XGBoost Ensemble Signals for Intraday Stock Trading

Summary

This document describes a Chinese stock strategy that combines classification and regression models to forecast intraday returns. It uses XGBoost and an ensemble of machine learning models, drawing on 15 features related to overnight returns, opening auction activity, order imbalance, price limits, and early price movement. The document reports which features were associated with positive or negative intraday returns, and notes a cyclical relationship for one feature measuring changes in the best bid and ask midpoint.

For each day, the strategy selects the strongest 2% of signals, buys equally weighted positions at the open, and sells at the close. The reported out-of-sample period is January through October 2019, with a stated two-sided transaction cost of 0.2%. The document reports a 57.24% win rate, 130.2% annualized return, 4.31 Sharpe ratio, and 18.9% maximum drawdown. These are results from a limited historical sample; the summary provides no further detail on validation design, market impact, or whether the results generalize to other periods.

Key ideas

• The strategy combines XGBoost classification and regression models to estimate intraday stock returns.
• It uses 15 features, including overnight returns, auction activity, order imbalance, and early price behavior.
• The strategy buys the strongest 2% of daily signals at the open and sells at the close using equal weights.
• The document reports out-of-sample results for January to October 2019 with two-sided transaction costs included.
• The reported performance comes from a short historical period and may not generalize to other market conditions.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.