Intraday Multi-Factor Optimization Shows a Sharp Post-Lunch Score Drop

Summary

The post reports an observation from more than six months of intraday modeling in Chinese equities using five-minute bars. At each time point, the author independently tuned parameters with tree-structured Parzen estimation. The reported optimization objective peaks in the morning, then falls sharply after the midday break; the afternoon values are described as roughly one percent of the morning peak. The factor set includes bar-level measures such as maximum gain, volume concentration, breakout, and relative strength.

The author says that fewer factor dimensions are available in the afternoon, but considers this insufficient to explain the size of the gap. The post poses possible explanations, including lower post-lunch liquidity or volatility, and asks whether the effect is specific to the Chinese market. It also raises separate afternoon factor sets or stop parameters as possibilities. These are questions rather than established conclusions: no controlled comparison, out-of-sample validation, or causal analysis is provided, and the optimized objective values alone do not demonstrate trading profitability.

Key ideas

• The author reports a pronounced decline in optimized objective values after the midday break in an intraday Chinese equity model.
• The study uses five-minute bars and independently tunes parameters at each time point with Bayesian optimization.
• The factor set includes bar-level gain, volume concentration, breakout, and relative-strength measures.
• The post proposes liquidity, volatility, and time-specific model design as questions for further investigation, not proven explanations.

Tags

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