金融工程 | 金工定期报告 配置方向上，我们的行业配置模型显示，中期角度行业配置继续推荐困境 反转型板块，推荐港股创新药和证券保险，上行趋势仍在延续，此外受益 政策驱动板块方面，光伏以及化工有望保持上行；TWO BETA 模型继续推 荐科技板块，继续关注军工算力以及电池。短期信号显示，黄金股有望在 调整后迎来反弹。 从估值指标来看，wind 全 A 指数整体 PE 位于 85 分位点附近，属于中等水 平，PB 位于 50 分位点附近，属于较低水平，结合短期趋势判断，根据我 们的仓位管理模型，当前以 wind 全 A 为股票配置主体的绝对收益产品建 议仓位至 80%。 金融工程 证券研究报告 量化择时周报：牛市思维，行业如何配置？ 牛市思维，行业如何配置？ 上周周报(20250817)认为：短期而言，上周市场放量加速上攻，短期日线 或有震荡，但仍建议逢低加仓；当前 WIND 全 A 趋势线位于 5625 点附近， 赚钱效应值为 3.73%，显著为正，在赚钱效应转负之前，建议耐心持有，保 持高仓位。最终 wind 全 A 继续大涨，全周上涨 3.87%。市值维度上，上周 代表小市值股票的中证 2000 上涨 3. ...

广发证券首席金工分析师 安宁宁 SAC: S0260512020003 anningning@gf.com.cn 广发证券资深金工分析师 张钰东 SAC: S0260522070006 zhangyudong@gf.com.cn 广发金工安宁宁陈原文团队 摘要 最近5个交易日，科创50指数涨13.31%，创业板指涨5.85%，大盘价值涨1.56%，大盘成长涨4.77%，上证50涨3.38%，国证2000代表的小盘涨3.47%，通 信、电子表现靠前，房地产、煤炭表现靠后。 风险溢价，中证全指静态PE的倒数EP减去十年期国债收益率，权益与债券资产隐含收益率对比，历史数次极端底部该数据均处在均值上两倍标准差区 域，比如2012/2018/2020年（疫情突发），2022/04/26达到4.17%，2022/10/28风险溢价再次上升到4.08%，市场迅速反弹，2024/01/19指标4.11%，自2016年 以来第五次超过4%。截至2025/08/22指标3.03%，两倍标准差边界为4.77%。 估值水平，截至2025/08/22，中证全指PETTM分位数76%，上证50与沪深300分别为72%、68%，创业板指接 ...

大盘指数 投资要点 面对当前"多变量撕裂"的复杂市场,传统择时方法效力减弱。 通过构建"宏观-资金情绪-市场状态"三维分析框架,利用HMM 模型 将不可观测的市场环境量化为 4 种可感知的隐藏状态 (趋势上涨/震 荡上涨/震荡下跌/趋势下跌）,实现了对市场运行模式的精准刻画。 动态调制机制突破传统趋势模型瓶颈,实现"预见性"择时。传 统 HMM 模型是静态的,难以适应瞬息万变的市场,创新性地引入了基 于宏观经济脉冲(PMI+信贷脉冲)和资金情绪(恐慌指数+融资盘+ETF 流向+散户情绪)的双因子动态调制矩阵,可在外生环境变化时主动 干预模型输出,显著提升了模型对市场拐点的反应速度甚至提供了一 定预判能力。 资料来源:聚源,中邮证券研究所 研究所 分析师:黄子签 SAC 登记编号:S1340523090002 Email : huangziyin@cnpsec. com 8000 2000 策略观点 市场脉搏 (1)：基于隐马尔科夫链与动态调制的 量化择时方案 近期研究报告 《节奏和方向同样重要》 - 2025. 08. 11 凯利公式动态仓位优化,实现风险收益比最大化。有别于简单的 固定仓位模式,将HMM 状 ...

Group 1 - Market sentiment remains high with an index value of 3.2, showing signs of potential decline, suggesting further observation is needed [3][9] - The trading volume across the A-share market has significantly increased, with daily trading exceeding 2 trillion RMB for three consecutive days, indicating strong market activity [15][17] - The industry trend indicators show an upward breakout, reflecting a narrowing of funding viewpoint discrepancies [21][23] Group 2 - The small-cap and growth styles are currently favored, with the electronic and computer sectors showing the strongest short-term trend scores, particularly with scores reaching 100 [30][31] - The model indicates a high degree of trading concentration in sectors like machinery, electronics, and construction decoration, which may pose valuation and sentiment risks [39][41] - The report highlights that sectors with lower trading concentration, such as beauty care and public utilities, may present opportunities for gradual long-term positioning as risk appetite increases [39][41]

Quantitative Models and Construction Methods 1. Model Name: Timing System Signal (Wind All A Moving Average Distance Model) - **Model Construction Idea**: This model uses the distance between the short-term moving average (20-day) and the long-term moving average (120-day) of the Wind All A Index to determine the market's overall trend. A positive and expanding distance indicates an upward trend[2][9]. - **Model Construction Process**: 1. Calculate the 20-day moving average (short-term) and the 120-day moving average (long-term) of the Wind All A Index. - Latest values: 20-day MA = 5658, 120-day MA = 5241[2][9]. 2. Compute the percentage difference between the two moving averages: $ \text{Distance} = \frac{\text{20-day MA} - \text{120-day MA}}{\text{120-day MA}} \times 100\% $ - Current distance = 7.96%[2][9]. 3. Interpret the signal: If the distance is greater than 3% and positive, the market is in an upward trend[2][9]. - **Model Evaluation**: The model effectively captures the market's upward momentum and provides a clear signal for maintaining high equity positions during positive trends[2][9]. 2. Model Name: Industry Allocation Model - **Model Construction Idea**: This model identifies industries with potential for medium-term outperformance based on factors such as policy support, valuation, and growth trends[2][10]. - **Model Construction Process**: 1. Analyze industry-specific drivers, including policy incentives and growth catalysts. 2. Identify sectors with "distressed reversal" characteristics or benefiting from policy-driven growth. 3. Recommend sectors such as innovative pharmaceuticals, securities insurance, photovoltaics, coal, and non-ferrous metals. 4. Use the TWO BETA model to emphasize technology-related sectors, including military, computing power, and batteries[2][10]. - **Model Evaluation**: The model provides actionable insights for sector rotation, aligning with macroeconomic and policy trends[2][10]. 3. Model Name: Position Management Model - **Model Construction Idea**: This model determines optimal equity allocation levels based on valuation metrics and market trends[3][10]. - **Model Construction Process**: 1. Assess valuation levels of the Wind All A Index using PE and PB ratios. - Current PE: 70th percentile (moderate level). - Current PB: 30th percentile (low level)[3][10]. 2. Combine valuation analysis with timing signals (e.g., moving average distance and profit-making effect). 3. Recommend equity allocation levels based on the above factors. - Current recommendation: 80% equity allocation[3][10]. - **Model Evaluation**: The model balances valuation and trend analysis, providing a systematic approach to equity allocation[3][10]. --- Model Backtesting Results 1. Timing System Signal - Moving average distance: 7.96% (greater than the 3% threshold, indicating an upward trend)[2][9]. 2. Industry Allocation Model - Recommended sectors: Innovative pharmaceuticals, securities insurance, photovoltaics, coal, non-ferrous metals, military, computing power, and batteries[2][10]. 3. Position Management Model - PE: 70th percentile (moderate level)[3][10]. - PB: 30th percentile (low level)[3][10]. - Recommended equity allocation: 80%[3][10]. --- Quantitative Factors and Construction Methods 1. Factor Name: Profit-Making Effect - **Factor Construction Idea**: This factor measures the market's ability to generate profits for investors, serving as a key indicator of market sentiment and potential capital inflows[2][10]. - **Factor Construction Process**: 1. Calculate the profit-making effect value based on market performance. - Current value: 3.73% (positive)[2][10]. 2. Interpret the signal: A positive value indicates sustained investor confidence and potential for further capital inflows[2][10]. - **Factor Evaluation**: The factor is a reliable indicator of market sentiment, supporting timing and allocation decisions[2][10]. --- Factor Backtesting Results 1. Profit-Making Effect - Current value: 3.73% (positive, indicating sustained market confidence)[2][10].

Quantitative Models and Construction Methods 1. Model Name: "Three-Dimensional Composite Timing Signal" - **Model Construction Idea**: This model integrates three key timing indicators—"Credit Impulse, Beta Dispersion, and Trading Volume"—to represent three core timing dimensions: economic fundamentals, overall sentiment, and structural risk. It aims to balance high probability and high payoff indicators for superior timing performance[5][12]. - **Model Construction Process**: - **Credit Impulse**: Measures the month-on-month change in credit balance percentile, reflecting economic fundamentals[5][15]. - **Beta Dispersion**: Captures the dispersion of stock betas, representing market sentiment and structural risk[5][12]. - **Trading Volume**: Quantifies market activity and liquidity, serving as a sentiment indicator[5][12]. - The composite signal combines these three indicators to generate timing signals, with historical backtesting showing strong in-sample and out-of-sample performance[12][14]. - **Model Evaluation**: The model demonstrates excellent timing performance in both in-sample and out-of-sample tests, effectively capturing market uptrends[12][14]. 2. Model Name: "Short-Term Timing Strategy" - **Model Construction Idea**: This model uses macroeconomic, valuation, sentiment, and liquidity indicators to generate weekly timing signals[20][23]. - **Model Construction Process**: - **Macroeconomic Indicators**: Includes PMI (>50 for optimism), credit impulse percentile (62.71%), and M1 growth rate percentile (96.61%)[20][23]. - **Valuation Indicators**: PE and PB percentiles (99.59% and 96.36%, respectively) are used to assess valuation levels[21][23]. - **Sentiment Indicators**: Beta dispersion (69.49%), trading volume sentiment (93.80%), and volatility (11.00%) are analyzed for market sentiment[21][23]. - **Liquidity Indicators**: Monetary rate (37.29%), exchange rate expectations (74.58%), and financing data (97.11%) are used to evaluate liquidity conditions[22][23]. - Signals are aggregated to determine overall market positioning[23]. - **Model Evaluation**: The strategy has consistently outperformed the benchmark, with significant annualized returns and lower drawdowns[22][23]. 3. Model Name: "Growth-Value Style Rotation Model" - **Model Construction Idea**: This model evaluates macroeconomic, valuation, and sentiment factors to determine the optimal allocation between growth and value styles[29][30]. - **Model Construction Process**: - **Macroeconomic Factors**: Profit cycle slope (4.17), interest rate cycle level (14.17), and credit cycle changes (-3.33) are analyzed[31]. - **Valuation Factors**: PE and PB valuation spreads (23.99% and 39.00%, respectively) are used to assess relative attractiveness[31]. - **Sentiment Factors**: Turnover and volatility spreads (38.13% and 19.97%, respectively) are considered for sentiment analysis[31]. - Signals are combined to recommend allocations between growth and value styles[31]. - **Model Evaluation**: The model has delivered significant excess returns over the benchmark since 2012, though it underperformed in 2025 YTD[30][32]. 4. Model Name: "Small-Cap vs. Large-Cap Style Rotation Model" - **Model Construction Idea**: This model evaluates macroeconomic, valuation, and sentiment factors to determine the optimal allocation between small-cap and large-cap styles[33][34]. - **Model Construction Process**: - **Macroeconomic Factors**: Profit cycle slope (4.17), interest rate cycle level (14.17), and credit cycle changes (-3.33) are analyzed[35]. - **Valuation Factors**: PE and PB valuation spreads (93.88% and 97.67%, respectively) are used to assess relative attractiveness[35]. - **Sentiment Factors**: Turnover and volatility spreads (81.01% and 51.58%, respectively) are considered for sentiment analysis[35]. - Signals are combined to recommend allocations between small-cap and large-cap styles[35]. - **Model Evaluation**: The model has consistently outperformed the benchmark since 2012, though it underperformed in 2025 YTD[34][36]. 5. Model Name: "Four-Style Rotation Model" - **Model Construction Idea**: This model integrates the conclusions of the growth-value and small-cap-large-cap rotation models to recommend allocations across four styles: small-cap growth, small-cap value, large-cap growth, and large-cap value[37]. - **Model Construction Process**: - Combines the signals from the growth-value and small-cap-large-cap models to allocate weights across the four styles[37]. - Current recommended allocation: small-cap growth (37.5%), small-cap value (12.5%), large-cap growth (37.5%), and large-cap value (12.5%)[37]. - **Model Evaluation**: The model has delivered significant excess returns over the benchmark since 2012, though it underperformed in 2025 YTD[37][38]. --- Model Backtesting Results 1. "Three-Dimensional Composite Timing Signal" - Annualized Return: 21.26% - Annualized Volatility: 14.46% - Maximum Drawdown: 12.80% - Sharpe Ratio: 1.2676 - Annualized Excess Return: 13.39%[14] 2. "Short-Term Timing Strategy" - Annualized Return: 17.83% - Annualized Volatility: 15.87% - Maximum Drawdown: 22.44% - Sharpe Ratio: 0.9874 - Annualized Excess Return: 13.24%[22][27] 3. "Growth-Value Style Rotation Model" - Annualized Return: 11.76% - Annualized Volatility: 20.77% - Maximum Drawdown: 43.07% - Sharpe Ratio: 0.5438 - Annualized Excess Return: 4.73%[30][32] 4. "Small-Cap vs. Large-Cap Style Rotation Model" - Annualized Return: 12.45% - Annualized Volatility: 22.65% - Maximum Drawdown: 50.65% - Sharpe Ratio: 0.5441 - Annualized Excess Return: 5.21%[34][36] 5. "Four-Style Rotation Model" - Annualized Return: 13.37% - Annualized Volatility: 21.51% - Maximum Drawdown: 47.91% - Sharpe Ratio: 0.5988 - Annualized Excess Return: 5.72%[37][38]

Core Viewpoint - The recent market performance shows a significant increase in the ChiNext and Sci-Tech 50 indices, while large-cap value stocks have declined, indicating a shift in investor sentiment towards growth sectors [1][2]. Market Performance - In the last five trading days, the Sci-Tech 50 index rose by 5.53%, the ChiNext index increased by 8.48%, while the large-cap value index fell by 0.76%. The large-cap growth index rose by 3.63%, and the Shanghai 50 index increased by 1.57%. Small-cap stocks represented by the CSI 2000 index rose by 3.86% [1]. - The communication and electronics sectors performed well, while the banking and steel sectors lagged behind [1]. Risk Premium Analysis - The risk premium, measured as the difference between the inverse of the static PE of the CSI All Share Index and the yield of ten-year government bonds, has reached historical extremes. As of October 28, 2022, the risk premium was at 4.08%, indicating a potential market rebound [1]. - The risk premium has exceeded 4% for the fifth time since 2016, with the latest reading on January 19, 2024, at 4.11% [1]. Valuation Levels - As of August 15, 2025, the CSI All Share Index's TTM PE is at the 72nd percentile, with the Shanghai 50 and CSI 300 at 69% and 63%, respectively. The ChiNext index is at a relatively low valuation level of approximately 33% [2]. - The long-term view of the Deep 100 index suggests a cyclical pattern of bear and bull markets every three years, with the current adjustment phase starting in Q1 2021 showing sufficient time and space for a potential upward cycle [2]. Fund Flow and Trading Activity - In the last five trading days, there was an outflow of 10.4 billion yuan from ETFs, while margin financing increased by approximately 41.8 billion yuan. The average daily trading volume across both markets was 20,767 billion yuan [3]. AI and Trend Observation - The use of convolutional neural networks (CNN) for modeling price and volume data has been explored, with the latest focus on mapping learned features to industry themes, particularly in the communication sector [8].

- The Black-Litterman model (BL model) is used for asset allocation, combining investor views with market equilibrium[17][20] - The construction process of the BL model involves adjusting the expected returns based on investor views and then optimizing the portfolio using mean-variance optimization[17][20] - The Risk Parity model aims to allocate risk equally across all assets in a portfolio, rather than allocating capital equally[27][30] - The construction process of the Risk Parity model involves calculating the risk contribution of each asset and solving an optimization problem to equalize these contributions[28][29][30] - The Counter-Cyclical Allocation model adjusts asset allocation based on economic cycles, aiming to reduce risk during downturns and increase exposure during upturns[11][43] - The Macro Momentum Timing model uses macroeconomic indicators to time market entries and exits, aiming to capture trends and avoid downturns[11][60] - The Sentiment Timing model uses investor sentiment indicators to time market entries and exits, aiming to capitalize on market overreactions[67] Model Performance Metrics - **Black-Litterman Model**: Annualized return 6.58%, maximum drawdown 3.18%, annualized volatility 2.15%, Sharpe ratio 1.86, Calmar ratio 2.07[22][24] - **Risk Parity Model**: Annualized return 6.07%, maximum drawdown 3.78%, annualized volatility 2.26%, Sharpe ratio 1.58, Calmar ratio 1.61[31] - **Counter-Cyclical Allocation Model**: Annualized return 7.36%, maximum drawdown 8.85%, annualized volatility 6.12%, Sharpe ratio 1.13, Calmar ratio 0.85[43][47] - **Macro Momentum Timing Model**: Annualized return 7.06%, maximum drawdown 6.60%, annualized volatility 6.06%, Sharpe ratio 1.13, Calmar ratio 1.97[60] - **Sentiment Timing Model**: Annualized return 7.74%, maximum drawdown 24.91%, annualized volatility 17.49%, Sharpe ratio 1.01, Calmar ratio 0.62[67][87]

Quantitative Models and Construction Methods 1. Model Name: Timing Strategy Based on Dynamic Macro Event Factors for CSI Southbound Consumer Index - **Model Construction Idea**: The model aims to explore the impact of China's macroeconomic environment on the overall performance and trends of Hong Kong-listed consumer companies. It uses dynamic macro event factors to construct a timing strategy framework[3][4][21] - **Model Construction Process**: 1. **Macro Data Selection**: Over 20 macro indicators across four dimensions (economy, inflation, monetary, and credit) were tested, including PMI, PPI, M1, etc.[22][24] 2. **Data Preprocessing**: - Align data frequency to monthly - Fill missing values using the formula: $$ X_{t} = X_{t-1} + Median_{diff12} $$ - Apply filtering (e.g., one-sided HP filter): $$ \hat{t}_{t|t,\lambda} = \sum_{s=1}^{t} \omega_{t|t,s,\lambda} \cdot y_{s} = W_{t|t,\lambda}(L) \cdot y_{t} $$ - Derive factors using transformations like YoY, MoM, and moving averages[28][29][30] 3. **Event Factor Construction**: - Identify event breakout directions based on the correlation between data and asset returns - Generate event factors using methods like data breaking through moving averages, medians, or directional changes - Construct 28 different event factors per indicator[31][33] 4. **Factor Evaluation and Selection**: - Use metrics like "win rate of returns" and "volatility-adjusted returns" for screening - Select the top-performing factors based on statistical significance, win rate (>55%), and occurrence frequency[32][34] 5. **Final Macro Factor Selection**: - Five macro factors were selected based on their performance in the backtest, including "PMI: Raw Material Prices" and "YoY Growth of Aggregate Financing"[35][36] 6. **Timing Signal Construction**: - If >2/3 of factors signal bullish, the category signal is marked as 1 - If <1/3 signal bullish, the category signal is marked as 0 - Intermediate proportions are marked accordingly - Aggregate category scores determine the timing position signal[4][36][38] - **Model Evaluation**: The strategy effectively captures systematic opportunities and mitigates risks, outperforming benchmarks in most years and controlling drawdowns during market downturns[12][21] --- Model Backtest Results 1. Timing Strategy Based on Dynamic Macro Event Factors - **Annualized Return**: 9.31% (2018/11–2025/7)[11][23] - **Maximum Drawdown**: -29.72%[11][23] - **Sharpe Ratio**: 0.54[11][23] - **Return-to-Drawdown Ratio**: 0.31[11][23] - **Average Position**: 43%[11] - **Monthly Return (2025/7)**: 2.79% (vs. benchmark 2.48%)[11][13] --- Quantitative Factors and Construction Methods 1. Factor Name: PMI: Raw Material Prices - **Factor Construction Idea**: Captures inflationary pressures and their impact on consumer sector performance[36] - **Factor Construction Process**: - Data Type: Original data - Rolling Window: 96 months[36] 2. Factor Name: US-China 10Y Bond Spread - **Factor Construction Idea**: Reflects monetary policy divergence and its influence on capital flows[36] - **Factor Construction Process**: - Data Type: Original data - Rolling Window: 72 months[36] 3. Factor Name: YoY Growth of Aggregate Financing (12M Rolling) - **Factor Construction Idea**: Measures credit expansion and its implications for economic growth[36] - **Factor Construction Process**: - Data Type: Original data - Rolling Window: 96 months[36] 4. Factor Name: M1 YoY Growth - **Factor Construction Idea**: Tracks monetary liquidity and its correlation with asset prices[36] - **Factor Construction Process**: - Data Type: Original data - Rolling Window: 48 months[36] 5. Factor Name: YoY Growth of Medium- to Long-Term Loans (12M Rolling) - **Factor Construction Idea**: Indicates long-term credit trends and their impact on investment[36] - **Factor Construction Process**: - Data Type: Original data - Rolling Window: 48 months[36] --- Factor Backtest Results 1. PMI: Raw Material Prices - **Rolling Window**: 96 months[36] 2. US-China 10Y Bond Spread - **Rolling Window**: 72 months[36] 3. YoY Growth of Aggregate Financing (12M Rolling) - **Rolling Window**: 96 months[36] 4. M1 YoY Growth - **Rolling Window**: 48 months[36] 5. YoY Growth of Medium- to Long-Term Loans (12M Rolling) - **Rolling Window**: 48 months[36]

Group 1 - Market sentiment indicators show a slight increase to 3.25, maintaining a high level and a bullish outlook, although there is a need to monitor for potential turning points as scores show a slight decline during the week [9][12][30] - The price-volume consistency indicator remains elevated, indicating high levels of market activity, while the PCR combined with VIX has shifted from positive to negative, suggesting a change in market sentiment [12][23][24] - Total trading volume for the week showed a slight decline but remained strong, with daily trading volumes exceeding 1.6 trillion RMB on most days, indicating robust market activity [17][30] Group 2 - The report highlights that sectors with high trading congestion, such as machinery, defense, and non-ferrous metals, have seen significant price increases, but caution is advised due to potential valuation and sentiment corrections [30][34][36] - The report identifies that the small-cap growth style is currently favored, with the RSI model indicating a preference for growth stocks, although the 5-day RSI shows a rapid decline compared to the 20-day RSI, warranting further observation [30][39][41] - The report provides a detailed analysis of sector performance, with machinery, light industry, and defense showing the strongest short-term trends, particularly machinery scoring a perfect 100 [30][31][32]