[Table_Title] 证券研究报告 /金融工程研究报告 "量化+主观"灵活资产配置方案 --量化资产配置系列之四 报告摘要： 本文为东北金工量化资产配置系列第四篇，主要介绍按照哈佛捐赠基金 的基于因子的灵活非决定性资产配置(FIFAA)思路对国内可投资产进行 组合配置的过程和结果。 哈佛捐赠基金提出 FIFAA 是希望将量化方案的学术性/规范性与主观方 案的前瞻性/灵活性进行组合，使用基于历史数据得到的量化结果(ex post)与主观观点转换成资产-因子暴露的预期结果(ex ante)进行叠加，得 到兼具量化和主观优点的组合结果。同时，与本系列此前介绍的方案不 同，FIFAA 要求在构建宏观因子时需要具有可投资性/简洁性，以降低量 化计算中的误差和主观判断的不确定性。 由于在复制方法论的过程中使用投资者主观观点进行 beta 调整和不同阶 段敞口选择在回溯时较难实践，本文进行了一定简化，具体操作过程如 下： 综合结果显示，无论是使用历史风险载荷还是调整风险载荷，两种优化 结果相对于多资产等权均提供了更可观的收益和风险回报。在实操中， 如果持有单资产对多因子的联合关系观点，叠加预期收益的判断，或可 得到更 ...

Core Insights - The report predicts changes in the yield curve using the Nelson-Siegel model, which describes the curve's dynamics through three factors: level, slope, and curvature [3][29]. - An improvement in the model for predicting the level factor has been made by incorporating policy rates, market benchmark rates, slope, and curvature factors, which enhances the predictive accuracy [4][56]. - The duration rotation strategy based on yield curve predictions shows robust performance, consistently outperforming benchmarks and achieving significant excess returns [5][91]. Duration Rotation Strategy - The latest signal from the duration rotation strategy, as of October 31, 2025, indicates a strong preference for long-duration interest rate bonds, with a signal value of 10 [6][96]. - The strategy is designed to capitalize on the natural "risk-return-liquidity" trade-offs present in different maturity bonds, where short-term bonds offer lower duration and volatility but higher reinvestment risk, while long-term bonds provide higher coupon protection but are more exposed to interest rate risk [10][14]. Yield Curve Construction - The report establishes the yield curve using historical spot rate data from 2006 to 2025, showing that the average yield curve is monotonically upward over the entire period [21][22]. - Principal component analysis of historical spot rates reveals three main components that represent the level, slope, and curvature of the yield curve, providing insights into its dynamics [26][41]. Statistical Characteristics of Spot Rates - The statistical characteristics of spot rates indicate that as the maturity increases, the mean yield rises while volatility decreases, with the average yield curve showing a consistent upward trend [21][22]. - The report provides detailed statistics on various maturities, including total returns, annualized returns, annualized volatility, Sharpe ratios, and maximum drawdowns, highlighting the performance of different maturity segments [12][95]. Model Improvements - The report discusses enhancements to the predictive model for the level factor by integrating external variables such as policy rates and market rates, which have shown to improve the direction prediction accuracy [56][62]. - The introduction of additional factors, including slope and curvature, aims to refine predictions during periods of yield curve inversion, thereby increasing the model's robustness [70][75]. Backtesting Results - Backtesting results demonstrate that the improved duration rotation strategy yields a total return of 110.37% over the evaluation period, significantly outperforming various maturity indices and equal-weighted indices [91][95]. - The strategy's maximum drawdown is reported at 5.36%, which is lower than the maximum drawdown of 7.23% for the 7-10 year index, indicating a more stable performance [95].

- **Domestic Asset BL Model** - **Model Name**: Black-Litterman (BL) Model - **Construction Idea**: The BL model integrates subjective views with quantitative asset allocation using Bayesian theory, optimizing asset weights based on market analysis and expected returns. It addresses the sensitivity of mean-variance models to expected returns and provides higher fault tolerance compared to purely subjective investments [26][27][33] - **Construction Process**: 1. Use historical returns of assets over the past five years to estimate market equilibrium returns (Π) 2. Specify a risk aversion coefficient (e.g., λ = 10), which corresponds to a target volatility 3. Alternatively, assign fixed weights (e.g., stock:bond:convertible bond:commodity:gold = 10:80:5:2.5:2.5) and reverse calculate the risk aversion coefficient dynamically for each period [33] - **Evaluation**: The BL model effectively combines subjective views with quantitative methods, providing robust asset allocation solutions [26][27] - **Domestic Asset Risk Parity Model** - **Model Name**: Risk Parity Model - **Construction Idea**: The model aims to equalize the risk contribution of each asset to the overall portfolio, optimizing asset weights based on expected volatility and correlation [32][35] - **Construction Process**: 1. Select appropriate underlying assets 2. Calculate each asset's risk contribution to the portfolio 3. Solve optimization problems to determine final asset weights 4. Use daily returns over the past five years to estimate the covariance matrix for stability [35] - **Evaluation**: The model provides stable returns across economic cycles and is well-suited for domestic investors [32][35] - **Macro Factor-Based Asset Allocation Strategy** - **Model Name**: Macro Factor-Based Strategy - **Construction Idea**: The strategy bridges macroeconomic research with asset allocation by constructing high-frequency macro factors (e.g., growth, inflation, interest rates, credit, exchange rates, liquidity) and aligning asset weights with subjective macroeconomic views [41][46] - **Construction Process**: 1. Calculate factor exposure levels for assets monthly 2. Use risk parity portfolios as benchmarks to compute baseline factor exposures 3. Adjust factor exposure targets based on subjective macroeconomic views (e.g., inflation up = positive deviation) 4. Solve for asset weights using the model [41][46] - **Evaluation**: The strategy effectively incorporates macroeconomic insights into asset allocation, enhancing adaptability to changing economic conditions [41][46] - **Backtest Results for Models** - **Domestic Asset BL Model 1**: - Annualized return: 3.58% - Max drawdown: 1.31% - Annualized volatility: 2.19% [31][33] - **Domestic Asset BL Model 2**: - Annualized return: 3.18% - Max drawdown: 1.06% - Annualized volatility: 1.99% [31][33] - **Domestic Asset Risk Parity Model**: - Annualized return: 3.12% - Max drawdown: 0.76% - Annualized volatility: 1.34% [39][40] - **Macro Factor-Based Strategy**: - Annualized return: 3.42% - Max drawdown: 0.65% - Annualized volatility: 1.32% [46][47]

Group 1 - The article emphasizes a balanced approach to factor selection, integrating macroeconomic quantitative insights with factor momentum to identify resonant factors while adjusting for discrepancies between macro and micro indicators [1] - Current macroeconomic indicators show signs of economic recovery, slightly loose liquidity, and improved credit metrics, leading to a revised outlook of economic improvement, weak liquidity, and loose credit [1] - The article presents a table summarizing the performance of various factors across different indices, indicating that growth factors remain strong in the CSI 300, while the CSI 500 shows a more balanced factor selection [2][3] Group 2 - Economic leading indicators are beginning to rise, with the PMI and new orders showing increases, suggesting a slight upward trend in economic activity expected to continue into early 2026 [5][9] - The liquidity environment is assessed as slightly loose despite rising interest rates, with a comprehensive liquidity signal indicating a mixed outlook [11][15] - Credit indicators are showing weakness, with a slight positive shift in overall credit metrics, indicating a complex credit environment [15][16] Group 3 - The article suggests a preference for asset allocation towards gold due to strong momentum, while equity allocations are slightly reduced, reflecting a cautious stance on A-shares [16] - The focus of market attention is shifting from liquidity to economic indicators, with recent trends indicating a growing concern for economic performance over liquidity conditions [17] - Industry selection is advised to favor sectors sensitive to economic changes but less affected by liquidity, with public utilities and coal being highlighted as top sectors based on their sensitivity scores [19]

Group 1 - The report indicates that the economic leading indicators are showing signs of a slight recovery, with liquidity remaining slightly loose and credit indicators improving [3][12][19] - The economic forecast model suggests that October 2025 is at a turning point, with expectations for a slight upward trend over the next three months before entering a plateau [12][13] - The report highlights that the focus of the market is shifting towards economic indicators, surpassing liquidity concerns, with increased attention on economic and PPI-related factors [26][27] Group 2 - The liquidity environment is characterized by rising interest rates, with long-term rates exceeding the average, while overall liquidity remains slightly loose due to positive monetary supply signals [19][22] - Credit indicators have shown a slight positive trend, although the overall credit volume and structure remain low, indicating a mixed outlook for credit conditions [23][24] - The asset allocation perspective suggests a high allocation to gold due to strong momentum, while equity allocations have been slightly reduced [24][25] Group 3 - The industry selection is leaning towards sectors that are sensitive to economic conditions but less sensitive to liquidity, with a notable increase in defensive and consumer attributes [28][29] - The report identifies specific industries with the highest sensitivity to economic changes, including utilities and coal, while also highlighting sectors like media and consumer electronics for credit sensitivity [28][29] - The overall balance in industry selection reflects a decline in growth attributes, emphasizing a more defensive investment strategy [29]

Group 1 - The report indicates a slight recovery in economic indicators, with liquidity remaining slightly loose and credit indicators showing improvement. The macroeconomic dimensions suggest an overall direction of economic improvement, weak liquidity, and loose credit [3][6][8] - The economic leading indicators are expected to show a slight upward trend over the next three months, indicating a bottoming out in October 2025, with a prolonged period of slight recovery compared to last month [12][13] - The liquidity environment is characterized by rising interest rates, with long-term rates exceeding the moving average, while overall liquidity remains slightly loose due to positive monetary supply signals [19][22] Group 2 - The report emphasizes a high allocation to gold, with a weakening view on bonds and a slight reduction in A-share allocation. The current economic upturn, tight liquidity, and favorable credit conditions support this allocation strategy [24][26] - Market focus has shifted towards economic indicators, surpassing liquidity concerns, with a notable increase in attention to economic and PPI-related factors since September [26][28] - The industry selection is inclined towards sectors sensitive to economic changes, less sensitive to liquidity, and sensitive to credit conditions. The report highlights a decrease in growth attributes and an increase in defensive and consumer attributes, indicating a balanced approach [28][30][29]

Group 1 - The report highlights that domestic equity assets performed well in July 2025, with the risk parity strategy achieving a year-to-date return of 2.65% [2][5][20] - The report provides a summary of various asset allocation strategies, indicating that the domestic asset BL strategy 1 and 2 yielded returns of 2.40% and 2.34% respectively, while the risk parity strategy and macro factor-based strategy returned 2.65% and 2.59% respectively [21][41][42] - The report notes that the domestic equity market saw significant gains, with the CSI 1000 index rising by 4.8% and the Hang Seng Index increasing by 2.78% in July [8][9][10] Group 2 - The report discusses the correlation between different asset classes, indicating that the correlation between the CSI 300 and the total wealth index of government bonds was -38.08%, suggesting a potential for diversification [15][16] - The report outlines the performance of various asset allocation models, with the domestic risk parity strategy showing a maximum drawdown of 0.76% and an annualized volatility of 1.46% [41][42] - The macroeconomic outlook suggests downward risks for growth factors, while inflation expectations may stabilize due to recent policy measures [45][47]

- The report references the Fama-MacBeth method to simulate macro risk factors, transforming the logic of configuring macro risks through asset allocation into a logic of configuring macro risks by configuring assets[1][12][18] - Real macro factor data uses forecast values of relevant monthly macro indicators or asset monthly returns (interest rates/credit), performing univariate time series regression with each asset to obtain risk loadings, and applying a half-life weighting to historical loadings to smooth out instability caused by asset volatility[1][18][22] - The macro factor risk is decomposed into underlying asset portfolios to construct a macro factor risk parity portfolio[1][18][22] - The optimization results of risk parity for macro factors show certain economic growth elasticity, with both returns and volatility higher than those based on asset risk parity[2][39] - The report also discusses the factors influencing stock-bond correlation, referencing AQR's approach, which decomposes stock-bond correlation into economic growth volatility, inflation volatility, and the correlation between economic growth and inflation[3][42][43] - The study finds that economic growth volatility negatively contributes to stock-bond correlation, while interest rate volatility positively contributes, and the correlation between economic growth and interest rates is a positive contributing variable in domestic asset research[3][42][48] - Adding the inflation level factor further improves the explanatory power, with domestic data showing that the inflation level is a significant positive variable for stock-bond correlation[3][48][51] - Using a three-year historical window to calculate the coefficients of each variable, the study combines real values and consensus forecast data to calculate the change in stock-bond correlation for the next month, showing that the estimated and predicted values have the same trend and consistent signs with the real values[3][48][54] Quantitative Models and Construction Methods 1. **Model Name**: Macro-Factor Mimicking - **Construction Idea**: Transform the logic of configuring macro risks through asset allocation into configuring macro risks by configuring assets[1][12][18] - **Construction Process**: - Use forecast values of relevant monthly macro indicators or asset monthly returns (interest rates/credit) - Perform univariate time series regression with each asset to obtain risk loadings - Apply a half-life weighting to historical loadings to smooth out instability caused by asset volatility - Decompose macro factor risk into underlying asset portfolios to construct a macro factor risk parity portfolio[1][18][22] - **Formula**: $$ r_{t}=\alpha_{t}+B\cdot f_{t}+\varepsilon_{t} $$ $$ \Sigma=B\cdot F\cdot B^{T}+E $$ $$ \sigma_{P}{}^{2}\ =\ w^{T}\cdot\Sigma\ \cdot w=\ (w^{T}\cdot B)\cdot F\cdot(B^{T}\cdot w)+w^{T}\cdot E\cdot w $$ $$ \%\text{RC}\ =(w^{T}\cdot B)_{i}\cdot\frac{\partial\sigma_{P}}{\partial(w^{T}\cdot B)_{i}}/\sigma_{P}=\frac{(w^{T}\cdot B)_{i}\cdot(F\cdot(B^{T}\cdot w))_{i}}{w^{T}\cdot\Sigma\ \cdot w} $$ where B is the time-series calculated risk loadings, f is the factor returns, Σ is the asset risk matrix, and F is the macro factor return risk matrix[23][24] - **Evaluation**: The optimization results of risk parity for macro factors show certain economic growth elasticity, with both returns and volatility higher than those based on asset risk parity[2][39] Model Backtest Results 1. **Macro-Factor Mimicking Model**: - **Annualized Return**: 9.86% (12-month half-life), 9.46% (no half-life)[29] - **Annualized Volatility**: 9.55% (12-month half-life), 9.44% (no half-life)[29] - **Maximum Drawdown**: -14.30% (12-month half-life), -15.20% (no half-life)[29] - **2016 Return**: 37.24% (12-month half-life), 18.65% (no half-life)[29] - **2017 Return**: 2.17% (12-month half-life), 7.29% (no half-life)[29] - **2018 Return**: -5.02% (12-month half-life), -7.45% (no half-life)[29] - **2019 Return**: 14.61% (12-month half-life), 14.23% (no half-life)[29] - **2020 Return**: 12.20% (12-month half-life), 7.57% (no half-life)[29] - **2021 Return**: 14.63% (12-month half-life), 10.27% (no half-life)[29] - **2022 Return**: 0.36% (12-month half-life), 8.15% (no half-life)[29] - **2023 Return**: 5.41% (12-month half-life), 3.68% (no half-life)[29] - **2024 Return**: 6.83% (12-month half-life), 15.40% (no half-life)[29] - **2025.07.31 Return**: 7.44% (12-month half-life), 11.53% (no half-life)[29]

Group 1 - The report emphasizes that growth has become a resonant factor in the current economic environment, with a focus on selecting factors that are insensitive to economic conditions but sensitive to credit [2][7][9] - The report suggests that the current economic indicators are weak, leading to a preference for growth-oriented stocks in the investment strategy, particularly in the CSI 300 and CSI 1000 indices [2][9][10] - The macroeconomic outlook indicates a potential short-term recovery in economic indicators, with a forecasted slight increase in the economic leading indicators in August 2025 [12][13][14] Group 2 - The liquidity environment is described as relatively stable, with interest rates showing slight increases but remaining below historical averages, indicating a slightly loose liquidity condition [19][20][22] - Credit indicators are noted to be weak, with a decline in credit volume and structure, although the overall credit indicators remain positive [23][24] - The report advocates for an increase in stock allocation, reflecting a positive trend in equity markets, while reducing allocations in other asset classes [2][24][25] Group 3 - The report identifies liquidity as the primary focus of market attention, especially following recent market movements driven by liquidity conditions [26][27] - In terms of industry selection, the report recommends focusing on sectors that are less sensitive to economic fluctuations but more responsive to credit conditions, highlighting industries with growth attributes [4][31][28] - The report lists specific industries with high scores for economic insensitivity and credit sensitivity, including electronics, media, and beauty care, indicating a strategic focus on growth-oriented sectors [28][31]

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].