Research on Financial Data Prediction Based on Deep Autoencoder
Download as PDF
Ziming Gao, Jinchang Zhang, Boyang Nan
As a revolutionary achievement in machine learning in the era of big data, deep learning has developed rapidly since its introduction, and it has created an upsurge of research and application in the Internet field.From the perspective of market microstructure, the formation and change of stock prices are determined by the trading behavior of buyers and sellers. Therefore, the mining of high-frequency market data may have a predictive ability for future stock price movements.In this paper, the deep learning prediction model is trained by a large amount of historical data in the sample, and the price index of the stock index futures is predicted at 1 sec. The accuracy of the model outside the sample is over 73%.At the same time, based on the prediction of stock price changes by deep learning stock price forecasting model, this paper proposes the intraday trading strategy of stock index futures.The trading strategy has accumulated 99.6% yield since 2013, with an annualized rate of return of 77.6% and a maximum retracement of -5.86%.Through the empirical study of the high-frequency price forecasting model of stock index futures, this paper verifies the effectiveness of the machine learning tool in the big data era of deep learning in stock price forecasting, and proposes a stock index futures trading strategy based on the forecasting model, which has achieved good results.
Deep Learning, Stock Index Futures, Trend Forecasting, Trading Strategy, Deep Autoencoder