基于机器学习的基金收益预测与投资组合研究

Research on Fund Return Prediction and Investment Portfolio Based on Machine Learning

文章使用公募基金定期披露的基本信息构建了15个基金基本特征,以2009—2023年的权益型公募基金为样本,分别采用传统线性模型、决策树模型、随机森林模型来预测基金的未来收益,并通过置换检验与信息系数分析考察特征有效性。在此基础上,采用分组及多空检验对模型性能进行评估分析。实证结果表明,机器学习模型能够有效发掘基金基本特征中蕴含的非线性收益预测信息,根据模型预测基金收益构建投资组合,多头组合可以实现年化收益率达17.12%;相较于传统线性模型,机器学习模型在特征筛选与组合优化维度更具优势。

The article constructs 15 basic features of fund using the bas ic information regularly disclosed by mutual funds.Using equity mutual funds from 2009 to 2023 as samples,traditional linear models,decision tree models,and random forest models are employed to predict the future returns of the fund respectively,and the validity of the features is examined through permutation tests and information coefficients.On this basis,grouping and multi-space testing are used to evaluate the performance of the model.The empirical results suggest that the machine learning model can effectively explore the non-linear return prediction information in the basic characteristics of funds,and construct investment portfolios based on the model's predicted fund returns.Multiple portfolios can achieve an annualized return of up to 17.12%;Compared to traditional linear models,machine learning models are more advantageous in terms of feature screening and portfolio optimization.