Quantitative Models and Construction Methods 1. Model Name: Dividend Timing Model - **Model Construction Idea**: This model uses macroeconomic indicators such as the 10-year US Treasury yield, domestic M2 growth, and the M1-M2 scissors difference to predict the relative excess return of the CSI Dividend Index compared to the Wind All A Index[8][12] - **Model Construction Process**: - The model incorporates three key indicators: 1. **Global Liquidity**: 10-year US Treasury yield 2. **Internal Liquidity**: Domestic M2 year-on-year growth 3. **Domestic Economic Expectations**: Domestic M1-M2 year-on-year scissors difference - Historical data from 2010 onward is used to calculate the annualized excess return of the timing strategy, which is 8.14%[8] - **Model Evaluation**: The model demonstrates strong predictive power for excess returns, but its performance in 2025 YTD shows a negative excess return of -5.36%, indicating potential short-term challenges[8] 2. Model Name: Regression-Based Valuation Model - **Model Construction Idea**: This model uses the CSI Dividend Index's absolute and relative PETTM valuation levels to predict future absolute and excess returns[19][21] - **Model Construction Process**: - **Absolute Valuation**: - The absolute PETTM valuation of the CSI Dividend Index is calculated using a weighted factor adjustment to align with its dividend yield characteristics - Historical data shows a correlation coefficient of -29.66% between the absolute PETTM percentile and future absolute returns, with a regression T-statistic of -15.61[19] - Regression formula: $ y = -0.281x + 0.2635 $ - $y$: Future absolute return - $x$: Absolute PETTM percentile[23] - **Relative Valuation**: - The relative PETTM is calculated as the ratio of the CSI Dividend Index's PETTM to the Wind All A Index's PETTM - Historical data shows a correlation coefficient of -34.10% between the relative PETTM percentile and future excess returns, with a regression T-statistic of -18.23[21] - Regression formula: $ y = -0.1233x + 0.0984 $ - $y$: Future excess return - $x$: Relative PETTM percentile[30] - **Model Evaluation**: The model effectively identifies valuation extremes, with higher PETTM levels indicating greater downside risk. However, the current valuation levels suggest limited upside potential[19][22] 3. Model Name: Price-Volume Regression Model - **Model Construction Idea**: This model uses price and volume metrics, such as the weight of stocks above the 120-day moving average and trading volume percentiles, to predict future returns[25][31] - **Model Construction Process**: - **Price Dimension**: - The weight of CSI Dividend Index constituents above the 120-day moving average is calculated - Historical data shows a correlation coefficient of -43.92% between this weight and future absolute returns, with a regression T-statistic of -20.70[25] - Regression formula: $ y = -0.2344x + 0.2115 $ - $y$: Future absolute return - $x$: Weight above the 120-day moving average[27] - **Volume Dimension**: - Absolute trading volume percentiles are calculated for the CSI Dividend Index - Historical data shows a correlation coefficient of -39.91% between trading volume percentiles and future absolute returns, with a regression T-statistic of -21.87[31] - Regression formula: $ y = -0.3821x + 0.3434 $ - $y$: Future absolute return - $x$: Trading volume percentile[31] - **Model Evaluation**: The model highlights the importance of price and volume extremes in predicting returns. Current metrics suggest moderate upside potential[25][31] 4. Model Name: Dividend 50 Optimized Portfolio - **Model Construction Idea**: This portfolio combines high dividend yield stocks with a linear multi-factor model to enhance capital gains while maintaining a stable dividend style exposure[45] - **Model Construction Process**: - High dividend yield stocks are selected as the base - A linear multi-factor model is applied to optimize capital gains - Barra style factor constraints are used to ensure consistent dividend style exposure - Timing adjustments are made based on the three-dimensional dividend timing model to further enhance returns[45] - **Model Evaluation**: The portfolio demonstrates strong performance, with significant excess returns over the CSI Dividend Index[45] --- Model Backtest Results 1. Dividend Timing Model - Annualized excess return since 2010: 8.14%[8] - 2025 YTD excess return: -5.36%[8] 2. Regression-Based Valuation Model - **Absolute Valuation**: - Current absolute PETTM: 9.35x - 3-year percentile: 98.53% - Predicted future absolute return: -1.34%[19][22] - **Relative Valuation**: - Current relative PETTM: 0.49x - 3-year percentile: 72.36% - Predicted future excess return: 0.92%[22][30] 3. Price-Volume Regression Model - **Price Dimension**: - Weight above 120-day moving average: 57.03% - Predicted future absolute return: 7.78%[25][27] - **Volume Dimension**: - Absolute trading volume percentile: 47.40% - Predicted future absolute return: 16.23%[31] - Relative trading volume percentile: 7.21% - Predicted future excess return: 0.81%[32] 4. Dividend 50 Optimized Portfolio - **Performance Metrics**: - 1-year absolute return: 9.53% - 1-year excess return: 6.20% - 3-month absolute return: 6.04% - 3-month excess return: 2.91%[46]