Quantitative Models and Construction Methods - **Model Name**: Maximized Factor Exposure Portfolio (MFE) **Model Construction Idea**: The MFE portfolio is designed to test the effectiveness of single factors under real-world constraints, such as industry exposure, style exposure, stock weight limits, and turnover rate. This approach ensures that factors deemed "effective" can genuinely contribute to return prediction in the final portfolio[39][40]. **Model Construction Process**: The MFE portfolio is constructed using the following optimization model: $ \begin{array}{ll} max & f^{T} w \\ s.t. & s_{l} \leq X(w-w_{b}) \leq s_{h} \\ & h_{l} \leq H(w-w_{b}) \leq h_{h} \\ & w_{l} \leq w-w_{b} \leq w_{h} \\ & b_{l} \leq B_{b}w \leq b_{h} \\ & \mathbf{0} \leq w \leq l \\ & \mathbf{1}^{T} w = 1 \end{array} $ - **Objective Function**: Maximize single-factor exposure, where \( f \) represents factor values, \( f^{T}w \) is the weighted exposure of the portfolio to the factor, and \( w \) is the stock weight vector. - **Constraints**: 1. **Style Exposure**: \( X \) is the factor exposure matrix for stocks, \( w_b \) is the benchmark weight vector, and \( s_l, s_h \) are the lower and upper bounds for style factor exposure. 2. **Industry Exposure**: \( H \) is the industry exposure matrix, where \( H_{ij} = 1 \) if stock \( i \) belongs to industry \( j \), and \( h_l, h_h \) are the lower and upper bounds for industry deviation. 3. **Stock Deviation**: \( w_l, w_h \) are the lower and upper bounds for individual stock deviations from the benchmark. 4. **Constituent Weight**: \( B_b \) is a 0-1 vector indicating whether a stock is a benchmark constituent, and \( b_l, b_h \) are the lower and upper bounds for constituent weights. 5. **No Short Selling**: Ensures non-negative weights and limits individual stock weights to \( l \). 6. **Full Investment**: Ensures the portfolio is fully invested with \( \mathbf{1}^{T}w = 1 \)[39][40][41]. **Model Evaluation**: The MFE portfolio effectively tests factor performance under realistic constraints, making it a robust tool for evaluating factor predictability in practical scenarios[39][40]. --- Quantitative Factors and Construction Methods - **Factor Name**: Standardized Unexpected Earnings (SUE) **Factor Construction Idea**: Measures the deviation of actual quarterly net profit from expected profit, standardized by the standard deviation of expected profit. This factor captures earnings surprises[17]. **Factor Construction Process**: $ SUE = \frac{\text{Actual Quarterly Net Profit} - \text{Expected Quarterly Net Profit}}{\text{Standard Deviation of Expected Net Profit}} $ **Factor Evaluation**: SUE is a growth-related factor and has shown strong performance in certain market conditions, particularly in capturing earnings surprises[17]. - **Factor Name**: One-Year Momentum **Factor Construction Idea**: Measures the momentum of stock prices over the past year, excluding the most recent month, to avoid short-term reversals[17]. **Factor Construction Process**: $ \text{One-Year Momentum} = \text{Cumulative Return Over the Past Year (Excluding the Last Month)} $ **Factor Evaluation**: This factor is widely used in momentum strategies and has demonstrated consistent performance in various market environments[17]. - **Factor Name**: Three-Month Earnings Revision **Factor Construction Idea**: Tracks the net number of analyst upgrades versus downgrades over the past three months, normalized by the total number of analysts covering the stock[17]. **Factor Construction Process**: $ \text{Three-Month Earnings Revision} = \frac{\text{Number of Upgrades} - \text{Number of Downgrades}}{\text{Total Number of Analysts}} $ **Factor Evaluation**: This factor reflects changes in market sentiment and has shown strong predictive power for short-term stock performance[17]. --- Backtesting Results of Models - **MFE Portfolio Performance**: - **CSI 300 Index**: Weekly excess return: -0.14%; YTD excess return: 3.07%[5][14]. - **CSI 500 Index**: Weekly excess return: -0.27%; YTD excess return: -0.57%[5][14]. - **CSI 1000 Index**: Weekly excess return: -0.69%; YTD excess return: 3.24%[5][14]. - **CSI A500 Index**: Weekly excess return: 0.12%; YTD excess return: 3.43%[5][14]. --- Backtesting Results of Factors - **Standardized Unexpected Earnings (SUE)**: - **CSI 300 Index**: Weekly excess return: 0.31%; Monthly excess return: -0.50%; YTD excess return: 0.16%[19]. - **CSI 500 Index**: Weekly excess return: 0.77%; Monthly excess return: -0.02%; YTD excess return: 0.11%[21]. - **CSI 1000 Index**: Weekly excess return: 0.31%; Monthly excess return: 0.40%; YTD excess return: -1.04%[23]. - **CSI A500 Index**: Weekly excess return: 0.65%; Monthly excess return: -0.68%; YTD excess return: 0.46%[25]. - **One-Year Momentum**: - **CSI 300 Index**: Weekly excess return: 0.54%; Monthly excess return: 0.74%; YTD excess return: 0.36%[19]. - **CSI 500 Index**: Weekly excess return: 0.08%; Monthly excess return: -0.56%; YTD excess return: -1.95%[21]. - **CSI 1000 Index**: Weekly excess return: -0.33%; Monthly excess return: -0.12%; YTD excess return: 1.52%[23]. - **CSI A500 Index**: Weekly excess return: 0.66%; Monthly excess return: -0.96%; YTD excess return: -1.32%[25]. - **Three-Month Earnings Revision**: - **CSI 300 Index**: Weekly excess return: 0.19%; Monthly excess return: -0.47%; YTD excess return: -0.04%[19]. - **CSI 500 Index**: Weekly excess return: 1.02%; Monthly excess return: 2.06%; YTD excess return: 0.80%[21]. - **CSI 1000 Index**: Weekly excess return: 0.31%; Monthly excess return: 2.78%; YTD excess return: 3.88%[23]. - **CSI A500 Index**: Weekly excess return: 0.02%; Monthly excess return: 0.53%; YTD excess return: 0.56%[25].

Group 1 - The core viewpoint of the article emphasizes the performance of the "brokerage golden stocks" and their ability to track the performance of mixed equity funds effectively, showcasing the analytical capabilities of brokerage firms [2][9] - In January 2026, the top-performing stocks in the brokerage golden stock pool included Zhuoyi Information, Hongjing Technology, and Shenghui Integration, with significant monthly increases [3][4] - The top three brokerages in terms of returns for January 2026 were Zhongyou Securities, Shenwan Hongyuan Securities, and Hualong Securities, with monthly returns of 18.34%, 17.37%, and 16.64% respectively, outperforming the mixed equity fund index and the CSI 300 index [6][7] Group 2 - The brokerage golden stock pool showed a high allocation in the electronics (12.83%), machinery (9.44%), and basic chemicals (8.47%) sectors, while there was a relative increase in transportation, building materials, and retail sectors [26][20] - The performance of selection factors in the brokerage golden stock pool indicated that total market capitalization, quarterly net profit growth, and analyst net upgrade ratios performed well, while day-to-day returns and quarterly ROE showed weaker performance [19][15] - The brokerage golden stock performance enhancement portfolio achieved an absolute return of 11.47% from January 5 to January 30, 2026, exceeding the mixed equity fund index by 4.15% [35][30] Group 3 - The article highlights that the brokerage golden stock index has a strong correlation with the mixed equity fund index, indicating that stocks recommended by multiple analysts tend to gain higher market attention [9][21] - The brokerage golden stock pool consists of 273 A-shares after deduplication, with 41 brokerages publishing their golden stocks for February 2026 [20][23] - The article also notes that stocks with lower prior market attention can see significant increases in focus once recommended as golden stocks by analysts [23][28]

Investment Rating - The report does not explicitly state an investment rating for the industry or specific companies [3]. Core Insights - The top five sectors with the highest excess returns for long positions in January 2026 (excluding comprehensive) are Defense and Military, Communication, Agriculture, Home Appliances, and Electric Equipment & New Energy, with excess returns of 11.41%, 8.40%, 7.85%, 6.01%, and 4.98% respectively [2]. - Over the past year, the sectors with the highest average monthly excess returns (excluding comprehensive) are Real Estate, Home Appliances, Retail, Construction, and Defense and Military, with excess returns of 2.17%, 2.09%, 1.69%, 1.69%, and 1.58% respectively [2]. - The GAN_GRU factor has shown a mean Information Coefficient (IC) of 0.1107 and an annualized excess return of 22.36% from January 2019 to January 2026 [41]. - As of January 28, 2026, the latest IC for the GAN_GRU factor is 0.0003, with a one-year mean IC of 0.0553 [41]. - The top five sectors based on the recent IC performance of the GAN_GRU factor are Defense and Military, Construction, Real Estate, Banking, and Communication, with IC values of 0.3498, 0.2478, 0.2165, 0.1993, and 0.1976 respectively [41]. - The long position combination based on the GAN_GRU factor has shown the highest excess returns in the sectors of Defense and Military, Communication, Agriculture, Home Appliances, and Electric Equipment & New Energy [45]. Summary by Sections GAN_GRU Model Overview - The GAN_GRU model utilizes Generative Adversarial Networks (GAN) for processing time-series features and GRU for encoding these features into stock selection factors [4][13]. GAN_GRU Factor Performance - The GAN_GRU factor has demonstrated significant performance metrics, including a mean IC of 0.1107 and an annualized excess return of 22.36% [41]. - The recent IC rankings for various sectors indicate strong performance in Defense and Military, Construction, and Real Estate [41][45]. Long Position Combinations - The report lists the top ten stocks selected based on the GAN_GRU factor, including companies like Xinhua Insurance, Guanghong Technology, and Guangdong Expressway [50].

Investment Insights - The report explores microstructure information contained in Level 2 order data, constructing stock selection factors with certain predictive capabilities based on order hand sizes and investor activity [2][8] - The main order hand size factors exhibit robust stock selection abilities, with small orders (1 hand, 100 shares) indicating strong institutional participation, while small orders (5, 10, 15 hands) from retail investors negatively impact stock prices [2][15] - The combined order buy hand size factor from 2015 to 2025 shows a T1-T6 VWAP RankIC of 0.048, with annualized excess returns of 18.6% for long positions and 30.6% for long-short strategies [2][19] Main Order Hand Size Factors - The report constructs a comprehensive order hand size factor by synthesizing significant hand sizes, including buy and sell orders, both executed and canceled [13][14] - The order buy hand size factor has a RankIC of 0.048 from 2015 to 2025, indicating a strong correlation with stock performance, while the order sell hand size factor has a RankIC of 0.040 [22][24] Investor Type Factors - The report identifies four types of investors based on order hand sizes: institutional investors, retail investors, quantitative traders, and speculative traders, each exhibiting distinct trading behaviors [27][29] - The buy-to-sell ratio for speculative investors shows a negative correlation with future returns, indicating that higher speculative buying may lead to lower future stock performance [29][33] Investor Activity - The report introduces a dynamic monitoring system for investor activity, quantifying the participation intensity of different investor types over time [56] - The analysis of specific stocks, such as Han's Laser and Neway, reveals that institutional and quantitative investor activity significantly influences stock price movements during certain periods [56][58] Speculative Stock Pool - The report constructs a "speculative stock pool" based on abnormal order sizes, aiming to capture stocks in the accumulation phase before price increases [69][72] - The enhanced strategy, incorporating machine learning factors, shows improved performance metrics, with annualized excess returns reaching 14.7% [2][72]

Quantitative Models and Factor Analysis Quantitative Models and Construction Methods Model Name: Guosen JinGong Index Enhanced Portfolio - **Model Construction Idea**: The model aims to outperform its respective benchmarks by constructing enhanced portfolios based on multiple factors[11][12] - **Model Construction Process**: 1. **Return Prediction**: Predicting the returns of stocks within the benchmark index 2. **Risk Control**: Implementing risk control measures to manage the portfolio's risk exposure 3. **Portfolio Optimization**: Optimizing the portfolio to maximize returns while adhering to risk constraints[12] - **Model Evaluation**: The model is designed to consistently outperform its benchmarks by leveraging multiple factors[11][12] Model Backtesting Results - **Guosen JinGong Index Enhanced Portfolio**: - **CSI 300 Index Enhanced Portfolio**: Weekly excess return 0.44%, annual excess return 0.44%[5][14] - **CSI 500 Index Enhanced Portfolio**: Weekly excess return -1.80%, annual excess return -1.80%[5][14] - **CSI 1000 Index Enhanced Portfolio**: Weekly excess return -2.20%, annual excess return -2.20%[5][14] - **CSI A500 Index Enhanced Portfolio**: Weekly excess return 0.61%, annual excess return 0.61%[5][14] Quantitative Factors and Construction Methods Factor Name: Single Factor MFE (Maximized Factor Exposure) Portfolio - **Factor Construction Idea**: The factor aims to maximize the exposure to a single factor while controlling for various constraints such as industry exposure, style exposure, and stock weight deviations[40][41] - **Factor Construction Process**: 1. **Optimization Model**: The optimization model is formulated as follows: $$ \begin{array}{ll} \text{max} & f^{T} w \\ \text{s.t.} & s_{l} \leq X(w - w_{b}) \leq s_{h} \\ & h_{l} \leq H(w - w_{b}) \leq h_{h} \\ & w_{l} \leq w - w_{b} \leq w_{h} \\ & b_{l} \leq B_{b} w \leq b_{h} \\ & \mathbf{0} \leq w \leq l \\ & \mathbf{1}^{T} w = 1 \end{array} $$ where \( f \) represents the factor values, \( w \) is the stock weight vector, and the constraints include style exposure, industry exposure, stock weight deviations, and component stock weight limits[40][41] 2. **Constraints**: The constraints include: - **Style Exposure**: \( X \) is the factor exposure matrix, \( w_{b} \) is the benchmark weight vector, \( s_{l} \) and \( s_{h} \) are the lower and upper bounds for style exposure[41] - **Industry Exposure**: \( H \) is the industry exposure matrix, \( h_{l} \) and \( h_{h} \) are the lower and upper bounds for industry exposure[41] - **Stock Weight Deviations**: \( w_{l} \) and \( w_{h} \) are the lower and upper bounds for stock weight deviations[41] - **Component Stock Weight Limits**: \( B_{b} \) is the 0-1 vector indicating whether a stock is a benchmark component, \( b_{l} \) and \( b_{h} \) are the lower and upper bounds for component stock weights[41] - **No Short Selling**: The weights are non-negative and sum to 1[41] 3. **Portfolio Construction**: The MFE portfolio is constructed by maximizing the factor exposure while adhering to the constraints[42][44] - **Factor Evaluation**: The MFE portfolio is used to test the effectiveness of single factors under realistic constraints, making it more likely to reflect the true predictive power of the factors in the final portfolio[40][41] Factor Backtesting Results - **CSI 300 Index**: - **Best Performing Factors (Weekly)**: Three-month institutional coverage (0.86%), DELTAROA (0.61%), DELTAROE (0.52%)[19] - **Worst Performing Factors (Weekly)**: Expected net profit QoQ (-0.78%), one-year momentum (-0.45%), idiosyncratic volatility (-0.42%)[19] - **CSI 500 Index**: - **Best Performing Factors (Weekly)**: Single-quarter net profit YoY growth (0.06%), expected net profit QoQ (0.33%), idiosyncratic volatility (0.22%)[21] - **Worst Performing Factors (Weekly)**: One-month volatility (-2.47%), EPTTM (-3.56%), single-quarter ROE (-0.67%)[21] - **CSI 1000 Index**: - **Best Performing Factors (Weekly)**: One-year momentum (1.94%), single-quarter revenue YoY growth (1.31%), standardized unexpected income (0.92%)[23] - **Worst Performing Factors (Weekly)**: EPTTM (-3.56%), dividend yield (-3.27%), expected EPTTM (-3.22%)[23] - **CSI A500 Index**: - **Best Performing Factors (Weekly)**: Single-quarter net profit YoY growth (1.14%), DELTAROE (0.88%), single-quarter operating profit YoY growth (0.70%)[25] - **Worst Performing Factors (Weekly)**: EPTTM (-1.29%), one-month volatility (-1.22%), three-month volatility (-1.09%)[25] - **Public Fund Heavy Index**: - **Best Performing Factors (Weekly)**: Single-quarter net profit YoY growth (1.14%), expected net profit QoQ (0.88%), three-month reversal (0.29%)[27] - **Worst Performing Factors (Weekly)**: Expected EPTTM (-0.74%), EPTTM (-1.29%), one-month volatility (-1.22%)[27]

- The report focuses on the performance of stock selection factors in 2025, highlighting the effectiveness of factors such as transaction count, liquidity, crowding, price stability, and reversal in stock selection across the market [1][5][15] - Factors are categorized into two main groups: volume-price factors and growth factors. Volume-price factors are further divided into two representative categories: price stability and reversal, while liquidity, crowding, and transaction count serve as average representatives of volume-price factors [6][24] - The construction of major factors involves market capitalization and industry neutrality, outlier removal, and standardization, followed by equal-weight synthesis into major factors [13] - Sub-factors are detailed with their calculation methods, such as residual volatility derived from the Fama-French three-factor model regression residual volatility, turnover rate variation coefficient calculated as turnover rate divided by the standard deviation over the mean, and entropy of transaction volume proportion using the entropy formula [13] - The report provides statistical data on the performance of major factors, including IC, ICIR, excess returns, maximum drawdowns, IR, long-short returns, long-short maximum drawdowns, and long-short Sharpe ratios. For example, liquidity factor achieved an IC of 9.72%, ICIR of 1.08, excess return of 23.67%, and IR of 3.43 [15][16] - Sub-factors with notable performance include short-term reversal (IC 6.27%, ICIR 1.21, excess return 4.86%), residual volatility (IC 9.42%, ICIR 1.22, excess return 1.53%), and turnover rate (IC 10.75%, ICIR 1.29, excess return 17.46%) [17] - The report highlights the time-series performance of factors, noting that the main profit periods for all factors were concentrated between January and November 2025, with significant drawdowns occurring between September and December 2025. Growth, SUE, and price stability factors had lower profit levels and higher drawdowns, while liquidity factors had higher profit levels and higher drawdowns [19][20][23] - The correlation analysis of excess returns among factors shows that price stability has a high correlation with other factors, while reversal has a low correlation with other factors. Liquidity, crowding, and transaction count factors exhibit low mutual correlation [21][24]

Quantitative Factors and Construction Methods - **Factor Name**: TPS (Turn20 conformed by PLUS) **Factor Construction Idea**: The TPS factor is designed to improve the traditional turnover rate factor by incorporating price information, specifically the shadow difference (representing intraday market sentiment), to address the limitations of the traditional turnover rate factor[6][9] **Factor Construction Process**: 1. The traditional turnover rate factor (Turn20) is calculated as the average turnover rate over the past 20 trading days, followed by cross-sectional market capitalization neutralization[6] 2. The TPS factor integrates the shadow difference as a price factor to complement the Turn20 factor, leveraging intraday price movements (e.g., opening and closing prices) to better capture market sentiment[9] **Factor Evaluation**: The TPS factor significantly outperforms traditional turnover rate factors in terms of performance and maintains strong stock selection ability even after removing common style and industry effects[6][9] - **Factor Name**: SPS (STR conformed by PLUS) **Factor Construction Idea**: Similar to TPS, the SPS factor improves the STR (Stable Turnover Rate) factor by incorporating price information (shadow difference) to enhance its effectiveness in stock selection[9] **Factor Construction Process**: 1. The STR factor was initially developed to address the limitations of the Turn20 factor, focusing on turnover rate stability rather than magnitude[7] 2. The SPS factor further integrates the shadow difference as a price factor to complement the STR factor, using intraday price movements to better reflect market sentiment[9] **Factor Evaluation**: The SPS factor demonstrates superior performance compared to both the traditional Turn20 and STR factors, with strong stock selection capabilities even after removing style and industry biases[9] Factor Backtesting Results - **TPS Factor**: - Annualized Return: 39.30% - Annualized Volatility: 15.71% - IR: 2.50 - Monthly Win Rate: 77.64% - Maximum Drawdown: 18.19%[1][11] - **SPS Factor**: - Annualized Return: 42.98% - Annualized Volatility: 13.15% - IR: 3.27 - Monthly Win Rate: 83.54% - Maximum Drawdown: 11.58%[1][12][14]

Quantitative Models and Construction Methods 1. Model Name: Maximized Factor Exposure Portfolio (MFE) - **Model Construction Idea**: The MFE portfolio is designed to test the effectiveness of single factors under real-world constraints, such as industry exposure, style exposure, stock weight limits, and turnover constraints. This approach ensures that factors deemed effective can genuinely contribute to return prediction in the final portfolio[41][42]. - **Model Construction Process**: - The optimization model aims to maximize single-factor exposure while adhering to various constraints: $$ \begin{array}{ll} \text{max} & f^{T}w \\ \text{s.t.} & s_{l} \leq X(w-w_{b}) \leq s_{h} \\ & h_{l} \leq H(w-w_{b}) \leq h_{h} \\ & w_{l} \leq w-w_{b} \leq w_{h} \\ & b_{l} \leq B_{b}w \leq b_{h} \\ & \mathbf{0} \leq w \leq l \\ & \mathbf{1}^{T}w = 1 \end{array} $$ - **Objective Function**: Maximize single-factor exposure, where \( f \) represents factor values, \( f^{T}w \) is the weighted exposure of the portfolio to the factor, and \( w \) is the stock weight vector[42]. - **Constraints**: - **Style Exposure**: \( X \) is the factor exposure matrix for stocks, \( w_b \) is the benchmark weight vector, and \( s_l, s_h \) are the lower and upper bounds for style factor exposure[42]. - **Industry Exposure**: \( H \) is the industry exposure matrix, and \( h_l, h_h \) are the lower and upper bounds for industry deviations[42]. - **Stock Weight Deviation**: \( w_l, w_h \) are the lower and upper bounds for individual stock weight deviations from the benchmark[42]. - **Constituent Stock Weight**: \( B_b \) is a 0-1 vector indicating whether a stock is a benchmark constituent, and \( b_l, b_h \) are the lower and upper bounds for constituent stock weights[42]. - **No Short Selling**: Ensures non-negative weights and limits individual stock weights to \( l \)[42]. - **Full Investment**: Ensures the portfolio is fully invested with \( \mathbf{1}^{T}w = 1 \)[43]. - The MFE portfolio is constructed monthly, and historical returns are calculated after accounting for transaction costs of 0.3% on both sides[45]. - **Model Evaluation**: The MFE portfolio effectively tests factor performance under realistic constraints, making it a robust tool for evaluating factor predictability in practical scenarios[41][42]. --- Factor Construction and Methods 1. Factor Name: Momentum (1-Year Momentum) - **Factor Construction Idea**: Measures the momentum effect by capturing the price trend over the past year, excluding the most recent month[18]. - **Factor Construction Process**: - Formula: \( \text{1-Year Momentum} = \text{Cumulative Return over the past 12 months (excluding the last month)} \)[18]. - **Factor Evaluation**: Momentum factors generally perform well in capturing price trends, as evidenced by their positive performance in multiple sample spaces[20][22][24]. 2. Factor Name: DELTAROE - **Factor Construction Idea**: Measures the change in return on equity (ROE) compared to the same quarter in the previous year, reflecting profitability improvement[18]. - **Factor Construction Process**: - Formula: \( \text{DELTAROE} = \text{Current Quarter ROE} - \text{ROE of the Same Quarter Last Year} \)[18]. - **Factor Evaluation**: DELTAROE is effective in identifying companies with improving profitability, as shown by its strong performance in various sample spaces[22][24][26]. 3. Factor Name: Standardized Unexpected Earnings (SUE) - **Factor Construction Idea**: Measures the deviation of actual earnings from expected earnings, standardized by the standard deviation of expected earnings, to capture earnings surprises[18]. - **Factor Construction Process**: - Formula: \( \text{SUE} = \frac{\text{Actual Quarterly Net Profit} - \text{Expected Net Profit}}{\text{Standard Deviation of Expected Net Profit}} \)[18]. - **Factor Evaluation**: SUE is a reliable indicator of earnings surprises and is particularly effective in growth-oriented sample spaces[18][24]. --- Factor Backtesting Results 1. 1-Year Momentum - **Performance in Different Sample Spaces**: - **CSI 300**: Positive performance in the past week but underperformed in the past month and year-to-date[20]. - **CSI 500**: Strong performance in the past week and year-to-date, with weaker results in the past month[22]. - **CSI 1000**: Underperformed year-to-date but showed strong weekly performance[24]. - **CSI A500**: Mixed results, with strong weekly performance but weaker year-to-date performance[26]. - **Public Fund Heavyweight Index**: Positive weekly performance but underperformed year-to-date[28]. 2. DELTAROE - **Performance in Different Sample Spaces**: - **CSI 300**: Strong year-to-date performance, with mixed results in the past week and month[20]. - **CSI 500**: Positive weekly and year-to-date performance, with weaker results in the past month[22]. - **CSI 1000**: Strong weekly and year-to-date performance, with weaker results in the past month[24]. - **CSI A500**: Positive weekly and year-to-date performance, with weaker results in the past month[26]. - **Public Fund Heavyweight Index**: Positive weekly and year-to-date performance, with weaker results in the past month[28]. 3. SUE - **Performance in Different Sample Spaces**: - **CSI 300**: Not explicitly mentioned in the report[18]. - **CSI 500**: Not explicitly mentioned in the report[18]. - **CSI 1000**: Not explicitly mentioned in the report[18]. - **CSI A500**: Not explicitly mentioned in the report[18]. - **Public Fund Heavyweight Index**: Not explicitly mentioned in the report[18]. --- Quantitative Model Backtesting Results 1. MFE Portfolio - **Performance in Different Sample Spaces**: - **CSI 300**: Weekly excess return of 0.64%, year-to-date excess return of 17.85%[15]. - **CSI 500**: Weekly excess return of 0.00%, year-to-date excess return of 7.07%[15]. - **CSI 1000**: Weekly excess return of 0.21%, year-to-date excess return of 14.89%[15]. - **CSI A500**: Weekly excess return of 0.44%, year-to-date excess return of 8.26%[15].

Research Background - The stock market exhibits overnight correlation characteristics, where daily returns can be decomposed into overnight and intraday returns. This report characterizes the correlation features of similar stocks based on recent academic findings [1][9]. Overnight Price Change Correlation Research - The study separates long and short signals from trading execution to capture cross-stock information effects. A correlation matrix is constructed based on overnight and intraday returns, identifying leading (Leader) and lagging (Lagger) groups. Trading strategies are developed to generate signals only from the leading group and trade within the lagging group [2][10][16]. Empirical Research - The analysis shows that the leading-lagging effect in A-shares presents a reversal effect, where a bullish signal from the leading group results in stronger performance from the short positions, and vice versa. The strategy is particularly applicable to small-cap stocks [2][35][44]. Factor Research - Weekly and monthly stock selection factors are constructed based on overnight correlation information. The introduction of conventional correlation improves the distinction of stock selection, with the combined factor showing a monthly RANK_IC of 8.13% and an annualized return of 18.2% [2][57][79]. Correlation Analysis - The internal correlation among factors is relatively low, indicating that the correlation factors provide marginal incremental value. The correlation factor shows some similarity with style factors, such as residual volatility [2][90]. Group Identification - The report attempts to identify groups within the A-share market, including the CSI 300 and the CSI 1000. The results indicate that the method of classifying leading and lagging groups based on correlation matrix features yields stable results [30][34]. Portfolio Construction Process - The portfolio construction framework separates signal generation from execution, capturing cross-stock information effects. The process includes constructing a correlation matrix, identifying leading and lagging groups, and extracting trading signals based on the leading group's average impact score [27][35]. Factor Construction and Backtesting - The report explores the performance of factors based on overnight correlation, with results indicating that conventional correlation factors outperform overnight correlation factors in terms of predictive effectiveness [57][72]. Performance Metrics - The backtesting results show that the strategy can achieve an annualized return of approximately 10.51% when focusing on small-cap stocks, while the distinction between long and short groups is less pronounced in large-cap stocks [44][72].

Quantitative Models and Construction Methods 1. Model Name: Guosen Quantitative Index Enhanced Portfolio - **Model Construction Idea**: The model aims to construct enhanced portfolios benchmarked against indices such as CSI 300, CSI 500, CSI 1000, and CSI A500, with the goal of consistently outperforming their respective benchmarks [10][11]. - **Model Construction Process**: 1. **Revenue Prediction**: Predict stock returns using multiple factors. 2. **Risk Control**: Apply constraints on industry exposure, style exposure, stock weight deviation, and turnover rate. 3. **Portfolio Optimization**: Optimize the portfolio to maximize single-factor exposure while adhering to constraints. The optimization model is as follows: $ \begin{array}{ll} max & f^{T} w \\ s.t. & s_{l} \leq X(w-w_{b}) \leq s_{h} \\ & h_{l} \leq H(w-w_{b}) \leq h_{h} \\ & w_{l} \leq w-w_{b} \leq w_{h} \\ & b_{l} \leq B_{b}w \leq b_{h} \\ & \mathbf{0} \leq w \leq l \\ & \mathbf{1}^{T} w = 1 \end{array} $ - **Objective Function**: Maximize single-factor exposure, where \( f \) represents factor values, \( w \) is the stock weight vector, and \( f^{T}w \) is the weighted exposure to the factor. - **Constraints**: - **Style Exposure**: \( X \) is the factor exposure matrix, \( w_b \) is the benchmark weight vector, and \( s_l, s_h \) are the lower and upper bounds for style exposure. - **Industry Exposure**: \( H \) is the industry exposure matrix, and \( h_l, h_h \) are the lower and upper bounds for industry deviation. - **Stock Weight Deviation**: \( w_l, w_h \) are the lower and upper bounds for stock weight deviation. - **Component Stock Weight**: \( B_b \) is a 0-1 vector indicating whether a stock is a benchmark component, and \( b_l, b_h \) are the lower and upper bounds for component stock weight. - **No Short Selling**: Ensure non-negative weights and limit individual stock weights. - **Full Investment**: Ensure the portfolio is fully invested with weights summing to 1 [40][41][42]. 4. **Backtesting**: Rebalance the portfolio monthly, calculate historical returns, and evaluate performance metrics such as excess returns and risk statistics [44]. 2. Model Name: Public Fund Heavyweight Index - **Model Construction Idea**: Construct an index based on the holdings of public funds to evaluate factor performance under "institutional style" [42][43]. - **Model Construction Process**: 1. **Sample Selection**: Include ordinary equity funds and partial equity hybrid funds with a minimum size of 50 million RMB and at least six months of listing history. Exclude recently transformed funds or those with insufficient data. 2. **Data Collection**: Use fund periodic reports (annual, semi-annual, or quarterly) to gather holding information. 3. **Weight Calculation**: Average the stock weights across eligible funds. 4. **Index Construction**: Sort stocks by weight in descending order and select those accounting for 90% of cumulative weight to form the index [43]. --- Model Backtesting Results 1. Guosen Quantitative Index Enhanced Portfolio - **CSI 300 Enhanced Portfolio**: - Weekly excess return: -0.71% - Year-to-date excess return: 16.74% [13] - **CSI 500 Enhanced Portfolio**: - Weekly excess return: 0.12% - Year-to-date excess return: 6.85% [13] - **CSI 1000 Enhanced Portfolio**: - Weekly excess return: -0.94% - Year-to-date excess return: 14.08% [13] - **CSI A500 Enhanced Portfolio**: - Weekly excess return: -1.37% - Year-to-date excess return: 7.55% [13] 2. Public Fund Heavyweight Index - **CSI 300 Index Enhanced Products**: - Weekly excess return: Max 0.70%, Min -1.26%, Median 0.09% - Year-to-date excess return: Max 9.92%, Min -4.53%, Median 2.58% [31] - **CSI 500 Index Enhanced Products**: - Weekly excess return: Max 1.17%, Min -1.13%, Median 0.11% - Year-to-date excess return: Max 13.14%, Min -9.17%, Median 3.94% [33] - **CSI 1000 Index Enhanced Products**: - Weekly excess return: Max 0.89%, Min -1.38%, Median -0.05% - Year-to-date excess return: Max 19.12%, Min -1.84%, Median 8.24% [36] - **CSI A500 Index Enhanced Products**: - Weekly excess return: Max 0.71%, Min -0.86%, Median -0.04% - Year-to-date excess return: Max 2.67%, Min -4.14%, Median -0.76% [39] --- Quantitative Factors and Construction Methods 1. Factor Name: Maximized Factor Exposure (MFE) - **Factor Construction Idea**: Evaluate factor effectiveness under real-world constraints by maximizing single-factor exposure in a portfolio [40][41]. - **Factor Construction Process**: 1. Define constraints for style exposure, industry exposure, stock weight deviation, and component stock weight. 2. Optimize the portfolio to maximize single-factor exposure while adhering to constraints. 3. Rebalance monthly and calculate historical returns [40][41][44]. 2. Factor Name: Public Fund Heavyweight Factors - **Factor Construction Idea**: Test factor performance in the public fund heavyweight index to reflect institutional preferences [42][43]. - **Factor Construction Process**: 1. Use public fund holdings to construct the index. 2. Evaluate factor performance within this index using metrics such as excess returns and risk-adjusted returns [42][43]. --- Factor Backtesting Results 1. Maximized Factor Exposure (MFE) - **CSI 300 Sample Space**: - Best-performing factors (weekly): One-month volatility (0.83%), one-month turnover (0.68%), three-month volatility (0.65%) - Worst-performing factors (weekly): Single-quarter profit growth (-0.26%), three-month institutional coverage (-0.24%), one-year momentum (-0.24%) [18] - **CSI 500 Sample Space**: - Best-performing factors (weekly): Three-month institutional coverage (1.09%), one-month reversal (1.01%), three-month reversal (0.99%) - Worst-performing factors (weekly): Standardized unexpected earnings (-1.00%), DELTAROA (-0.81%), DELTAROE (-0.81%) [20] - **CSI 1000 Sample Space**: - Best-performing factors (weekly): One-month turnover (1.08%), three-month institutional coverage (1.06%), single-quarter ROA (1.04%) - Worst-performing factors (weekly): Single-quarter SP (-1.29%), expected PEG (-1.25%), SPTTM (-1.22%) [22] - **CSI A500 Sample Space**: - Best-performing factors (weekly): One-month turnover (0.82%), three-month turnover (0.75%), one-month volatility (0.74%) - Worst-performing factors (weekly): Expected net profit QoQ (-0.91%), single-quarter net profit growth (-0.61%), expected PEG (-0.41%) [24] - **Public Fund Heavyweight Index**: - Best-performing factors (weekly): One-month volatility (1.32%), one-month turnover (1.23%), three-month turnover (0.89%) - Worst-performing factors (weekly): Single-quarter revenue growth (-0.89%), single-quarter profit growth (-0.88%), single-quarter ROE (-0.81%) [26]