科创板因子测试与组合构建

Quantitative Models and Construction Methods - Model Name: Full A Linear Regression Multi-Factor Model Construction Idea: This model uses multiple factors such as market capitalization, dividend yield, low volatility, turnover rate, fundamentals, expectations, and high-frequency factors to predict stock returns across the A-share market. However, its performance on the STAR Market is relatively weak[6][7] Construction Process: The model calculates composite scores based on factor values and evaluates the correlation between these scores and future stock returns using IC and RankIC metrics. The STAR Market's IC and RankIC averages are significantly lower than other A-shares[7] Evaluation: The model's effectiveness is limited on the STAR Market, with weaker IC and RankIC performance compared to other A-shares[6][7] - Model Name: Multi-Factor Top100 Combination Construction Idea: Select stocks with the highest composite factor scores to build a portfolio[37][40] Construction Process: 1. Exclude stocks with a market capitalization below 2 billion RMB 2. Select the top 100 stocks with the highest composite factor scores 3. Construct a market-cap-weighted portfolio with a single stock weight cap of 10% Evaluation: The model demonstrates strong performance with high IC, monthly win rates, and significant excess returns relative to the STAR Market Composite Index[37][40] - Model Name: Linear Optimization Combination Construction Idea: Build an enhanced portfolio under specific constraints to optimize risk-return characteristics[42] Construction Process: 1. Apply constraints such as individual stock deviation (1%), market cap deviation (0.2%), beta deviation (0.5%), momentum deviation (0.5%), and industry deviation (5%) 2. Construct the portfolio based on these constraints and evaluate its excess returns relative to the STAR Market Composite Index[42] Evaluation: The model achieves lower excess returns compared to the Top100 Combination but exhibits reduced volatility and drawdowns[42] - Model Name: Composite Combination Construction Idea: Combine the Top100 Combination and Linear Optimization Combination to balance risk and return[45] Construction Process: 1. Allocate 20% weight to the Top100 Combination and 80% weight to the Linear Optimization Combination 2. Rebalance monthly to maintain the weight distribution Evaluation: The model achieves higher excess returns and a better risk-return ratio compared to individual combinations[45] Model Backtesting Results - Full A Linear Regression Multi-Factor Model: - STAR Market monthly average IC: 4.62% - STAR Market monthly average RankIC: 7.53% - Other A-shares monthly average IC: 6.74% - Other A-shares monthly average RankIC: 9.88%[7] - Multi-Factor Top100 Combination: - Annualized return: 18.6% - Relative annualized excess return: 22.8% - Monthly win rate: 74.0% - Excess return annualized volatility: 10.3% - Information ratio (IR): 2.27[40][41] - Linear Optimization Combination: - Annualized excess return: 10.2% - Excess return annualized volatility: 5.6% - Information ratio (IR): 2.0 - Monthly win rate: 72.0%[42][43] - Composite Combination: - Annualized excess return: 12.7% - Excess return annualized volatility: 6.0% - Information ratio (IR): 2.25 - Monthly win rate: 70.0%[45][46] Quantitative Factors and Construction Methods - Factor Name: Basic Negative Exclusion Factor Construction Idea: Combine factors with strong negative effects to exclude underperforming stocks[23][24] Construction Process: 1. Combine SUE, SUE_Rev, and revenue growth rate factors equally 2. Identify stocks in the bottom 20% of factor rankings as "short positions" 3. Evaluate the excess returns of these short positions relative to the STAR Market average[23][24] Evaluation: The factor demonstrates strong negative excess returns and statistical significance[23][24] - Factor Name: Market Correlation Factor Construction Idea: Measure the correlation between individual stock returns and the STAR Market Composite Index to identify high-risk premium stocks[25][27] Construction Process: 1. Use the past 12 months' stock returns to regress against the STAR Market Composite Index returns 2. Use regression coefficients as the market correlation factor 3. Divide stocks into quintiles based on factor values and evaluate excess returns for each group[27][28] Evaluation: The factor shows statistically significant IC and cumulative IC trends, with higher correlation stocks outperforming[27][28] - Factor Name: Improved Momentum Factor Construction Idea: Adjust traditional momentum factors to account for downside risks and sector trends[30][31] Construction Process: 1. Extract trading days where the sector's average return is positive over the past 3 months 2. Calculate the 20% quantile of individual stock excess returns on these trading days 3. Use the quantile value as the momentum factor[30][31] Evaluation: The factor demonstrates strong IC and win rates, with statistical significance even after orthogonalization[30][34] Factor Backtesting Results - Basic Negative Exclusion Factor: - Annualized excess return: -6.75% - Monthly win rate: 32.0% - Statistical significance: t-value -4.03, p-value 0.000[23][24] - Market Correlation Factor: - Monthly average IC: 2.5% - Monthly win rate: 58% - Statistical significance: 10% confidence level[27][29] - Improved Momentum Factor: - Original monthly average IC: 3.91% - Orthogonalized monthly average IC: 2.64% - Monthly win rate: 66.0% (original), 64.0% (orthogonalized)[31][34]