基于机器学习预测的M-SV投资组合决策研究
Predicted Mean Semi-Variance Portfolio Optimization Based on Machine Learning
张鹏 1李璟欣 1崔淑琳1
作者信息
- 1. 华南师范大学经济与管理学院,广东 广州 510006
- 折叠
摘要
运用随机森林(Random forest,RF)、极端梯度提升(eXtreme Gradient boosting,XGBoost)、支持向量回归(Support vector regression,SVR)三种机器学习方法预测股票的收益率.运用预测的平均股票收益率,计算投资组合的期望收益、下半方差和下半协方差.考虑交易成本、借贷约束和交易量限制等,构建了基于机器学习预测的均值-下半方差(M-SV)投资组合模型,并运用不等式的旋转算法求解.最后,运用中证100指数成分股,进行样本内和样本外实证分析.研究结果表明,在样本内检验中SVR+M-SV模型的有效前沿水平相较于RF+M-SV、XGBoost+M-SV模型更高;在样本外检验中,三种机器学习混合方法+M-SV模型在收益、夏普比率、索提诺比率等的表现优于M-SV和等比例模型.
Abstract
In this paper,the predicted returns rates are forecast by Random Forest,XG-Boost,and SVR.The mean,semi-variance and semi-covariance of the portfolio model cal-culated by using the average predicted returns rates.Considering the threshold constraints,borrowing constraints and transaction costs,the mean semi-variance portfolio model is proposed.The model is solved by the pivoting algorithms.Finally,this paper uses the constituent stocks of CSI 100 index as a sample for empirical analysis.The results show that SVR+M-SV model has a higher level of efficient frontier compared to RF+M-SV and XGBoost+M-SV model in the in-sample test;In the out-of-sample test,the three machine learning hybrid algorithms+M-SV model significantly outperforms the M-SV and 1/N mod-els in terms of return,Sharpe ratio,and Sortino ratio.
关键词
投资组合/均值-下半方差/随机森林/极端梯度提升/支持向量回归Key words
portfolio selection/mean semi-variance/random forest/Xgboost/support vector regression引用本文复制引用
基金项目
国家自然科学基金(71271161)
广东省自然科学基金(2024A1515011808)
出版年
2024
NETL
NSTL
万方数据