Quantitative Models and Construction Methods DTW Timing Model - **Model Name**: DTW Timing Model - **Model Construction Idea**: The model is based on the principle of similarity and the DTW algorithm, focusing on price and volume timing[1][5][14] - **Model Construction Process**: - The model examines the similarity between current index trends and historical trends, selecting several historical segments with high similarity as references[25] - It calculates the weighted average future price change and weighted standard deviation of the selected historical segments (weights are the inverse of the distance)[25] - Based on the average future price change and standard deviation, trading signals are generated[25] - The model uses the DTW distance algorithm instead of the Euclidean distance for similarity measurement, as DTW distance can better handle time series mismatches[27] - Improved DTW algorithms such as Sakoe-Chiba and Itakura Parallelogram are introduced to overcome the "over-bending" issue in traditional DTW algorithms[29][30][35] - **Model Evaluation**: The model has shown stable excess returns in general market conditions, although it faced some drawdowns during periods of sudden macroeconomic policy changes[16] Foreign Capital Timing Model - **Model Name**: Foreign Capital Timing Model - **Model Construction Idea**: The model is based on the divergence between foreign and domestic related assets[1][14] - **Model Construction Process**: - The model uses two foreign-listed assets related to A-shares: FTSE China A50 Index Futures (Singapore market) and Southern A50 ETF (Hong Kong market)[34] - It constructs two indicators from FTSE China A50 Index Futures: premium and price divergence, forming the FTSE China A50 Index Futures timing signal[34] - It constructs a price divergence indicator from Southern A50 ETF, forming the Southern A50 ETF timing signal[34] - The timing signals from both assets are combined to form the foreign capital timing signal[34] - **Model Evaluation**: The model has shown good performance with high annualized returns and low maximum drawdowns[20][23] Model Backtest Results DTW Timing Model - **Absolute Return**: 25.79% since November 2022[5][16] - **Excess Return**: 16.83% relative to CSI 300[5][16] - **Maximum Drawdown**: 21.32%[5][16] - **Absolute Return (2024)**: 23.98% on CSI 300[18] - **Excess Return (2024)**: 2.76%[18] - **Maximum Drawdown (2024)**: 21.36%[18] - **Win Rate (2024)**: 53.85%[18] - **Profit-Loss Ratio (2024)**: 2.93[18] Foreign Capital Timing Model - **Absolute Return (2024)**: 29.11% for long strategy[5][23] - **Maximum Drawdown (2024)**: 8.32% for long strategy[5][23] - **Annualized Return (2014-2024)**: 18.96% for long-short strategy, 14.19% for long strategy[20] - **Maximum Drawdown (2014-2024)**: 25.69% for long-short strategy, 17.27% for long strategy[20] - **Daily Win Rate (2014-2024)**: Nearly 55%[20] - **Profit-Loss Ratio (2014-2024)**: Both exceed 2.5[20]

Quantitative Models and Construction Methods 1. Model Name: DTW Timing Model - **Model Construction Idea**: Based on the principle of similarity and the Dynamic Time Warping (DTW) algorithm, this model identifies timing signals by comparing current market trends with historical patterns[4][25] - **Model Construction Process**: - The model calculates the similarity between the current index trend and historical market trends using the DTW distance metric, which is more flexible than Euclidean distance for time-series data[27] - Historical segments with high similarity are selected as references - The weighted average future returns and standard deviations of these historical segments are calculated (weights are the inverse of the DTW distance) - Trading signals are generated based on the average future returns and standard deviations[25] - Improved DTW algorithms, such as those with Sakoe-Chiba and Itakura boundary constraints, are used to address issues like "over-warping" in traditional DTW[29] - **Model Evaluation**: The DTW algorithm is particularly suitable for time-series problems and outperforms other methods due to its flexibility in handling temporal misalignments[27] 2. Model Name: Foreign Capital Timing Model - **Model Construction Idea**: This model leverages the price movements of offshore assets related to A-shares to generate timing signals for the A-share market[32] - **Model Construction Process**: - Two offshore assets are used: FTSE China A50 Index Futures (Singapore market) and Southern A50 ETF (Hong Kong market) - For FTSE China A50 Index Futures, two indicators are constructed: basis and price deviation - For Southern A50 ETF, a price deviation indicator is constructed - Timing signals from these two assets are combined to form the overall foreign capital timing signal[32] --- Model Backtesting Results 1. DTW Timing Model - **Out-of-Sample Performance (Since November 2022)**: - Absolute return: 25.38% - Excess return over CSI 300: 19.02% - Maximum drawdown: 20.07% - Weekly win rate: Over 60% - Weekly win rate in 2024: Over 70%[4][16] - **Performance in 2024 (CSI 300)**: - Absolute return: 23.58% - Excess return: 5.27% - Maximum drawdown: 14.88% - Trading win rate: 63.64% - Profit-loss ratio: 2.64[16] 2. Foreign Capital Timing Model - **Full-Sample Performance (2014-2024)**: - Annualized return: 18.96% (long-short), 14.19% (long-only) - Maximum drawdown: 25.69% (long-short), 17.27% (long-only) - Daily win rate: Nearly 55% - Profit-loss ratio: Above 2.5[18] - **Out-of-Sample Performance in 2024**: - Long-only strategy: - Absolute return: 28.05% - Maximum drawdown: 8.32% - Long-short strategy: - Absolute return: 21.66% - Maximum drawdown: 13.14%[21]