金融工程研究报告：利率择时：短周期价量策略

Quantitative Models and Construction Methods 1. Model Name: Short-Cycle Interest Rate Timing Model - Model Construction Idea: Combines price-volume patterns and interest rate trend states to predict short-term fluctuations and optimize timing decisions[1][3][21] - Model Construction Process: - Short-term Fluctuation Prediction: - Utilizes futures and stock index trend models instead of traditional technical analysis indicators - Employs time-series networks to process price-volume and trend features, capturing nonlinear relationships for short-term fluctuation predictions[25][27] - Features include: - Price-volume features (e.g., closing price, trading volume, intraday returns, upward/downward amplitude)[26] - Trend features (e.g., adjusted returns over 1 month, 3 months, 1 year)[26] - Volatility features (e.g., short-term and medium-term volatility)[26] - Pattern features (e.g., K-line patterns with a 10-day lookback window)[26] - Time-series data is processed with a rolling window approach, retraining quarterly using historical data up to the current date[26] - Interest Rate Trend State Recognition: - Constructs state variables based on the term structure of government bond yields - Identifies trends through yield curve "deformations" using three metrics: 1. Translation Trend Strength: Measures the uniformity of yield movements across maturities using information entropy 2. Twist Degree: Assesses steepening or flattening of the yield curve using KL divergence 3. Deformation Magnitude: Measures yield changes relative to the previous period[30][31][32] - Preprocessed features include translation, twist, and deformation metrics for various maturities[34] - Signal Synthesis: - Combines price-volume and trend state features into a feature vector - Inputs the vector into a biLSTM-attn time-series network to generate timing signals (positive for long positions, non-positive for short positions)[41] - Model Evaluation: Demonstrates significant improvement in risk-return and timing accuracy compared to traditional technical analysis and price-volume-only models[27][42] --- Model Backtesting Results 1. Short-Cycle Interest Rate Timing Model - Annualized Return: 6.63% - Annualized Excess Return: 3.93% - Annualized Volatility: 0.0252 - Timing Accuracy: 58.68% - Average Profit-Loss Ratio: 1.02 - Annualized Sharpe Ratio: 2.56 - Benchmark Sharpe Ratio: 0.71[43] 2. Price-Volume Trend Model (Baseline) - Annualized Return: 4.68% - Annualized Excess Return: 2.03% - Annualized Volatility: 0.0249 - Timing Accuracy: 55.29% - Average Profit-Loss Ratio: 1.03 - Annualized Sharpe Ratio: 1.85 - Benchmark Sharpe Ratio: 0.71[29] 3. Traditional Technical Analysis - Annualized Return: 2.92% - Annualized Excess Return: 0.32% - Annualized Volatility: 0.0250 - Timing Accuracy: 53.40% - Average Profit-Loss Ratio: 0.96 - Annualized Sharpe Ratio: 1.16 - Benchmark Sharpe Ratio: 0.71[29] --- Quantitative Factors and Construction Methods 1. Factor Name: Price-Volume Features - Construction Idea: Reflects market sentiment and risk preferences through short-term price-volume dynamics[3][21] - Construction Process: - Closing Price: Standardized using time-series z-score - Trading Volume: Standardized using time-series z-score - Intraday Returns: Calculated as np.log(close/open) - Upward Amplitude: Calculated as np.log(high/open) - Downward Amplitude: Calculated as np.log(open/low)[26] 2. Factor Name: Interest Rate Trend State Features - Construction Idea: Captures slow and continuous changes in the central tendency of interest rates through yield curve deformations[3][30] - Construction Process: - Translation Trend Strength: Measured using information entropy - Twist Degree: Measured using KL divergence - Deformation Magnitude: Measures yield changes relative to the previous period - Preprocessed features include metrics for various maturities (e.g., 1, 3, 5, 10, 15, 20, 30 years)[34] --- Factor Backtesting Results 1. Price-Volume Features - Annualized Return: 4.68% - Annualized Excess Return: 2.03% - Annualized Volatility: 0.0249 - Timing Accuracy: 55.29% - Average Profit-Loss Ratio: 1.03 - Annualized Sharpe Ratio: 1.85 - Benchmark Sharpe Ratio: 0.71[29] 2. Interest Rate Trend State Features - Annualized Return: 6.63% - Annualized Excess Return: 3.93% - Annualized Volatility: 0.0252 - Timing Accuracy: 58.68% - Average Profit-Loss Ratio: 1.02 - Annualized Sharpe Ratio: 2.56 - Benchmark Sharpe Ratio: 0.71[43]