模型中,能够在已有交易数据和技术指标的基础上提升预测精确度,与传统的机器学习模型SVM和单一的LSTM模型相比,基于注意力机制的LSTM模型具有更好的预测能力。

Attention-Based LSTM for Stock Price Movements Prediction

This paper addresses problem of stock price movements prediction for china stock markets.We present an Attention-Based LSTM approach to predict stock price movements.Nine years of historical data from 2009 to 2017 of 42 stocks of SSE 50 are selected for experimental evaluation.According to the empirical rules generally accepted by the stock market,the stock technical indicators are quantified to obtain the stock price movements prediction and together with the trading data as input to the prediction model.Then,we use Attention-Based LSTM to extract important features for prediction.The experimental results suggest that introducing stock discrete trend data into the prediction model can achieve higher prediction accuracy based on trading data and technical indicators.Experimental results also show that the Attention-Based LSTM model outperforms both traditional machine learning model SVM and the single LSTM model on overall performance.