主动量化研究系列：指增超额回撤控制：波动率分域视角

Quantitative Models and Construction Methods Model Name: Residual Volatility Domain Management - Model Construction Idea: The model aims to manage excess drawdowns by segmenting stocks based on residual volatility, relaxing constraints on low-volatility stocks while tightening constraints on high-volatility stocks. This approach seeks to optimize the risk-return profile of the portfolio[3][11][44] - Model Construction Process: 1. Define residual volatility as the unexplained portion of stock returns after accounting for country, industry, and style factors 2. Use the following formula for stock returns: $r_{n}=f_{c}+\sum_{i}X_{n i}f_{i}+\sum_{s}X_{n s}f_{s}+u_{n}$ - $r_{n}$: Stock excess return - $f_{c}$: Country factor - $f_{i}$: Industry factor - $f_{s}$: Style factor - $u_{n}$: Residual term[34][36] 3. Optimize portfolio weights using the following formula: $w_{i}=\lambda^{-1}F_{i}\,/\,\sigma_{i}$ - $w_{i}$: Active weight of stock $i$ - $F_{i}$: Risk-adjusted signal - $\sigma_{i}$: Residual volatility of stock $i$[45][46] 4. Segment stocks into three groups based on residual volatility (low, medium, high) using the 30% and 70% quantiles 5. Apply different weight constraints for each group: - Low-volatility stocks: [-0.2%, 0.4%] - Medium-volatility stocks: [-0.2%, 0.3%] - High-volatility stocks: [-0.1%, 0.2%][62] - Model Evaluation: The model effectively reduces portfolio drawdowns while maintaining or improving excess returns, especially during high-volatility periods[11][44][68] --- Model Backtesting Results Residual Volatility Domain Management Model - Annualized Excess Return: 4.66% (compared to 4.30% for the benchmark portfolio) - Maximum Excess Drawdown: -6.78% (compared to -10.47% for the benchmark portfolio) - Information Ratio (IR): 1.15 (compared to 0.82 for the benchmark portfolio)[67] --- Quantitative Factors and Construction Methods Factor Name: Alpha Factors (e.g., Growth, Momentum, Surprise) - Factor Construction Idea: Alpha factors are designed to predict stock returns by capturing specific characteristics such as growth, momentum, and earnings surprises. These factors are often correlated with style factors like volatility, liquidity, and market capitalization[1][17][18] - Factor Construction Process: 1. Use individual factors as signals to generate excess returns while constraining industry, style, and stock deviations 2. Calculate the correlation between alpha factors and style factors to understand their intrinsic relationships 3. Example correlations: - Growth factor positively correlates with momentum (23.9%) and volatility (23.3%) - Surprise factor positively correlates with momentum (35.3%) but negatively correlates with valuation (-9.7%)[17][19] - Factor Evaluation: Alpha factors show strong correlations with style factors, but their predictive power for stock returns is relatively weak and unstable, especially during high-volatility periods[20][43][68] --- Factor Backtesting Results Alpha Factors - IC Mean: Mostly within ±10%, indicating limited predictive power for stock returns[20][22] - Correlation with Style Factors: - Growth factor: Momentum (23.9%), Volatility (23.3%) - Surprise factor: Momentum (35.3%), Valuation (-9.7%)[19] Residual Volatility Factor - Residual Volatility and Market Cap: Negative correlation observed, with smaller-cap stocks exhibiting higher residual volatility[38][40] - Residual Volatility Predictability: - Residual return predictability: Low (1.9% median correlation with a 1-day lag) - Residual volatility predictability: High (66% median correlation with a 21-day lag)[49][51] --- Key Observations and Insights - Residual volatility plays a critical role in managing excess drawdowns, with high-volatility stocks contributing disproportionately to portfolio risk[3][44][56] - Alpha factors exhibit weak and unstable predictive power for stock returns, particularly during periods of market turbulence[20][43][68] - Segmenting stocks by residual volatility and applying differentiated constraints can effectively balance risk and return, as demonstrated by the improved performance of the optimized portfolio[62][67][68]