高频选股因子周报（20260224- 20260227） 高频整体表现优异，多粒度因子维持正收益。AI 增强组合 超额持续回撤。 本报告导读： 上周（20260224-20260227，下同）高频整体表现优异，多粒度因子多头回撤。AI增 强组合超额持续回撤。 投资要点： | [Table_Authors] | 郑雅斌(分析师) | | --- | --- | | | 021-23219395 | | | zhengyabin@gtht.com | | 登记编号 | S0880525040105 | | | 余浩淼(分析师) | | | 021-23185650 | | | yuhaomiao@gtht.com | | 登记编号 | S0880525040013 | [Table_Report] 相关报告 低频选股因子周报（2026.02.13-2026.02.27） 2026.03.01 量化择时和拥挤度预警周报（20260227） 2026.02.28 红利风格择时周报（0209-0213） 2026.02.24 量化择时和拥挤度预警周报（20260220） 2026.02.22 高频选股因子周报（2026 ...

Investment Rating - The report does not explicitly provide an investment rating for the industry Core Insights - The deep learning factor achieved an excess return of 1.50% in February, with the highest heat change observed in the steel and environmental protection sectors [1] - The top 5 industries with positive heat change this week include steel, environmental protection, public utilities, computer, and building materials [2] - The rolling 5-year historical valuation percentiles for major indices are as follows: CSI 300 at 91%, CSI 500 at 100%, and CSI 1000 at 100% [3] Summary by Sections Deep Learning Factor Tracking - A long position portfolio based on the DecompGRU model has generated a cumulative absolute return of 84.37% and an excess return of 43.11% relative to the equal-weighted CSI All Share Index since its inception [10] - The ETF rotation portfolio has achieved a cumulative absolute return of 53.40% and an excess return of 17.47% relative to the Wind ETF Index [13] Market Sentiment Tracking - The CSI 1000 index experienced the highest heat change rate this week, increasing by 3.63%, while the CSI 2000 index saw a decrease of 2.85% [24] - The top 5 concepts with the largest heat change include fertilizers, phosphorus chemicals, cultivated diamonds, rare earth permanent magnets, and rural e-commerce [36] Market Valuation Monitoring - The current valuation of several primary industries is above the historical 80th percentile, including power equipment, electronics, building materials, light industry manufacturing, environmental protection, and steel [45][46] - Industries with valuations below the historical 20th percentile include food and beverage, comprehensive, and non-bank financials [45] Event Tracking - This week, there were 9 stock incentive events, 18 significant shareholder buy/sell events, and 5 private placement events [50][51][53] - Analysts initiated coverage on 12 stocks and upgraded ratings for 3 stocks this week [55][57]

金融工程 证 券 研 究 报 告 市场情绪监控周报（20260224-20260227） 深度学习因子 2 月超额 1.50%，本周热度变化最大 行业为钢铁、环保 深度学习因子跟踪 基于 DecompGRU 模型得分 TOP200 构建周度多头选股组合，组合样本外累计 绝对收益 84.37%，相对全指等权超额 43.11%；2 月组合绝对收益为 5.41%， 超额为 1.50%。 将个股得分聚合为 ETF 轮动组合，组合样本外累计绝对收益 53.40%，相对万 得主题 ETF 指数超额为 17.47%；2 月组合绝对收益为 9.51%，超额为 8.18%。 本周情绪因子跟踪 本周宽基热度变化方面：热度变化率最大的中证 1000，相比上周提高 3.63%， 最小的为中证 2000，相比上周降低 2.85%；宽基热度动量组合 26 年累计收益 为 5.6%。 本周申万行业热度变化方面，一级行业中热度变化率正向变化前 5 的一级行 业分别为钢铁、环保、公用事业、计算机、建筑材料，负向变化前 5 的一级行 业分别为食品饮料、商贸零售、纺织服饰、美容护理、社会服务；申万二级行 业中，热度正向变化率最大的 5 个行业是农 ...

Investment Rating - The report does not explicitly state an investment rating for the industry [1] Core Insights - The report highlights that the deep learning factor tracking has shown a cumulative absolute return of 74.91% since its inception, with a relative excess return of 38.96% compared to the benchmark [10] - The sentiment factor tracking indicates that the top five industries with positive sentiment changes are petroleum and petrochemicals, non-ferrous metals, food and beverage, coal, and textiles and apparel [34] - The market valuation tracking shows that the rolling 5-year historical percentiles for major indices are at 91% for CSI 300, 100% for CSI 500, and 100% for CSI 1000, indicating high valuation levels [44] Summary by Sections Deep Learning Factor Tracking - A long-only portfolio was constructed based on the DecompGRU model, with a cumulative absolute return of 74.91% and a maximum drawdown of 10.08% since March 31, 2025 [10] - An ETF rotation portfolio was also created, achieving a cumulative absolute return of 40.08% since March 18, 2025, with a maximum drawdown of 7.82% [13] Sentiment Factor Tracking - The report tracks sentiment across broad indices, with the CSI 300 showing the highest increase in sentiment by 11.05% compared to the previous week [3] - The top five industries with positive sentiment changes include petroleum and petrochemicals, non-ferrous metals, food and beverage, coal, and textiles and apparel [34] Market Valuation Monitoring - The report indicates that several primary industries are currently above the 80% historical percentile for valuations, including electronics, power equipment, light industry manufacturing, and construction materials [46] - Conversely, industries like food and beverage and non-bank financials are below the 20% historical percentile, suggesting potential undervaluation [46] Event Tracking - The report details various corporate events, including stock incentive plans, significant shareholder buybacks, and analyst coverage updates, which may influence market sentiment and stock performance [48][56][57]

- The DecompGRU model was used to construct a weekly long-only stock selection portfolio, holding the top 200 stocks with the highest integrated scores equally weighted The portfolio is rebalanced weekly based on the updated factor values from the previous Friday's closing prices Stocks with price limits or suspension are excluded, and transaction costs are not considered The benchmark is the CSI All Share Equal Weight Index[8][10] - The DecompGRU model's individual stock scores were aggregated to construct an ETF rotation portfolio The ETF pool is limited to industry and thematic ETFs, retaining only the ETF with the highest average daily trading volume over the past five days if multiple ETFs track the same index The portfolio is rebalanced weekly, holding 2-6 ETFs per period, with a benchmark of the Wind Thematic ETF Index[11][13] - A sentiment factor was constructed using user behavior data from Tonghuashun, aggregating stock-level heat indicators (browsing, watchlist, and click counts) normalized as a percentage of the total market and scaled by 10,000 This aggregated heat indicator serves as a proxy for "sentiment heat" at the broad-based index, industry, and concept levels[15][19][28] - A simple rotation strategy was built based on the weekly heat change rate (MA2) of broad-based indices, buying the index with the highest heat change rate on the last trading day of each week If the "Others" group has the highest change rate, the strategy remains in cash The strategy achieved an annualized return of 8.74% since 2017, with a maximum drawdown of 23.5%[21][24] - A concept-level sentiment strategy was constructed by selecting the top 5 concepts with the highest weekly heat change rates, excluding the bottom 20% of stocks by market capitalization within each concept From each concept, the top 10 stocks by total heat were equally weighted to form the "TOP" portfolio, while the bottom 10 stocks formed the "BOTTOM" portfolio The BOTTOM portfolio achieved an annualized return of 15.71% with a maximum drawdown of 28.89%[39][41][42] - The DecompGRU TOP200 portfolio achieved a cumulative absolute return of 74.91% and an excess return of 38.96% relative to the CSI All Share Equal Weight Index since its inception on March 31, 2025 The portfolio's maximum drawdown was 10.08%, with a weekly win rate of 68.18% and a monthly win rate of 100% In January 2026, the portfolio's absolute return was 8.99%, with an excess return of 0.98%[10] - The ETF rotation portfolio achieved a cumulative absolute return of 40.08% and an excess return of 5.93% relative to the Wind Thematic ETF Index since its inception on March 18, 2025 The portfolio's maximum drawdown was 7.82%, with a weekly win rate of 64.44% and a monthly win rate of 70% In January 2026, the portfolio's absolute return was 10.98%, with an excess return of 3.37%[13][14] - The broad-based index heat momentum strategy achieved a cumulative return of 6.6% in 2026[24] - The concept-level sentiment BOTTOM portfolio achieved a cumulative return of 3.7% in 2026[42]

- The "Buy Intention Factor" showed strong performance at the beginning of the year, with intraday high-frequency skewness factor, intraday downside volatility proportion factor, post-opening buy intention proportion factor, post-opening buy intention strength factor, post-opening large order net buy proportion factor, post-opening large order net buy strength factor, intraday return factor, end-of-day trading proportion factor, average single outflow amount proportion factor, and large order push-up factor all being evaluated[5][6][9] - The "Multi-Granularity Factor" showed average performance, with GRU(10,2)+NN(10) factor, GRU(50,2)+NN(10) factor, multi-granularity model (5-day label) factor, and multi-granularity model (10-day label) factor being evaluated[5][6][9] - The "AI Enhanced Portfolio" had a poor start to the year, with significant drawdowns observed in the weekly rebalanced CSI 500 AI enhanced wide constraint portfolio, CSI 500 AI enhanced strict constraint portfolio, CSI 1000 AI enhanced wide constraint portfolio, and CSI 1000 AI enhanced strict constraint portfolio[5][6][9] Quantitative Factors and Construction Methods 1. **Factor Name: Intraday High-Frequency Skewness Factor** - **Construction Idea**: Measures the skewness of intraday returns to capture the asymmetry in return distribution[5][6] - **Construction Process**: Calculated using high-frequency data to determine the skewness of returns within a trading day[5][6] - **Evaluation**: Demonstrated strong performance at the beginning of the year[5][6] 2. **Factor Name: Intraday Downside Volatility Proportion Factor** - **Construction Idea**: Measures the proportion of downside volatility in intraday returns[5][6] - **Construction Process**: Calculated using high-frequency data to determine the proportion of downside volatility within a trading day[5][6] - **Evaluation**: Showed moderate performance[5][6] 3. **Factor Name: Post-Opening Buy Intention Proportion Factor** - **Construction Idea**: Measures the proportion of buy intentions after market opening[5][6] - **Construction Process**: Calculated using high-frequency data to determine the proportion of buy intentions after the market opens[5][6] - **Evaluation**: Demonstrated strong performance at the beginning of the year[5][6] 4. **Factor Name: Post-Opening Buy Intention Strength Factor** - **Construction Idea**: Measures the strength of buy intentions after market opening[5][6] - **Construction Process**: Calculated using high-frequency data to determine the strength of buy intentions after the market opens[5][6] - **Evaluation**: Showed moderate performance[5][6] 5. **Factor Name: Post-Opening Large Order Net Buy Proportion Factor** - **Construction Idea**: Measures the proportion of net buy orders of large size after market opening[5][6] - **Construction Process**: Calculated using high-frequency data to determine the proportion of net buy orders of large size after the market opens[5][6] - **Evaluation**: Demonstrated weak performance[5][6] 6. **Factor Name: Post-Opening Large Order Net Buy Strength Factor** - **Construction Idea**: Measures the strength of net buy orders of large size after market opening[5][6] - **Construction Process**: Calculated using high-frequency data to determine the strength of net buy orders of large size after the market opens[5][6] - **Evaluation**: Showed weak performance[5][6] 7. **Factor Name: Intraday Return Factor** - **Construction Idea**: Measures the return within a trading day[5][6] - **Construction Process**: Calculated using high-frequency data to determine the return within a trading day[5][6] - **Evaluation**: Demonstrated strong performance at the beginning of the year[5][6] 8. **Factor Name: End-of-Day Trading Proportion Factor** - **Construction Idea**: Measures the proportion of trading activity at the end of the day[5][6] - **Construction Process**: Calculated using high-frequency data to determine the proportion of trading activity at the end of the day[5][6] - **Evaluation**: Showed strong performance[5][6] 9. **Factor Name: Average Single Outflow Amount Proportion Factor** - **Construction Idea**: Measures the proportion of average single outflow amounts[5][6] - **Construction Process**: Calculated using high-frequency data to determine the proportion of average single outflow amounts[5][6] - **Evaluation**: Demonstrated moderate performance[5][6] 10. **Factor Name: Large Order Push-Up Factor** - **Construction Idea**: Measures the impact of large orders on price increases[5][6] - **Construction Process**: Calculated using high-frequency data to determine the impact of large orders on price increases[5][6] - **Evaluation**: Showed moderate performance[5][6] 11. **Factor Name: GRU(10,2)+NN(10) Factor** - **Construction Idea**: Combines GRU and neural network models to capture complex patterns in data[5][6] - **Construction Process**: Utilizes GRU with 10 units and 2 layers, followed by a neural network with 10 units[5][6] - **Evaluation**: Demonstrated average performance[5][6] 12. **Factor Name: GRU(50,2)+NN(10) Factor** - **Construction Idea**: Combines GRU and neural network models to capture complex patterns in data[5][6] - **Construction Process**: Utilizes GRU with 50 units and 2 layers, followed by a neural network with 10 units[5][6] - **Evaluation**: Showed weak performance[5][6] 13. **Factor Name: Multi-Granularity Model (5-Day Label) Factor** - **Construction Idea**: Uses multi-granularity approach to capture patterns over different time frames[5][6] - **Construction Process**: Trained using a 5-day label to capture short-term patterns[5][6] - **Evaluation**: Demonstrated average performance[5][6] 14. **Factor Name: Multi-Granularity Model (10-Day Label) Factor** - **Construction Idea**: Uses multi-granularity approach to capture patterns over different time frames[5][6] - **Construction Process**: Trained using a 10-day label to capture longer-term patterns[5][6] - **Evaluation**: Showed weak performance[5][6] Factor Backtest Results 1. **Intraday High-Frequency Skewness Factor**: IC -0.007, e^(-rank mae) 0.312, long-short return 0.29%, long-only excess return 0.99%, monthly win rate 1/1[9][10] 2. **Intraday Downside Volatility Proportion Factor**: IC -0.001, e^(-rank mae) 0.313, long-short return 0.22%, long-only excess return 0.95%, monthly win rate 1/1[9][10] 3. **Post-Opening Buy Intention Proportion Factor**: IC 0.032, e^(-rank mae) 0.324, long-short return 1.04%, long-only excess return -0.41%, monthly win rate 0/1[9][10] 4. **Post-Opening Buy Intention Strength Factor**: IC 0.027, e^(-rank mae) 0.323, long-short return 0.65%, long-only excess return 0.62%, monthly win rate 1/1[9][10] 5. **Post-Opening Large Order Net Buy Proportion Factor**: IC -0.006, e^(-rank mae) 0.306, long-short return -0.52%, long-only excess return -0.53%, monthly win rate 0/1[9][10] 6. **Post-Opening Large Order Net Buy Strength Factor**: IC 0.004, e^(-rank mae) 0.308, long-short return -0.07%, long-only excess return -0.66%, monthly win rate 0/1[9][10] 7. **Intraday Return Factor**: IC 0.037, e^(-rank mae) 0.328, long-short return 1.77%, long-only excess return 1.89%, monthly win rate 1/1[9][10] 8. **End-of-Day Trading Proportion Factor**: IC 0.084, e^(-rank mae) 0.334, long-short return 2.67%, long-only excess return 1.35%, monthly win rate 1/1[9][10] 9. **Average Single Outflow Amount Proportion Factor**: IC 0.013, e^(-rank mae) 0.319, long-short return 0.45%, long-only excess return 0.14%, monthly win rate 1/1[9][10] 10. **Large Order Push-Up Factor**: IC -0.007, e^(-rank mae) 0.327, long-short return 0.22%, long-only excess return 0.43%, monthly win rate 1/1[9][10] 11. **GRU(10,2

- The DecompGRU model was used to construct a weekly long-only stock selection portfolio, holding the top 200 stocks with the highest integrated scores based on the model. The portfolio is rebalanced weekly on the first trading day, using factor values updated after the previous Friday's close. Stocks from the CSI All Share Index are selected, excluding stocks with trading halts or price limits, and transaction costs are not considered. The benchmark for comparison is the CSI All Share equal-weighted index[7][9] - The DecompGRU model's stock scores were aggregated to construct an ETF rotation portfolio. The ETF pool is limited to industry and thematic ETFs, retaining only the ETF with the highest 5-day average trading volume if multiple ETFs track the same index. ETFs must meet minimum trading volume criteria (5-day average > 20 million and 20-day average > 10 million). The portfolio holds 2-6 ETFs per period and is rebalanced weekly without a fixed schedule. The benchmark for comparison is the Wind ETF Index[10][12] - A sentiment factor was constructed using user behavior data from Tonghuashun, aggregating stock-level heat metrics (browsing, watchlist, and click counts) normalized by market share on the same day and multiplied by 10,000. The sentiment factor serves as a proxy for "emotional heat" at the broader index, industry, and concept levels[14] - A simple rotation strategy was built based on weekly heat change rates (MA2 smoothed) for broad-based indices. On the last trading day of each week, the strategy buys the index with the highest heat change rate. If the "Others" group has the highest rate, the strategy remains in cash. The strategy's annualized return since 2017 is 8.74%, with a maximum drawdown of 23.5%. In 2025, the strategy achieved a return of 36.8%, compared to the benchmark's 35%[20][23] - A concept-level sentiment strategy was constructed by selecting the top 5 concepts with the highest weekly heat change rates. Stocks within these concepts were filtered to exclude the bottom 20% by market capitalization. Two portfolios were created: a "TOP" portfolio holding the top 10 stocks by total heat within each concept, and a "BOTTOM" portfolio holding the bottom 10 stocks. The BOTTOM portfolio achieved an annualized return of 15.71% with a maximum drawdown of 28.89%. In 2025, the BOTTOM portfolio returned 41.8%[40][41] - The DecompGRU TOP200 portfolio achieved a cumulative absolute return of 60.48% and an excess return of 34.62% relative to the CSI All Share equal-weighted index since its inception on March 31, 2025. The portfolio's maximum drawdown was 10.08%, with weekly and monthly win rates of 67.50% and 100%, respectively. In December 2025, the portfolio's absolute return was 7.57%, with an excess return of 5.46%[9] - The ETF rotation portfolio achieved a cumulative absolute return of 26.23% and an excess return of 1.56% relative to the Wind ETF Index since its inception on March 18, 2025. The portfolio's maximum drawdown was 7.82%, with weekly and monthly win rates of 60.98% and 66.67%, respectively. In December 2025, the portfolio's absolute return was 2.35%, with an excess return of -1.51%[12][13]

- The report analyzes the performance of Alpha factors in various market segments, including the entire market, CSI 300, CSI A500, CSI 500, CSI 1000, and the ChiNext board, with monthly and weekly rebalancing[5] - The deep learning factor agru_dailyquote shows RankIC averages of 5.18%, 12.44%, 14.42%, and 13.94% over the past week, month, year, and historically, respectively, with a historical win rate of 91.63%[5] - The DL_1 factor shows RankIC averages of 4.00%, 19.68%, 16.48%, and 14.08% over the past week, month, year, and historically, respectively, with a historical win rate of 87.97%[5] - The fimage factor shows RankIC averages of -0.17%, 3.95%, 3.92%, and 5.17% over the past week, month, year, and historically, respectively, with a historical win rate of 78.11%[5] - The integrated_bigsmall_longshort factor, a Level-2 high-frequency factor, shows RankIC averages of -4.74%, 15.18%, 9.78%, and 11.10% over the past week, month, year, and historically, respectively, with a historical win rate of 75.86%[5] - The Amihud_illiq factor, a minute-frequency factor, shows RankIC averages of -3.21%, 16.88%, 13.34%, and 11.17% over the past week, month, year, and historically, respectively, with a historical win rate of 74.95%[5] - The report includes detailed performance analysis of 29 Level-2 high-frequency factors and 55 minute-frequency factors[5] - The deep learning factor agru_dailyquote shows an excess return of 9.01% in the CSI 300 index, 9.68% in the CSI A500 index, 5.82% in the CSI 500 index, 11.18% in the CSI 800 index, 10.75% in the CSI 1000 index, and 6.58% in the ChiNext index, with maximum drawdowns of 1.96%, 1.23%, 3.47%, 1.49%, 1.58%, and 1.95%, respectively, for the year-to-date period[5]

Quantitative Models and Construction Methods Model Name: GRU(10,2)+NN(10) - **Model Construction Idea**: This model combines Gated Recurrent Units (GRU) with a neural network (NN) to capture temporal dependencies in high-frequency data[4] - **Model Construction Process**: The model uses a GRU with 10 units and 2 layers, followed by a neural network with 10 units. The GRU processes sequential data, and the NN captures non-linear relationships[4] - **Model Evaluation**: The model shows strong performance in capturing temporal patterns and generating significant returns[4] Model Name: GRU(50,2)+NN(10) - **Model Construction Idea**: Similar to the GRU(10,2)+NN(10) model but with more GRU units to capture more complex temporal dependencies[4] - **Model Construction Process**: The model uses a GRU with 50 units and 2 layers, followed by a neural network with 10 units. This setup allows for deeper temporal feature extraction[4] - **Model Evaluation**: The model is effective in capturing complex temporal patterns and generating significant returns[4] Model Name: Multi-Granularity Model (5-day label) - **Model Construction Idea**: This model uses multiple granularities of data to improve prediction accuracy[4] - **Model Construction Process**: The model labels data with a 5-day horizon and uses a combination of features from different time scales to enhance prediction[4] - **Model Evaluation**: The model shows strong performance in capturing multi-scale patterns and generating significant returns[4] Model Name: Multi-Granularity Model (10-day label) - **Model Construction Idea**: Similar to the 5-day label model but with a 10-day horizon to capture longer-term dependencies[4] - **Model Construction Process**: The model labels data with a 10-day horizon and combines features from different time scales to enhance prediction[4] - **Model Evaluation**: The model is effective in capturing longer-term patterns and generating significant returns[4] Model Backtesting Results - **GRU(10,2)+NN(10)**: - Multi-Period Return: -1.32% (last week), -0.71% (November), 44.83% (2025)[4] - Excess Return: -0.77% (last week), -1.01% (November), 7.21% (2025)[4] - **GRU(50,2)+NN(10)**: - Multi-Period Return: -1.5% (last week), -1.23% (November), 44.56% (2025)[4] - Excess Return: -0.83% (last week), -0.92% (November), 7.9% (2025)[4] - **Multi-Granularity Model (5-day label)**: - Multi-Period Return: 0.75% (last week), 2.56% (November), 63.15% (2025)[4] - Excess Return: 1.07% (last week), 2.36% (November), 24.44% (2025)[4] - **Multi-Granularity Model (10-day label)**: - Multi-Period Return: 0.91% (last week), 2.55% (November), 57.7% (2025)[4] - Excess Return: 0.98% (last week), 2.27% (November), 24.14% (2025)[4] Quantitative Factors and Construction Methods Factor Name: Intraday Skewness Factor - **Factor Construction Idea**: This factor captures the skewness of intraday returns to identify asymmetric return distributions[4] - **Factor Construction Process**: The factor is calculated using the skewness of intraday returns over a specified period[4] - **Factor Evaluation**: The factor is effective in identifying stocks with asymmetric return distributions[4] Factor Name: Downside Volatility Proportion Factor - **Factor Construction Idea**: This factor measures the proportion of downside volatility to capture risk characteristics[4] - **Factor Construction Process**: The factor is calculated as the proportion of downside volatility relative to total volatility over a specified period[4] - **Factor Evaluation**: The factor is effective in identifying stocks with higher downside risk[4] Factor Name: Post-Open Buy Intention Proportion Factor - **Factor Construction Idea**: This factor measures the proportion of buy intentions after market open to capture investor sentiment[4] - **Factor Construction Process**: The factor is calculated as the proportion of buy orders relative to total orders after market open[4] - **Factor Evaluation**: The factor is effective in capturing investor sentiment and predicting stock movements[4] Factor Name: Post-Open Buy Intensity Factor - **Factor Construction Idea**: This factor measures the intensity of buy intentions after market open to capture investor sentiment strength[4] - **Factor Construction Process**: The factor is calculated as the intensity of buy orders relative to total orders after market open[4] - **Factor Evaluation**: The factor is effective in capturing the strength of investor sentiment and predicting stock movements[4] Factor Backtesting Results - **Intraday Skewness Factor**: - Multi-Period Return: -0.26% (last week), 0.49% (November), 22.76% (2025)[4] - Excess Return: 0.42% (last week), 1.46% (November), 6.14% (2025)[4] - **Downside Volatility Proportion Factor**: - Multi-Period Return: 0.38% (last week), 1.35% (November), 20.32% (2025)[4] - Excess Return: 0.41% (last week), 1.08% (November), 3.54% (2025)[4] - **Post-Open Buy Intention Proportion Factor**: - Multi-Period Return: 0.28% (last week), -0.01% (November), 19.33% (2025)[4] - Excess Return: 0.47% (last week), 0.28% (November), 8.78% (2025)[4] - **Post-Open Buy Intensity Factor**: - Multi-Period Return: 0.27% (last week), 0.57% (November), 26.36% (2025)[4] - Excess Return: -0.22% (last week), -0.55% (November), 10.06% (2025)[4]

Quantitative Models and Construction - **Model Name**: DecompGRU **Model Construction Idea**: The model improves information interaction between time-series and cross-sectional data by introducing two simple de-mean modules on the GRU baseline model[18] **Model Construction Process**: 1. The DecompGRU model architecture is based on GRU as the baseline 2. Two de-mean modules are added to enhance the interaction between time-series and cross-sectional data 3. The model is trained using IC and weighted MSE loss functions[18] **Model Evaluation**: The model demonstrates improved interaction between time-series and cross-sectional data, enhancing prediction accuracy[18] Model Backtesting Results - **DecompGRU TOP200 Portfolio**: - Cumulative absolute return: 41.11% - Excess return relative to WIND All A equal-weight index: 13.98% - Maximum drawdown: 10.08% - Weekly win rate: 64.52% - Monthly win rate: 100% - October absolute return: 1.78%, excess return: -0.07%[11] - **ETF Rotation Portfolio**: - Cumulative absolute return: 19.06% - Excess return relative to benchmark: -2.00% - Maximum drawdown: 7.82% - Weekly win rate: 62.50% - Monthly win rate: 57.14% - October absolute return: -2.04%, excess return: -1.18%[14][15] Quantitative Factors and Construction - **Factor Name**: Sentiment Heat Factor **Factor Construction Idea**: The factor aggregates stock-level sentiment heat metrics (e.g., browsing, self-selection, and clicks) to represent broader market sentiment[19] **Factor Construction Process**: 1. Individual stock sentiment heat is calculated as the sum of browsing, self-selection, and click counts 2. The sentiment heat is normalized by dividing by the total market sentiment on the same day and multiplying by 10,000 3. Aggregated sentiment heat is used as a proxy for market sentiment at the index, industry, and concept levels[19] **Factor Evaluation**: The factor effectively captures market sentiment and its impact on pricing errors[19] Factor Backtesting Results - **Broad-based Index Sentiment Heat Rotation Strategy**: - Annualized return since 2017: 8.74% - Maximum drawdown: 23.5% - 2025 portfolio return: 38.5% - Benchmark return: 32.9%[28] - **Concept Sentiment Heat BOTTOM Portfolio**: - Annualized return: 15.71% - Maximum drawdown: 28.89% - 2025 portfolio return: 42.1%[41][44]