Workflow
Fama五因子模型
icon
Search documents
金融工程专题研究:FOF系列专题之十:隐性风险视角下的选基因子统一改进框架
Guoxin Securities· 2025-06-17 14:28
Quantitative Models and Construction Methods Hidden Risk Model - **Model Name**: Hidden Risk Model - **Construction Idea**: Funds with high correlation in net asset value (NAV) trends are likely exposed to similar risks (explicit or hidden). By regressing fund factors weighted by correlation with similar funds, hidden risks can be stripped away [3][68][69] - **Construction Process**: 1. Identify funds with high NAV correlation over the past year (top N funds, N=20) [70] 2. Calculate weighted daily returns of similar funds based on correlation [70] 3. Perform time-series regression of the target fund's daily returns against the weighted returns of similar funds. The intercept term represents the adjusted alpha factor [70] - Formula: $$R_{p}=\alpha+\beta\cdot SimiRet+\varepsilon_{p}$$ [90] - **Evaluation**: Provides higher explanatory power for fund returns compared to traditional models like Fama-Five-Factor [3][86][90] Hidden Risk-Adjusted Comprehensive Selection Factor - **Factor Name**: Hidden Risk-Adjusted Comprehensive Selection Factor - **Construction Idea**: Combine factors improved by hidden risk adjustments (e.g., Sharpe ratio, hidden trading ability) with original factors that do not require adjustment (e.g., reverse investment ability) [115][118] - **Construction Process**: 1. Adjust factors like Sharpe ratio and hidden trading ability using hidden risk regression [94][101] 2. Combine adjusted factors with original factors using equal weighting [115] - **Evaluation**: Improves stability and predictive power of selection factors, with RankICIR increasing significantly [107][118] Sharpe Ratio Factor Adjustment - **Factor Name**: Hidden Risk-Adjusted Sharpe Ratio Factor - **Construction Idea**: Adjust the original Sharpe ratio factor by regressing it against the weighted Sharpe ratios of similar funds [94] - **Construction Process**: 1. Calculate weighted Sharpe ratios of similar funds based on correlation [94] 2. Perform cross-sectional regression of the original Sharpe ratio against the weighted Sharpe ratios [94] - Formula: $$Sharpe=\alpha+\beta\cdot SimiSharpe+\varepsilon$$ [94] - **Evaluation**: Stability significantly improved, with RankICIR increasing from 0.77 to 1.99 [96][98] Hidden Trading Ability Factor Adjustment - **Factor Name**: Hidden Risk-Adjusted Hidden Trading Ability Factor - **Construction Idea**: Adjust the original hidden trading ability factor by regressing it against the weighted hidden trading ability of similar funds [101] - **Construction Process**: 1. Calculate weighted hidden trading ability of similar funds based on correlation [101] 2. Perform cross-sectional regression of the original hidden trading ability against the weighted hidden trading ability [101] - **Evaluation**: Stability significantly improved, with RankICIR increasing from 1.68 to 2.23 [102][106] --- Model Backtesting Results Hidden Risk Model - **RankIC Mean**: 92.32% [86] - **RankICIR**: Not explicitly mentioned - **RankIC Win Rate**: Not explicitly mentioned Hidden Risk-Adjusted Comprehensive Selection Factor - **RankIC Mean**: 13.99% [118][121] - **RankICIR**: 3.18 [118][121] - **RankIC Win Rate**: 93.01% [118][121] - **Annualized Excess Information Ratio**: 2.4 [3] Sharpe Ratio Factor Adjustment - **RankIC Mean**: 7.70% [98] - **RankICIR**: 1.99 [98] - **RankIC Win Rate**: 87.41% [98] Hidden Trading Ability Factor Adjustment - **RankIC Mean**: 7.21% [102] - **RankICIR**: 2.23 [102] - **RankIC Win Rate**: 90.21% [102] --- Factor Backtesting Results Hidden Risk-Adjusted Comprehensive Selection Factor - **RankIC Mean**: 13.99% [118][121] - **RankICIR**: 3.18 [118][121] - **RankIC Win Rate**: 93.01% [118][121] - **Quarterly Excess Return (Top Decile)**: 1.46% [118][121] - **Quarterly Long-Short Return**: 2.74% [118][121] Sharpe Ratio Factor Adjustment - **RankIC Mean**: 7.70% [98] - **RankICIR**: 1.99 [98] - **RankIC Win Rate**: 87.41% [98] - **Quarterly Excess Return (Top Decile)**: 0.86% [98] - **Quarterly Long-Short Return**: Not explicitly mentioned Hidden Trading Ability Factor Adjustment - **RankIC Mean**: 7.21% [102] - **RankICIR**: 2.23 [102] - **RankIC Win Rate**: 90.21% [102] - **Quarterly Excess Return (Top Decile)**: 0.92% [102] - **Quarterly Long-Short Return**: Not explicitly mentioned --- FOF Portfolio Construction Results Hidden Risk-Adjusted Comprehensive Selection Factor-Based Portfolio - **Annualized Excess Return**: 8.86% [147] - **Annualized Tracking Error**: 3.52% [147] - **Excess Information Ratio**: 2.31 [147] - **Maximum Relative Drawdown**: 3.40% [147] - **Relative Return-to-Drawdown Ratio**: 2.61 [147] - **Monthly Win Rate**: 75.91% [147]