“一句话”自动回测框架

Quantitative Models and Construction Methods - Model Name: "One-sentence" automated backtesting framework Model Construction Idea: The framework leverages AI programming tools and a rules-driven workflow to transform natural language strategy descriptions into structured data queries, stock screening, portfolio construction, and backtesting results[7][10][13] Model Construction Process: 1. Natural Language Input: Users describe strategies in plain language, e.g., "monthly strategy, select stocks with market cap < 40 billion, ROE and ROA in the top 50%, and choose the 30 stocks with the lowest PE"[24] 2. Data Mapping: The system uses a standardized database query interface and a WIND data dictionary to map strategy elements (e.g., market cap, ROE, ROA, PE) to specific database tables and fields[7][13][22] - Example tables: - AShareEODDerivativeIndicator: Fields include S_VAL_MV (market cap) and S_VAL_PE (PE ratio)[24] - AShareFinancialIndicator: Fields include S_FA_ROE (ROE) and S_FA_ROA (ROA)[24] 3. Portfolio Construction: The system generates standardized portfolio data with three key elements: date, asset, and weight. It automatically adjusts for user-defined rebalancing frequencies (daily, weekly, monthly, quarterly)[13][15] 4. Backtesting: The framework runs backtests using the constructed portfolio and outputs performance metrics, risk analysis, and detailed reports[7][24] Model Evaluation: The framework is innovative in bridging natural language and structured data, enabling rapid strategy validation. However, its reliance on WIND data quality and AI model accuracy may introduce risks[7][13][24] Model Backtesting Results - "One-sentence" automated backtesting framework: - Annualized Return: 2020: 18.96%, 2021: 16.70%, 2022: 8.21%, 2023: 15.66%, 2024: 7.15%, 2025: 37.70%[29] - Annualized Volatility: 2020: 20.95%, 2021: 23.41%, 2022: 24.94%, 2023: 13.94%, 2024: 33.71%, 2025: 17.02%[29] - Sharpe Ratio: 2020: 0.91, 2021: 0.71, 2022: 0.33, 2023: 1.12, 2024: 0.21, 2025: 2.21[29] - Maximum Drawdown: 2020: -5.90%, 2021: -9.59%, 2022: -18.31%, 2023: -6.72%, 2024: -14.18%, 2025: -3.76%[29] - Win Rate: 2020: 66.67%, 2021: 66.67%, 2022: 66.67%, 2023: 50.00%, 2024: 50.00%, 2025: 75.00%[29] - Calmar Ratio: 2020: 3.21, 2021: 1.74, 2022: 0.45, 2023: 2.33, 2024: 0.50, 2025: 10.03[29] Quantitative Factors and Construction Methods - Factor Name: Small-cap value factor Factor Construction Idea: Select stocks with small market capitalization and strong financial performance, then rank by valuation metrics[24] Factor Construction Process: 1. Stock Pool Definition: Limit to stocks listed on Shanghai and Shenzhen exchanges with market cap < 40 billion[24] 2. Financial Screening: Filter stocks with ROE and ROA in the top 50% of the defined pool[24] 3. Valuation Ranking: Rank remaining stocks by ascending PE ratio and select the top 30[24] Factor Evaluation: The factor effectively combines size, profitability, and valuation metrics, aligning with traditional value investing principles[24] Factor Backtesting Results - Small-cap value factor: - Annualized Return: 2020: 18.96%, 2021: 16.70%, 2022: 8.21%, 2023: 15.66%, 2024: 7.15%, 2025: 37.70%[29] - Annualized Volatility: 2020: 20.95%, 2021: 23.41%, 2022: 24.94%, 2023: 13.94%, 2024: 33.71%, 2025: 17.02%[29] - Sharpe Ratio: 2020: 0.91, 2021: 0.71, 2022: 0.33, 2023: 1.12, 2024: 0.21, 2025: 2.21[29] - Maximum Drawdown: 2020: -5.90%, 2021: -9.59%, 2022: -18.31%, 2023: -6.72%, 2024: -14.18%, 2025: -3.76%[29] - Win Rate: 2020: 66.67%, 2021: 66.67%, 2022: 66.67%, 2023: 50.00%, 2024: 50.00%, 2025: 75.00%[29] - Calmar Ratio: 2020: 3.21, 2021: 1.74, 2022: 0.45, 2023: 2.33, 2024: 0.50, 2025: 10.03[29]