首页|多约束条件放松下机构投资者资产组合变动与业绩检验和预测

多约束条件放松下机构投资者资产组合变动与业绩检验和预测

扫码查看
本文研究了我国资本市场改革不断深化、投资环境日益改善背景下,机构投资者业绩是否"与时俱进"及如何预测问题.首先将经典资产组合模型作为基准模型,分两种情况分别给出有效证券数量增加时有效前沿变动的数理解析;然后以开放式基金为机构投资者代表,使用周夏普比率(SR)检验其业绩,并借助双向门控循环单元模型(BiGRU)预测其业绩.研究发现:随着有效证券数量增加,最优资产组合业绩在理论上应不断向好;实践中机构投资者业绩在纵向上并无显著增大趋势,也无长短期自相关性,该结论在不同市场周期、不同模态和不同样本组中都稳健;基于深度学习方法的业绩预测模型性能良好,预测结果相比传统机器学习方法的预测结果更优.
Portfolio Movement,Performance Test and Prediction of Institutional Investors under Multi-Constraints Relaxation
With the reform measures being implemented successively and the investment environment being optimized constantly in our capital market,this paper studies whether the performance of insti-tutional investors has being improved synchronously and how to predict it.We firstly use the classic portfolio selection model as benchmark model,and give the mathematical connotation of the effective frontier movement while the effective securities increase in two cases,which provides a theoretical expla-nation for portfolio optimization.Then we choose the open-ended funds as samples,test the statistical significance of their performance by using the data of weekly Sharpe ratio,and predict their performance by using the bidirectional gated recurrent unit model.Some conclusions are drawn as following.In theory,as the effective securities increasing,the effective frontier necessarily moves to the left and the optimal portfolio should continue to improve.However,in reality,there is not any obvious statistical evidence proving that the performance of institutional investors has sustained upward trend during the survey period,or has any long-term or short-term autocorrelation,which is robust across the test of different market cycles,modalities,and sample groups.The prediction model based on deep learning method has better performance than the traditional machine learning methods.

portfolio theoryeffective securitiesSharpe ratiobidirectional gated recurrent unitinsti-tutional investors

刘广、刘汉中

展开 >

广州大学经济与统计学院,广东 广州 510006

资产组合理论 有效证券 夏普比率 BiGRU 机构投资者

2024

数理统计与管理
中国现场统计研究会

数理统计与管理

CSTPCDCSSCICHSSCD北大核心
影响因子:1.114
ISSN:1002-1566
年,卷(期):2024.43(6)