Group 1: Main Points of the Article - The article emphasizes the importance of intraday trading behaviors of major funds, particularly during specific market moments characterized by significant price drops, low stock prices, and high trading volumes [1][3][4] - A standardized average transaction amount factor (SATD) is introduced to capture the trading behavior of major funds, which is derived from the average transaction amount during special moments divided by the average transaction amount for the entire day [1][17][18] Group 2: Trading Behavior Based on Price Movements - The SATD factor shows a strong predictive ability for future stock returns, especially during moments of price decline, with a higher performance observed as the decline deepens [1][54] - The construction of the SATD factor is improved by incorporating tick-by-tick transaction data, allowing for a distinction between "main buy" and "main sell" transactions [1][59][62] Group 3: Trading Behavior Based on Price Levels - The SATD factor constructed during the lowest price moments demonstrates a strong predictive capability for future returns, outperforming factors constructed during the highest price moments [1][82][88] - The performance of the SATD factor improves as the threshold for defining low price moments becomes stricter [1][82] Group 4: Trading Behavior Based on Trading Volume - The SATD factor derived from the highest trading volume moments also exhibits strong predictive power, with a RankIC mean of 10.69% and a monthly win rate of 86% [1][3] - The article highlights the effectiveness of the composite factor constructed from various SATD factors across different market conditions and stock pools [1][3][4] Group 5: Composite Factor Performance - The composite factor, which combines various SATD factors, achieves a monthly RankIC mean of 10.33% and an annualized RankICIR of 4.32, indicating robust stock selection effectiveness across different indices and styles [1][3][4] - The composite factor maintains strong predictive capabilities even after traditional factors are stripped away, demonstrating its reliability in forecasting future stock returns [1][3][4]

Quantitative Models and Factor Construction Quantitative Models and Construction Process - **Model Name**: Standardized Average Transaction Amount Factor (SATD) **Construction Idea**: This factor captures the trading behavior of major funds by normalizing the average transaction amount during specific time periods against the daily average transaction amount[1][25][26] **Construction Process**: 1. Calculate the average transaction amount for specific time periods: $ ATD_{P} = \frac{\sum_{t \in P} Amt_{t}}{\sum_{t \in P} DealNum_{t}} $ Here, $ ATD_{P} $ represents the average transaction amount for the specific time period $ P $, $ Amt_{t} $ is the transaction amount at time $ t $, and $ DealNum_{t} $ is the number of transactions at time $ t $[26][27] 2. Calculate the daily average transaction amount: $ ATD_{T} = \frac{\sum_{t \in T} Amt_{t}}{\sum_{t \in T} DealNum_{t}} $ Here, $ ATD_{T} $ represents the daily average transaction amount[27] 3. Normalize the specific time period's average transaction amount: $ SATD_{P} = \frac{ATD_{P}}{ATD_{T}} $ Here, $ SATD_{P} $ is the standardized average transaction amount factor for the specific time period $ P $[28][29] Quantitative Factors and Construction Process - **Factor Name**: Downward Price Movement Factor **Construction Idea**: This factor identifies the predictive power of major fund activity during periods of price decline[39][40] **Construction Process**: 1. Classify minute-level price movements into upward, downward, and flat periods using the following formulas: $ UpFlag_{t} = \begin{cases} 1, & if\ ret_{i} > 0 \\ 0, & if\ ret_{i} \leq 0 \end{cases} $ $ DownFlag_{t} = \begin{cases} 0, & if\ ret_{i} \geq 0 \\ 1, & if\ ret_{i} < 0 \end{cases} $ $ ZeroFlag_{t} = \begin{cases} 0, & if\ ret_{i} \neq 0 \\ 1, & if\ ret_{i} = 0 \end{cases} $ Here, $ ret_{i} $ represents the minute-level return[39][40] 2. Calculate the average transaction amount for downward periods and normalize it against the daily average transaction amount: $ SATDDown = \frac{ATD_{DownFlag}}{ATD_{T}} $[43][44] - **Factor Name**: Maximum Downward Price Movement Factor **Construction Idea**: This factor focuses on the periods with the largest price declines, hypothesizing that major funds are more active during these times[59][60] **Construction Process**: 1. Rank minute-level price movements by their magnitude of decline 2. Select the top N% of minutes with the largest price declines 3. Calculate the average transaction amount for these periods and normalize it against the daily average transaction amount[59][60] - **Factor Name**: Lowest Price Factor **Construction Idea**: This factor identifies periods when the stock price is at its lowest, hypothesizing that major funds are more active during these times[87][89] **Construction Process**: 1. Rank minute-level prices from lowest to highest 2. Select the bottom N% of minutes with the lowest prices 3. Calculate the average transaction amount for these periods and normalize it against the daily average transaction amount[89][91] - **Factor Name**: Highest Volume Factor **Construction Idea**: This factor identifies periods with the highest trading volume, hypothesizing that these periods contain more significant information[109][110] **Construction Process**: 1. Rank minute-level trading volumes from highest to lowest 2. Select the top N% of minutes with the highest volumes 3. Calculate the average transaction amount for these periods and normalize it against the daily average transaction amount[109][110] - **Factor Name**: Volume-Price Divergence Factor **Construction Idea**: This factor identifies periods where trading volume and price movements are negatively correlated, hypothesizing that these periods contain more significant information[128][129] **Construction Process**: 1. Calculate the correlation coefficient between transaction prices and volumes for each minute 2. Rank minutes by their correlation coefficients 3. Select the bottom N% of minutes with the lowest correlation coefficients 4. Calculate the average transaction amount for these periods and normalize it against the daily average transaction amount[129][134] - **Factor Name**: Composite Factor **Construction Idea**: This factor combines the most effective factors (e.g., maximum downward price movement, lowest price, and highest volume factors) to enhance predictive power[160][161] **Construction Process**: 1. Combine the selected factors using equal weighting: $ CompositeFactor = DownwardFactor + LowestPriceFactor + HighestVolumeFactor $[160][161] Backtesting Results for Factors - **Downward Price Movement Factor**: RankIC Mean = 6.84%, Annualized RankICIR = 3.23, Monthly Win Rate = 83.93%[46][48] - **Maximum Downward Price Movement Factor**: RankIC Mean = 7.31%, Annualized RankICIR = 4.04, Monthly Win Rate = 86.49%[60][61] - **Lowest Price Factor**: RankIC Mean = 7.21%, Annualized RankICIR = 4.52, Monthly Win Rate = 91.96%[91][92] - **Highest Volume Factor**: RankIC Mean = 9.70%, Annualized RankICIR = 3.67, Monthly Win Rate = 83.04%[110][113] - **Volume-Price Divergence Factor**: RankIC Mean = 5.41%, Annualized RankICIR = 3.20, Monthly Win Rate = 81.25%[134][135] - **Composite Factor**: RankIC Mean = 10.33%, Annualized RankICIR = 4.32, Monthly Win Rate = 90.18%[160][161]