Quantitative Models and Construction 国证 2000 Index Enhancement Strategy - **Model Name**: 国证 2000 Index Enhancement Strategy - **Model Construction Idea**: Focused on the small-cap stock rotation phenomenon in A-shares, aiming to select stocks effectively within 国证 2000 index components to enhance returns [11] - **Model Construction Process**: - Selected factors such as technical, reversal, and idiosyncratic volatility, which showed strong performance on 国证 2000 index components [12] - Addressed high correlation among factors by regressing volatility factors on technical and reversal factors to obtain residual volatility factors [12] - Combined all major factors equally and performed industry and market capitalization neutralization to construct the 国证 2000 enhancement factor [12] - Formula: Residual volatility factor = Volatility factor - Regression(Technical factor, Reversal factor) [12] - **Model Evaluation**: Demonstrated strong predictive performance with an IC mean of 12.63% and T-statistic of 12.70 [12] - **Strategy Construction**: - Monthly rebalancing at the end of each month, buying the top 10% ranked stocks based on factor values, constructing an equal-weighted long portfolio [15] - Backtesting period: April 2014 to present, benchmarked against 国证 2000 index, with a transaction fee rate of 0.2% per side [15] Machine Learning Index Enhancement Strategy - **Model Name**: TSGRU+LGBM Machine Learning Index Enhancement Strategy - **Model Construction Idea**: Improved machine learning stock selection model by integrating TimeMixer framework with GRU and LightGBM, leveraging multi-scale mixing and seasonal/trend decomposition mechanisms [21] - **Model Construction Process**: - Original strategy used GBDT and NN models trained on different feature datasets and prediction labels, but showed signs of failure due to market style adjustments [21] - Enhanced model incorporated TimeMixer framework into GRU, combined LightGBM with TSGRU latent vectors and traditional quantitative factors [21] - Optimized portfolio construction by controlling tracking error and individual stock weight deviation to maximize factor exposure [25] - **Model Evaluation**: Improved ability to capture recent market information, showing strong performance [21] Dividend Style Timing + Dividend Stock Selection Strategy - **Model Name**: Dividend Style Timing + Dividend Stock Selection Strategy - **Model Construction Idea**: Leveraged the long-term stability and high dividend characteristics of dividend stocks to reduce risk during weak market conditions [36] - **Model Construction Process**: - Used 10 indicators related to economic growth and monetary liquidity to construct a dynamic event factor system for dividend index timing [36] - Applied AI models to test stock selection within 中证红利 index components, achieving stable excess returns [36] - **Model Evaluation**: Demonstrated significant stability improvement compared to 中证红利 index total return [36] --- Model Backtesting Results 国证 2000 Index Enhancement Strategy - **IC Mean**: 12.63% [12] - **Latest Month IC**: 25.34% [12] - **Annualized Excess Return**: 13.30% [16] - **Information Ratio (IR)**: 1.73 [16] - **Tracking Error**: 7.68% [19] - **October Excess Return**: 2.92% [16] TSGRU+LGBM Machine Learning Index Enhancement Strategy - **沪深 300 Index**: - **Annualized Excess Return**: 6.96% [26] - **Information Ratio (IR)**: 1.40 [26] - **Tracking Error**: 4.97% [26] - **October Excess Return**: 2.25% [26] - **中证 500 Index**: - **Annualized Excess Return**: 10.11% [30] - **Information Ratio (IR)**: 1.96 [30] - **Tracking Error**: 5.16% [30] - **October Excess Return**: -0.59% [30] - **中证 1000 Index**: - **Annualized Excess Return**: 13.52% [35] - **Information Ratio (IR)**: 2.37 [35] - **Tracking Error**: 5.70% [35] - **October Excess Return**: 2.63% [35] Dividend Style Timing + Dividend Stock Selection Strategy - **Stock Selection Strategy**: - **Annualized Return**: 18.98% [38] - **Sharpe Ratio**: 0.90 [38] - **October Return**: 2.52% [38] - **Timing Strategy**: - **Annualized Return**: 13.83% [38] - **Sharpe Ratio**: 0.90 [38] - **October Return**: 3.28% [38] - **固收+ Strategy**: - **Annualized Return**: 7.39% [38] - **Sharpe Ratio**: 2.19 [38] - **October Return**: 0.92% [38]

Core Insights - The article discusses the use of a multi-factor model suitable for the CSI 300 index constituents, combined with a small-cap high-growth satellite strategy, to enhance the performance of the CSI 300 enhanced strategy [1][2] - Since 2016, the CSI 300 enhanced strategy has achieved an annualized excess return of 12.6% with a tracking error of 5.2% under a satellite allocation of 30% domestic and 10% foreign [1][2] Summary by Sections - **Performance Analysis**: The CSI 300 enhanced strategy has shown an annualized excess return of at least 10% since 2016, with an information ratio exceeding 2.0. The internal component of the strategy has lower tracking error and relative drawdown, while the external component offers greater return elasticity but with higher tracking error and drawdown [1][2] - **Model Construction**: The multi-factor model is constructed based on fundamental and momentum indicators, which has demonstrated better stock selection robustness compared to the all-A multi-factor model [1] - **Satellite Strategy**: The external component can be replaced with small-cap high-growth or GARP strategies. The optimal satellite allocation depends on the risk-return preference, with the most extreme case showing an annualized excess return of 17.5% when fully utilizing satellite strategies [2]

Core Insights - The report emphasizes the development of a deep learning factor selection system for stock prediction and portfolio optimization, shifting from traditional logic-driven methods to data-driven approaches [7][10]. - The system integrates diverse data sources, including daily and minute market data, to enhance the performance of alpha signals [7][10]. - The report outlines the construction of multiple models that utilize different network architectures to extract unique alpha signals, demonstrating low correlation among them [8][54]. Data and Network - The input data consists of three categories: daily market data, minute market data, and manually crafted features, with neural networks independently extracting alpha features from each dataset [11]. - The report describes the use of Long Short-Term Memory (LSTM) networks combined with self-attention mechanisms to capture long-term dependencies in time series data [19]. - A Graph Attention Network (GAT) is employed to model the complex relationships between stocks, providing a global analysis perspective [20]. Alpha Models - The report presents various alpha models, including simple equal-weight, tree model weighting, and network weighting, with a focus on combining multiple signals to enhance robustness [3][3.1][3.2]. - The average Information Coefficient (IC) for the combined factors since 2019 is reported as 11.3% for 5-day IC and 12.4% for 10-day IC, indicating strong predictive power [31][32]. Risk Models - The report highlights the use of neural networks to identify high-dimensional non-linear risk patterns directly from raw price and volume data, enhancing risk control in portfolio construction [9]. Index Enhancement Strategies - The report details the performance of enhanced index strategies based on deep learning alpha signals, with annualized returns reported as follows: - CSI 300 enhanced portfolio: 18.2% annualized return, 14.2% excess return over the index [3][5.1]. - CSI 500 enhanced portfolio: 22.4% annualized return, 17.2% excess return over the index [3][5.2]. - CSI 1000 enhanced portfolio: 29.8% annualized return, 24.5% excess return over the index [3][5.3].