孙悦萌 本报（chinatimes.net.cn）记者张玫 北京报道 她用数学的尺规丈量市场的波动，将最终的投资答案，写进每一位客户真实的生活场景里。 建信基金数量投资部总经理助理孙悦萌日前接受了《华夏时报》记者的专访。"我们交付的不是模型， 而是与客户生活场景深度匹配的解决方案。"这句话背后，是一位数学背景的基金经理，将冰冷数据转 化为温暖陪伴的长期主义实践。 当数学语言翻译成资产密码 孙悦萌的世界始于数学的纯粹与严谨。本科数学、海外运筹学与金融工程的学术路径，赋予她解构复杂 系统的能力。然而从象牙塔到投资战场，她完成的第一次重要跨越，是认识到"学术追求'点'的深度， 投资更需要'面'的广度"。 她坦言，许多在教科书上优美的模型，例如经典的"均值方差模型"，在真实市场中会因参数过度敏感 而"水土不服"。这教会她的不是放弃量化，而是为理性框架注入现实的弹性。 她建立起"三层调整"的纪律化框架，却不为模型所限制。每个月，她和团队会像检修精密仪器般审视组 合，依据量化信号进行战术调整，但在风格切换上，她更倾向于"均衡中的适度偏离"。 她将基金投顾的角色理解为"解决方案提供者"而非"产品销售者"。在净值化时代，她认 ...

Quantitative Models and Construction Methods 1. Model Name: Mean-Variance Model - **Model Construction Idea**: The model determines the optimal portfolio by balancing expected returns and risks, based on the mean and variance of asset returns[8] - **Model Construction Process**: 1. Define the portfolio return as a random variable 2. Use the expected return ($E[R]$) and variance ($Var[R]$) to measure the portfolio's performance 3. Solve the optimization problem to maximize expected return for a given level of risk or minimize risk for a given level of return - Formula: $ \text{Minimize: } \sigma_p^2 = \sum_{i=1}^n \sum_{j=1}^n w_i w_j \sigma_{ij} $ $ \text{Subject to: } \sum_{i=1}^n w_i = 1 $ Where $w_i$ is the weight of asset $i$, $\sigma_{ij}$ is the covariance between assets $i$ and $j$[8] - **Model Evaluation**: Flexible in adjusting portfolios based on expected returns and risks, but struggles to incorporate new market dynamics and subjective views[8] 2. Model Name: Black-Litterman Model - **Model Construction Idea**: Combines the Bayesian framework with the mean-variance model to incorporate subjective views into the portfolio optimization process[8] - **Model Construction Process**: 1. Start with a prior distribution of expected returns based on market equilibrium 2. Incorporate subjective views as additional constraints 3. Use the Bayesian approach to update the prior distribution with subjective views to form a posterior distribution - Formula: $ \Pi = \tau \Sigma w_{mkt} $ $ E[R] = \left( \tau \Sigma^{-1} + P^T \Omega^{-1} P \right)^{-1} \left( \tau \Sigma^{-1} \Pi + P^T \Omega^{-1} Q \right) $ Where $\Pi$ is the implied equilibrium return, $\tau$ is a scaling factor, $\Sigma$ is the covariance matrix, $w_{mkt}$ is the market portfolio weights, $P$ is the view matrix, $\Omega$ is the uncertainty matrix, and $Q$ is the view vector[8] - **Model Evaluation**: Flexible and allows integration of subjective views, but requires strong assumptions about return distributions and is computationally complex[8] 3. Model Name: Risk Parity Model - **Model Construction Idea**: Focuses on balancing the risk contribution of each asset in the portfolio rather than their weights[7] - **Model Construction Process**: 1. Calculate the risk contribution of each asset: $RC_i = w_i \cdot \sigma_i \cdot \rho_{i,p}$ 2. Adjust weights to equalize the risk contributions across all assets - Formula: $ RC_i = w_i \cdot \sigma_i \cdot \rho_{i,p} $ Where $RC_i$ is the risk contribution of asset $i$, $w_i$ is the weight of asset $i$, $\sigma_i$ is the standard deviation of asset $i$, and $\rho_{i,p}$ is the correlation between asset $i$ and the portfolio[7] - **Model Evaluation**: Enhances risk control and can incorporate multiple risk dimensions, but lacks a mechanism to optimize returns and may struggle with unrecognized risks[7] 4. Model Name: All-Weather Model (Bridgewater) - **Model Construction Idea**: Aims to achieve stable performance across all economic environments by focusing on risk parity under growth and inflation sensitivity[11] - **Model Construction Process**: 1. Classify assets based on their sensitivity to growth and inflation 2. Allocate weights to achieve risk parity across these dimensions - Formula: Not explicitly provided, but the model emphasizes balancing risk rather than returns[11] - **Model Evaluation**: Stable allocation structure with a focus on low-risk assets, but may underperform in specific market conditions due to its heavy reliance on bonds and cash[15] --- Model Backtesting Results 1. Mean-Variance Model - **Maximum Drawdown**: Exceeded 4% in some periods (e.g., 2018-2019), but quickly recovered[57] - **Sharpe Ratio**: Higher than benchmarks in optimistic scenarios, demonstrating strong risk-adjusted returns[57] 2. Black-Litterman Model - **Maximum Drawdown**: Similar to the mean-variance model, with better adaptability to subjective views[57] - **Sharpe Ratio**: Improved compared to the mean-variance model due to the integration of subjective views[57] 3. Risk Parity Model - **Maximum Drawdown**: Generally lower than the mean-variance model, reflecting its focus on risk control[57] - **Sharpe Ratio**: Moderate, as the model does not explicitly optimize returns[57] 4. All-Weather Model - **Maximum Drawdown**: Comparable to fixed-ratio models, with a focus on stability[15] - **Sharpe Ratio**: Similar to benchmarks, reflecting its conservative allocation[15] --- Quantitative Factors and Construction Methods 1. Factor Name: Monthly Frequency Slicing - **Factor Construction Idea**: Use historical slices of monthly data to reflect maximum drawdown and market sentiment[41] - **Factor Construction Process**: 1. Extract rolling 20-day returns for each year 2. Use the bottom 20% quantile to estimate pessimistic scenarios and maximum drawdown - Formula: $ \text{Max Drawdown} = \text{Min} \left( \frac{P_t - P_{peak}}{P_{peak}} \right) $ Where $P_t$ is the price at time $t$, and $P_{peak}$ is the peak price[41] - **Factor Evaluation**: Effective in capturing extreme market conditions, but limited in predicting long-term trends[41] 2. Factor Name: BootStrap State Space - **Factor Construction Idea**: Use BootStrap sampling to create a state space of asset returns under different scenarios[45] - **Factor Construction Process**: 1. Sample historical data with replacement to create new sequences 2. Calculate return distributions for pessimistic, neutral, and optimistic scenarios - Formula: $ F = B - \alpha \cdot C $ Where $F$ is the objective function, $B$ is the expected return under risk constraints, $C$ is the penalty for exceeding risk constraints, and $\alpha$ is the penalty parameter[50] - **Factor Evaluation**: Provides a robust framework for scenario analysis, but computationally intensive[45] --- Factor Backtesting Results 1. Monthly Frequency Slicing - **Maximum Drawdown**: Successfully captured extreme drawdowns in historical data, with 90% coverage for A-shares and Hong Kong stocks[40] - **Sharpe Ratio**: Not explicitly provided, but the factor is more focused on risk control[40] 2. BootStrap State Space - **Maximum Drawdown**: Achieved a 4% maximum drawdown target in most scenarios, with only minor deviations in extreme conditions[57] - **Sharpe Ratio**: Optimized under different scenarios, with higher ratios in optimistic environments[57]

Core Insights - The report emphasizes the importance of asset allocation strategies based on both long-term and short-term perspectives, utilizing historical data to optimize investment portfolios [21][22][23]. Asset Performance Overview - Equity assets showed strong performance, with the CSI 300 Index and Nasdaq 100 Index rising by 3.5% and 2.4% respectively over the past month, while the Indian Sensex 30 Index declined by 2.9% [7][6]. - In the bond market, government bond yields increased, leading to a 0.2% decline in the government bond index, while corporate bond indices remained stable due to narrowing credit spreads [12][11]. - Commodity assets experienced a 3.8% increase in the South China Commodity Index in July, although gold prices fluctuated, ending the month nearly flat [17][16]. Asset Allocation Strategies - The report suggests a debt-oriented asset allocation strategy comprising 10% Asia-Pacific emerging market stocks, 80% corporate bonds, and 10% gold [28]. - A mixed asset allocation strategy is recommended, including 23% Nasdaq 100 Index, 7% CSI 300 Index, 40% corporate bonds, and 30% commodities [28]. Strategy Performance Tracking - From April 2015 to July 2025, the mean-variance model strategy achieved an annualized return of 6.81% with a maximum drawdown of 3.6% and a Sharpe ratio of 2.76 [25]. - The strategy's performance from January 2025 to July 2025 yielded a cumulative return of 1.97%, with a notable return of -0.15% in July due to insufficient bond contributions and declines in the Indian market index [25][27]. Model Utilization - The report employs a mean-variance model for long-term asset allocation, which outperforms constant mix strategies, and integrates the Black-Litterman model to enhance return stability by combining historical and recent performance data [22][23][24].