- The reversal factor performed relatively well, with the GARP portfolio achieving a weekly return of 3.28% from August 1, 2025, to August 8, 2025[1] - The cumulative return of the GARP portfolio in 2025 was 28.2%[1] - The PB-profit combination had a weekly return of 2.86%, with a cumulative return of 20.53% in 2025[5][9] - The small-cap growth portfolio had a weekly return of 4.87%, with a cumulative return of 56.37% in 2025[5][9] - The small-cap value preferred portfolio 1 had a weekly return of 3.67%, with a cumulative return of 48.10% in 2025[5][9] - The small-cap value preferred portfolio 2 had a weekly return of 5.00%, with a cumulative return of 56.61% in 2025[5][9] - The performance of the multi-factor portfolios showed that the aggressive portfolio and the balanced portfolio had weekly returns of 3.37% and 3.19%, respectively[10][11] - The aggressive portfolio and the balanced portfolio had cumulative returns of 61.10% and 49.08% in 2025, respectively[11] - The enhanced CSI 300 portfolio had a weekly return of 1.43%, with a cumulative return of 11.18% in 2025[14][15] - The enhanced CSI 500 portfolio had a weekly return of 2.17%, with a cumulative return of 14.96% in 2025[14][15] - The enhanced CSI 1000 portfolio had a weekly return of 2.01%, with a cumulative return of 22.07% in 2025[14][15] - The performance of the style factors showed that small-cap stocks outperformed large-cap stocks, and high-valuation stocks outperformed low-valuation stocks[5][43] - The performance of the technical factors showed that the reversal factor contributed positive returns, with a weekly long-short return of 0.98%[5][46][48] - The performance of the fundamental factors showed that the SUE factor and the expected net profit adjustment factor contributed positive returns, with weekly long-short returns of 0.51% and 0.34%, respectively[5][50][52]

Core Viewpoint - The article emphasizes the potential of smart beta index funds, which utilize more sophisticated stock selection rules compared to traditional index funds, to achieve long-term excess returns in the market [3][4][11]. Group 1: Smart Beta Index Funds - Smart beta index funds represent a small portion of the market, with only 1.7 trillion yuan, accounting for approximately 0.5% of the total public fund size of 32.24 trillion yuan in China by the end of 2024 [2]. - These funds employ stock selection based on proven financial metrics or price characteristics, rather than just market capitalization [4][5]. - Common factors used in smart beta strategies include dividend yield, quality, value, low volatility, and momentum [15]. Group 2: Performance of Smart Beta Strategies - Historical data from 1927 to 2023 indicates that smart beta strategies can outperform the market, with various factors showing significant annualized returns above the overall market return of 9.5% [17][18]. - The long-term performance of factor-based strategies demonstrates that almost all factor long portfolios yield returns significantly higher than the market index, suggesting that holding a good smart beta fund is likely to provide better returns than traditional indices like the CSI 300 [20][23]. Group 3: Challenges and Considerations - Despite the effectiveness of smart beta strategies, they can experience prolonged periods of underperformance, which may lead to investor skepticism [24][26]. - Historical data shows that some factors can have long periods of underperformance, with the longest being four years for several factors [28][29]. - Diversifying across multiple factors can help mitigate risks associated with individual factor underperformance, as different factors may perform well at different times [30]. Group 4: Insights from Historical Data - Long-term data supports the reliability of smart beta index funds, indicating that missing out on these investment opportunities could be regrettable [32]. - Investors are advised to construct multi-factor portfolios to balance risk and return, incorporating defensive and aggressive strategies [35]. - A long-term investment horizon is essential for realizing the excess returns from smart beta strategies, as they may require enduring periods of underperformance [37][39]. - Risk management is crucial, as smart beta funds are still subject to market fluctuations and can decline during bear markets [40][41].

Quantitative Models and Construction Methods - **Model Name**: GRU **Model Construction Idea**: GRU is used for industry rotation and stock selection based on historical data and market trends[3][5][7] **Model Construction Process**: GRU utilizes gated recurrent units to process sequential data, capturing temporal dependencies in stock price movements and industry performance. It incorporates multiple factors such as momentum, volatility, and valuation metrics to predict future trends[3][5][7] **Model Evaluation**: GRU demonstrates strong performance in multi-factor combinations and industry rotation strategies, with notable differentiation across different stock pools[3][5][7] - **Model Name**: Barra **Model Construction Idea**: Barra focuses on style factors to explain stock returns and risks[14][15][16] **Model Construction Process**: Barra includes multiple style factors such as Beta, Size, Momentum, Volatility, Non-linear Size, Valuation, Liquidity, Profitability, Growth, and Leverage. Each factor is calculated using specific formulas: - **Beta**: Historical beta - **Size**: Natural logarithm of total market capitalization - **Momentum**: Mean of historical excess return series - **Volatility**: $0.74 \times \text{historical excess return volatility} + 0.16 \times \text{cumulative excess return deviation} + 0.1 \times \text{historical residual return volatility}$ - **Non-linear Size**: Cubic transformation of market capitalization - **Valuation**: Reciprocal of price-to-book ratio - **Liquidity**: $0.35 \times \text{monthly turnover rate} + 0.35 \times \text{quarterly turnover rate} + 0.3 \times \text{annual turnover rate}$ - **Profitability**: Weighted combination of analyst forecast earnings-price ratio, reciprocal of cash flow ratio, reciprocal of trailing twelve-month P/E ratio, and forecasted growth rates - **Growth**: Weighted combination of earnings growth rate and revenue growth rate - **Leverage**: Weighted combination of market leverage, book leverage, and debt-to-asset ratio[15] **Model Evaluation**: Barra style factors provide a comprehensive framework for analyzing stock returns, with strong differentiation in multi-factor strategies[14][15][16] Model Backtesting Results - **GRU Model**: - **open1d**: Weekly excess return 0.61%, monthly 1.56%, yearly 7.78% - **close1d**: Weekly excess return 0.02%, monthly 1.45%, yearly 7.28% - **barra1d**: Weekly excess return -0.24%, monthly -0.07%, yearly 3.61% - **barra5d**: Weekly excess return 0.06%, monthly 1.35%, yearly 8.63% - **Multi-factor combination**: Weekly excess return 0.61%, monthly 0.82%, yearly 3.22%[31][32][33] Quantitative Factors and Construction Methods - **Factor Name**: Beta **Factor Construction Idea**: Measures historical sensitivity to market movements[15] **Factor Construction Process**: Calculated as historical beta using regression analysis of stock returns against market returns[15] - **Factor Name**: Size **Factor Construction Idea**: Captures the impact of market capitalization on stock returns[15] **Factor Construction Process**: Natural logarithm of total market capitalization[15] - **Factor Name**: Momentum **Factor Construction Idea**: Reflects the persistence of stock price trends[15] **Factor Construction Process**: Mean of historical excess return series[15] - **Factor Name**: Volatility **Factor Construction Idea**: Measures risk associated with stock price fluctuations[15] **Factor Construction Process**: $0.74 \times \text{historical excess return volatility} + 0.16 \times \text{cumulative excess return deviation} + 0.1 \times \text{historical residual return volatility}$[15] - **Factor Name**: Non-linear Size **Factor Construction Idea**: Captures non-linear effects of market capitalization on returns[15] **Factor Construction Process**: Cubic transformation of market capitalization[15] - **Factor Name**: Valuation **Factor Construction Idea**: Reflects the relative attractiveness of stock prices[15] **Factor Construction Process**: Reciprocal of price-to-book ratio[15] - **Factor Name**: Liquidity **Factor Construction Idea**: Measures ease of trading stocks[15] **Factor Construction Process**: $0.35 \times \text{monthly turnover rate} + 0.35 \times \text{quarterly turnover rate} + 0.3 \times \text{annual turnover rate}$[15] - **Factor Name**: Profitability **Factor Construction Idea**: Captures earnings quality and growth potential[15] **Factor Construction Process**: Weighted combination of analyst forecast earnings-price ratio, reciprocal of cash flow ratio, reciprocal of trailing twelve-month P/E ratio, and forecasted growth rates[15] - **Factor Name**: Growth **Factor Construction Idea**: Reflects revenue and earnings growth trends[15] **Factor Construction Process**: Weighted combination of earnings growth rate and revenue growth rate[15] - **Factor Name**: Leverage **Factor Construction Idea**: Measures financial risk associated with debt levels[15] **Factor Construction Process**: Weighted combination of market leverage, book leverage, and debt-to-asset ratio[15] Factor Backtesting Results - **Beta**: Weekly excess return -0.24%, monthly -0.07%, yearly 3.61%[31][32][33] - **Size**: Weekly excess return 0.02%, monthly 1.45%, yearly 7.28%[31][32][33] - **Momentum**: Weekly excess return 0.61%, monthly 1.56%, yearly 7.78%[31][32][33] - **Volatility**: Weekly excess return 0.06%, monthly 1.35%, yearly 8.63%[31][32][33] - **Non-linear Size**: Weekly excess return 0.61%, monthly 0.82%, yearly 3.22%[31][32][33] - **Valuation**: Weekly excess return 0.61%, monthly 0.82%, yearly 3.22%[31][32][33] - **Liquidity**: Weekly excess return 0.61%, monthly 0.82%, yearly 3.22%[31][32][33] - **Profitability**: Weekly excess return 0.61%, monthly 0.82%, yearly 3.22%[31][32][33] - **Growth**: Weekly excess return 0.61%, monthly 0.82%, yearly 3.22%[31][32][33] - **Leverage**: Weekly excess return 0.61%, monthly 0.82%, yearly 3.22%[31][32][33]

证券研究报告：金融工程报告 研究所 分析师:肖承志 SAC 登记编号:S1340524090001 Email:xiaochengzhi@cnpsec.com 研究助理:金晓杰 SAC 登记编号:S1340124100010 Email:jinxiaojie@cnpsec.com 2025.06.16 《结合基本面和量价特征的 GRU 模 型》 - 2025.06.05 近期研究报告 《稳定币应用场景及行业研究》 - 2025.07.18 《Grok 4 发布 ，通 义 开源 智能 体 WebSailor——AI 动态汇总 20250714》 - 2025.07.16 《beta 风格显著，高波占优——中邮 因子周报 20250629》 - 2025.06.30 《反转风格显著，小市值回撤——中 邮因子周报 20250622》 - 2025.06.23 《关注基本面支撑，高波风格占优— —中邮因子周报 20250615》 - 《Claude 4 系列发布，谷歌上线编程 《证监会修改《重组办法》，深化并购 重组改革——微盘股指数周报 20250518》 - 2025.05.19 《通义千问发布 Qwen-3 模 ...

证券研究报告：金融工程报告 发布时间：2025-06-30 研究所 分析师:肖承志 SAC 登记编号:S1340524090001 Email:xiaochengzhi@cnpsec.com 研究助理:金晓杰 SAC 登记编号:S1340124100010 Email:jinxiaojie@cnpsec.com 近期研究报告 《基于相对强弱视角下的扩散指数择 时模型》 - 2025.06.25 《调整仍不充分——微盘股指数周报 20250622》 - 2025.06.23 《短期上涨动能枯竭，控制仓位做好 防御——微盘股指数周报 20250615》 - 2025.06.16 《为何微盘股基金仓位下降指数却不 断新高？——微盘股指数周报 20250608》 - 2025.06.09 《小盘股成交占比高意味着拥挤度高 吗？——微盘股指数周报 20250601》 - 2025.06.02 《微盘股容易被忽略的"看空成本" ——微盘股指数周报 20250525》 - 2025.05.26 《证监会修改《重组办法》，深化并购 重组改革——微盘股指数周报 20250518》 - 2025.05.19 《微盘股会涨到什么时 ...

Quantitative Models and Construction 1. Model Name: GRU Model - **Model Construction Idea**: The GRU model integrates fundamental and technical features to predict stock performance[3][19] - **Model Construction Process**: The GRU model is a recurrent neural network (RNN) variant designed to handle sequential data. It uses gating mechanisms to control the flow of information, allowing it to capture temporal dependencies in financial data. Specific details on the input features or training process are not provided in the report[3][19] - **Model Evaluation**: The GRU model shows mixed performance, with significant drawdowns in certain market segments[3][19] 2. Model Name: Barra1d - **Model Construction Idea**: A short-term factor model based on the Barra framework, focusing on daily data[3][19] - **Model Evaluation**: Barra1d exhibits significant drawdowns in multiple market segments, indicating weaker performance[3][19] 3. Model Name: Barra5d - **Model Construction Idea**: A medium-term factor model based on the Barra framework, focusing on 5-day data[3][19] - **Model Evaluation**: Barra5d demonstrates strong performance, achieving positive returns in various market segments[3][19] 4. Model Name: Close1d - **Model Construction Idea**: A short-term model focusing on daily closing prices[3][19] - **Model Evaluation**: Close1d performs well in certain market segments, achieving positive returns[3][19] 5. Model Name: Open1d - **Model Construction Idea**: A short-term model focusing on daily opening prices[3][19] - **Model Evaluation**: Open1d shows weaker performance, with significant drawdowns in certain market segments[3][19] --- Model Backtesting Results 1. GRU Model - **Weekly Excess Return**: -0.08% to -0.54% relative to the CSI 1000 Index[7][30] 2. Barra1d - **Weekly Excess Return**: -0.54%[31] - **Year-to-Date Excess Return**: 3.75%[31] 3. Barra5d - **Weekly Excess Return**: -0.31%[31] - **Year-to-Date Excess Return**: 7.42%[31] 4. Close1d - **Weekly Excess Return**: -0.40%[31] - **Year-to-Date Excess Return**: 5.73%[31] 5. Open1d - **Weekly Excess Return**: -0.08%[31] - **Year-to-Date Excess Return**: 6.68%[31] --- Quantitative Factors and Construction 1. Factor Name: Beta - **Factor Construction Idea**: Measures historical beta to capture market sensitivity[15] 2. Factor Name: Market Capitalization - **Factor Construction Idea**: Logarithm of total market capitalization[15] 3. Factor Name: Momentum - **Factor Construction Idea**: Average historical excess returns[15] 4. Factor Name: Volatility - **Factor Construction Process**: $ Volatility = 0.74 * \text{Historical Excess Return Volatility} + 0.16 * \text{Cumulative Excess Return Deviation} + 0.1 * \text{Residual Return Volatility} $ - **Parameters**: - Historical Excess Return Volatility: Measures the standard deviation of excess returns - Cumulative Excess Return Deviation: Captures deviations in cumulative returns - Residual Return Volatility: Measures the volatility of residual returns[15] 5. Factor Name: Nonlinear Market Capitalization - **Factor Construction Idea**: Cubic transformation of market capitalization[15] 6. Factor Name: Valuation - **Factor Construction Idea**: Inverse of price-to-book ratio[15] 7. Factor Name: Liquidity - **Factor Construction Process**: $ Liquidity = 0.35 * \text{Monthly Turnover} + 0.35 * \text{Quarterly Turnover} + 0.3 * \text{Annual Turnover} $ - **Parameters**: - Monthly Turnover: Measures trading activity over a month - Quarterly Turnover: Measures trading activity over a quarter - Annual Turnover: Measures trading activity over a year[15] 8. Factor Name: Profitability - **Factor Construction Process**: $ Profitability = 0.68 * \text{Analyst Forecast Earnings Yield} + 0.21 * \text{Inverse Price-to-Cash Flow} + 0.11 * \text{Inverse Price-to-Earnings (TTM)} $ $ + 0.18 * \text{Analyst Long-Term Growth Forecast} + 0.11 * \text{Analyst Short-Term Growth Forecast} $ - **Parameters**: - Analyst Forecast Earnings Yield: Measures expected earnings relative to price - Inverse Price-to-Cash Flow: Captures cash flow efficiency - Analyst Growth Forecasts: Reflects expected growth rates[15] 9. Factor Name: Growth - **Factor Construction Process**: $ Growth = 0.24 * \text{Earnings Growth Rate} + 0.47 * \text{Revenue Growth Rate} $ - **Parameters**: - Earnings Growth Rate: Measures growth in earnings - Revenue Growth Rate: Measures growth in revenue[15] 10. Factor Name: Leverage - **Factor Construction Process**: $ Leverage = 0.38 * \text{Market Leverage} + 0.35 * \text{Book Leverage} + 0.27 * \text{Debt-to-Asset Ratio} $ - **Parameters**: - Market Leverage: Measures leverage based on market value - Book Leverage: Measures leverage based on book value - Debt-to-Asset Ratio: Captures the proportion of debt in total assets[15] --- Factor Backtesting Results 1. Momentum Factors - **120-Day Momentum**: Weekly return -2.37%[28] - **60-Day Momentum**: Weekly return -2.17%[28] - **20-Day Momentum**: Weekly return -1.69%[28] 2. Volatility Factors - **60-Day Volatility**: Weekly return -1.53%[28] - **20-Day Volatility**: Weekly return -0.96%[28] - **120-Day Volatility**: Weekly return 0.78%[28] 3. Median Deviation - **Weekly Return**: -0.40%[28]

- The report tracks style factors including profitability, volatility, and momentum, which showed strong long positions, while nonlinear market capitalization, valuation, and leverage factors demonstrated strong short positions[3][16] - Barra style factors include Beta (historical beta), market capitalization (logarithm of total market capitalization), momentum (mean of historical excess return series), volatility (weighted combination of historical excess return volatility, cumulative excess return deviation, and residual return volatility), nonlinear market capitalization (third power of market capitalization style), valuation (inverse of price-to-book ratio), liquidity (weighted turnover rates across monthly, quarterly, and yearly periods), profitability (weighted combination of analyst forecast earnings-price ratio, inverse cash flow ratio, and inverse trailing twelve-month PE ratio), growth (weighted combination of earnings growth rate and revenue growth rate), and leverage (weighted combination of market leverage, book leverage, and debt-to-asset ratio)[15] - GRU factors demonstrated strong multi-directional performance across various stock pools, with models like barra5d showing particularly strong positive returns[4][5][7] - GRU long-only portfolio outperformed the CSI 1000 index with excess returns ranging from 0.06% to 0.95% this week, while the barra5d model achieved a year-to-date excess return of 7.75%[8][30][31]

Quantitative Models and Construction 1. Model Name: GRU (Gated Recurrent Unit) Models - **Model Construction Idea**: GRU models are used to capture sequential patterns in stock price movements and combine fundamental and technical features for prediction[3][4] - **Model Construction Process**: - The GRU models are trained on historical stock data, incorporating both fundamental and technical indicators as input features - Different variations of GRU models are used, such as `open1d`, `close1d`, and `barra1d`, which focus on specific aspects of stock price movements (e.g., open prices, close prices, or Barra-style factor adjustments)[4][5][6] - **Model Evaluation**: GRU models show mixed performance, with some models like `barra1d` performing well, while others like `close1d` exhibit significant drawdowns[5][6][8] --- Backtesting Results of Models GRU Models - **open1d**: Weekly excess return: -0.23%, Monthly: 2.34%, YTD: 6.70%[31][32] - **close1d**: Weekly excess return: 0.06%, Monthly: 3.83%, YTD: 5.55%[31][32] - **barra1d**: Weekly excess return: 0.00%, Monthly: 0.34%, YTD: 3.33%[31][32] - **barra5d**: Weekly excess return: 0.10%, Monthly: 2.88%, YTD: 7.01%[31][32] --- Quantitative Factors and Construction 1. Factor Name: Beta - **Factor Construction Idea**: Measures the historical sensitivity of a stock's returns to market returns[15] - **Factor Construction Process**: Calculated as the historical beta of the stock relative to the market[15] 2. Factor Name: Momentum - **Factor Construction Idea**: Captures the average historical excess returns of a stock over a specific period[15] - **Factor Construction Process**: - Momentum = Mean of historical excess return series[15] 3. Factor Name: Volatility - **Factor Construction Idea**: Measures the variability of a stock's excess returns over time[15] - **Factor Construction Process**: - Volatility = 0.74 * Historical excess return volatility + 0.16 * Cumulative excess return deviation + 0.10 * Residual return volatility[15] 4. Factor Name: Valuation - **Factor Construction Idea**: Represents the inverse of the price-to-book ratio, indicating undervaluation[15] - **Factor Construction Process**: - Valuation = 1 / Price-to-Book Ratio[15] 5. Factor Name: Growth - **Factor Construction Idea**: Measures the growth potential of a stock based on earnings and revenue growth[15] - **Factor Construction Process**: - Growth = 0.24 * Earnings Growth Rate + 0.47 * Revenue Growth Rate[15] 6. Factor Name: Profitability - **Factor Construction Idea**: Combines various profitability metrics to assess a stock's financial health[15] - **Factor Construction Process**: - Profitability = 0.68 * Analyst Forecast Earnings Yield + 0.21 * Inverse of Price-to-Cash Flow Ratio + 0.11 * Inverse of Price-to-Earnings Ratio (TTM) + 0.18 * Analyst Forecast Long-Term Growth Rate + 0.11 * Analyst Forecast Short-Term Growth Rate[15] --- Backtesting Results of Factors Fundamental Factors - **Static Financial Factors**: Weekly excess return: Negative[4][6] - **Growth Factors**: Weekly excess return: Positive[4][6] - **Surprise Growth Factors**: Weekly excess return: Positive[4][6] Technical Factors - **Short-Term Momentum**: Weekly excess return: Negative[4][6] - **Long-Term Momentum**: Weekly excess return: Positive[4][6] - **Volatility**: Weekly excess return: Positive[4][6] GRU Factors - **open1d**: Weekly excess return: Positive[4][6] - **close1d**: Weekly excess return: Negative[5][6] - **barra1d**: Weekly excess return: Positive[5][6]

Quantitative Models and Construction Methods 1. Model Name: GRU (Generalized Recurrent Unit) - **Model Construction Idea**: GRU models are used to capture temporal dependencies and patterns in financial data, leveraging recurrent neural network structures to predict stock performance or factor returns[3][4][5] - **Model Construction Process**: The GRU model is trained on historical stock data, incorporating features such as price movements, volume, and other technical indicators. Specific GRU-based models mentioned include: - **open1d**: Focuses on daily opening prices - **close1d**: Focuses on daily closing prices - **barra1d**: Integrates Barra-style risk factors for daily predictions - **barra5d**: Extends Barra-style risk factors to a 5-day horizon[5][6][25] - **Model Evaluation**: GRU models show mixed performance, with some models like open1d performing well, while others like barra1d and barra5d experience significant drawdowns in certain market conditions[5][6][25] --- Model Backtesting Results GRU Model Performance - **open1d**: - Weekly excess return: 1.22% - Monthly excess return: 2.58% - Year-to-date excess return: 6.08%[29][30] - **close1d**: - Weekly excess return: 1.89% - Monthly excess return: 2.91% - Year-to-date excess return: 4.14%[29][30] - **barra1d**: - Weekly excess return: 0.85% - Monthly excess return: 1.50% - Year-to-date excess return: 3.48%[29][30] - **barra5d**: - Weekly excess return: 0.84% - Monthly excess return: 2.25% - Year-to-date excess return: 5.59%[29][30] --- Quantitative Factors and Construction Methods 1. Factor Name: Barra Style Factors - **Factor Construction Idea**: Barra factors are designed to capture systematic risk exposures across various dimensions such as size, value, momentum, and volatility[13][14] - **Factor Construction Process**: - **Beta**: Historical beta of the stock - **Size**: Natural logarithm of total market capitalization - **Momentum**: Weighted average of historical excess returns, combining volatility, cumulative deviation, and residual volatility $ Momentum = 0.74 \cdot \text{Volatility} + 0.16 \cdot \text{Cumulative Deviation} + 0.1 \cdot \text{Residual Volatility} $ - **Volatility**: Weighted average of historical residual return volatilities - **Valuation**: Inverse of price-to-book ratio - **Liquidity**: Weighted average of turnover ratios (monthly, quarterly, yearly) - **Profitability**: Weighted average of analyst forecasted earnings yield, cash flow yield, and other profitability metrics - **Growth**: Weighted average of earnings and revenue growth rates - **Leverage**: Weighted average of market leverage, book leverage, and debt-to-asset ratio[14][15] - **Factor Evaluation**: Barra factors demonstrate varying performance across different market conditions, with some factors like volatility and liquidity showing strong returns, while others like size and growth exhibit weaker performance[15][16] 2. Factor Name: Technical Factors - **Factor Construction Idea**: Technical factors aim to capture price and volume-based patterns, focusing on momentum and volatility metrics[17][20][24] - **Factor Construction Process**: - **Momentum**: Calculated over different time horizons (e.g., 20-day, 60-day, 120-day) - **Volatility**: Measured as the standard deviation of returns over specific periods (e.g., 20-day, 60-day, 120-day) - **Median Deviation**: Captures the median absolute deviation of returns[27] - **Factor Evaluation**: High-momentum and high-volatility stocks generally outperform, but certain periods show negative returns for these factors, especially in the 120-day horizon[17][27] 3. Factor Name: Fundamental Factors - **Factor Construction Idea**: Fundamental factors are derived from financial statements, focusing on profitability, growth, and valuation metrics[17][20][24] - **Factor Construction Process**: - **Static Financial Metrics**: Return on equity (ROE), return on assets (ROA), and profit margins - **Growth Metrics**: Earnings growth, revenue growth, and cash flow growth - **Surprise Metrics**: Earnings and revenue surprises relative to analyst expectations[19][21][23] - **Factor Evaluation**: Growth and surprise factors perform well, while static financial metrics like ROA and ROE show weaker performance in certain periods[19][21][23] --- Factor Backtesting Results Barra Factors - **Volatility**: Weekly return: 0.75%, Monthly return: 2.73% - **Liquidity**: Weekly return: 0.68%, Monthly return: 1.37% - **Size**: Weekly return: -1.45%, Monthly return: -3.60%[15][16] Technical Factors - **20-day Momentum**: Weekly return: -1.81%, Monthly return: -6.16% - **60-day Volatility**: Weekly return: -1.79%, Monthly return: -0.74% - **120-day Momentum**: Weekly return: -1.68%, Monthly return: -0.80%[27] Fundamental Factors - **ROA Growth**: Weekly return: 0.23%, Monthly return: 1.31% - **Earnings Surprise**: Weekly return: 0.20%, Monthly return: 1.11% - **Revenue Growth**: Weekly return: 0.17%, Monthly return: 0.77%[19][21][23]

Core Viewpoint - The article discusses the integration of return factors into an orthogonal layer within deep learning models to enhance stock selection effectiveness while maintaining low correlation with existing return factors [1][2]. Group 1: Deep Learning Model Enhancements - By incorporating return factors into the orthogonal layer, deep learning factors can maintain good stock selection performance while ensuring low correlation with these return factors [1]. - The deep learning model's black-box nature makes it challenging to manually adjust factor weights during significant market style shifts; thus, the orthogonal layer allows for easier manual adjustments without compromising stock selection effectiveness [1]. Group 2: Performance Metrics - After adding return factors to the orthogonal layer, deep learning factors still exhibit strong stock selection capabilities, achieving an Information Coefficient (IC) above 0.02 and an IC Information Ratio (IR) exceeding 6 [2]. - The combination of deep learning factors with manually constructed return factors leads to significant improvements in overall market long positions compared to using deep learning factors alone, although the enhancement varies across different index-enhanced portfolios [2]. Group 3: Correlation and Performance - The correlation between deep learning factors and multi-granularity factors remains low after integrating return factors into the orthogonal layer, with high-frequency data inputs showing a correlation of no more than 0.01 [2]. - Utilizing deep learning factors alongside multi-granularity factors can significantly enhance the performance of overall market long positions, although the deep learning factors show limited predictive capability for mid to large-cap stock returns, resulting in less noticeable improvements for index-enhanced portfolios [2].