丨证券研究报告丨 金融工程丨专题报告 [Table_Title] 质量风格占优，攻守兼备红利组合持续跑出 超额 [Table_Summary] 本周 A 股市场整体回暖，科创板块表现强势，红利资产也实现正收益；红利内细分板块来看， 红利增长和红利质量相对纯粹红利资产存在超额收益。从市场表现来看，本周 A 股内各板块均 实现正收益，但收益分化较大，TMT 板块持续领涨，而医疗板块涨幅较小；聚焦电子板块内部， 半导体赛道表现活跃，集成电路设计周度涨幅大幅领先电子其他子赛道。策略表现上，本周红 利增强组合均跑赢中证红利全收益，而电子增强组合受周五半导体快速拉升行情影响，均未能 跑赢电子行业指数。 分析师及联系人 [Table_Author] 冷旭晟 蔡文捷 覃川桃 SAC：S0490524080001 SAC：S0490523120001 SAC：S0490513030001 SFC：BUT353 请阅读最后评级说明和重要声明 %% %% %% %% 报告要点 [Table_Title 质量风格占优，攻守兼备红利组合持续跑出超 2] 额 [Table_Summary2] 自 2023 年 7 月以来，长江金工团队先 ...

[Table_StockNameRptType] 金融工程 专题报告 基于图形派与基本面派的股市信息效率模型 ——学海拾珠系列之二百四十六 [Table_RptDate] 报告日期：2025-08-20 主要观点： [Table_Summary] 本篇是学海拾珠系列第二百四十六篇，文章利用图表派—基本面派 模型，研究信息效率股票市场的局限性。研究结果表明，市场可能存在 两种不同的状态并存：恒定的错误定价和振荡性的错误定价。 分析师：严佳炜 执业证书号：S0010520070001 邮箱：yanjw@hazq.com ⚫ 信息效率与市场动态的理论分歧 学界对市场信息效率存分歧：Fama（1970）主张价格完全反映信 息；Grossman 和 Stiglitz（1980）认为完全有效会削弱信息收集动 机，导致恒定错误定价；Lo 和 Farmer（1999）强调效率动态变化， 错误定价随行为调整变化。 ⚫ 图表派—基本面派模型的设定与行为机制 本文构建的图表派—基本面派模型同时解释恒定与振荡性错误定 价，兼容各方观点。本模型基于图表派与基本面派的互动机制，做市商 根据超额需求调整价格，图表派通过趋势外推预测，基本面 ...

证券研究报告 | 2025年08月19日 **Acknowledgments** I would like to thank my supervisor, for his kind of support. I would like to thank my supervisor, for his kind of support. 市场表现：今日(20250819) 规模指数中中证 2000 指数表现较好，板块指数 中北证 50 指数表现较好，风格指数中中证 500 成长指数表现较好。综合、 通信、综合金融、纺织服装、食品饮料行业表现较好，国防军工、非银、石 油石化、医药、煤炭行业表现较差。液态金属、消费电子代工、光模块(CPO)、 家纺、白酒等概念表现较好，PVC 地板、保险精选、毫米波、PEEK 材料、 光刻机等概念表现较差。 市场情绪：今日市场情绪较为高涨，收盘时有 104 只股票涨停，有 10 只股 票跌停。昨日涨停股票今日收盘收益为 3.61%，昨日跌停股票今日收盘收益 为-6.26%。今日封板率 69%，较前日下降 1%，连板率 27%，较前日下降 3%。 市场资金流向：截至 20250818 两融余 ...

Quantitative Models and Factor Construction Quantitative Models and Construction Methods - **Model Name**: Market Cap Segmented Linear Regression Model **Construction Idea**: Adjust the weights of factor regressions based on market cap segmentation to address the performance differences of factors across different market cap groups [7][10][12] **Construction Process**: 1. Factors are divided into five categories: Dividend, ROE_SUE, Daily Volume-Price, High-Frequency Volume-Price, and a final composite factor [7][10] 2. Use OLS regression with IC or ICIR weighting to combine sub-factors into composite factors [7] 3. Apply KMedian clustering on the log of market cap to divide stocks into 11 groups [7] 4. Assign weights to each group using the formula: $ w_{i}=w_{base}+(1-w_{base})*|i-I|/n $ where $w_{base}$ is the minimum weight (set to 0.9, 0.5, or 0), $n$ is the number of groups, and $I$ is the group with the highest weight [7] 5. Train 11 models with different weight assignments and evaluate the composite factor's IC, RankMAE, long-short returns, and long-only returns [7] **Evaluation**: This model improves factor performance in specific market cap segments, particularly for small-cap stocks, but extreme weighting can increase volatility [7][12] - **Model Name**: Market Cap Weighted Composite Factor Model **Construction Idea**: Reweight composite factors based on market cap distribution to enhance factor performance in specific indices [48][49][65] **Construction Process**: 1. Use market cap weights from benchmark indices (e.g., CSI 300, CSI 500, CSI 1000) to reweight composite factors [48] 2. Construct enhanced portfolios with weekly rebalancing and constraints on individual stock weights, industry weights, and turnover [48] **Evaluation**: Significant performance improvement in CSI 300 and CSI 500 indices, with annualized excess returns increasing by over 1% in some cases. However, the method is less effective for CSI 1000 [49][65][79] - **Model Name**: Market Cap Weighted Cross-Composite Factor Model **Construction Idea**: Match factor weights to the market cap group of each stock to reduce parameter sensitivity [80][81] **Construction Process**: 1. Assign factor values based on the stock's market cap group: $ F_{i}=F_{l_{i}}\;\;i\in I $ where $i$ belongs to market cap group $I$ [80] 2. Evaluate single-factor performance and construct enhanced portfolios for different indices [81][85] **Evaluation**: Performance improvement is observed in CSI 300 and CSI 500 indices, but the method is less effective for CSI 1000. Parameter sensitivity is reduced compared to other methods [85][92][96] - **Model Name**: Multi-Style Factor Weighted Composite Factor Model **Construction Idea**: Incorporate style factors (e.g., value-growth, industry) into the weighting process to address factor performance differences across styles [98][99] **Construction Process**: 1. Cluster stocks based on style factors using Manhattan distance [98] 2. Construct 11 composite factor models centered on each style cluster [98] 3. Use cross-composite and component-composite methods to evaluate performance in enhanced portfolios [100][101] **Evaluation**: Performance improvement is limited compared to market cap-based methods. Cross-composite weighting shows better results than component-composite weighting in some cases [101][115][132] Backtest Results of Models - **Market Cap Segmented Linear Regression Model**: - IC: 0.057 (all-market), 0.037 (CSI 300), 0.040 (CSI 500), 0.052 (CSI 1000), 0.060 (small-cap) [7][81][84] - RankMAE: 1.090 (all-market), 1.119 (CSI 300), 1.111 (CSI 500), 1.106 (CSI 1000), 1.092 (small-cap) [7][81][84] - Long-Short Returns: 1.07% (all-market), 0.38% (CSI 300), 0.49% (CSI 500), 0.92% (CSI 1000), 1.19% (small-cap) [7][81][84] - **Market Cap Weighted Composite Factor Model**: - CSI 300: Annualized Return 8.21%, IR 0.966, Max Drawdown 15.67% (base_w=0) [49] - CSI 500: Annualized Return 14.64%, IR 1.385, Max Drawdown 12.60% (base_w=0.5) [59] - CSI 1000: Annualized Return 18.95%, IR 1.585, Max Drawdown 16.59% (equal weight) [70] - **Market Cap Weighted Cross-Composite Factor Model**: - CSI 300: Annualized Return 7.36%, IR 0.901, Max Drawdown 16.33% (base_w=0) [85] - CSI 500: Annualized Return 15.06%, IR 1.409, Max Drawdown 13.14% (base_w=0.5) [92] - CSI 1000: Annualized Return 18.95%, IR 1.585, Max Drawdown 16.59% (equal weight) [92] - **Multi-Style Factor Weighted Composite Factor Model**: - CSI 300: Annualized Return 7.24%, IR 0.926, Max Drawdown 16.32% (base_w=0.9, component-composite) [103] - CSI 500: Annualized Return 14.17%, IR 1.377, Max Drawdown 12.65% (base_w=0, cross-composite) [115] - CSI 1000: Annualized Return 18.63%, IR 1.570, Max Drawdown 16.47% (base_w=0, component-composite) [132]

- The report introduces several active quantitative strategies launched by the Changjiang Quantitative Team since July 2023, including the Dividend Selection Strategy and the Industry High Winning Rate Strategy[6][13] - The "Dividend Quality" segment showed relatively active performance with a weekly average return of approximately 1.64%, indicating excess returns compared to pure dividend assets[6][16] - The "Central State-Owned Enterprises High Dividend 30 Portfolio" and the "Balanced Dividend 50 Portfolio" both outperformed the CSI Dividend Total Return Index this week, with excess returns of approximately 0.61% and 1.51%, respectively[6][22] - The "Balanced Dividend 50 Portfolio" achieved positive returns this week[6][22] - The "Electronic Balanced Allocation Enhanced Portfolio" and the "Electronic Sector Preferred Enhanced Portfolio" both achieved positive returns this week, although they did not outperform the electronic industry index[7][31] - The "Electronic Sector Preferred Enhanced Portfolio" had a weekly return of approximately 6.20%, outperforming the median of technology-themed fund products[7][31] Quantitative Models and Construction Methods 1. Model Name: Dividend Selection Strategy; Model Construction Idea: Focuses on selecting stocks with high dividend yields and quality; Model Construction Process: The strategy involves screening stocks based on dividend yield, payout ratio, and other fundamental factors to construct a portfolio that aims to provide stable and high returns; Model Evaluation: The strategy has shown to provide excess returns compared to pure dividend assets[6][13][16] 2. Model Name: Industry High Winning Rate Strategy; Model Construction Idea: Focuses on selecting stocks within high-performing industries; Model Construction Process: The strategy involves identifying industries with strong performance and selecting stocks within those industries based on various fundamental and technical factors; Model Evaluation: The strategy aims to provide alternative perspectives and investment choices for investors by tracking market hotspots and selecting individual stocks within high-performing industries[6][13] Model Backtesting Results 1. Dividend Selection Strategy, Excess Return: 1.64%[6][16] 2. Central State-Owned Enterprises High Dividend 30 Portfolio, Excess Return: 0.61%[6][22] 3. Balanced Dividend 50 Portfolio, Excess Return: 1.51%[6][22] 4. Electronic Sector Preferred Enhanced Portfolio, Weekly Return: 6.20%[7][31] Quantitative Factors and Construction Methods 1. Factor Name: Dividend Quality; Factor Construction Idea: Focuses on stocks with high dividend quality; Factor Construction Process: The factor involves screening stocks based on dividend yield, payout ratio, and other fundamental factors to identify stocks with high dividend quality; Factor Evaluation: The factor has shown to provide excess returns compared to pure dividend assets[6][16] 2. Factor Name: Industry Performance; Factor Construction Idea: Focuses on stocks within high-performing industries; Factor Construction Process: The factor involves identifying industries with strong performance and selecting stocks within those industries based on various fundamental and technical factors; Factor Evaluation: The factor aims to provide alternative perspectives and investment choices for investors by tracking market hotspots and selecting individual stocks within high-performing industries[6][13] Factor Backtesting Results 1. Dividend Quality Factor, Weekly Average Return: 1.64%[6][16] 2. Industry Performance Factor, Weekly Return: 6.20%[7][31]

Group 1 - The article outlines a comprehensive research framework training program titled "洞察价值，共创未来" (Insight Value, Co-create Future) scheduled for August 18-19 and August 25-26, 2025, focusing on various sectors including macroeconomics, consumption, finance, cycles, medicine, technology, and manufacturing [18][19]. - The training sessions will cover a wide range of topics, with specific time slots allocated for each area of research, such as food and beverage, internet applications, and renewable energy [14][15][16]. - The event will take place at the Guotai Junan Financial Bund Plaza in Shanghai, emphasizing the importance of in-depth analysis across all sectors [18]. Group 2 - The training program is designed to enhance the research capabilities of analysts and is led by various chief analysts specializing in different fields, ensuring a comprehensive approach to industry analysis [8][10]. - Participants will have the opportunity to engage with experts in macroeconomic research, strategy, fixed income, and various sector-specific studies, fostering a collaborative learning environment [14][15][16]. - The program aims to equip analysts with the necessary tools and insights to navigate the complexities of the financial markets and identify potential investment opportunities [18].

- The report introduces a quantitative model for estimating dividend points in stock indices, which is crucial for accurately assessing the premium or discount in stock index futures contracts. The model incorporates factors such as component stock weights, dividend amounts, total market capitalization, and index closing prices[38][44][46] - The model calculates the dividend points for a stock index during the period from the current date (t) to the futures contract expiration date (T) using the formula: $$ \text{Dividend Points} = \sum_{n=1}^{N} \left( \frac{\text{Dividend Amount of Stock n}}{\text{Total Market Cap of Stock n}} \times \text{Weight of Stock n} \times \text{Index Closing Price} \right) $$ This formula ensures that only stocks with ex-dividend dates between t and T are included[38][44] - Component stock weights are dynamically adjusted using the formula: $$ W_{n,t} = \frac{w_{n0} \times (1 + r_{n})}{\sum_{i=1}^{N} w_{i0} \times (1 + r_{i})} $$ Here, \( w_{n0} \) represents the weight of stock \( n \) at the last disclosed date, and \( r_{n} \) is the non-adjusted return of stock \( n \) between the last disclosed date and the current date[45] - The model estimates net profit for stocks without disclosed data by categorizing companies into stable and unstable profit distribution groups. Stable companies are predicted based on historical patterns, while unstable ones use the previous year's profit as a proxy[47][50] - Dividend payout ratios are estimated using historical averages. If a company paid dividends in the previous year, that ratio is used; otherwise, a three-year average is applied. Companies with no dividend history are assumed not to pay dividends[51][53] - Ex-dividend dates are predicted using a linear extrapolation method based on historical intervals between announcement and ex-dividend dates. Default dates are applied if historical data is insufficient or inconsistent[51][56] - The model's accuracy was validated by comparing predicted dividend points with actual values for the Shanghai 50, CSI 300, and CSI 500 indices in 2023 and 2024. The Shanghai 50 and CSI 300 predictions showed errors within 5 points, while the CSI 500 had slightly larger errors, around 10 points[57][61][66]

Quantitative Models and Construction Methods - **Model Name**: Dividend Enhanced Portfolio **Model Construction Idea**: Focuses on high-dividend stocks, aiming to capture excess returns from dividend-paying assets through a systematic approach[6][15] **Model Construction Process**: 1. Selects stocks with high dividend yields from the market. 2. Constructs two portfolios: "Central SOE High Dividend 30 Portfolio" and "Balanced Growth Dividend 50 Portfolio". 3. Applies a "steady + growth" style for the Central SOE portfolio and a "balanced offensive and defensive" style for the Dividend 50 portfolio[14][21] **Model Evaluation**: The model underperformed the benchmark this week, indicating weaker relative performance in the current market environment[21] - **Model Name**: Electronics Enhanced Portfolio **Model Construction Idea**: Focuses on the electronics sector, leveraging quantitative methods to identify outperforming sub-sectors and stocks[7][14] **Model Construction Process**: 1. Constructs two portfolios: "Electronics Balanced Allocation Enhanced Portfolio" and "Electronics Sector Preferred Enhanced Portfolio". 2. The Balanced Allocation Portfolio adopts a diversified approach across the electronics sector. 3. The Preferred Portfolio targets leading companies in mature sub-sectors of the electronics industry[14][31] **Model Evaluation**: Both portfolios outperformed the electronics sector index, demonstrating strong relative performance[31] Model Backtesting Results - **Dividend Enhanced Portfolio**: - Central SOE High Dividend 30 Portfolio: Weekly return of approximately 4.49%, underperforming the benchmark[15][21] - Balanced Growth Dividend 50 Portfolio: Weekly return of approximately 2.57%, also underperforming the benchmark[15][21] - **Electronics Enhanced Portfolio**: - Electronics Balanced Allocation Enhanced Portfolio: Weekly excess return of approximately 0.59%, ranking in the 15th percentile among active technology funds[7][31] - Electronics Sector Preferred Enhanced Portfolio: Weekly excess return of approximately 0.47%, ranking in the 17th percentile among active technology funds[7][31]

Group 1: Core Insights - The report introduces a quantitative model solution for addressing the value-growth style switching issue based on odds and win rates [1][8] - The recent performance shows that the growth style portfolio achieved a return of 2.54%, while the value style portfolio returned 2.24% [1][8] Group 2: Odds - The relative valuation levels of market styles are key factors influencing expected odds, which are negatively correlated [2][14] - The current estimated odds for the growth style is 1.11, while for the value style it is 1.09 [2][14] Group 3: Win Rates - Among seven win rate indicators, four point to growth and three to value, resulting in a current win rate of 53.87% for growth and 46.13% for value [3][16] Group 4: Investment Expectations and Strategy Returns - The investment expectation for the growth style is calculated at 0.14, while for the value style it is -0.04, leading to a recommendation for the growth style [4][18] - Since 2013, the annualized return of the style rotation model based on investment expectations is 27.62%, with a Sharpe ratio of 1.02 [4][19]

Quantitative Factors and Models Summary Quantitative Factors and Construction Process - **Factor Name**: Intraday Skewness Factor **Construction Idea**: This factor captures the skewness of intraday stock returns, reflecting the asymmetry in return distribution[13][16][18] **Construction Process**: The factor is calculated based on the third moment of intraday return distribution, normalized by the cube of standard deviation. The detailed methodology is referenced in the report "Stock Selection Factor Series Research (19) - High-Frequency Factors on Stock Return Distribution Characteristics"[13][16][18] - **Factor Name**: Downside Volatility Proportion Factor **Construction Idea**: This factor measures the proportion of downside volatility in the total realized volatility of a stock[18][19][20] **Construction Process**: The factor is derived by decomposing realized volatility into upside and downside components. The methodology is detailed in the report "Stock Selection Factor Series Research (25) - High-Frequency Factors on Realized Volatility Decomposition"[18][19][20] - **Factor Name**: Post-Open Buying Intention Proportion Factor **Construction Idea**: This factor quantifies the proportion of buying intention in the early trading period after market open[22][23][24] **Construction Process**: The factor is constructed using high-frequency data to identify and aggregate buying signals in the post-open period. The methodology is detailed in the report "Stock Selection Factor Series Research (64) - Low-Frequency Applications of High-Frequency Data Based on Intuitive Logic and Machine Learning"[22][23][24] - **Factor Name**: Post-Open Buying Intensity Factor **Construction Idea**: This factor measures the intensity of buying activity in the early trading period after market open[27][28][29] **Construction Process**: Similar to the proportion factor, this factor aggregates the magnitude of buying signals during the post-open period, normalized by trading volume[27][28][29] - **Factor Name**: Post-Open Large Order Net Buying Proportion Factor **Construction Idea**: This factor captures the proportion of large order net buying in the early trading period after market open[32][34][35] **Construction Process**: The factor is calculated by summing the net buying of large orders during the post-open period and dividing by total trading volume[32][34][35] - **Factor Name**: Post-Open Large Order Net Buying Intensity Factor **Construction Idea**: This factor measures the intensity of large order net buying in the early trading period after market open[37][39][40] **Construction Process**: The factor aggregates the net buying of large orders during the post-open period, normalized by the total number of large orders[37][39][40] - **Factor Name**: Improved Reversal Factor **Construction Idea**: This factor captures the reversal effect in stock returns, adjusted for high-frequency data characteristics[40][43][44] **Construction Process**: The factor is constructed by identifying stocks with extreme short-term returns and measuring their subsequent reversal performance[40][43][44] - **Factor Name**: Deep Learning Factor (Improved GRU(50,2)+NN(10)) **Construction Idea**: This factor leverages a deep learning model combining GRU and neural networks to predict stock returns[63][65][66] **Construction Process**: The model uses 50 GRU units and 10 neural network layers, trained on historical high-frequency data to predict short-term stock returns[63][65][66] - **Factor Name**: Deep Learning Factor (Residual Attention LSTM(48,2)+NN(10)) **Construction Idea**: This factor employs an LSTM model with residual attention mechanisms to enhance prediction accuracy[65][66][68] **Construction Process**: The model uses 48 LSTM units and 10 neural network layers, incorporating residual connections to capture long-term dependencies in high-frequency data[65][66][68] - **Factor Name**: Multi-Granularity Model Factor (5-Day Label) **Construction Idea**: This factor predicts stock returns over a 5-day horizon using a multi-granularity deep learning model[68][69][70] **Construction Process**: The model is trained using bidirectional AGRU (Attention-Gated Recurrent Unit) to capture multi-scale temporal patterns in stock data[68][69][70] - **Factor Name**: Multi-Granularity Model Factor (10-Day Label) **Construction Idea**: Similar to the 5-day label factor, this factor predicts stock returns over a 10-day horizon[69][70][71] **Construction Process**: The model uses the same AGRU architecture as the 5-day label factor but is trained with a 10-day prediction horizon[69][70][71] Factor Backtesting Results - **Intraday Skewness Factor**: - IC: 0.024 (2025), 0.019 (historical) - e^(-RankMAE): 0.327 (2025), 0.324 (historical) - Long-Short Return: 16.90% (2025 YTD), -0.66% (last week) - Long-Only Excess Return: 1.84% (2025 YTD), -0.79% (last week)[9][10][13] - **Downside Volatility Proportion Factor**: - IC: 0.020 (2025), 0.016 (historical) - e^(-RankMAE): 0.325 (2025), 0.323 (historical) - Long-Short Return: 12.93% (2025 YTD), -1.19% (last week) - Long-Only Excess Return: -0.12% (2025 YTD), -1.07% (last week)[9][10][18] - **Post-Open Buying Intention Proportion Factor**: - IC: 0.026 (2025), 0.026 (historical) - e^(-RankMAE): 0.322 (2025), 0.321 (historical) - Long-Short Return: 13.98% (2025 YTD), 0.27% (last week) - Long-Only Excess Return: 7.20% (2025 YTD), 0.28% (last week)[9][10][22] - **Post-Open Buying Intensity Factor**: - IC: 0.029 (2025), 0.030 (historical) - e^(-RankMAE): 0.327 (2025), 0.326 (historical) - Long-Short Return: 18.53% (2025 YTD), 0.05% (last week) - Long-Only Excess Return: 7.09% (2025 YTD), 0.43% (last week)[9][10][27] - **Post-Open Large Order Net Buying Proportion Factor**: - IC: 0.027 (2025), 0.036 (historical) - e^(-RankMAE): 0.319 (2025), 0.322 (historical) - Long-Short Return: 18.25% (2025 YTD), 0.31% (last week) - Long-Only Excess Return: 9.48% (2025 YTD), 0.43% (last week)[9][10][32] - **Post-Open Large Order Net Buying Intensity Factor**: - IC: 0.019 (2025), 0.025 (historical) - e^(-RankMAE): 0.318 (2025), 0.321 (historical) - Long-Short Return: 10.50% (2025 YTD), 0.31% (last week) - Long-Only Excess Return: 7.08% (2025 YTD), 0.24% (last week)[9][10][37] - **Improved Reversal Factor**: - IC: 0.025 (2025), 0.031 (historical) - e^(-RankMAE): 0.331 (2025), 0.330 (historical) - Long-Short Return: 17.44% (2025 YTD), 0.12% (last week) - Long-Only Excess Return: 6.14% (2025 YTD), 0.33% (last week)[9][10][40] - **Deep Learning Factor (Improved GRU(50,2)+NN(10))**: - IC: 0.045 (2025), 0.066 (historical) - e^(-RankMAE): 0.335 (2025), 0.336 (historical) - Long-Short Return: 28.86% (2025 YTD), 1.36% (last week) - Long-Only Excess Return: 2.19% (2025 YTD), 0.06% (last week)[9][10][63] - **Deep Learning Factor (Residual