基于深度强化学习下的股票量化交易算法设计

Design of stock quantitative trading algorithm based on deep reinforcement learning

孔荫莹 ¹黄志花 ¹邓浩东 ¹唐毅康¹

扫码查看

作者信息

1. 广东财经大学统计与数学学院,广东广州 510320
折叠

摘要

针对股票量化交易中有限数据预测未来价格趋势和智能资产组合配置等难题,采用DeepAR模型来预测股票价格的未来涨跌趋势,根据这些趋势计算涨跌幅精选了 16支有潜力的股票,并运用SAC模型进行智能资产配置.结果表明,DeepAR模型的股票选择有助于SAC模型实现智能资产组合配置,而SAC模型的量化决策也取得了理想的效果.在4个月的时间内,实现了 10.79％的收益率和32.37％的年化收益率.相较于上证指数和沪深300指数有显著的超额收益率,分别为12.47％和21.48％.此外,2016-2022年回测中达到了 1.3％的夏普比率和29％的最大回撤率.

Abstract

In view of the difficulties of predicting future price trend and intelligent asset portfolio allocationwith limited da-tain quantitative stock trading,theDeepAR model was used to predict the futureprice trend of stocks.Based on these trends,16 promising stocks were selected based on their potential price changesand the SAC model was used for intelligent asset alloca-tion.The results showed thatthe stock selection of theDeepAR model can help the SAC model realize intelligent asset portfolio allocation and the quantitative decisions of the SAC model can also achieve desirable results.Over a period of 4month,a return rate of 10.79％and an annualized return rate of 32.37％were achieved.Compared withthe SSE Index andthe CSI 300 Index,there were significant excess return rates of 12.47％and 21.48％,respectively.In addition,a Sharpe ratio of 1.3％and a maxi-mum retracement rate of 29％were achieved in the 2016-2022 retest.

关键词

深度强化学习/量化交易/超额收益/股票预测

Key words

deep reinforcement learning/quantitative trading/excess return/stock forecast

引用本文复制引用

基金项目

广东省基础与应用基础研究基金(2022A1515012429)

广东省教育厅创新团队项目(2022WCXTD009)

广州市科技计划(202201020345)

出版年

2024

南昌大学学报(理科版)

南昌大学

南昌大学学报(理科版)

CSTPCD

影响因子：0.418

ISSN：1006-0464

参考文献量17

段落导航