金融工程定期：开源交易行为因子绩效月报（2026年3月）-20260331

Quantitative Models and Construction Methods - Model Name: Barra Style Factors Construction Idea: The model tracks the performance of common Barra style factors, focusing on dimensions such as size, value, growth, and profitability[3][13] Construction Process: The model calculates the monthly returns of specific factors, including market capitalization, book-to-market ratio, growth, and earnings expectations[3][13] Evaluation: Provides insights into the relative performance of different style factors in the market[3][13] - Model Name: Open-source Trading Behavior Composite Factor Construction Idea: Combines multiple trading behavior factors to monitor dynamic performance[4][29] Construction Process: 1. Normalize individual trading behavior factors within industries 2. Use the past 12 periods' ICIR values as weights to form the composite factor 3. Apply industry market capitalization neutrality to the composite factor[29][33] Evaluation: Demonstrates robust performance across various stock pools, with better results in small-cap indices like CSI 1000 compared to CSI 800[29][30][33] Factor Construction Methods - Factor Name: Ideal Reversal Factor Construction Idea: Captures reversal strength by analyzing large transaction days[4][38] Construction Process: 1. Retrieve the past 20 days' data for selected stocks 2. Calculate daily average transaction amounts (transaction amount/number of transactions) 3. Identify the top 10 days with the highest transaction amounts and sum their returns (M_high) 4. Identify the bottom 10 days with the lowest transaction amounts and sum their returns (M_low) 5. Compute the factor as M = M_high - M_low[38][40] Evaluation: Highlights the micro-level reversal dynamics in A-shares[4][38] - Factor Name: Smart Money Factor Construction Idea: Identifies institutional trading activity using minute-level price-volume data[4][39] Construction Process: 1. Retrieve the past 10 days' minute-level data for selected stocks 2. Calculate the indicator $S_t = |R_t| / V_t^{0.25}$, where $R_t$ is the return at minute $t$, and $V_t$ is the trading volume at minute $t$[39] 3. Sort minute-level data by $S_t$ in descending order and select the top 20% cumulative trading volume minutes as smart money trades 4. Compute VWAPsmart (volume-weighted average price of smart money trades) and VWAPall (volume-weighted average price of all trades) 5. Calculate the factor as $Q = \text{VWAPsmart} / \text{VWAPall}$[39][41] Evaluation: Effectively tracks institutional trading patterns[4][39] - Factor Name: APM Factor Construction Idea: Measures behavioral differences between morning (or overnight) and afternoon trading[4][40] Construction Process: 1. Retrieve the past 20 days' data for selected stocks 2. Record daily overnight stock returns ($r$) and index returns ($R$), as well as afternoon stock returns ($r$) and index returns ($R$) 3. Perform regression $r = \alpha + \beta R + \epsilon$ to obtain residuals $\epsilon$ 4. Calculate the difference between overnight and afternoon residuals $\delta_t = \epsilon_{\text{overnight}} - \epsilon_{\text{afternoon}}$ 5. Compute the statistic $\text{stat} = \frac{\mu(\delta_t)}{\sigma(\delta_t) / \sqrt{N}}$, where $\mu$ is the mean, $\sigma$ is the standard deviation, and $N$ is the sample size[42] 6. Perform cross-sectional regression to remove momentum effects, using $\text{stat} = \text{Ret20} + \epsilon$, where Ret20 represents the past 20-day momentum factor[43] 7. Use the residual $\epsilon$ as the APM factor[40][42][43] Evaluation: Captures intraday reversal dynamics effectively[4][40] - Factor Name: Ideal Amplitude Factor Construction Idea: Differentiates amplitude information between high and low price states[4][45] Construction Process: 1. Retrieve the past 20 days' data for selected stocks 2. Calculate daily amplitude as $(\text{High Price}/\text{Low Price}) - 1$ 3. Select the top 25% high-price days and compute the average amplitude (V_high) 4. Select the bottom 25% low-price days and compute the average amplitude (V_low) 5. Compute the factor as $V = V_{\text{high}} - V_{\text{low}}$[45] Evaluation: Reveals structural differences in amplitude information across price states[4][45] Backtesting Results - Barra Style Factors: - Market Capitalization Factor: Return -0.18%[3][13] - Book-to-Market Ratio Factor: Return 0.45%[3][13] - Growth Factor: Return -0.60%[3][13] - Earnings Expectations Factor: Return -0.46%[3][13] - Open-source Trading Behavior Factors: - Ideal Reversal Factor: - IC Mean -0.048, rankIC Mean -0.060, IR 2.37, Monthly Win Rate 77.1% (historical)[5][14] - March Return -0.47%, 12-month Win Rate 50.0%[6][14] - Smart Money Factor: - IC Mean -0.037, rankIC Mean -0.062, IR 2.68, Monthly Win Rate 80.5% (historical)[5][19] - March Return 1.35%, 12-month Win Rate 66.7%[6][19] - APM Factor: - IC Mean 0.028, rankIC Mean 0.034, IR 2.26, Monthly Win Rate 76.0% (historical)[5][23] - March Return 1.50%, 12-month Win Rate 41.7%[6][23] - Ideal Amplitude Factor: - IC Mean -0.053, rankIC Mean -0.073, IR 2.98, Monthly Win Rate 82.7% (historical)[5][26] - March Return 2.08%, 12-month Win Rate 66.7%[6][26] - Composite Factor: - IC Mean 0.065, rankIC Mean 0.093, IR 3.24, Monthly Win Rate 79.3% (historical)[5][29] - March Return 2.45%, 12-month Win Rate 58.3%[6][29] - Outperforms in CSI 1000 and CSI 2000 indices with IRs of 2.61 and 2.83, respectively[30]