Training a Stock Ranker with an Overnight Return Factor

Summary

The article describes training BigQuant’s StockRanker machine-learning model using an overnight-return factor, following earlier factor research that links this signal to informed trading. Training uses all A-share stocks from 2013 through the start of 2021. The reported evaluation covers February 2021 to May 2022, comparing a broad-market universe with selected transportation, chemical, nonferrous-metal, and auto-related industries. The post says the industry-focused selection performed better, but supplies no numerical performance statistics in the text.

The author suggests combining the factor with industry selection or rotation and pairing it with momentum or reversal signals. The rationale is that the overnight-return measure reflects informed-trading behavior over roughly the preceding 20 trading days; the author also highlights sharp moves around late April 2022 as a period to examine. The comparison is limited: the industry universe was chosen using research reports and expectations, ST stocks were excluded, and the single-factor setup differs from a broader multi-factor selection process. The article notes that it concerns an older platform version, further limiting direct reproducibility.

Key ideas

• The model trains a stock-ranking system with an overnight-return factor on A-share data.
• The stated training window runs from 2013 to early 2021, with evaluation through May 2022.
• The author reports stronger results for selected industries than for the broad market, without giving numerical metrics.
• The article recommends testing the factor alongside momentum or reversal signals and industry rotation.
• Industry choice involved judgment, and the article describes an older platform version.

Tags

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