市场风格轮动系列：如何从赔率和胜率看大小盘

Quantitative Models and Construction Methods 1. Model Name: Size Rotation Model Based on Odds and Win Rates - Model Construction Idea: The model integrates the concepts of odds and win rates to capture the rotation between large-cap and small-cap stocks. Odds are derived from valuation differences, while win rates are calculated using multiple indicators[4][30][40] - Model Construction Process: - Odds Calculation: - Define odds as the ratio of average positive returns to the absolute value of average negative returns - Formula: $ \mathbb{R}{\mathbb{B}}^{\pm},\mathbb{R}=-\frac{\sum{i=1}^{n}r e t u r n_{i},/n}{\sum_{j=1}^{m}r e t u r n_{j},/m} $ where $ \mathbb{R}{\mathbb{B}}^{+} $ represents positive returns and $ \mathbb{R}{\mathbb{B}}^{-} $ represents negative returns[30][31] - Use historical valuation differences between large-cap (CSI 300) and small-cap (CSI 2000) indices to estimate odds through linear regression[32][36] - Win Rate Calculation: - Combine multiple indicators (e.g., Shibor, short-term credit spread, market trend, market volatility, style momentum, style crowding, and calendar effects) to derive a composite win rate signal - Assign scores: 1 for large-cap signals, 0 for small-cap signals, and 0.5 for neutral signals. The average score represents the win rate[40][72] - Kelly Formula for Allocation: - Use the Kelly formula to calculate optimal allocation weights for large-cap and small-cap stocks based on odds and win rates - Formula: $ x = \frac{p*b - (1-p)}{b} $ where $ p $ is the win rate, $ b $ is the odds, and $ x $ is the allocation proportion[77] - Adjust weights to ensure they sum to 1 and avoid negative values, forming a complete rotation strategy[77][78] - Model Evaluation: The model effectively captures the rotation between large-cap and small-cap stocks, achieving significant excess returns and risk-adjusted performance[78] 2. Model Name: Weighted Size Rotation Strategy - Model Construction Idea: Adjust allocation weights between large-cap and small-cap stocks based on the difference in configuration scores derived from odds and win rates[82] - Model Construction Process: - Calculate the difference in configuration scores between large-cap and small-cap stocks - Standardize the score difference using a Z-score over the past 250 weeks - Map the standardized score to allocation weights using a predefined mapping table[83] - Model Evaluation: This strategy reduces maximum drawdown while maintaining a high level of excess returns and information ratio[84] 3. Model Name: Detailed Style Rotation Model - Model Construction Idea: Combine the size rotation model with a growth-value rotation model to form a detailed style rotation strategy, targeting large-cap growth, large-cap value, small-cap growth, and small-cap value[87] - Model Construction Process: - Use the size rotation model to determine the size preference (large-cap or small-cap) - Use the growth-value rotation model to determine the style preference (growth or value) - Combine the two signals to allocate to one of the four detailed styles[87] - Model Evaluation: The model demonstrates outstanding rotation effects, achieving the highest excess returns and information ratio among all strategies[90][92] --- Model Backtesting Results 1. Size Rotation Model Based on Odds and Win Rates - Total Return: 531.87% - Annualized Return: 23.70% - Annualized Volatility: 23.03% - Maximum Drawdown: 25.25% - Information Ratio (IR): 2.27 - Return-to-Drawdown Ratio: 2.79[79] 2. Weighted Size Rotation Strategy - Total Return: 204.13% - Annualized Return: 13.69% - Annualized Volatility: 22.02% - Maximum Drawdown: 29.17% - Information Ratio (IR): 2.47 - Return-to-Drawdown Ratio: 4.66[84] 3. Detailed Style Rotation Model - Total Return: 1329.51% - Annualized Return: 35.91% - Annualized Volatility: 23.97% - Maximum Drawdown: 23.37% - Information Ratio (IR): 3.11 - Return-to-Drawdown Ratio: 3.87[92] --- Quantitative Factors and Construction Methods 1. Factor Name: Shibor Signal - Construction Idea: Reflects the impact of liquidity conditions on small-cap and large-cap stocks[42] - Construction Process: - Calculate the historical percentile of the latest Shibor rate over the past year - Signal: If the percentile > 50%, favor large-cap; otherwise, favor small-cap[42] - Backtesting Results: - Annualized Excess Return: 11.46% - Information Ratio (IR): 1.23[43] 2. Factor Name: Short-Term Credit Spread - Construction Idea: Captures the impact of short-term credit market conditions on size rotation[47] - Construction Process: - Calculate the spread between 1-year and 7-day AAA+ short-term bond yields - Signal: If the 20-day average spread > 250-day average, favor large-cap; otherwise, favor small-cap[47] - Backtesting Results: - Annualized Excess Return: 7.41% - Information Ratio (IR): 0.79[48] 3. Factor Name: Market Trend - Construction Idea: Reflects the impact of market activity on size rotation[51] - Construction Process: - Compare the 5-day and 20-day moving averages of the CSI All Share Index - Signal: If the 5-day MA > 20-day MA and market volume is increasing, favor small-cap; otherwise, favor large-cap[51] - Backtesting Results: - Annualized Excess Return: 3.52% - Information Ratio (IR): 0.48[52] 4. Factor Name: Market Volatility - Construction Idea: Reflects the impact of market stability on size rotation[54] - Construction Process: - Compare the 20-day market volatility with its 3-year average - Signal: If volatility < average, favor large-cap; otherwise, favor small-cap[54] - Backtesting Results: - Annualized Excess Return: 13.18% - Information Ratio (IR): 1.42[55] 5. Factor Name: Style Momentum - Construction Idea: Captures the momentum effect in size rotation[57] - Construction Process: - Compare the past 4-week returns of CSI 300 and CSI 2000 indices - Signal: If CSI 300 return > CSI 2000 return, favor large-cap; otherwise, favor small-cap[57] - Backtesting Results: - Annualized Excess Return: 8.16% - Information Ratio (IR): 0.87[58] 6. Factor Name: Style Crowding - Construction Idea: Reflects the risk of style overcrowding and potential reversals[60] - Construction Process: - Calculate the historical percentile of the 20-day trading volume of the largest 20% and smallest 20% stocks - Signal: If large-cap volume > 75th percentile, favor small-cap; if small-cap volume > 75th percentile, favor large-cap[60] - Backtesting Results: - Annualized Excess Return: 6.63% - Information Ratio (IR): 0.93[61] 7. Factor Name: Calendar Effect - Construction Idea: Reflects the impact of periodic events on size rotation[63] - Construction Process: - Calculate the historical win rate of large-cap over small-cap for each calendar month - Signal: If the win rate > 50%, favor large-cap; otherwise, favor small-cap[66] - Backtesting Results: - Annualized Excess Return: 4.73% - Information Ratio (IR): 0.50[67] 8. Factor Name: Composite Win Rate Signal - Construction Idea: Combines all individual factors into a single composite signal[72] - Construction Process: - Average the scores of all individual factors to derive the composite win rate - Signal: If the composite score > 0.5, favor large-cap; otherwise, favor small-cap[72] - Backtesting Results: - Annualized Excess Return: 19.72% - Information Ratio (IR): 2.17[73]