Quantitative Models and Construction Methods Model Name: Price Anomaly Detection - **Model Construction Idea**: Identify stocks with price movements that deviate significantly from the benchmark index by calculating the correlation coefficient between the stock's intraday price series and the benchmark index's price series [11][13] - **Model Construction Process**: 1. Calculate the minute-level closing price series for individual stocks and the benchmark index (e.g., Wind All A Index) [13] 2. Compute the correlation coefficient between the stock's price series and the benchmark index's price series for the trading day [13] 3. If the correlation coefficient < 0, the stock is considered to have experienced a price anomaly on that day [13] - **Model Evaluation**: The model captures stocks with price movements deviating from the market but fails to generate effective alpha signals as the future excess return win rate is below 50% [15][19] Model Name: Upward and Downward Price Anomalies - **Model Construction Idea**: Refine the price anomaly detection by incorporating the direction of excess returns relative to the benchmark index [17][18] - **Model Construction Process**: 1. Define "Upward Price Anomaly" as stocks with a correlation coefficient < 0 and stock return > benchmark return on the same day [18] 2. Define "Downward Price Anomaly" as stocks with a correlation coefficient < 0 and stock return < benchmark return on the same day [18] 3. Test the model using the CSI 800 index constituents over the period 2016/01/01–2026/02/28 [18] - **Model Evaluation**: Both upward and downward price anomalies fail to provide significant excess returns, with win rates below 50% and average excess returns near zero [19][23] Model Name: Anomaly Radar Event Cluster - **Model Construction Idea**: Extend the anomaly detection framework by incorporating multi-dimensional capital flow indicators and systematically producing event-driven signals [28][29] - **Model Construction Process**: 1. **Correlation Coefficient Calculation**: Compute the correlation coefficient between intraday capital flow indicators (e.g., transaction volume, transaction amount) of individual stocks and the benchmark index. If the correlation coefficient < 0, the stock is considered anomalous [29][30] 2. **Excess Return Direction**: Incorporate the direction of excess returns relative to the benchmark index to classify anomalies as "upward" or "downward" [39] 3. **Signal Screening and Synthesis**: Batch-produce event signals, evaluate their effectiveness and correlation, and synthesize effective signals into a stable event-driven strategy [42][43] 4. Construct a capital channel strategy using selected signals, with a 20-day holding period and weekly rebalancing [42][43] - **Model Evaluation**: The synthesized anomaly radar signals demonstrate strong performance, with an annualized excess return of 7.51% and an IR of 2.48 during the backtest period [45][48] --- Model Backtest Results Price Anomaly Detection - Annualized excess return: Near zero [15] - Win rate: Below 50% across all time horizons [15] Upward Price Anomalies - Annualized excess return: Near zero [19] - Win rate: Below 50% across all time horizons [19] Downward Price Anomalies - Annualized excess return: Near zero [23] - Win rate: Below 50% across all time horizons [23] Anomaly Radar Event Cluster - Annualized excess return: 7.51% [45][48] - IR: 2.48 [45][48] - Maximum drawdown: 4.13% [45][48] Anomaly Radar + Negative Signal Filtering - Annualized excess return: 9.77% [51][53] - IR: 2.92 [51][53] - Maximum drawdown: 2.85% [51][53] --- Quantitative Factors and Construction Methods Factor Name: Industry Anomaly Factor - **Factor Construction Idea**: Map individual stock anomaly signals to industry-level factors for use in sector rotation strategies [66][67] - **Factor Construction Process**: 1. Calculate the number of stocks triggering anomaly signals within each industry daily [66] 2. Normalize the number of triggered stocks by the total number of stocks in the industry and compute a 20-day rolling average [66] 3. Calculate the historical percentile of the rolling average to define the industry anomaly factor [66] - **Factor Evaluation**: The factor demonstrates moderate predictive power with a monthly IC of 0.03 and low correlation (11%) with traditional industry trend factors [68] --- Factor Backtest Results Industry Anomaly Factor - Monthly IC: 0.03 [68] - Multi-group backtest: Top quintile significantly outperforms other groups [68] Sector Rotation Strategy: "Anomaly + Strong Trend + Low Crowding, Exclude Low Prosperity" - Annualized excess return: 9.50% (vs. 6.78% without anomaly factor) [70][73] - IR: 1.09 (vs. 0.74 without anomaly factor) [70][73] - Maximum drawdown: 9.62% (vs. 18.64% without anomaly factor) [70][73] Sector Rotation Strategy: "Anomaly + Strong Trend + High Prosperity, Exclude High Crowding" - Annualized excess return: 9.04% (vs. 5.97% without anomaly factor) [75][78] - IR: 0.80 (vs. 0.47 without anomaly factor) [75][78] - Maximum drawdown: 18.66% (vs. 30.72% without anomaly factor) [75][78]