高频选股因子周报（20251208- 20251212）：高频因子走势分化，多粒度因子显著回撤。AI 增强组合均大幅度回撤。-20251214

• The report discusses the performance of high-frequency factors, deep learning factors, and AI-enhanced portfolios, highlighting significant drawdowns in multi-granularity factors and AI-enhanced combinations during the week of December 8-12, 2025 [1][2][5] - High-frequency factors include intraday skewness, downside volatility proportion, post-open buying intention proportion, post-open buying intention intensity, post-open large-order net buying proportion, post-open large-order net buying intensity, improved reversal, end-of-day transaction proportion, average single-outflow amount proportion, and large-order-driven price increase [6][8][9] - Deep learning factors include GRU(50,2)+NN(10), residual attention LSTM(48,2)+NN(10), multi-granularity models with 5-day and 10-day labels, trained using bidirectional AGRU [6][8][65][66] - AI-enhanced portfolios are constructed based on deep learning factors, including CSI 500 and CSI 1000 indices under wide and strict constraint conditions. Optimization aims to maximize expected returns using the formula: $max\sum\mu_{i}w_{i}$, where $w_i$ is the weight of stock $i$ and $\mu_i$ is the expected excess return of stock $i$ [70][71][72] - Intraday skewness factor achieved historical IC of 0.026 and 2025 IC of 0.030, with 2025 multi-long-short returns of 16.44% and monthly win rate of 7/12 [9][10] - Downside volatility proportion factor achieved historical IC of 0.024 and 2025 IC of 0.025, with 2025 multi-long-short returns of 12.84% and monthly win rate of 8/12 [9][10] - Post-open buying intention proportion factor achieved historical IC of 0.031 and 2025 IC of 0.023, with 2025 multi-long-short returns of 10.49% and monthly win rate of 7/12 [9][10] - Post-open buying intention intensity factor achieved historical IC of 0.035 and 2025 IC of 0.025, with 2025 multi-long-short returns of 13.52% and monthly win rate of 10/12 [9][10] - Post-open large-order net buying proportion factor achieved historical IC of 0.040 and 2025 IC of 0.029, with 2025 multi-long-short returns of 17.93% and monthly win rate of 10/12 [9][10] - Post-open large-order net buying intensity factor achieved historical IC of 0.032 and 2025 IC of 0.023, with 2025 multi-long-short returns of 13.05% and monthly win rate of 9/12 [9][10] - Improved reversal factor achieved historical IC of 0.031 and 2025 IC of 0.011, with 2025 multi-long-short returns of 4.38% and monthly win rate of 6/12 [9][10] - End-of-day transaction proportion factor achieved historical IC of 0.049 and 2025 IC of 0.027, with 2025 multi-long-short returns of 16.82% and monthly win rate of 9/12 [9][10] - Average single-outflow amount proportion factor achieved historical IC of 0.018 and 2025 IC of -0.006, with 2025 multi-long-short returns of -2.57% and monthly win rate of 5/12 [9][10] - Large-order-driven price increase factor achieved historical IC of 0.015 and 2025 IC of 0.001, with 2025 multi-long-short returns of 4.72% and monthly win rate of 7/12 [9][10] - GRU(50,2)+NN(10) factor achieved historical IC of 0.066 and 2025 IC of 0.045, with 2025 multi-long-short returns of 45.90% and weekly win rate of 40/50 [12][60] - Residual attention LSTM(48,2)+NN(10) factor achieved historical IC of 0.062 and 2025 IC of 0.043, with 2025 multi-long-short returns of 45.73% and weekly win rate of 45/50 [12][62] - Multi-granularity model (5-day label) factor achieved historical IC of 0.080 and 2025 IC of 0.064, with 2025 multi-long-short returns of 65.67% and weekly win rate of 44/50 [12][65] - Multi-granularity model (10-day label) factor achieved historical IC of 0.073 and 2025 IC of 0.059, with 2025 multi-long-short returns of 60.45% and weekly win rate of 44/50 [12][66] - CSI 500 AI-enhanced wide constraint portfolio achieved weekly returns of -2.80%, December returns of -3.04%, and 2025 YTD returns of 5.03%, with weekly win rate of 28/50 [13][76] - CSI 500 AI-enhanced strict constraint portfolio achieved weekly returns of -2.42%, December returns of -2.51%, and 2025 YTD returns of 8.24%, with weekly win rate of 32/50 [13][79] - CSI 1000 AI-enhanced wide constraint portfolio achieved weekly returns of -3.54%, December returns of -4.18%, and 2025 YTD returns of 13.63%, with weekly win rate of 33/50 [13][80] - CSI 1000 AI-enhanced strict constraint portfolio achieved weekly returns of -2.35%, December returns of -2.88%, and 2025 YTD returns of 17.29%, with weekly win rate of 32/50 [13][86]