行业轮动周报：融资盘被动爆仓导致大幅净流出，GRU模型仍未配置成长-20250414

• The report discusses two main quantitative models: the Diffusion Index Model and the GRU Factor Model[3][5][6] Diffusion Index Model 1. Model Name: Diffusion Index Model 2. Model Construction Idea: The model is based on the principle of price momentum, capturing industry trends by observing the diffusion index of various sectors[5][6] 3. Model Construction Process: - Calculate the diffusion index for each industry based on price momentum - Rank industries by their diffusion index values - Select top-ranked industries for portfolio construction - Formula: $ \text{Diffusion Index} = \frac{\text{Number of stocks with positive momentum}}{\text{Total number of stocks}} $ 4. Model Evaluation: The model has shown varying performance over different years, with notable returns in some periods and significant drawdowns in others[24][25] 5. Model Test Results: - 2025 YTD excess return: -3.78%[23] - April 2025 average return: -4.42%, excess return: -0.59%[28] GRU Factor Model 1. Model Name: GRU Factor Model 2. Model Construction Idea: The model leverages GRU (Gated Recurrent Unit) deep learning networks to analyze minute-level price and volume data, aiming to capture short-term trading signals[6][13] 3. Model Construction Process: - Collect minute-level price and volume data - Train a GRU network to predict industry performance - Rank industries based on GRU factor scores - Select top-ranked industries for portfolio construction 4. Model Evaluation: The model has been effective in capturing trading signals in short cycles but has shown mixed performance in longer cycles[30][31] 5. Model Test Results: - 2025 YTD excess return: -3.82%[30] - April 2025 excess return: 0.33%[33] Model Performance Metrics 1. Diffusion Index Model: - 2025 YTD excess return: -3.78%[23] - April 2025 average return: -4.42%, excess return: -0.59%[28] 2. GRU Factor Model: - 2025 YTD excess return: -3.82%[30] - April 2025 excess return: 0.33%[33]