Quantitative Models and Factor Construction Quantitative Models - Model Name: Extended CAPM with Sentiment Beta Model Construction Idea: Incorporates turnover rate changes as a proxy for investor sentiment to measure the sensitivity of asset returns to sentiment changes, extending the traditional CAPM framework[35][36] Model Construction Process: The model is expressed as: $R = \alpha + \beta_{MRT} \times MKT + \beta_{TO} \times \Delta TO + \varepsilon$ - $R$: Asset daily return (e.g., stocks, industries, funds) - $MKT$: Market factor, represented by CSI All Share Index daily return - $\Delta TO$: Change in turnover rate, calculated as: - $\beta_{TO}$: Sentiment Beta, representing the sensitivity of asset returns to sentiment changes[35][36] Model Evaluation: Demonstrates strong predictive power for future asset performance, with lower Sentiment Beta assets generally outperforming higher Sentiment Beta assets[42][43][50] Quantitative Factors - Factor Name: Sentiment Beta (Stock Level) Factor Construction Idea: Measures the sensitivity of stock returns to changes in investor sentiment, represented by turnover rate changes[35][36] Factor Construction Process: - Calculate Sentiment Beta for each stock using the extended CAPM model - Neutralize the factor for industry and market capitalization effects using the following regression: - $\beta_{i,TO}$: Neutralized Sentiment Beta for stock $i$ - $mktcap$: Market capitalization of the stock - $lnd_{j,i}$: Dummy variable for industry classification[60][61] Factor Evaluation: RankIC mean of -2.75%, annualized RankICIR of -0.49, indicating strong predictive power for future stock returns[43][45] - Factor Name: Sentiment Beta (Industry Level) Factor Construction Idea: Measures the sensitivity of industry index returns to sentiment changes[47] Factor Construction Process: - Calculate Sentiment Beta for each industry index using the extended CAPM model - Group industries by Sentiment Beta and analyze future performance differences[47] Factor Evaluation: RankIC mean of -4.44%, annualized RankICIR of -0.29, with low Sentiment Beta industries outperforming high Sentiment Beta industries[47][51] - Factor Name: Sentiment Beta (Fund Level) Factor Construction Idea: Quantifies fund managers' contrarian investment ability based on the sensitivity of fund returns to sentiment changes[50] Factor Construction Process: - Calculate Sentiment Beta for each fund using the extended CAPM model - Group funds by Sentiment Beta and analyze future performance differences[50][52] Factor Evaluation: RankIC mean of -5.78%, annualized RankICIR of -0.48, with low Sentiment Beta funds outperforming high Sentiment Beta funds[52][55] - Factor Name: Contrarian Investment Ability Factor (Fund Holdings-Based) Factor Construction Idea: Aggregates the Sentiment Beta of stocks held by a fund to measure the fund manager's contrarian ability[58] Factor Construction Process: - Calculate stock-level Sentiment Beta using an extended Fama-French five-factor model with turnover rate changes - Aggregate stock-level Sentiment Beta weighted by fund holdings: - $w_i$: Normalized weight of stock $i$ in the fund's holdings - $\widehat{\beta}{i,TO}$: Neutralized Sentiment Beta for stock $i$[63][64] Factor Evaluation: RankIC mean of -7.30%, annualized RankICIR of -0.92, win rate of 67.21%[64][65] - Factor Name: Contrarian Investment Ability Factor (Fund Returns-Based) Factor Construction Idea: Directly measures fund managers' contrarian ability using fund return data[69] Factor Construction Process: - Extend the Carhart four-factor model by adding turnover rate changes: - $\beta{i,TO}$: Sentiment Beta for fund $i$[70][71] Factor Evaluation: RankIC mean of -8.92%, annualized RankICIR of -1.04, win rate of 75.41%[71][77] - Factor Name: Contrarian Investment Ability Factor (Composite) Factor Construction Idea: Combines holdings-based and returns-based factors to create a comprehensive measure of contrarian ability[75] Factor Construction Process: - Equal-weight the holdings-based and returns-based factors to form the composite factor[75] Factor Evaluation: RankIC mean of -10.85%, annualized RankICIR of -1.39, win rate of 78.69%[75][79] Backtesting Results of Factors - Sentiment Beta (Stock Level): RankIC mean -2.75%, annualized RankICIR -0.49[43][45] - Sentiment Beta (Industry Level): RankIC mean -4.44%, annualized RankICIR -0.29[47][51] - Sentiment Beta (Fund Level): RankIC mean -5.78%, annualized RankICIR -0.48[52][55] - Contrarian Investment Ability Factor (Holdings-Based): RankIC mean -7.30%, annualized RankICIR -0.92, win rate 67.21%[64][65] - Contrarian Investment Ability Factor (Returns-Based): RankIC mean -8.92%, annualized RankICIR -1.04, win rate 75.41%[71][77] - Contrarian Investment Ability Factor (Composite): RankIC mean -10.85%, annualized RankICIR -1.39, win rate 78.69%[75][79]
金融工程专题研究FOF 系列专题之九:基金经理逆向投资能力与投资业绩
Guoxin Securities·2025-06-04 08:25