宏观视角下的港股择时模型

Quantitative Models and Construction Methods 1. Model Name: Hang Seng Index R Timing Model - Model Construction Idea: The model integrates macroeconomic indicators, liquidity, inflation, and market sentiment to predict the monthly rise or fall of the Hang Seng Index R using a Logit regression approach. It transforms continuous variables into binary variables for prediction purposes[3][8][91] - Model Construction Process: - Selected four key variables as inputs: 1. Directional variable: USDHKD expectation revision lag_2 2. Threshold variable: CPI expectation revision lag_1 3. Directional variable: Citi China Economic Surprise Index monthly average lag_3 4. Volatility variable: Hang Seng Index put-call ratio change[8][85][87] - Applied Logit regression to train the model using 80% of the data as the training set and 20% as the test set - The regression equation is as follows: $ Return+1 = α + β1 + β2−1 + β3−2 + β4−3 + β5 + β6ℎ + β7 + β8 $ - Where Return is the binary variable representing the rise (1) or fall (0) of the Hang Seng Index R - Var represents the continuous variable transformed into a binary state variable - Duration is the state duration variable - Change is the state change variable (1 for change, 0 for no change) - Entropy is the information entropy over the past six periods - Quarter encodes the quarter of the variable (0 to 3 for Q1 to Q4)[39][91] - The model generates buy (1) or hold (0) signals based on the predicted value of the dependent variable[93] - Model Evaluation: The model demonstrates strong predictive ability with an out-of-sample AUC of approximately 0.70. Overfitting is controlled, with a degree of overfitting around 9.15%. The USDHKD expectation revision lag_2 is identified as the most significant variable, indicating that marginal changes in liquidity expectations dominate the prediction of Hang Seng Index R movements[8][91][95] --- Model Backtesting Results 1. Hang Seng Index R Timing Model - Annualized Excess Return: Approximately 10.74% from 2015 to January 2026[3][8][95] - Monthly Win Rate: About 81.95%[3][8][95] - Annual Win Rate: About 81.82%[3][8][95] - Out-of-Sample AUC: Approximately 0.70[3][8][91] - Performance by Year: - 2015: Annualized return 6.46%, excess annualized return 11.34%, IR 0.64 - 2016: Annualized return 9.16%, excess annualized return 4.66%, IR 0.28 - 2017: Annualized return 36.35%, excess annualized return -3.50%, IR -0.74 - 2018: Annualized return -8.06%, excess annualized return 4.16%, IR 0.45 - 2019: Annualized return 17.23%, excess annualized return 2.33%, IR 0.13 - 2020: Annualized return 21.20%, excess annualized return 21.55%, IR 1.39 - 2021: Annualized return 10.12%, excess annualized return 24.89%, IR 1.79 - 2022: Annualized return 5.59%, excess annualized return 20.74%, IR 0.78 - 2023: Annualized return 18.39%, excess annualized return 32.21%, IR 1.70 - 2024: Annualized return 37.07%, excess annualized return 11.50%, IR 0.78 - 2025: Annualized return 21.74%, excess annualized return -8.13%, IR -0.70 - 2026 (up to January): Annualized return 6.88%, excess annualized return 0.00%, IR 1.10[95][96] --- Quantitative Factors and Construction Methods 1. Factor Name: USDHKD Expectation Revision (lag_2) - Factor Construction Idea: Captures the marginal changes in liquidity expectations driven by USDHKD exchange rate revisions[3][8][46] - Factor Construction Process: - Calculated as the difference between the current month's forecast and the previous month's forecast for the USDHKD exchange rate - Transformed into a binary variable using a directional approach: positive values are set to 1, and negative values to 0[42][46] - Factor Evaluation: Demonstrates strong predictive ability with an out-of-sample AUC of 0.75 and a beta coefficient of -2.41, indicating that a slowdown in USD appreciation or an acceleration in depreciation increases the probability of Hang Seng Index R rising[46][61][63] 2. Factor Name: CPI Expectation Revision (lag_1) - Factor Construction Idea: Reflects inflation expectations and their impact on market sentiment and economic conditions[3][8][46] - Factor Construction Process: - Calculated as the difference between the current month's forecast and the previous month's forecast for CPI - Transformed into a binary variable using a threshold approach: values below the 20th percentile of the past 12 months are set to 1, and others to 0[42][46] - Factor Evaluation: Shows good predictive ability with an out-of-sample AUC of 0.66 and a beta coefficient of -1.42, indicating that a slowdown in inflation expectations increases the probability of Hang Seng Index R rising[46][53][55] 3. Factor Name: Citi China Economic Surprise Index Monthly Average (lag_3) - Factor Construction Idea: Measures the extent to which economic data exceeds or falls short of expectations, indicating economic momentum[3][8][56] - Factor Construction Process: - Calculated as the monthly average of the Citi China Economic Surprise Index - Transformed into a binary variable using a directional approach: positive values are set to 1, and negative values to 0[42][56] - Factor Evaluation: Demonstrates strong predictive ability with an out-of-sample AUC of 0.69 and a beta coefficient of 1.55, indicating that better-than-expected economic data increases the probability of Hang Seng Index R rising[56][61][64] 4. Factor Name: Hang Seng Index Put-Call Ratio Change - Factor Construction Idea: Captures market sentiment and risk aversion through changes in the put-call ratio of Hang Seng Index options[3][8][83] - Factor Construction Process: - Calculated as the change in the put-call ratio over the past six periods - Transformed into a binary variable using a volatility approach: values falling below the 20th percentile of the past 12 periods are set to 1, and others to 0[42][83] - Factor Evaluation: Provides incremental information for risk management, particularly in avoiding rapid market downturns. However, it has limited predictive ability for medium- to long-term trends, with an out-of-sample AUC of 0.48 and a beta coefficient of -1.71[83][85] --- Factor Backtesting Results 1. USDHKD Expectation Revision (lag_2) - Out-of-Sample AUC: 0.75 - Beta: -2.41 - P-value: 0.0006[61][63] 2. CPI Expectation Revision (lag_1) - Out-of-Sample AUC: 0.66 - Beta: -1.42 - P-value: 0.0333[45][46] 3. Citi China Economic Surprise Index Monthly Average (lag_3) - Out-of-Sample AUC: 0.69 - Beta: 1.55 - P-value: 0.0027[61][64] 4. Hang Seng Index Put-Call Ratio Change - Out-of-Sample AUC: 0.48 - Beta: -1.71 - P-value: 0.0053[80][83]