量化投资周报：AI行业轮动模型看好石油石化、家电等

Quantitative Models and Construction Methods AI Industry Rotation Model - Model Name: AI Industry Rotation Model - Model Construction Idea: The model uses a full-spectrum volume-price fusion factor to score 32 primary industries and constructs a weekly rebalancing strategy, selecting the top 5 industries for equal-weight allocation[1][16][23] - Model Construction Process: 1. Industry Pool: Includes 32 primary industries, with some industries split into subcategories (e.g., food and beverage split into food, beverage, and liquor)[23] 2. Factor: Full-spectrum volume-price fusion factor, derived from deep learning models extracting information from multi-frequency volume-price data[16][23] 3. Strategy Rules: - Select the top 5 industries with the highest scores on the last trading day of each week - Equal-weight allocation - Buy at the next week's first trading day's closing price - Weekly rebalancing, no transaction costs considered[23] - Model Evaluation: The model leverages AI's feature extraction capabilities to identify patterns in multi-frequency volume-price data, complementing top-down strategies[16] AI Theme Index Rotation Model - Model Name: AI Theme Index Rotation Model - Model Construction Idea: The model uses a full-spectrum volume-price fusion factor to score 133 thematic indices and constructs a weekly rebalancing strategy, selecting the top 10 indices for equal-weight allocation[2][6][9] - Model Construction Process: 1. Index Pool: Includes 133 thematic indices tracked by thematic ETFs, based on Wind's ETF classification[9] 2. Factor: Full-spectrum volume-price fusion factor, scoring each thematic index based on its constituent stocks[9] 3. Strategy Rules: - Select the top 10 indices with the highest scores on the last trading day of each week - Equal-weight allocation - Buy at the next week's first trading day's opening price - Weekly rebalancing, transaction costs set at 0.04% for both sides[9] - Model Evaluation: The model effectively identifies high-performing thematic indices using AI-driven factor scoring[6] AI Concept Index Rotation Model - Model Name: AI Concept Index Rotation Model - Model Construction Idea: The model uses a full-spectrum volume-price fusion factor to score 72 concept indices and constructs a weekly rebalancing strategy, selecting the top 10 indices for equal-weight allocation[11][15] - Model Construction Process: 1. Index Pool: Includes 72 concept indices based on Wind's popular concept indices[15] 2. Factor: Full-spectrum volume-price fusion factor, scoring each concept index based on its constituent stocks[15] 3. Strategy Rules: - Select the top 10 indices with the highest scores on the last trading day of each week - Equal-weight allocation - Buy at the next week's first trading day's opening price - Weekly rebalancing, transaction costs set at 0.04% for both sides[15] - Model Evaluation: The model efficiently captures trends in concept indices using AI-based factor scoring[11] AI CSI 1000 Enhanced Portfolio - Model Name: AI CSI 1000 Enhanced Portfolio - Model Construction Idea: The portfolio is constructed using the full-spectrum volume-price fusion factor to enhance the CSI 1000 index[3][26][29] - Model Construction Process: 1. Factor: Full-spectrum volume-price fusion factor, combining high-frequency deep learning factors and low-frequency multi-task learning factors[26] 2. Portfolio Construction Rules: - Constituent stock weight ≥ 80% - Individual stock weight deviation limit: 0.8% - Barra exposure < 0.3 - Weekly rebalancing, turnover rate controlled at 30% - Transaction costs set at 0.4% for both sides[29] - Model Evaluation: The portfolio demonstrates strong enhancement capabilities relative to the CSI 1000 index, with high IR and controlled tracking error[28] Text-based FADT_BERT Stock Selection Portfolio - Model Name: Text-based FADT_BERT Portfolio - Model Construction Idea: The portfolio is based on the forecast_adjust_txt_bert factor, which upgrades text factors in earnings forecast adjustment scenarios[32] - Model Construction Process: 1. Factor: Forecast_adjust_txt_bert factor, derived from text analysis of earnings forecast adjustments[32] 2. Portfolio Construction Rules: - Top 25 stocks from the long side of the factor's base stock pool - Active quantitative stock selection strategy[32] - Model Evaluation: The portfolio effectively integrates text-based factors into stock selection, achieving high long-term returns[32] --- Model Backtesting Results AI Industry Rotation Model - Annualized return: 24.95% - Annualized excess return: 20.80% - Maximum drawdown of excess return: 12.43% - Excess Sharpe ratio: 2.00 - YTD return: 4.88% - YTD excess return: 1.11%[1][22][25] AI Theme Index Rotation Model - Annualized return: 16.03% - Annualized excess return: 13.10% - Maximum drawdown of excess return: 16.55% - Excess Sharpe ratio: 1.02 - YTD return: 9.86% - YTD excess return: 10.43%[2][8][9] AI Concept Index Rotation Model - Annualized return: 22.42% - Annualized excess return: 12.68% - Maximum drawdown of excess return: 17.96% - Excess Sharpe ratio: 1.07 - YTD return: 11.48% - YTD excess return: 7.61%[11][13][15] AI CSI 1000 Enhanced Portfolio - Annualized return: 17.31% - Annualized excess return: 22.17% - Annualized tracking error: 6.07% - Maximum drawdown of excess return: 7.55% - IR: 3.65 - Calmar ratio: 2.93[3][28][30] Text-based FADT_BERT Stock Selection Portfolio - Annualized return since inception: 39.29% - Annualized excess return since inception: 31.74% - Maximum drawdown: 48.69% - Sharpe ratio: 1.36 - Calmar ratio: 0.81[32][36][38]