Quantitative Models and Factor Construction Quantitative Models and Construction Methods 1. **Model Name**: Nonlinear Multi-Factor Alpha Model **Model Construction Idea**: This model addresses the nonlinear relationship between factors and stock returns by transforming nonlinear factors into a linear form through polynomial functions[3][23]. **Model Construction Process**: - The model uses a third-degree polynomial transformation for factors with nonlinear characteristics. - The polynomial function is expressed as: $ r_{i}=aF^{3}+bF^{2}+cF+d $ where $a$, $b$, $c$, and $d$ are coefficients determined through least squares fitting[24][25]. **Model Evaluation**: This method is simple and intuitive but relies heavily on historical data and lacks rigorous economic logic[23][99]. 2. **Model Name**: Alpha Model with Additional Factors **Model Construction Idea**: This model introduces auxiliary factors to explain the nonlinear characteristics of primary factors, improving their predictive power[4][75]. **Model Construction Process**: - Identify auxiliary factors that interact with primary factors. For example, market capitalization is used as an auxiliary factor for turnover rate. - Construct a dummy variable $d_{high\_cap}$, which takes values of 0 or 1 depending on whether the stock belongs to a high or low market capitalization group. - The adjusted model is expressed as: $ r_{i}=v_{i0}+v_{i1}F_{i}+v_{i2}d_{i}F_{i}+\varepsilon_{i} $ where $d_{i}$ represents the dummy variable[78][79]. **Model Evaluation**: This method provides a more reasonable economic explanation and achieves significant improvements in factor effectiveness. However, it requires extensive pairwise comparisons and is labor-intensive[5][99]. Model Backtesting Results 1. **Nonlinear Multi-Factor Alpha Model**: - **Annualized Return**: 19.36% (in-sample), 19.19% (out-of-sample) - **Annualized Volatility**: 13.07% (in-sample), 8.97% (out-of-sample) - **IR**: 1.48 (in-sample), 2.14 (out-of-sample) - **Maximum Drawdown**: 15.33% (in-sample), 2.26% (out-of-sample)[63][67][76] 2. **Alpha Model with Additional Factors**: - **Annualized Return**: 30.90% (in-sample), 17.20% (out-of-sample) - **Annualized Volatility**: 15.43% (in-sample), 11.01% (out-of-sample) - **IR**: 2.00 (in-sample), 1.56 (out-of-sample) - **Maximum Drawdown**: 8.23% (in-sample), 1.70% (out-of-sample)[96][97][98] --- Quantitative Factors and Construction Methods 1. **Factor Name**: Debt-to-Asset Ratio **Factor Construction Idea**: This factor exhibits a nonlinear relationship with stock returns, where both high and low values are suboptimal[27]. **Factor Construction Process**: - Transform the factor using a third-degree polynomial function. - Compare cumulative returns before and after transformation. - Annualized return improved from -0.5% to 3% after transformation[27][28]. 2. **Factor Name**: Turnover Rate **Factor Construction Idea**: The factor shows a nonlinear characteristic at low turnover levels[30]. **Factor Construction Process**: - Apply a polynomial transformation to enhance linearity. - Annualized return increased from 21.7% to 28.4% after transformation[30][36]. 3. **Factor Name**: Earnings-to-Price Ratio (EP) **Factor Construction Idea**: The factor demonstrates nonlinear behavior at low EP levels[34]. **Factor Construction Process**: - Use a polynomial transformation to improve linearity. - Annualized return increased from 11% to 13.5% after transformation[34][40]. 4. **Factor Name**: Total Assets **Factor Construction Idea**: The factor is generally linear but shows nonlinear characteristics at low asset levels[38]. **Factor Construction Process**: - Apply a polynomial transformation. - Annualized return improved from 8.1% to 10% after transformation[38][44]. 5. **Factor Name**: Fixed Ratio **Factor Construction Idea**: The factor exhibits a "middle is better" nonlinear pattern[42]. **Factor Construction Process**: - Transform the factor using a polynomial function. - Annualized return increased from 2.75% to 6.57% after transformation[42][48]. 6. **Factor Name**: Current Ratio **Factor Construction Idea**: Similar to the fixed ratio, this factor also shows a "middle is better" pattern[46]. **Factor Construction Process**: - Apply a polynomial transformation. - Annualized return improved from 3.76% to 5.32% after transformation[46][52]. Factor Backtesting Results 1. **Debt-to-Asset Ratio**: Annualized return improved from -0.5% to 3%[27][28] 2. **Turnover Rate**: Annualized return improved from 21.7% to 28.4%[30][36] 3. **Earnings-to-Price Ratio (EP)**: Annualized return improved from 11% to 13.5%[34][40] 4. **Total Assets**: Annualized return improved from 8.1% to 10%[38][44] 5. **Fixed Ratio**: Annualized return improved from 2.75% to 6.57%[42][48] 6. **Current Ratio**: Annualized return improved from 3.76% to 5.32%[46][52]

Quantitative Models and Factor Construction Quantitative Models and Construction Methods - **Model Name**: Factor Timing Strategy Based on Mean Reversion **Model Construction Idea**: The model leverages the observed mean reversion characteristics of factor ICIR to dynamically adjust factor weights in a multi-factor portfolio[3][56][59] **Model Construction Process**: 1. Identify 9 alpha factors, including "1-month turnover", "turnover rate", "1-month price reversal", "3-month price reversal", "6-month price reversal", "market capitalization", "EP", "SP", and "BP"[10][11][58] 2. Calculate the ICIR for each factor based on its historical IC performance over its optimal observation period (e.g., 8 months for "1-month turnover", 10 months for "turnover rate")[31][58][59] 3. Rank factors by ICIR in descending order and assign scores: 1 point for the top tier, 2 points for the middle tier, and 4 points for the bottom tier[3][59] 4. Adjust factor weights dynamically based on their scores, where the weight of each factor equals its score divided by the total score of all factors[59][61] **Model Evaluation**: The model effectively captures the mean reversion characteristics of factors and improves portfolio performance by dynamically adjusting factor weights[56][72][73] Quantitative Factors and Construction Methods - **Factor Name**: ICIR-Based Alpha Factors **Factor Construction Idea**: ICIR is used as a measure of factor effectiveness, considering both the magnitude and volatility of IC over a specific observation period[25][31][58] **Factor Construction Process**: 1. Calculate IC for each factor as the correlation between factor exposure and future stock returns[10][11] 2. Compute ICIR as the ratio of the mean IC to its standard deviation over the observation period[25][31] 3. Determine the optimal observation period for each factor based on the significant negative correlation between ICIR and IC (e.g., 6 months for "EP" and "SP", 8 months for "BP")[31][58] **Factor Evaluation**: ICIR effectively reflects the time-varying effectiveness of factors and provides a robust basis for factor timing strategies[25][31][58] Backtesting Results Model Backtesting Results - **Factor Timing Strategy**: - **Average Return**: 21.88% (long-short hedge), 15.79% (index futures hedge)[74] - **Volatility**: 12.11% (long-short hedge), 12.84% (index futures hedge)[74] - **IR**: 1.806 (long-short hedge), 1.230 (index futures hedge)[74] - **Maximum Drawdown**: 7.33% (long-short hedge), 8.89% (index futures hedge)[74] - **Performance Improvement**: IR increased by over 11% compared to equal-weighted strategy, with a win rate of 60.76%[72][73] Factor Backtesting Results - **IC and ICIR Statistics for 9 Factors**: - "1-month turnover": IC = -5.50%, ICIR = -1.09[11] - "Turnover rate": IC = -6.13%, ICIR = -1.57[11] - "1-month price reversal": IC = -4.67%, ICIR = -0.93[11] - "3-month price reversal": IC = -3.85%, ICIR = -0.67[11] - "6-month price reversal": IC = -2.24%, ICIR = -0.37[11] - "Market capitalization": IC = -3.16%, ICIR = -0.43[11] - "EP": IC = 4.21%, ICIR = 1.37[11] - "SP": IC = 2.97%, ICIR = 1.37[11] - "BP": IC = 4.36%, ICIR = 1.27[11]