首页|期刊导航|杭州师范大学学报(自然科学版)|基于机器学习和多因子模型的主动型基金收益预测研究

基于机器学习和多因子模型的主动型基金收益预测研究OA

Research on the predictive ability of active fund returns based on machine learning and multi-factor models

中文摘要英文摘要

为评估业绩排序法、线性多因子模型法和非线性轻量级梯度提升机(light gradient boosting machine,LightGBM)法在主动型基金业绩预测中的有效性,在接近真实市场环境下构建含多种风格因子的基金投资组合,并比较预测表现.结果显示,LightGBM 法在 3 种方法中略占优势.LightGBM 法在最高收益组合上表现最优,市场上涨周期累计净值增长最快,体现了非线性模型对复杂模式更强的捕捉能力.特征重要性与传统因子模型的预测性能结果表明,LightGBM 法能更充分挖掘短周期行为因子等高信息含量因子的价值,并在因子缺失时保持更稳健的预测表现.研究表明,结合高信息含量因子与非线性模型有助于优化基金选择与资产配置.

To evaluate the effectiveness of the performance ranking method,the linear multi-factor model,and the nonlinear light gradient boosting machine(LightGBM)method in predicting the performance of actively managed funds,this study constructed fund portfolios incorporating multiple style factors in a near-real-market environment and compared their predictive performance.The results showed that the LightGBM method had a slight overall advantage among the three approaches.It performed best in the top-return portfolio and achieved the fastest cumulative net asset value growth during market upcycles,demonstrating the stronger capability of nonlinear models in capturing complex patterns.The predictive performance results of feature importance analysis and the traditional factor model indicated that LightGBM more effectively exploited the value of high-information-content factors such as short-horizon behavioral factors,and maintained more robust predictive performance when such factors were removed.The study suggests that combining high-information-content factors with nonlinear models helps optimize fund selection and asset allocation.

林思凝;朱一鸣;潘奇

杭州师范大学信息科学与技术学院,浙江 杭州杭州师范大学信息科学与技术学院,浙江 杭州杭州师范大学信息科学与技术学院,浙江 杭州

管理科学

基金业绩LightGBM多因子模型主动型基金

fund performanceLightGBMmulti-factor modelactive fund

《杭州师范大学学报(自然科学版)》 2026 (5)

476-485,10

10.19926/j.cnki.issn.1674-232X.2024.12.021

评论