摘要
近两年来,Google团队提出的BERT模型被越来越多地应用于文本分类任务中.在BERT模型的基础上,文章提出了一个基于新闻文本挖掘的股指期货高频预测模型,进而设计了相应的高频交易策略.文章基于股指期货的高频价格波动为每条新闻赋予涨跌平标签,利用所提出模型对新闻进行分类,从而预测三大股指期货价格的涨跌平方向,并完成股指期货模拟高频交易.基于5年半以来的三大股指期货的高频数据及证券新闻文本的实证研究显示,文章提出的预测模型和交易策略取得了较高的准确率和收益率,且在中证500股指期货上表现最好.
In the past two years,the BERT model proposed by Google has been increasingly used in text classification tasks.Based on the BERT model,this article proposes a high-frequency prediction model for stock index futures based on news text mining,and then designs the corresponding high-frequency trading strategy.This paper marks each news with rising,fall or flat label by the price fluctuations of stock index futures,and train the model with the news in the training set.The trained model is applied to classify the news in the test set and conduct simulated trading of stock index futures.In order to verify the performance of the model on different test set lengths and different stock index futures,this paper sets up 18 independent experiments based on three test set lengths and three stock index futures.The news in the experimental study is published by the four major securities newspapers and Xinhua News Agency in the last five and a half years.The empirical results show that the proposed forecasting model and trading strategy have achieved high accuracy and yield,and the model works best on CSI 500 index futures.
作者
徐维军
付志能
李茂昌
张卫国
XU Weijun;FU Zhineng;LI Maochang;ZHANG Weiguo(School of Business Administration,South China University of Technology,Guangzhou 510641;Ping An Life Insurance of China,Shenzhen 518000)
出处
《系统科学与数学》
CSCD
北大核心
2021年第7期1856-1875,共20页
Journal of Systems Science and Mathematical Sciences
基金
国家自然科学基金面上项目(71771091)
国家自然科学基金重点国际(地区)合作与交流项目(71720107002)
广东省基础与应用基础研究基金(2019A1515011752)
科技部科技创新2030“新一代人工智能”重大项目(2020AAA0108404)资助课题
关键词
股指期货
高频预测
证券新闻
BERT模型
文本挖掘
Stock index futures
high-frequency forecasting
securities news
BERT
text mining
