用机器学习解锁量化投资新边界

Core Insights - The article highlights the successful implementation of a day trading strategy using machine learning and pressure factors to identify trading opportunities in the futures market [2][4]. Strategy Overview - The day trading strategy, initiated in February 2023, employs a machine learning framework to analyze market data and predict daily returns for selected futures [2]. - The strategy focuses on 40 to 50 mainstream commodity futures but trades only the top five predicted by the model, optimizing for a higher Sharpe ratio during backtesting [2]. - Position allocation follows an "equal market value" principle, which has shown comparable performance to a "strong signal high position" model while simplifying operations [2]. Data Utilization - The strategy captures short-term fluctuations using 1-minute K-line data while also considering daily data for long-term trends, generating signals twice a day [3]. - The approach avoids frequent predictions to reduce model complexity, especially in the absence of significant incremental information during trading hours [3]. Risk Management - A multi-layered risk control framework is established, including mandatory position closing before market close to avoid overnight risks and immediate liquidation in case of market reversals [4]. - The strategy has demonstrated strong drawdown control, with a real trading drawdown rate of 5% to 6% and a maximum drawdown of 28.95% during a specific competition [4]. - The strategy is best suited for volatile markets, relying on a reversal effect, and incorporates traditional trend sub-strategies to mitigate risks [4]. Future Plans - The company plans to expand the trading universe to 10 to 15 products, which will enhance capital capacity while maintaining profitability through diversified order placements [5]. - A new product based on the day trading strategy has been registered, indicating a move towards a broader asset management market [5]. - The focus will remain on the commodity futures CTA sector, with ongoing investments in factor exploration and model optimization to ensure robust performance for clients [5].