AI复盘之精选30策略组合：深度学习研究报告

Quantitative Models and Construction Methods - Model Name: AI Look Chart Factor Construction Idea: Utilizes image recognition technology to predict future stock price trends based on standardized price-volume data charts, replacing traditional human-based technical analysis[17][21][25] Construction Process: 1. Input standardized price-volume data charts, which include 60 trading days' data consisting of K-line charts (open, high, low, close prices), daily trading volume, and MACD information[17][18] 2. Use a convolutional neural network (CNN) based on ResNet residual structure to process the input data[21] 3. Train the model using Adam optimizer and validate with external datasets to determine the optimal early stopping point[21] Evaluation: Demonstrates strong differentiation and monotonicity in factor performance, with significant long-short returns[52][55] - Model Name: Large-Cap Select 30 Strategy Construction Idea: Constructs a portfolio based on AI Look Chart Factor to achieve superior returns while controlling turnover[55][57] Construction Process: 1. Select stocks from the top 1000 market-cap stocks in the market[55][56] 2. Exclude delisted, ST/*ST stocks, and stocks with price limits[56] 3. Limit annual turnover to below 5 times[56] Evaluation: Achieved high cumulative and annualized returns with controlled risk metrics[69][73] Model Backtesting Results - AI Look Chart Factor: - Factor performance shows clear differentiation across 50 partitions, with significant long-short returns[52][54] - Large-Cap Select 30 Strategy: - 2020-2024: Cumulative return 244.56%, annualized return 29.42%, max drawdown 19.80%, annualized volatility 18.24%, IR 1.61[69][70] - 2024: Annualized return 40.88%, max drawdown 8.82%, annualized volatility 20.47%, IR 2.00[69][70] - 2020: Annualized return 33.20%, max drawdown 13.95%, annualized volatility 21.84%, IR 1.52[70] - 2021: Annualized return 40.13%, max drawdown 8.54%, annualized volatility 13.53%, IR 2.97[70] - 2022: Annualized return 15.16%, max drawdown 19.80%, annualized volatility 21.36%, IR 0.71[70] - 2023: Annualized return 20.40%, max drawdown 8.30%, annualized volatility 11.59%, IR 1.76[70] Quantitative Factors and Construction Methods - Factor Name: AI Look Chart Factor Construction Idea: Predicts future stock price trends using image recognition technology applied to standardized price-volume data charts[17][21][25] Construction Process: 1. Standardized price-volume data charts include K-line charts, trading volume, and MACD information[17][18] 2. Processed using ResNet-based CNN structure[21] 3. Optimized with Adam optimizer and validated with external datasets[21] Evaluation: Exhibits strong differentiation and monotonicity in factor performance, with significant long-short returns[52][55] Factor Backtesting Results - AI Look Chart Factor: - Factor performance across 50 partitions shows clear differentiation and monotonicity, with significant long-short returns[52][54] Strategy Performance Metrics - Large-Cap Select 30 Strategy: - 2020-2024: Cumulative return 244.56%, annualized return 29.42%, max drawdown 19.80%, annualized volatility 18.24%, IR 1.61[69][70] - 2024: Annualized return 40.88%, max drawdown 8.82%, annualized volatility 20.47%, IR 2.00[69][70] - 2020: Annualized return 33.20%, max drawdown 13.95%, annualized volatility 21.84%, IR 1.52[70] - 2021: Annualized return 40.13%, max drawdown 8.54%, annualized volatility 13.53%, IR 2.97[70] - 2022: Annualized return 15.16%, max drawdown 19.80%, annualized volatility 21.36%, IR 0.71[70] - 2023: Annualized return 20.40%, max drawdown 8.30%, annualized volatility 11.59%, IR 1.76[70]