国泰海通｜金工：国泰海通量化选股系列（一）——基于PLS模型复合因子预期收益信号的应用研究

Group 1 - The article examines the application of PLS model expected factor returns in factor weighting, focusing on both single-factor multi-strategy and multi-factor single-strategy dimensions [1] - In the top 100 combinations of 20 single factors, using the PLS model for the five most volatile factor combinations resulted in an annualized return increase of approximately 4.0% compared to mean-weighted returns, and 6.6% compared to equal-weighted returns [2] - The article constructs six basic combinations including one dividend selection, one growth selection, two small-cap combinations, and two relatively balanced style combinations, achieving an annualized return increase of 3.3% over excess return mean weighting and 3.9% over equal weighting for volatile combinations [2] Group 2 - In multi-factor models, using PLS expected returns to determine factor weights can improve the expected IC and performance of top 100 combinations, although this improvement is not consistent across all cross-sections [3] - The PLS weighting method is noted to be more robust overall, but may underperform compared to mean IC weighting and ICIR weighting when factor momentum is strong, as observed in 2023 [3] - A composite quantitative fixed income + strategy using PLS expected return weighted multi-factor model for the stock side and the China Bond Short-term Index for the bond side achieved an annualized return of 8.1% with a volatility of 5.6% and a maximum drawdown of 5.4% from January 2018 to November 2025 [3]