金融工程定期：资产配置月报（2025年7月）-20250630

Quantitative Models and Construction Methods 1. Model Name: High-Frequency Macroeconomic Factors - Construction Idea: Utilize asset portfolio simulations to construct a high-frequency macroeconomic factor system, capturing market expectations of macroeconomic changes[11] - Construction Process: 1. Combine real macroeconomic indicators to synthesize low-frequency macroeconomic factors 2. Select assets leading low-frequency macroeconomic factors 3. Use rolling multivariate regression with asset returns as independent variables and low-frequency macroeconomic factors as dependent variables to determine asset weights and simulate macroeconomic factor trends - Example: - High-frequency economic growth: Rolling regression fitting with Hang Seng Index and CRB Metal Spot Index[12] - High-frequency consumer inflation: Rolling regression fitting with Food Price Index and Pork Price Index[12] - High-frequency production inflation: Rolling regression fitting with Production Material Price Index, CRB Industrial Spot Index, and CRB Composite Spot Index[12] - Evaluation: High-frequency indicators show a leading effect compared to low-frequency macroeconomic factors[12][16] 2. Model Name: Duration Timing Model - Construction Idea: Predict yield curves using an improved Diebold2006 model and map expected returns for bonds of different durations[18] - Construction Process: 1. Predict level, slope, and curvature factors 2. Level factor prediction: Based on macroeconomic variables and policy rate tracking 3. Slope and curvature factors: Predicted using AR(1) model[18] - Evaluation: The model provides actionable insights for short-duration bond holdings, though recent performance shows slight underperformance against benchmarks[19] 3. Model Name: Convertible Bond Allocation Model - Construction Idea: Compare convertible bonds with equities and credit bonds, and implement style rotation within convertible bonds[23] - Construction Process: 1. Relative valuation with equities: Construct "100-yuan conversion premium rate" and calculate rolling historical percentiles[23] 2. Relative valuation with credit bonds: Use "Adjusted YTM - Credit Bond YTM" to measure relative value[23] 3. Style rotation: Construct factors like conversion premium deviation and theoretical value deviation to exclude overvalued bonds; use 20-day momentum and volatility deviation to capture market sentiment[25] - Evaluation: The model effectively captures style rotation opportunities, achieving strong annualized returns and high IR[27] 4. Model Name: Gold Expected Return Model - Construction Idea: Link gold's forward real return to U.S. TIPS (Treasury Inflation-Protected Securities) real return[30] - Construction Process: - Formula: $ E[Real_Return^{gold}] = k \times E[Real_Return^{Tips}] $ $ E[R^{gold}] = \pi^{e} + k \times E[Real_Return^{Tips}] $ - Parameters: - $ k $ estimated via extended window OLS - $ \pi^{e} $ approximated using the Federal Reserve's long-term inflation target of 2%[30] - Evaluation: The model has consistently issued bullish signals for gold, with strong historical returns[32] 5. Model Name: Active Risk Budget Model - Construction Idea: Combine risk parity with active signals to dynamically adjust equity and bond weights based on three dimensions: cross-asset valuation, equity valuation, and market liquidity[35][36] - Construction Process: 1. Cross-asset valuation: Use the Fed model to calculate equity risk premium (ERP) - Formula: $ ERP = \frac{1}{PE_{ttm}} - YTM_{TB}^{10Y} $[37] 2. Equity valuation: Calculate rolling historical percentiles of equity valuations over the past five years[40] 3. Market liquidity: Use the M2-M1 scissors difference to measure marginal changes in liquid funds[41] 4. Aggregate signals and convert them into risk budget weights using the softmax function: $ softmax(x) = \frac{\exp(\lambda x)}{\exp(\lambda x) + \exp(-\lambda x)} $[45] - Evaluation: The model provides stable returns with low drawdowns, outperforming risk parity and equal-weighted strategies over the long term[49] 6. Model Name: Industry Rotation Model 3.0 - Construction Idea: Construct sub-models across six dimensions (trading behavior, prosperity, capital flow, chip structure, macro drivers, technical analysis) and dynamically synthesize signals for bi-weekly industry selection[50] - Construction Process: 1. Trading behavior: Capture intraday momentum and overnight reversal effects 2. Prosperity: Capture earnings momentum effects 3. Capital flow: Capture active buying and passive selling behaviors 4. Chip structure: Capture holding returns and resistance-support effects 5. Macro drivers: Map high-frequency macro expectations to industries 6. Technical analysis: Capture trading signals from trends, oscillations, and volume indicators of industry constituents[50] - Evaluation: The model effectively identifies industry rotation opportunities, though recent performance shows challenges in achieving excess returns[56][59] --- Model Backtest Results 1. High-Frequency Macroeconomic Factors - High-frequency economic growth: Leading low-frequency indicators as of June 27, 2025[12][16] - High-frequency consumer inflation: Leading low-frequency indicators as of June 27, 2025[16] - High-frequency production inflation: Leading low-frequency indicators as of June 27, 2025[16] 2. Duration Timing Model - June 2025 return: 31.6bp (benchmark: 33.7bp, excess: -2.1bp)[19] - 1-year return: 1.78% (benchmark: 4.74%, excess: -2.96%)[19] 3. Convertible Bond Allocation Model - Annualized return (2018-2025): 23.87% - Maximum drawdown: 16.67% - IR: 1.43 - Monthly win rate: 64.77% - 2025 YTD return: 25.21%[27] 4. Gold Expected Return Model - 1-year expected return: 23.0% (as of June 30, 2025)[30] - 1-year absolute return: 40.72% (based on TIPS timing model)[32] 5. Active Risk Budget Model - June 2025 portfolio return: 0.92% - Annualized return (full sample): 6.51% - Maximum drawdown: 4.89% - Return-to-volatility ratio: 1.64 - Return-to-drawdown ratio: 1.33[49] 6. Industry Rotation Model 3.0 - June 2025 long portfolio return: 1.05% - Short portfolio return: 2.50% - Equal-weight benchmark return: 2.03% - Long excess return: -0.98% - Short excess return: -0.47% - Long-short return: -1.45%[56]