基于多维高频数据和LSTM模型的沪深300股指期货价格预测

Forecast of CSIF 300 price based on multi-dimensional&high-frequency data and LSTM model

以2010—2019年的沪深300股指期货为对象,收集日收盘价、5分钟收盘价,以及影响波动的5维度89个指标,采用维度删减、间隔采样方法,组合成多个不同维度和不同频率的LSTM深度学习模型对沪深300股指期货进行预测,并且从空间和时间角度分析维度和频率对股指期货价格波动的影响。研究表明:LSTM模型可以很好地描绘沪深300股指期货多维高频数据的特征;空间上,变量维度对沪深300股指期货价格的预测带来间接影响,预测精度最高的出现在10至20个交易日区间;时间上,数据频率的影响更为直接,频率越高预测精度越高。研究结论有助于股指期货参与各方分散和化解金融风险。

Taking the CSIF 300 from 2010 to 2019 as the object,this paper collects daily closing price,5-minute closing price,and 895-dimensional indicators affecting fluctuation,and uses the methods of dimension deletion and interval sampling to combine them into LSTM deep learning models with different dimensions and different frequencies to predict the closing price of CSIF 300.It also analyzes the impact of dimension and frequency on the price fluctuation of stock index futures from the perspective of space and time.The research shows that the LSTM model can well describe the characteristics of multidimensional high-frequency data of CSIF 300.Spatially,the variable dimension has an indirect impact on the prediction of the price of CSIF 300.The highest prediction accuracy occurs in the range of 10 to 20 trading days.In terms of time,the influence of data frequency is more direct.The higher the frequency,the higher the prediction accuracy.The research conclusion is helpful for all parties involved in stock index futures to disperse and resolve financial risks.