行业轮动周报：指数震荡反内卷方向领涨，ETF持续净流入金融地产-20250922

Quantitative Models and Construction Methods 1. Model Name: Diffusion Index Industry Rotation Model - Model Construction Idea: The model is based on the principle of price momentum, aiming to capture upward trends in industries through a diffusion index[26][27] - Model Construction Process: 1. Calculate the diffusion index for each industry based on price momentum 2. Rank industries by their diffusion index values 3. Select top industries for allocation based on their rankings 4. Adjust the portfolio monthly or weekly based on updated diffusion index rankings[26][27] - Model Evaluation: The model has shown stable performance in certain years (e.g., 2022 with an annual excess return of 6.12%) but struggled during market reversals or concentrated market themes, such as in 2024 and 2025[26][33] 2. Model Name: GRU Factor Industry Rotation Model - Model Construction Idea: This model leverages GRU (Gated Recurrent Unit) deep learning networks to process high-frequency volume and price data, aiming to identify industry rotation opportunities[38] - Model Construction Process: 1. Input high-frequency volume and price data into the GRU network 2. Train the GRU model on historical data to identify patterns in industry rotation 3. Generate factor scores for industries based on the GRU model's output 4. Rank industries by their GRU factor scores and allocate to top-ranked industries[38][34] - Model Evaluation: The model performs well in short cycles but struggles in long cycles or extreme market conditions. It has shown difficulty in capturing excess returns in concentrated market themes during 2025[33][38] --- Model Backtesting Results 1. Diffusion Index Industry Rotation Model - Weekly Average Return: -1.74%[30] - Excess Return (Weekly): -1.41%[30] - Excess Return (September 2025): -1.88%[30] - Excess Return (2025 YTD): 2.76%[25][30] 2. GRU Factor Industry Rotation Model - Weekly Average Return: -0.72%[36] - Excess Return (Weekly): -0.38%[36] - Excess Return (September 2025): -0.10%[36] - Excess Return (2025 YTD): -7.78%[33][36] --- Quantitative Factors and Construction Methods 1. Factor Name: Diffusion Index - Factor Construction Idea: Measures the breadth of price momentum across industries to identify upward trends[26][27] - Factor Construction Process: 1. Calculate the proportion of stocks in an industry with positive price momentum 2. Aggregate these proportions to derive the diffusion index for the industry 3. Rank industries based on their diffusion index values[27][28] - Factor Evaluation: Effective in capturing upward trends but vulnerable to reversals and underperformance in counter-trend markets[26][33] 2. Factor Name: GRU Factor - Factor Construction Idea: Utilizes GRU deep learning to analyze high-frequency trading data and generate predictive scores for industry rotation[38] - Factor Construction Process: 1. Input high-frequency trading data into the GRU network 2. Train the model to recognize patterns in industry rotation 3. Output factor scores for industries based on the model's predictions[38][34] - Factor Evaluation: Strong in short-term predictions but less effective in long-term or extreme market conditions[33][38] --- Factor Backtesting Results 1. Diffusion Index - Top Industries (Weekly): Non-ferrous Metals (0.978), Banking (0.968), Communication (0.946), Electronics (0.877), Automotive (0.874), Retail (0.873)[27] - Bottom Industries (Weekly): Food & Beverage (0.354), Real Estate (0.46), Coal (0.487), Transportation (0.543), Construction (0.574), Building Materials (0.618)[27] 2. GRU Factor - Top Industries (Weekly): Non-ferrous Metals (7.4), Petrochemicals (5.38), Coal (4.17), Steel (4.15), Building Materials (3.46), Non-banking Financials (3.08)[34] - Bottom Industries (Weekly): Comprehensive Finance (-19.42), Utilities (-13.41), Electronics (-13.18), Pharmaceuticals (-11.14), Automotive (-10.07), Consumer Services (-10.04)[34]