Quantitative Models and Construction Methods 1. Model Name: Timing Models for Broad-Based Indices - **Model Construction Idea**: The timing models are designed to capture the trends and turning points of six major equity indices, including Wind All A, CSI Dividend Total Return, Hang Seng Tech, STAR 50, Wind Microcap, and CSI 2000. These models aim to address the higher volatility and weaker momentum effects of equity indices compared to bonds and commodities[3][6]. - **Model Construction Process**: 1. **Factor Selection**: - Common factors across indices include capital flows, interest rates, commodity futures and spot prices, high-frequency daily indices (e.g., BDI), high-frequency fundamentals (e.g., daily consumption and production data), overseas factors (e.g., US stock and bond prices, volatility), and domestic equity market indicators (e.g., margin financing balance, stock buyback amounts)[7]. - Index-specific factors include trading volume, turnover, P/E ratio, P/B ratio, net capital inflow, and technical indicators (e.g., stochastic indicators, standard deviation, RSI, OBV)[7]. 2. **Factor Adjustments**: - Low-frequency factors (e.g., monthly or quarterly) are reduced due to their lagging nature and limited guidance for high-volatility assets[8]. - High-frequency factors are enriched by incorporating different parameter settings for technical indicators (e.g., moving averages, momentum indicators) to capture diverse market conditions[8]. 3. **Model Structure Adjustments**: - **Class Balance Mechanism**: Applied selectively based on the index's trend characteristics (e.g., Wind Microcap shows a clear upward trend, while Hang Seng Tech does not)[10]. - **Hidden Layers and Units**: For high-volatility assets, increasing hidden units improves model precision without overfitting, but adding layers may lead to overfitting[10]. - **Hyperparameter Tuning**: Adjustments include window length (to balance signal stability and responsiveness), regularization coefficients (to prevent overfitting), and learning rates (to ensure convergence without gradient explosion)[11]. 2. Model Name: CSI Dividend Total Return Timing Model - **Model Construction Idea**: This model focuses on capturing the high-frequency oscillations of the CSI Dividend Total Return Index, which lacks clear trends[15]. - **Model Construction Process**: - The model incorporates enriched factors and increased hidden units to improve performance, but the high-frequency oscillations of the index limit its effectiveness[15]. 3. Model Name: Hang Seng Tech Timing Model - **Model Construction Idea**: This model emphasizes Hong Kong stock market volume-price data and global liquidity factors, considering the unique trading day misalignment between Hong Kong and mainland China[22]. - **Model Construction Process**: - Factors are adjusted to account for the trading day misalignment and the high-frequency oscillations of the Hang Seng Tech Index[22]. 4. Model Name: STAR 50 Timing Model - **Model Construction Idea**: This model targets "innovative assets" in mainland China, focusing on volume-price data of STAR 50 stocks[28]. - **Model Construction Process**: - The model prioritizes mainland stock volume-price data to capture the high volatility and significant oscillations of the STAR 50 Index[28]. 5. Model Name: Wind Microcap Timing Model - **Model Construction Idea**: This model captures the trends of the microcap market, which exhibits a relatively fast upward trend and clear turning points[34]. - **Model Construction Process**: - The model leverages the clear upward trend of the Wind Microcap Index to identify turning points with higher sensitivity[34]. 6. Model Name: CSI 2000 Timing Model - **Model Construction Idea**: This model complements the Wind Microcap Timing Model by covering small-cap stocks, with a focus on the CSI 2000 Index's unique characteristics[41]. - **Model Construction Process**: - The model integrates data from CSI 2000 and Wind Microcap to provide comprehensive coverage of the small-cap market[41]. --- Model Backtesting Results 1. Wind All A Timing Model - Correct intervals: 23 - Incorrect intervals: 9 - Interval win rate: 71.88%[12] 2. CSI Dividend Total Return Timing Model - Correct intervals: 22 - Incorrect intervals: 4 - Interval win rate: 84.62%[15] 3. Hang Seng Tech Timing Model - Correct intervals: 23 - Incorrect intervals: 6 - Interval win rate: 79.31%[22] 4. STAR 50 Timing Model - Correct intervals: 19 - Incorrect intervals: 4 - Interval win rate: 82.61%[28] 5. Wind Microcap Timing Model - Correct intervals: 21 - Incorrect intervals: 5 - Interval win rate: 80.77%[34] 6. CSI 2000 Timing Model - Correct intervals: 20 - Incorrect intervals: 8 - Interval win rate: 71.43%[41]

- The report introduces multiple quantitative models for market timing, including short-term, medium-term, and long-term models, such as the "Volume Model," "Feature Institutional Model," "Feature Volume Model," "Intelligent Algorithm Model," "Limit-Up/Down Model," "Up/Down Return Difference Model," "Calendar Effect Model," and "Long-Term Momentum Model" [1][8][11] - The "Volume Model" is neutral for all broad-based indices in the short term, while the "Feature Institutional Model" is bullish, and the "Feature Volume Model" is bearish. The "Intelligent Algorithm Model" is neutral for both CSI 300 and CSI 500 indices [11][63] - The "Limit-Up/Down Model" and "Up/Down Return Difference Model" are bullish for all broad-based indices in the medium term, while the "Calendar Effect Model" remains neutral [12][64] - The "Long-Term Momentum Model" is bullish for the long term [13][65] - The "Comprehensive Weapon V3 Model" is bullish for A-shares, while the "Comprehensive Guozheng 2000 Model" is neutral [14][66] - For Hong Kong stocks, the "Turnover to Volatility Model" is bullish, while the "Up/Down Return Difference Model" is neutral in the medium term [15][67] - The report emphasizes that market timing requires a multi-cycle, multi-strategy model system, combining defensive and aggressive strategies to achieve a balanced approach [8] - Backtesting results for the models indicate that the "Double Bottom Pattern" portfolio outperformed the Shanghai Composite Index by 1.99% this week, with a cumulative return of 16.12% since December 31, 2020, compared to the index's 14.13% [38] - The "Cup and Handle Pattern" portfolio outperformed the Shanghai Composite Index by 1.32% this week but has underperformed the index by -1.36% cumulatively since December 31, 2020 [38]

- The report includes multiple quantitative models for market timing, such as short-term, mid-term, and long-term models, which are integrated into a multi-cycle, multi-strategy system[11][13][68] - Short-term models include the volume model (neutral for most broad-based indices), low volatility model (neutral), institutional feature model (bullish), feature volume model (bearish), and intelligent algorithm models for CSI 300 (bearish) and CSI 500 (neutral)[13][68][76] - Mid-term models include the limit-up-and-down model (neutral) and calendar effect model (neutral)[14][69] - Long-term models include the momentum model, which is bullish[15][70] - Comprehensive models such as the A-Share Comprehensive Weapon V3 model and the A-Share Comprehensive CSI 2000 model are bearish[16][71] - For Hong Kong stocks, the mid-term turnover inverse volatility model is bearish[17][72] Model Backtesting Results - The short-term institutional feature model is bullish for CSI 300 and CSI 500 indices, while the feature volume model is bearish for CSI 300[13][68] - The long-term momentum model shows bullish signals across broad-based indices[15][70] - Comprehensive models indicate bearish signals for CSI 2000 and other indices[16][71] - The turnover inverse volatility model for Hong Kong stocks has turned bearish[17][72]

Quantitative Models and Construction 1. Model Name: Volume Model - **Construction Idea**: This model uses trading volume data to predict short-term market trends[2][11] - **Construction Process**: The model evaluates trading volume changes across broad-based indices to generate buy or neutral signals. Specific thresholds or patterns in volume are used to determine the directional bias[11] - **Evaluation**: The model is partially optimistic for broad-based indices in the short term[11][66] 2. Model Name: Low Volatility Model - **Construction Idea**: This model focuses on the volatility of asset prices to assess market conditions[11] - **Construction Process**: The model calculates the historical volatility of indices and assigns a neutral signal when volatility remains within a predefined range[11] - **Evaluation**: The model is neutral for the short term[11][66] 3. Model Name: Institutional Feature Model (LHB) - **Construction Idea**: This model incorporates institutional trading data, such as large trades or block trades, to predict market movements[11] - **Construction Process**: The model analyzes institutional trading patterns, such as those from the "Dragon and Tiger List" (龙虎榜), to generate signals. A bearish signal is issued when institutional selling dominates[11] - **Evaluation**: The model is bearish for the short term[11][66] 4. Model Name: Intelligent Algorithm Models (HS300 and CSI500) - **Construction Idea**: These models use machine learning algorithms to analyze historical data and predict market trends[11] - **Construction Process**: The HS300 model generates a bullish signal for the CSI 300 index, while the CSI500 model remains neutral. The models likely use features such as price momentum, volume, and other technical indicators[11] - **Evaluation**: The HS300 model is optimistic, while the CSI500 model is neutral in the short term[11][66] 5. Model Name: Limit-Up/Limit-Down Model - **Construction Idea**: This model evaluates the frequency and distribution of limit-up and limit-down events to assess market sentiment[12] - **Construction Process**: The model calculates the ratio of stocks hitting daily price limits and assigns a neutral signal when no significant bias is observed[12] - **Evaluation**: The model is neutral for the medium term[12][67] 6. Model Name: Calendar Effect Model - **Construction Idea**: This model leverages seasonal or calendar-based patterns in market behavior[12] - **Construction Process**: The model analyzes historical performance around specific calendar dates (e.g., month-end or quarter-end) to generate signals. It remains neutral when no strong seasonal patterns are detected[12] - **Evaluation**: The model is neutral for the medium term[12][67] 7. Model Name: Long-Term Momentum Model - **Construction Idea**: This model uses long-term price momentum to predict market trends[13] - **Construction Process**: The model calculates momentum indicators over extended periods and assigns a neutral signal when no clear trend is identified[13] - **Evaluation**: The model is neutral for all broad-based indices in the long term[13][68] 8. Model Name: Comprehensive Weaponry V3 Model - **Construction Idea**: This composite model integrates multiple short-term, medium-term, and long-term signals to provide an overall market outlook[14] - **Construction Process**: The model aggregates signals from various sub-models (e.g., volume, volatility, momentum) and generates a bullish signal for the A-share market[14] - **Evaluation**: The model is optimistic for the A-share market[14][69] 9. Model Name: Comprehensive Guozheng 2000 Model - **Construction Idea**: This model focuses on the Guozheng 2000 index, combining multiple signals to assess market conditions[14] - **Construction Process**: Similar to the Weaponry V3 model, this model aggregates signals but remains neutral for the Guozheng 2000 index[14] - **Evaluation**: The model is neutral for the Guozheng 2000 index[14][69] 10. Model Name: Turnover-to-Volatility Model (Hong Kong Market) - **Construction Idea**: This model evaluates the ratio of turnover to price volatility to predict market trends in the Hong Kong market[15] - **Construction Process**: The model calculates the turnover-to-volatility ratio and generates a bullish signal when the ratio indicates strong market activity relative to volatility[15] - **Evaluation**: The model is optimistic for the medium term in the Hong Kong market[15][70] --- Backtesting Results of Models 1. Volume Model - **Signal**: Partially bullish for broad-based indices in the short term[11][66] 2. Low Volatility Model - **Signal**: Neutral for the short term[11][66] 3. Institutional Feature Model (LHB) - **Signal**: Bearish for the short term[11][66] 4. Intelligent Algorithm Models (HS300 and CSI500) - **Signal**: Bullish for HS300; neutral for CSI500 in the short term[11][66] 5. Limit-Up/Limit-Down Model - **Signal**: Neutral for the medium term[12][67] 6. Calendar Effect Model - **Signal**: Neutral for the medium term[12][67] 7. Long-Term Momentum Model - **Signal**: Neutral for all broad-based indices in the long term[13][68] 8. Comprehensive Weaponry V3 Model - **Signal**: Bullish for the A-share market[14][69] 9. Comprehensive Guozheng 2000 Model - **Signal**: Neutral for the Guozheng 2000 index[14][69] 10. Turnover-to-Volatility Model (Hong Kong Market) - **Signal**: Bullish for the medium term in the Hong Kong market[15][70]

Group 1 - The risk premium percentile of the CSI All Share Index is 71.95%, indicating that the market has returned from a high return area to a normal return area [1][8] - The price-to-book ratio percentile of the CSI All Share Index is 21.54%, suggesting that the market has returned from an undervalued state to a normal but slightly undervalued state [12] - The Shanghai Composite Index and CSI 800 are still in an undervalued state, warranting close attention [13] Group 2 - The CSI All Share Index's deviation rate is -0.03%, indicating that the overall price level of the market is in a normal range [16] - The ChiNext 50 has returned to a basic normal range after two months of recovery from an oversold state [19] - Over the past six months, the performance of value and growth styles has varied, and the value vs. growth style has yet to be defined, with future trends still to be observed [23] Group 3 - The performance of low valuation styles has generally been superior over the past six months, but high valuation styles have shown strong performance in the past month, suggesting investors should closely monitor the potential transition between high and low valuation styles [27] - Small-cap styles have significantly outperformed over the past six months, indicating a need for future focus on small-cap style targets [29] Group 4 - There has been a certain degree of excess return for convertible bonds relative to the CSI All Share Index over the past six months, suggesting that investors should consider convertible bond varieties from an asset allocation perspective [2][44] - Different types of convertible bonds have shown varying performance over the past six months, with a recommendation to focus on equity-oriented targets [48] Group 5 - The report emphasizes the importance of market style rotation, highlighting the differences in performance between value vs. growth, low vs. high valuation, and large vs. small capitalization stocks [20][21] - The report identifies that the performance of small-cap stocks has been notably superior, suggesting a focus on small-cap style targets moving forward [29] Group 6 - The report discusses industry/theme index rotation, focusing on low valuation rotation and dual momentum rotation strategies [33][34] - A selection of reference targets based on valuation factors or momentum factors is provided for investors to consider [37]

Key Points - The report indicates that the risk premium percentile of the CSI All Share Index is 80.41%, suggesting that the market is in a high return zone [1][5] - The report highlights that the price-to-book ratio percentile of the CSI All Share Index is 8.98%, indicating that the market is in a state of severe undervaluation [1][10] - The report notes that the deviation rate of the CSI All Share Index is -4.03%, suggesting that the overall price level of the market is in a normal range [1][13] - The report suggests that the performance of the value style has been significantly superior over the past six months, recommending a focus on value style targets [1][21] - The report also indicates that the performance of the low valuation style has been notably superior over the past six months, advising attention to low valuation style targets [1][24] - The report states that the performance of the small-cap style has been significantly superior over the past six months, recommending a focus on small-cap style targets [1][26] - The report identifies that there has been a certain degree of excess return for convertible bonds relative to the CSI All Share Index over the past six months, suggesting investors pay attention to convertible bond varieties from an asset allocation perspective [1][40]