Core Viewpoint - The report tracks the performance of a long-term stock selection strategy focusing on profitability and growth, which was initially published by the GF Financial Engineering team on August 26, 2020 [3][30]. Empirical Analysis - The backtesting period for the strategy spans from January 1, 2009, to October 31, 2025, with three rebalancing periods each year on April 30, August 31, and October 31 [5]. - The equal-weighted strategy achieved a cumulative return of 3458.94% and an annualized return of 23.55%, significantly outperforming the CSI 800 index, which had a cumulative return of 179.16% during the same period [6][31]. - The average number of stocks held in the portfolio was approximately 55, with an average market capitalization of around 14 billion [23][31]. - The strategy's annualized volatility relative to the CSI 800 index was 13.63%, with an information ratio of 1.19 [12][13]. Sector Distribution - The sectors with the highest frequency of stock selections included pharmaceuticals, chemicals, electronics, machinery, and food and beverages, while sectors like leisure services, construction, defense, steel, and non-bank financials were selected less frequently [26][31]. Market Capitalization Weighted Strategy - The market capitalization weighted strategy yielded a cumulative return of 2553.16% and an annualized return of 21.42%, with a relative annualized excess return of 13.88% compared to the CSI 800 index [14][21]. - The annualized volatility for the market capitalization weighted strategy was 14.17%, with an information ratio of 1.00 [21][22]. Summary - The report provides a comprehensive follow-up on the long-term stock selection strategy, emphasizing the importance of profitability and growth as key variables in stock selection, and highlights the strong performance of both equal-weighted and market capitalization weighted strategies [30][31].

Quantitative Models and Construction Methods - **Model Name**: Multi-cycle timing model for domestic interest rate price-volume trends **Model Construction Idea**: The model uses kernel regression algorithms to capture interest rate trend patterns, identifying support and resistance lines of interest rate data. It provides timing signals based on the shape of interest rate movements across different investment cycles [11][24][25] **Model Construction Process**: 1. **Data Input**: Utilize 5-year, 10-year, and 30-year government bond YTM data [11][24][25] 2. **Kernel Regression**: Apply kernel regression to identify support and resistance lines for interest rate trends [11][24][25] 3. **Cycle Analysis**: - Long cycle: Monthly frequency - Medium cycle: Bi-weekly frequency - Short cycle: Weekly frequency 4. **Signal Generation**: - If at least two cycles show downward breakthroughs of support lines and the trend is not upward, allocate fully to long-duration bonds - If at least two cycles show downward breakthroughs but the trend is upward, allocate 50% to medium-duration bonds and 50% to long-duration bonds - If at least two cycles show upward breakthroughs of resistance lines and the trend is not downward, allocate fully to short-duration bonds - If at least two cycles show upward breakthroughs but the trend is downward, allocate 50% to medium-duration bonds and 50% to short-duration bonds - Otherwise, allocate equally across short, medium, and long durations [24][25][29] **Model Evaluation**: The model demonstrates robust performance with high annualized returns and low drawdowns across different cycles [25][28][33] Model Backtesting Results - **5-Year YTM Model**: - Long-term annualized return: 5.5% - Maximum drawdown: 2.88% - Return-to-drawdown ratio: 1.91 - Short-term annualized return (since 2024): 2.21% - Maximum drawdown: 0.59% - Return-to-drawdown ratio: 3.74 - Long-term excess return: 1.07% - Short-term excess return: 0.87% - Historical win rate for annual absolute returns: 100% - Historical win rate for annual excess returns: 100% [25][37] - **10-Year YTM Model**: - Long-term annualized return: 6.09% - Maximum drawdown: 2.74% - Return-to-drawdown ratio: 2.22 - Short-term annualized return (since 2024): 2.64% - Maximum drawdown: 0.58% - Return-to-drawdown ratio: 4.57 - Long-term excess return: 1.65% - Short-term excess return: 1.43% - Historical win rate for annual absolute returns: 100% - Historical win rate for annual excess returns: 100% [28][37] - **30-Year YTM Model**: - Long-term annualized return: 7.37% - Maximum drawdown: 4.27% - Return-to-drawdown ratio: 1.73 - Short-term annualized return (since 2024): 3.28% - Maximum drawdown: 0.92% - Return-to-drawdown ratio: 3.59 - Long-term excess return: 2.41% - Short-term excess return: 2.68% - Historical win rate for annual absolute returns: 94.44% - Historical win rate for annual excess returns: 94.44% [33][37] Quantitative Factors and Construction Methods - **Factor Name**: Interest rate structure indicators (level, term, convexity) **Factor Construction Idea**: Transform YTM data into structural indicators to analyze the interest rate market from a mean-reversion perspective [8] **Factor Construction Process**: 1. **Level Structure**: - Formula: $ \text{Level} = \text{Average YTM across maturities} $ - Current reading: 1.61%, positioned at 21%, 12%, and 6% percentiles for 3, 5, and 10-year historical views, respectively [8] 2. **Term Structure**: - Formula: $ \text{Term} = \text{Difference between long and short maturity YTM} $ - Current reading: 0.41%, positioned at 27%, 17%, and 18% percentiles for 3, 5, and 10-year historical views, respectively [8] 3. **Convexity Structure**: - Formula: $ \text{Convexity} = \text{Second derivative of YTM curve} $ - Current reading: -0.04%, positioned at 10%, 6%, and 5% percentiles for 3, 5, and 10-year historical views, respectively [8] **Factor Evaluation**: These indicators provide a comprehensive view of the interest rate market's structural dynamics, aiding in timing and allocation decisions [8] Factor Backtesting Results - **Level Structure**: Current reading: 1.61% [8] - **Term Structure**: Current reading: 0.41% [8] - **Convexity Structure**: Current reading: -0.04% [8]

- Factor Name: EPD (Valuation Deviation Factor) - Construction Idea: Combining the Bollinger Bands mean reversion strategy commonly used in the CTA field with the logic of fundamental valuation repair, utilizing the mean reversion characteristic of the PE valuation indicator[7] - Construction Process: The EPD factor is constructed by using the mean reversion characteristic of the PE valuation indicator[7] - Evaluation: The EPD factor aims to capture valuation deviations and mean reversion in stock prices[7] - Factor Name: EPDS (Slow Deviation Factor) - Construction Idea: To eliminate the impact of changes in individual stock valuation logic, the EPD factor is used to remove the probability of individual stock valuation logic being altered (represented by the individual stock information ratio)[7] - Construction Process: The EPDS factor is constructed by using the EPD factor to remove the probability of individual stock valuation logic being altered[7] - Evaluation: The EPDS factor aims to provide a more stable measure of valuation deviations by accounting for changes in individual stock valuation logic[7] - Factor Name: EPA (Valuation Anomaly Factor) - Construction Idea: Removing the influence of Beta, growth, and value styles that affect the "valuation anomaly" logic[7] - Construction Process: The EPA factor is constructed by removing the influence of Beta, growth, and value styles from the EPD factor[7] - Evaluation: The EPA factor aims to capture valuation anomalies by eliminating the influence of common market factors[7] Factor Backtesting Results - EPD Factor - Annualized Return: 17.46%[2][8][12] - Annualized Volatility: 9.92%[2][8][12] - Information Ratio (IR): 1.76[2][8][12] - Monthly Win Rate: 70.37%[2][8][12] - Maximum Drawdown: 8.93%[2][8][12] - EPDS Factor - Annualized Return: 16.03%[2][8][12] - Annualized Volatility: 5.74%[2][8][12] - Information Ratio (IR): 2.79[2][8][12] - Monthly Win Rate: 78.31%[2][8][12] - Maximum Drawdown: 3.10%[2][8][12] - EPA Factor - Annualized Return: 17.15%[2][8][12] - Annualized Volatility: 5.16%[2][8][12] - Information Ratio (IR): 3.33[2][8][12] - Monthly Win Rate: 80.42%[2][8][12] - Maximum Drawdown: 3.12%[2][8][12]

Quantitative Models and Construction Methods - **Model Name**: "Day and Night Convergence" New Momentum Factor **Model Construction Idea**: The model is based on the price-volume relationship during intraday and overnight trading sessions. It improves traditional momentum factors by incorporating transaction volume information and separating the trading periods into day and night to explore their respective characteristics and logic[6][7] **Model Construction Process**: 1. The trading period is divided into intraday and overnight sessions 2. The price-volume relationship is analyzed separately for each session to identify distinct features 3. The improved intraday and overnight factors are synthesized into a new momentum factor 4. The factor is tested on the entire A-share market (excluding Beijing Stock Exchange stocks) from February 2014 to October 2025, using a 10-group long-short hedging strategy[7] **Model Evaluation**: The model demonstrates significant stock selection ability, outperforming traditional momentum factors in terms of stability and performance[6][7] Model Backtesting Results - **"Day and Night Convergence" New Momentum Factor**: - Annualized Return: 18.15% - Annualized Volatility: 8.68% - Information Ratio (IR): 2.09 - Monthly Win Rate: 78.01% - Maximum Drawdown: 9.07%[1][7][14] Quantitative Factors and Construction Methods - **Factor Name**: "Day and Night Convergence" New Momentum Factor **Factor Construction Idea**: The factor leverages the distinct characteristics of price-volume relationships during intraday and overnight trading sessions to enhance the signal strength of momentum factors[7] **Factor Construction Process**: 1. Separate the trading period into intraday and overnight sessions 2. Analyze the price-volume relationship for each session to identify unique features 3. Combine the improved intraday and overnight factors into a single momentum factor 4. Test the factor on the entire A-share market (excluding Beijing Stock Exchange stocks) from February 2014 to October 2025, using a 10-group long-short hedging strategy[7] **Factor Evaluation**: The factor significantly outperforms traditional momentum factors, with higher stability and better stock selection ability[6][7] Factor Backtesting Results - **"Day and Night Convergence" New Momentum Factor**: - Annualized Return: 18.15% - Annualized Volatility: 8.68% - Information Ratio (IR): 2.09 - Monthly Win Rate: 78.01% - Maximum Drawdown: 9.07%[1][7][14] - **Traditional Momentum Factor**: - Information Ratio (IR): 1.09 - Monthly Win Rate: 62.75% - Maximum Drawdown: 20.35%[6] October 2025 Performance - **"Day and Night Convergence" New Momentum Factor**: - Long Portfolio Return: 0.85% - Short Portfolio Return: -2.35% - Long-Short Hedging Return: 3.20%[1][10]

- The report focuses on the performance of various Smart beta portfolios, specifically the Value 50, Value Balanced 50, Growth 50, Growth Balanced 50, Small Cap 50, and Small Cap Balanced 50 portfolios[4][6] - The Value Balanced 50 portfolio outperformed last week with a weekly return of 2.28%, generating an excess return of 2.13% relative to the China Securities Value Index[1][4] - The Growth Balanced 50 portfolio achieved a weekly return of 1.52% last week, with an annual return of 31.07% year-to-date[4][6] - The Small Cap 50 portfolio had a negative weekly return of -0.21% last week, but it has an impressive annual return of 45.27% year-to-date[4][6] - The Small Cap Balanced 50 portfolio also had a negative weekly return of -0.50% last week, with an annual return of 41.50% year-to-date[4][6] - The report includes detailed performance metrics such as absolute returns, excess returns, and maximum relative drawdowns for each portfolio[7] - The Value 50 portfolio had a weekly return of 0.55% and an annual return of 16.22% year-to-date[4][7] - The Growth 50 portfolio had a weekly return of 0.08% and an annual return of 26.74% year-to-date[4][7] - The report provides visual representations of the weekly, monthly, and year-to-date performance of each portfolio[8][17][24]

- The "Dividend 50 Combination" strategy is designed to outperform the CSI Dividend Total Return Index by focusing on a balanced approach between growth and stability. The strategy includes stocks with high dividend yields and stable financial performance, aiming to capture excess returns in volatile markets[6][13][20] - The "Electronic Sector Preferred Enhanced Combination" strategy targets leading companies in mature sub-sectors within the electronic industry. It emphasizes stocks with strong fundamentals and growth potential, aiming to achieve positive excess returns relative to technology-themed funds[13][23][30] - The "Dividend 50 Combination" strategy achieved a weekly excess return of approximately 0.85% relative to the CSI Dividend Total Return Index, and a cumulative excess return of 7.35% since the beginning of 2025, placing it in the top 32% of all dividend-related fund products[20][22] - The "Electronic Sector Preferred Enhanced Combination" strategy delivered a weekly excess return of approximately 0.42%, outperforming the median return of active technology-themed funds during the same period[30][31]

Quantitative Models and Construction Quantitative Factors and Construction Process - **Factor Name**: Large Buy and Sell Factor **Construction Idea**: Reflects the market's active trading behavior and short-term price movement prediction[8][86][88] **Construction Process**: 1. Define "large orders" as transactions exceeding the rolling average by 1 standard deviation[8] 2. Calculate net buy amount as the difference between large buy and large sell orders[8] 3. Analyze the factor under three scenarios: full trading session, excluding the last 30 minutes, and only the first 30 minutes of trading[8] **Evaluation**: Strong short-term positive correlation with index returns due to momentum effects, but reverses over longer periods due to mean reversion[13][87][88] - **Factor Name**: Small Buy and Large Sell Factor **Construction Idea**: Captures the behavior of smaller investors and their impact on short-term market trends[8][86][88] **Construction Process**: 1. Define "large orders" as transactions exceeding the rolling average by 1 standard deviation[8] 2. Calculate the net buy amount for small buy and large sell orders[8] 3. Analyze the factor under three scenarios: full trading session, excluding the last 30 minutes, and only the first 30 minutes of trading[8] **Evaluation**: Strong short-term positive correlation with index returns due to momentum effects, but reverses over longer periods due to mean reversion[13][87][88] - **Factor Name**: Large Net Buy Factor **Construction Idea**: Represents the influence of large-scale net buying on market trends[8][86][88] **Construction Process**: 1. Define "large orders" as transactions exceeding the rolling average by 1 standard deviation[8] 2. Calculate net buy amount as the difference between large buy and large sell orders[8] 3. Analyze the factor under three scenarios: full trading session, excluding the last 30 minutes, and only the first 30 minutes of trading[8] **Evaluation**: Weak short-term negative correlation with index returns due to overbuying effects, but positive correlation over longer periods due to market support from large capital inflows[13][87][88] Optimal Parameters for Factors - **Large Buy and Sell Factor**: Optimal parameters are MA10-MA40 and MA10-MA60 for short-term and medium-term trends[9][32][88] - **Small Buy and Large Sell Factor**: Optimal parameters are MA5-MA20 and MA10-MA40 for short-term and medium-term trends[9][32][88] - **Large Net Buy Factor**: Optimal parameters are MA10-MA20 and MA10-MA40 for medium-term trends[9][32][88] --- Backtesting Results of Factors Single Factor Performance - **Large Buy and Sell Factor**: - **HS300**: Annualized return 12.2%-12.5%, Sharpe ratio 0.82-0.84, max drawdown -27.7%[36][38] - **CSI500**: Annualized return 10.6%, Sharpe ratio 0.60, max drawdown -32.0%[45] - **CSI1000**: Annualized return 11.4%, Sharpe ratio 0.64, max drawdown -45.0%[53] - **Small Buy and Large Sell Factor**: - **HS300**: Annualized return 12.5%, Sharpe ratio 0.84-0.85, max drawdown -24.4%[36][38] - **CSI500**: Annualized return 11.8%, Sharpe ratio 0.66, max drawdown -37.7%[45] - **CSI1000**: Annualized return 12.7%, Sharpe ratio 0.71, max drawdown -42.7%[53] - **Large Net Buy Factor**: - **HS300**: Annualized return 5.0%, Sharpe ratio 0.23, max drawdown -46.7%[36] - **CSI500**: Annualized return 6.8%, Sharpe ratio 0.27, max drawdown -65.2%[45] - **CSI1000**: Annualized return 5.1%, Sharpe ratio 0.22, max drawdown -57.9%[53] Composite Strategy Performance - **HS300**: - Aggressive long strategy: Annualized return 11.3%, Sharpe ratio 0.84, max drawdown -23.5%[64][66] - Conservative long strategy: Annualized return 10.1%, Sharpe ratio 0.85, max drawdown -29.9%[64][66] - Aggressive long-short strategy: Annualized return 17.2%, Sharpe ratio 0.84, max drawdown -32.1%[64][66] - Conservative long-short strategy: Annualized return 15.1%, Sharpe ratio 0.82, max drawdown -32.5%[64][66] - **CSI500**: - Aggressive long strategy: Annualized return 13.5%, Sharpe ratio 0.81, max drawdown -33.5%[69][72] - Conservative long strategy: Annualized return 16.1%, Sharpe ratio 1.21, max drawdown -15.0%[69][72] - Aggressive long-short strategy: Annualized return 16.1%, Sharpe ratio 0.69, max drawdown -53.3%[69][72] - Conservative long-short strategy: Annualized return 17.6%, Sharpe ratio 0.86, max drawdown -27.8%[69][72] - **CSI1000**: - Aggressive long strategy: Annualized return 12.1%, Sharpe ratio 0.65, max drawdown -50.3%[79][81] - Conservative long strategy: Annualized return 19.7%, Sharpe ratio 1.28, max drawdown -18.1%[79][81] - Aggressive long-short strategy: Annualized return 26.9%, Sharpe ratio 1.03, max drawdown -52.4%[79][81] - Conservative long-short strategy: Annualized return 28.5%, Sharpe ratio 1.25, max drawdown -38.8%[79][81] Annual Performance of Composite Strategies - **HS300**: Annual win rate exceeds 60%, with stable returns even during market downturns[66][67][70] - **CSI500**: Average annual win rate of 56%, higher elasticity compared to HS300, suitable for risk-tolerant strategies[75][76][77] - **CSI1000**: Annual win rate exceeds 70%, with the highest stability among all indices, especially for conservative strategies[82][83][84] --- Key Observations - Large Buy and Sell Factor and Small Buy and Large Sell Factor exhibit strong short-term positive correlation with index returns, while Large Net Buy Factor shows weak short-term negative correlation but positive long-term correlation[10][13][87] - Optimal parameters for high-frequency factors are concentrated in short-term (MA5, MA10) and medium-term (MA20, MA40) moving average distances[9][32][88] - Composite strategies outperform single-factor strategies in terms of stability and risk-adjusted returns, especially on indices with higher volatility like CSI500 and CSI1000[64][72][81] - Conservative strategies are more suitable for volatile indices, while aggressive strategies yield higher win rates on stable indices like HS300[85][89]

Group 1 - The report introduces a quantitative model solution for addressing the value-growth style switching issue, combining investment expectations based on odds and win rates [1][8] - The overall market growth style portfolio achieved a return of 4.58%, while the value style portfolio returned 2.24% in the last week [1][8] Group 2 - The estimated odds for the growth style is 1.08, while for the value style it is 1.12, indicating a negative correlation between relative valuation levels and expected odds [2][14] - The current win rate for the growth style is 63.24%, compared to 36.76% for the value style, based on seven win rate indicators [3][19] Group 3 - The latest investment expectation for the growth style is calculated to be 0.32, while the value style has an investment expectation of -0.22, leading to a recommendation for the growth style [4][21] - Since 2013, the annualized return of the style rotation model based on investment expectations is 27.99%, with a Sharpe ratio of 1.04 [4][22]

Quantitative Models and Construction Methods - **Model Name**: Net Index **Model Construction Idea**: The Net Index measures the difference between the number of long anomaly portfolios and short anomaly portfolios a stock belongs to in a given month[16][39][40] **Model Construction Process**: 1. Sort stocks monthly based on 130 anomaly characteristics derived from academic literature[38][39] 2. Define long and short ends of each anomaly strategy as the extreme quintiles from the sorting process[39] 3. Calculate the Net Index as the difference between the number of long anomaly portfolios and short anomaly portfolios a stock belongs to[39] **Formula**: $ Net_{t} = \text{Number of Long Portfolios}_{t} - \text{Number of Short Portfolios}_{t} $[39] **Model Evaluation**: The Net Index demonstrates high persistence across time and captures significant heterogeneity in extreme quintiles[40][41] Model Backtesting Results - **Net Index**: - Mean value: -1.30 - Standard deviation: 8.90 - Extreme quintile difference: 18.8[39][40][41] Quantitative Factors and Construction Methods - **Factor Name**: Retail Trading **Factor Construction Idea**: Retail trading is identified through sub-penny price improvements in transaction prices, reflecting individual investor activity[22][23][24] **Factor Construction Process**: 1. Calculate the fractional part of transaction prices: $ Z_{i t} = 100 \times mod(P_{i t}, 0.01) $ where $ P_{i t} $ is the transaction price[23] 2. Classify trades based on the fractional part and FINRA reporting codes: - Buy orders: $ Z_{i t} \in (0.6, 1) $ - Sell orders: $ Z_{i t} \in (0, 0.4) $[24] 3. Aggregate daily buy and sell proportions normalized by shares outstanding[25] **Factor Evaluation**: Retail trading reflects systematic errors by individual investors, often contrary to expected returns[19][25][27] - **Factor Name**: Short Seller Trading **Factor Construction Idea**: Short seller trading is measured by changes in short interest scaled by shares outstanding[33][34] **Factor Construction Process**: 1. Obtain monthly short interest data from stock exchanges[33] 2. Calculate short seller trading as: $ \text{Short Seller Trading} = \frac{\Delta \text{Short Interest}}{\text{Shares Outstanding}} $ where increases in short interest are negative and decreases are positive[33][34] **Factor Evaluation**: Short sellers are highly skilled in utilizing public information and aligning trades with expected returns[18][34][48] - **Factor Name**: Firm Trading **Factor Construction Idea**: Firm trading is measured by changes in shares outstanding due to issuance or repurchase, scaled by shares outstanding[35][36] **Factor Construction Process**: 1. Calculate monthly changes in shares outstanding adjusted for stock splits and dividends[35] 2. Define firm trading as: $ \text{Firm Trading} = \frac{\text{Issuance} - \text{Repurchase}}{\text{Shares Outstanding}} $ Positive values indicate net issuance, while negative values indicate net repurchase[35][36] **Factor Evaluation**: Firm trading reflects private information and aligns strongly with expected returns[16][35][48] Factor Backtesting Results - **Retail Trading**: - 1-year mean: 0.03% - 3-year mean: 0.05%[27][28] - **Short Seller Trading**: - 1-year mean: -0.18% - 3-year mean: -0.49%[34][44] - **Firm Trading**: - 1-year mean: -3.92% - 3-year mean: -11.40%[35][44] Predictive Results of Factors - **Retail Trading**: Negative correlation with future returns, indicating systematic errors by individual investors[19][66][70] - **Short Seller Trading**: Positive correlation with future returns, reflecting alignment with expected returns[18][66][70] - **Firm Trading**: Positive correlation with future returns, showcasing predictive power based on private information[16][66][70] Residual Analysis - **Retail Trading**: Residual predictive power remains significant for 3-year trading, indicating information orthogonal to anomaly variables[73][75][76] - **Short Seller Trading**: Predictive power largely explained by alignment with anomaly variables[76] - **Firm Trading**: Partial predictive power explained by anomaly alignment, with additional orthogonal information sources[76]

Core Insights - The report indicates that high-frequency factors continued to retract, while multi-granularity factors showed some rebound. The AI-enhanced portfolios have sustained a rebound, with the strictly constrained 1000 enhanced portfolio achieving a record high in excess returns [2][5]. Summary by Sections 1. High-Frequency Factors, Deep Learning Factors, and AI Enhanced Portfolio Performance Summary - The report summarizes the historical and 2025 performance of high-frequency stock selection factors, including multi-factor returns and excess returns for October and year-to-date [8]. - The high-frequency skew factor had a multi-directional return of -0.54% for the last week, -2.03% for October, and 20.66% year-to-date [10]. - The deep learning high-frequency factor (improved GRU(50,2)+NN(10)) reported a multi-directional return of 0.62% for the last week, 0.38% for October, and 43.14% year-to-date [12]. 2. Weekly Rebalancing of AI Index Enhanced Portfolios - The weekly rebalancing of the CSI 500 AI enhanced wide constraint portfolio achieved excess returns of 3.51%, 4.71%, and 4.65% for the last week, October, and year-to-date respectively [13]. - The weekly rebalancing of the CSI 1000 AI enhanced strict constraint portfolio achieved excess returns of 2.21%, 3.99%, and 17.63% for the last week, October, and year-to-date respectively [13]. 3. Performance of Specific Factors - The opening buy intention strength factor had a multi-directional return of -0.98% for the last week, -2.72% for October, and 23.09% year-to-date [10]. - The average single outflow amount factor reported a multi-directional return of -0.90% for the last week, -1.90% for October, and -2.44% year-to-date [10]. - The deep learning factor (multi-granularity model - 5-day label) achieved a multi-directional return of 2.04% for the last week, 2.53% for October, and 55.62% year-to-date [12].