- Model Name: Domestic Asset BL Model 1; Model Construction Idea: The BL model is an improvement of the traditional mean-variance model, combining subjective views with quantitative models using Bayesian theory; Model Construction Process: The model optimizes asset allocation weights based on investor market analysis and asset return forecasts, effectively addressing the sensitivity of the mean-variance model to expected returns; Model Evaluation: The BL model provides a higher fault tolerance compared to purely subjective investments, offering efficient asset allocation solutions[14][15] - Model Name: Domestic Asset BL Model 2; Model Construction Idea: Similar to Domestic Asset BL Model 1; Model Construction Process: The model is built on the same principles as Domestic Asset BL Model 1 but with different asset selections; Model Evaluation: Similar to Domestic Asset BL Model 1[14][15] - Model Name: Global Asset BL Model 1; Model Construction Idea: Similar to Domestic Asset BL Model 1; Model Construction Process: The model is built on the same principles as Domestic Asset BL Model 1 but targets global assets; Model Evaluation: Similar to Domestic Asset BL Model 1[14][15] - Model Name: Global Asset BL Model 2; Model Construction Idea: Similar to Global Asset BL Model 1; Model Construction Process: The model is built on the same principles as Global Asset BL Model 1 but with different asset selections; Model Evaluation: Similar to Global Asset BL Model 1[14][15] - Model Name: Domestic Asset Risk Parity Model; Model Construction Idea: The risk parity model aims to equalize the risk contribution of each asset in the portfolio; Model Construction Process: The model calculates the risk contribution of each asset and optimizes the deviation between actual and expected risk contributions to determine final asset weights; Model Evaluation: The model provides stable returns across different economic cycles[20][21] - Model Name: Global Asset Risk Parity Model; Model Construction Idea: Similar to Domestic Asset Risk Parity Model; Model Construction Process: The model is built on the same principles as Domestic Asset Risk Parity Model but targets global assets; Model Evaluation: Similar to Domestic Asset Risk Parity Model[20][21] - Model Name: Macro Factor-Based Asset Allocation Model; Model Construction Idea: The model constructs a macro factor system covering growth, inflation, interest rates, credit, exchange rates, and liquidity; Model Construction Process: The model uses the Factor Mimicking Portfolio method to construct high-frequency macro factors and optimizes asset weights based on subjective macro views; Model Evaluation: The model bridges macro research and asset allocation, reflecting subjective macro judgments in asset allocation[23][24][27] - Domestic Asset BL Model 1, Weekly Return: 0.02%, July Return: 0.61%, 2025 YTD Return: 2.46%, Annualized Volatility: 2.16%, Maximum Drawdown: 1.31%[17][19] - Domestic Asset BL Model 2, Weekly Return: -0.06%, July Return: 0.48%, 2025 YTD Return: 2.41%, Annualized Volatility: 1.93%, Maximum Drawdown: 1.06%[17][19] - Global Asset BL Model 1, Weekly Return: -0.09%, July Return: 0.56%, 2025 YTD Return: 0.95%, Annualized Volatility: 1.95%, Maximum Drawdown: 1.64%[17][19] - Global Asset BL Model 2, Weekly Return: -0.07%, July Return: 0.51%, 2025 YTD Return: 1.59%, Annualized Volatility: 1.7%, Maximum Drawdown: 1.28%[17][19] - Domestic Asset Risk Parity Model, Weekly Return: -0.02%, July Return: 0.36%, 2025 YTD Return: 2.7%, Annualized Volatility: 1.46%, Maximum Drawdown: 0.76%[22][23] - Global Asset Risk Parity Model, Weekly Return: -0.03%, July Return: 0.3%, 2025 YTD Return: 2.16%, Annualized Volatility: 1.66%, Maximum Drawdown: 1.2%[22][23] - Macro Factor-Based Asset Allocation Model, Weekly Return: -0.03%, July Return: 0.38%, 2025 YTD Return: 2.76%, Annualized Volatility: 1.36%, Maximum Drawdown: 0.64%[28][29]

Core Viewpoint - The report highlights the performance of various domestic asset allocation strategies in May 2025, indicating a mixed performance across different strategies and asset classes, with a notable focus on the risk parity strategy achieving the highest year-to-date return of 1.73% [1][3]. Group 1: Asset Strategy Performance - Domestic Asset BL Strategy 1 recorded a May return of -0.22% and a year-to-date return of 0.96% [1][3]. - Domestic Asset BL Strategy 2 had a May return of -0.1% and a year-to-date return of 1.05% [1][3]. - The Domestic Asset Risk Parity Strategy achieved a May return of 0.29% and a year-to-date return of 1.73% [1][3]. - The Macro Factor-Based Asset Allocation Strategy reported a May return of 0.27% and a year-to-date return of 1.45% [1][3]. Group 2: Major Asset Trends - In May 2025, domestic equity assets performed well, with the Hang Seng Index, CSI 300, and others showing significant gains, while gold experienced a pullback [2]. - The Hang Seng Index rose by 3.96%, CSI 300 by 1.85%, and the total wealth index of corporate bonds by 0.41% [2]. - The South China Commodity Index and SHFE gold saw declines of 2.4% and 1.39%, respectively [2]. - Correlation analysis indicated a -36.97% correlation between CSI 300 and the total wealth index of government bonds over the past year [2]. Group 3: Macroeconomic Insights - As of the end of May 2025, the manufacturing PMI was at 49.5%, indicating a slight improvement in manufacturing sentiment [4]. - The PPI for April showed a year-on-year decline of -2.7%, with expectations for May at -3.17%, indicating ongoing deflationary pressures [4]. - The central bank conducted a MLF operation of 550 billion yuan, net injecting 400 billion yuan to support special bond issuance [4]. - The social financing scale stood at 424 trillion yuan at the end of April 2025, reflecting the credit environment [4].

Quantitative Models and Construction Methods 1. Model Name: Black-Litterman (BL) Model - **Model Construction Idea**: The BL model integrates subjective views (e.g., analyst target prices) with market equilibrium returns to optimize portfolio weights. It naturally handles stocks without analyst opinions by redistributing weights among stocks with available views [1][7][23]. - **Model Construction Process**: 1. **Expected Return Calculation**: The expected return vector is calculated as: $$\mu_{p}=[(\tau\Sigma)^{-1}+P^{T}\Omega^{-1}P]^{-1}[(\tau\Sigma)^{-1}\pi+P^{T}\Omega^{-1}Q]$$ - \( \tau \): Weight of subjective views - \( \Sigma \): Covariance matrix of asset returns - \( \pi \): Equilibrium return vector - \( P \): View matrix indicating which assets are involved in each view - \( Q \): View return vector - \( \Omega \): Confidence matrix for views [7][8]. 2. **Covariance Matrix Adjustment**: The adjusted covariance matrix is: $$\Sigma_{p}^{*}=\Sigma+[(\tau\Sigma)^{-1}+P^{T}\Omega^{-1}P]^{-1}$$ [7]. 3. **Portfolio Weight Calculation**: Without constraints: $$w=(\delta\Sigma_{p}^{*})^{-1}\mu_{p}$$ - \( \delta \): Risk aversion coefficient [8]. With constraints, mean-variance optimization is applied [8]. 4. **Subjective Views**: - Views are derived from analyst target prices, forming long-short portfolios within industries. - Example: For Industry 1, buy stock 2 and sell stock 8, with an expected excess return of 5% [12][17]. 5. **Confidence Matrix**: - Confidence is based on the variance of analyst predictions. - Formula: $$\Omega=P Z P^{T}$$ - \( Z \): Diagonal matrix of standardized variances of analyst predictions [18][19]. - **Model Evaluation**: The BL model effectively integrates subjective views and market data, providing optimized portfolio weights superior to simpler weighting methods [25][35]. --- Model Backtesting Results 1. BL Model Performance in CSI 300 Index - **Cumulative Returns (2010-2021/5/31)**: - Full Portfolio: +186.52% - Positive View Portfolio: +215.15% - CSI 300 Index: +49.11% - Excess Returns: +137.41% (Full), +166.05% (Positive) [26][31]. - **Annualized Returns**: - Full Portfolio: Outperformed in most years except 2014, 2016-2017 [30][31]. - **Portfolio Composition**: - Full Portfolio: Average of 265 stocks - Positive View Portfolio: Average of 138 stocks [32][34]. - **Weighting Comparison**: BL weights outperformed market-cap and equal-weighted portfolios [35][38]. 2. BL Model Performance in CSI 500 Index - **Cumulative Returns (2010-2021/5/31)**: - Full Portfolio: +354.89% - Positive View Portfolio: +469.14% - CSI 500 Index: +50.02% - Excess Returns: +304.87% (Full), +419.12% (Positive) [40][45]. - **Annualized Returns**: - Full Portfolio: Outperformed in most years except 2014 [45]. - **Portfolio Composition**: - Full Portfolio: Average of 475 stocks - Positive View Portfolio: Average of 138 stocks [46]. 3. Comparison with Direct Sorting Methods - **Cumulative Returns (2010-2021/5/31)**: - BL Positive View Portfolio: +215.15% - Direct Sorting (Mixed, Market-Cap Weighted): +122.48% - Direct Sorting (Mixed, Equal Weighted): +120.51% - Direct Sorting (Industry, Market-Cap Weighted): +145.30% - Direct Sorting (Industry, Equal Weighted): +154.87% [47][48]. 4. Sensitivity to \( \tau \) - **Impact of \( \tau \)**: - Higher \( \tau \) values increase the weight of subjective views in portfolio construction. - In CSI 300, higher \( \tau \) values led to slightly higher returns, with stable parameter sensitivity [54][61]. - In CSI 500, \( \tau \) changes caused greater return volatility [58][59].

- Model Name: Black-Litterman (BL) Model - Model Construction Idea: The BL model combines subjective views with market equilibrium returns to determine asset allocation weights[1][7] - Model Construction Process: - Calculate equilibrium returns using reverse optimization: $$\Pi = \delta \Sigma \mathbf{w}_{m}$$ where $\Pi$ is the equilibrium return vector, $\delta$ is the risk aversion coefficient, $\Sigma$ is the covariance matrix of asset returns, and $\mathbf{w}_{m}$ is the market capitalization weights[10] - Incorporate subjective views into the model: $$\mu_{p} = [(\tau \Sigma)^{-1} + P^{T} \Omega^{-1} P]^{-1} [(\tau \Sigma)^{-1} \Pi + P^{T} \Omega^{-1} Q]$$ where $\mu_{p}$ is the posterior mean return vector, $\tau$ is a scalar reflecting the uncertainty in the equilibrium returns, $P$ is the matrix representing the assets involved in the views, $\Omega$ is the diagonal covariance matrix of the views, and $Q$ is the vector of view returns[7] - Calculate the posterior covariance matrix: $$\Sigma_{p}^{*} = \Sigma + [(\tau \Sigma)^{-1} + P^{T} \Omega^{-1} P]^{-1}$$ - Determine the asset weights using the posterior returns and covariance matrix: $$\mathbf{w} = (\delta \Sigma_{p}^{*})^{-1} \mu_{p}$$[7] - Model Evaluation: The BL model provides a stable and reasonable asset allocation by anchoring to market equilibrium weights and adjusting based on subjective views[3][13] Model Backtest Results - BL Model, Information Ratio (IR): 0.45[2] - BL Model, Sharpe Ratio: 0.35[2] - BL Model, Annualized Return: 8.5%[2] Factor Construction and Process - Factor Name: Subjective View Matrix (P) and View Return Vector (Q) - Factor Construction Idea: The subjective view matrix (P) and view return vector (Q) represent the investor's views on the expected returns of certain assets[16][17] - Factor Construction Process: - Define the subjective views: - View 1: Automotive industry expected return of 6% - View 2: Media industry expected return of 4% - View 3: Food and beverage expected to underperform home appliances by 1% - View 4: Technology and new energy sectors expected to outperform traditional sectors by 2%[17] - Construct the view matrix (P): $$P = \begin{pmatrix} 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 1 \\ 0 & 0 & 0 & 0 & 0 & 0 & 1 & 0 & 0 & 0 \\ 0 & 0 & 0 & 1 & 0 & 0 & 0 & -1 & 0 & 0 \\ 0 & 0.5 & -0.7 & 0 & 0.3 & -0.3 & 0 & 0 & 0.2 & 0 \end{pmatrix}$$[17][19] - Construct the view return vector (Q): $$Q = \begin{pmatrix} 0.06 \\ 0.04 \\ -0.01 \\ 0.02 \end{pmatrix}$$[20][21] - Factor Evaluation: The subjective view matrix and view return vector allow for the incorporation of investor views into the BL model, providing flexibility and customization in asset allocation[23][25] Factor Backtest Results - View 1 (Automotive), Expected Return: 6%, Implied Return: 4.30%, Difference: 1.70%[24] - View 2 (Media), Expected Return: 4%, Implied Return: 5.17%, Difference: -1.17%[24] - View 3 (Food and Beverage vs. Home Appliances), Expected Return: -1%, Implied Return: -0.29%, Difference: -0.71%[24] - View 4 (Technology vs. Traditional Sectors), Expected Return: 2%, Implied Return: 2.48%, Difference: -0.48%[24]

Quantitative Models and Construction Methods - **Model Name**: Black-Litterman (BL) Model **Model Construction Idea**: The BL model combines market equilibrium portfolio weights with subjective investor views using Bayesian theorem, aiming to improve stability and flexibility in asset allocation[2][3][8] **Model Construction Process**: 1. **Market Equilibrium Portfolio**: The starting point is the CAPM-based market equilibrium portfolio, where asset weights are determined by market capitalization. The equilibrium returns are calculated using the utility function: $U = w^{T}\Pi - \frac{\delta}{2}w^{T}\Sigma w$ Here, $\Pi$ represents equilibrium returns, $\Sigma$ is the covariance matrix, and $\delta$ is the risk aversion coefficient[12][13][14]. Alternatively, equilibrium returns can be derived as: $\Pi = \delta\Sigma_{eq}$[14][15]. 2. **Bayesian Integration**: Bayesian theorem is applied to combine prior information (market equilibrium returns) with subjective investor views. The posterior mean and covariance matrix are calculated as: $\mu_{p} = [(\tau\Sigma)^{-1} + P^{T}\Omega^{-1}P]^{-1}[(\tau\Sigma)^{-1}\Pi + P^{T}\Omega^{-1}Q]$ $\Sigma_{p} = [(\tau\Sigma)^{-1} + P^{T}\Omega^{-1}P]^{-1}$ Here, $\tau$ represents the uncertainty of prior returns, $P$ is the matrix indicating assets involved in subjective views, $Q$ is the vector of expected returns for subjective views, and $\Omega$ is the confidence matrix for subjective views[9][10][29]. 3. **Final Asset Weights**: Using the posterior mean and covariance matrix, asset weights are optimized via mean-variance optimization: $\mathbf{w} = (\delta\Sigma_{p}^{*})^{-1}\mu_{p}$ $\Sigma_{p}^{*}$ can be calculated using two methods: - $\Sigma_{p}^{*} = \Sigma_{p} + \Sigma$ (recommended for practical use)[30][31] - $\Sigma_{p}^{*} = \Sigma_{p}$ (used in specific cases)[31][32]. - **Model Evaluation**: The BL model improves stability by starting with market equilibrium weights and allows flexible incorporation of subjective views. It avoids the sensitivity issues of traditional mean-variance models and provides intuitive results[2][8][9]. Model Backtesting Results - **BL Model**: No specific numerical backtesting results are provided in the report. Quantitative Factors and Construction Methods - **Factor Name**: None explicitly mentioned in the report. Factor Backtesting Results - **Factor Results**: None explicitly mentioned in the report.