行业轮动周报：指数回撤下融资资金净流出，ETF资金大幅净流入，GRU调入传媒-20251125

Quantitative Models and Construction Methods 1. Model Name: Diffusion Index Model - Model Construction Idea: The model is based on the principle of price momentum, aiming to capture upward trends in industries and sectors[22][23] - Model Construction Process: The diffusion index is calculated for each industry based on its price momentum. The model ranks industries by their diffusion index values and selects the top-performing industries for portfolio allocation. The model has been tracking out-of-sample performance since 2021, with adjustments made monthly or weekly based on updated diffusion index rankings[22][23] - Model Evaluation: The model has shown strong performance in capturing industry trends during momentum-driven markets but struggles during market reversals[22][36] 2. Model Name: GRU Factor Model - Model Construction Idea: This model leverages minute-level price and volume data processed through a GRU (Gated Recurrent Unit) deep learning network to generate industry factors for rotation strategies[37] - Model Construction Process: The GRU model uses historical price and volume data as input to train a deep learning network. The network identifies patterns and generates factors that are used to rank industries. The top-ranked industries are selected for portfolio allocation. The model is updated weekly to reflect changes in the rankings[30][31][37] - Model Evaluation: The GRU model performs well in short-term trading environments but has shown limited effectiveness in long-term scenarios. It is also sensitive to extreme market conditions[37] --- Backtesting Results of Models 1. Diffusion Index Model - Weekly Average Return: -5.50% - Excess Return over Equal-Weighted CSI First-Level Industry Index: -0.42% - November-to-Date Excess Return: -1.13% - Year-to-Date Excess Return: 1.22%[26][22][23] 2. GRU Factor Model - Weekly Average Return: -4.71% - Excess Return over Equal-Weighted CSI First-Level Industry Index: 0.35% - November-to-Date Excess Return: 2.92% - Year-to-Date Excess Return: -2.74%[35][30][31] --- Quantitative Factors and Construction Methods 1. Factor Name: Diffusion Index - Factor Construction Idea: The diffusion index measures the momentum of industries by analyzing price trends and ranks industries based on their momentum[22][23] - Factor Construction Process: The diffusion index is calculated for each industry using price momentum data. Industries are ranked based on their diffusion index values, and the top-ranked industries are selected for portfolio allocation. The index is updated weekly or monthly to reflect changes in industry momentum[22][23] - Factor Evaluation: The factor effectively captures upward trends in industries but may underperform during market reversals[22][36] 2. Factor Name: GRU Industry Factor - Factor Construction Idea: The GRU industry factor is derived from minute-level price and volume data processed through a GRU deep learning network to identify patterns and rank industries[37] - Factor Construction Process: The GRU model processes historical price and volume data through a deep learning network. The network generates factors that are used to rank industries. The top-ranked industries are selected for portfolio allocation, with updates made weekly[30][31][37] - Factor Evaluation: The factor is effective in short-term trading environments but less so in long-term scenarios. It is also sensitive to extreme market conditions[37] --- Backtesting Results of Factors 1. Diffusion Index Factor - Weekly Average Return: -5.50% - Excess Return over Equal-Weighted CSI First-Level Industry Index: -0.42% - November-to-Date Excess Return: -1.13% - Year-to-Date Excess Return: 1.22%[26][22][23] 2. GRU Industry Factor - Weekly Average Return: -4.71% - Excess Return over Equal-Weighted CSI First-Level Industry Index: 0.35% - November-to-Date Excess Return: 2.92% - Year-to-Date Excess Return: -2.74%[35][30][31]