Quantitative Models and Construction Methods - **Model Name**: Minute Herding Effect Factor Cluster **Construction Idea**: Focus on the trading behavior of followers after significant actions by "trend funds" using minute-level data [13][14][18] **Construction Process**: 1. **Event Identification**: Detect actions of trend funds through anomalies in volume, price changes, volatility, and price-volume correlation [13][14] 2. **Factor Definition**: Measure herding strength by analyzing post-event price, volume, price-volume correlation, and other metrics [14][18] 3. **Data Frequency**: Use minute-level data to identify events and define factors [14][18] **Evaluation**: Effective in capturing herding behavior at the minute level [18] - **Model Name**: Tick-by-Tick Herding Effect Factor Cluster **Construction Idea**: Apply discrete factor definitions directly to tick-by-tick data to capture herding effects [1][11][20] **Construction Process**: 1. **Event Identification**: Identify liquidity shock events using tick-by-tick order and trade data, introducing the concept of "aggressiveness" for orders [21][22][25] 2. **Factor Definition**: Analyze post-event metrics such as order volume, trade volume, imbalance indicators, and price-volume correlation [30][31][61] 3. **Factor Production**: Generate approximately 20,000 factors, retaining the top 50 based on performance and low correlation [63][84] **Evaluation**: Demonstrates strong predictive power with annual ICIR values exceeding 2 [63][84] - **Model Name**: Tick-by-Tick Herding Effect Composite Factor **Construction Idea**: Combine the top 10 factors with the highest information ratio into a composite factor [67][85] **Construction Process**: 1. Select the top 10 factors based on information ratio from the tick-by-tick factor cluster [67][85] 2. Equally weight these factors to create the composite factor [67][85] **Evaluation**: Highly effective with robust performance metrics, even after neutralizing common style and industry factors [67][71][85] Model Backtesting Results - **Minute Herding Effect Composite Factor**: - Monthly IC Mean: 0.085 - Annual ICIR: 3.18 - Monthly RankIC Mean: 0.116 - Annual RankICIR: 4.10 - Annual Return: 41.59% - Annual Volatility: 12.56% - Information Ratio: 3.31 - Monthly Win Rate: 82.91% - Maximum Drawdown: 10.06% [18] - **Tick-by-Tick Herding Effect Factor Cluster**: - Annual ICIR Absolute Value: >2 for all 50 factors [63][65] - Example Factor (Factor 16): - Monthly IC Mean: 0.057 - Annual ICIR: 2.82 - Monthly RankIC Mean: 0.072 - Annual RankICIR: 3.01 - Annual Return: 25.86% - Annual Volatility: 9.11% - Information Ratio: 2.84 - Monthly Win Rate: 76.92% - Maximum Drawdown: 6.38% [64][65][66] - **Tick-by-Tick Herding Effect Composite Factor**: - Monthly IC Mean: 0.080 - Annual ICIR: 3.49 - Monthly RankIC Mean: 0.101 - Annual RankICIR: 3.74 - Annual Return: 44.26% - Annual Volatility: 10.90% - Information Ratio: 4.06 - Monthly Win Rate: 89.74% - Maximum Drawdown: 10.66% [67][85] - **Pure Tick-by-Tick Herding Effect Composite Factor** (Neutralized for Style and Industry): - Monthly IC Mean: 0.044 - Annual ICIR: 3.33 - Monthly RankIC Mean: 0.046 - Annual RankICIR: 3.03 - Annual Return: 19.53% - Annual Volatility: 6.36% - Information Ratio: 3.07 - Monthly Win Rate: 78.63% - Maximum Drawdown: 5.13% [71][85] Index Enhancement Portfolio Performance - **CSI 300 Index Enhancement Portfolio**: - Excess Annual Return: 8.89% - Tracking Error: 3.50% - Information Ratio: 2.54 - Monthly Win Rate: 77.78% - Maximum Drawdown: 2.96% [75][86] - **CSI 500 Index Enhancement Portfolio**: - Excess Annual Return: 13.46% - Tracking Error: 5.31% - Information Ratio: 2.54 - Monthly Win Rate: 79.49% - Maximum Drawdown: 5.15% [78][86] - **CSI 1000 Index Enhancement Portfolio**: - Excess Annual Return: 17.23% - Tracking Error: 4.78% - Information Ratio: 3.61 - Monthly Win Rate: 84.62% - Maximum Drawdown: 4.14% [80][86]