开源量化评论（111）：基于虚拟指数的另类增强方案

Quantitative Models and Construction Methods 1. Model Name: Virtual Index (虚拟指数) - Model Construction Idea: The Virtual Index is designed as an alternative to the official index, aiming to replicate the weight distribution of the original index while reducing concentration. This ensures that the enhanced strategy based on the Virtual Index can outperform the original index with greater stability[3][45][56] - Model Construction Process: - The Virtual Index is constructed by fitting the holdings of index-enhanced funds to approximate the weight distribution of the original index. - Two key metrics are used to evaluate the Virtual Index: - JS Divergence (Jensen-Shannon Divergence): Measures the similarity between the weight distributions of the Virtual Index and the original index. A lower JS divergence indicates higher similarity $J S(P||Q)={\frac{1}{2}}K L(P||M)+{\frac{1}{2}}K L(Q||M)$ $M={\frac{P+Q}{2}}$ $K L(P||Q)=\sum_{x\in X}P(x)l n{\frac{P(x)}{Q(x)}}$ - HHI (Herfindahl-Hirschman Index): Measures the concentration of weights in the Virtual Index. A lower HHI indicates lower concentration $H H I=\sum w_{i}^{2}$ - The Virtual Index is tested across three major indices: CSI 300, CSI 500, and CSI 1000[45][49][50] - Model Evaluation: The Virtual Index demonstrates weight distribution similarity to the original index (low JS divergence) and lower industry concentration (low HHI), making it a robust alternative for enhanced strategies[46][50][56] 2. Model Name: Enhanced Strategy Based on Virtual Index - Model Construction Idea: This strategy uses the Virtual Index as the baseline for constructing enhanced portfolios, aiming to outperform the original index by leveraging multi-factor models[57][63] - Model Construction Process: - Two groups of enhanced portfolios are constructed: - Reference Group: Based on the original index's weight distribution - Test Group: Based on the Virtual Index's weight distribution - Both groups use the same multi-factor framework and constraints for consistency[57][59] - Model Evaluation: The test group (Virtual Index-based) consistently outperforms the reference group (original index-based), particularly for CSI 300 and CSI 500. However, the performance improvement diminishes as the number of index constituents increases (e.g., CSI 1000)[59][61] --- Model Backtesting Results 1. Virtual Index - JS Divergence: - CSI 300: 0.0078 - CSI 500: 0.0062 - CSI 1000: 0.0062[46] - HHI (Industry Concentration): - CSI 300: Lower than the original index - CSI 500: Lower than the original index - CSI 1000: Lower than the original index, but with smaller deviations compared to CSI 300 and CSI 500[50][52] 2. Enhanced Strategy Based on Virtual Index - Cumulative Net Value: - CSI 300: - Test Group: 4.35 - Reference Group: 3.25[57] - CSI 500: - Test Group: 5.38 - Reference Group: 4.02[59] - CSI 1000: - Test Group: Slightly higher than the Reference Group, with divergence starting in 2024[59][61] --- Quantitative Factors and Construction Methods 1. Factor Name: RankIC and RankICIR Evaluation - Factor Construction Idea: Evaluate the predictive power and stability of individual factors within the Virtual Index and original index constituents[62][63] - Factor Construction Process: - RankIC and RankICIR are calculated for each factor within the Virtual Index and original index constituents - Factors are tested for their ability to differentiate stock performance and their stability over time[62][64] - Factor Evaluation: - Most factors show higher RankIC and RankICIR values within the Virtual Index constituents, indicating better predictive power and stability - Exceptions include factors like "long_momentum2" and "active_trading," which perform better in the original index constituents[62][64] --- Factor Backtesting Results 1. RankIC and RankICIR - RankIC: Higher for most factors in the Virtual Index constituents compared to the original index constituents - RankICIR: Higher for most factors in the Virtual Index constituents compared to the original index constituents[62][64]