Building and Evaluating Intraday Factors in a Stock Competition

Summary

This guide explains the workflow for a BigQuant contest focused on constructing 15-minute factors from three-second stock snapshot data. Participants are asked to write factor logic without using model-generated factor synthesis. The platform provides isolated workspace access, data and compute resources, and sample order-book pressure factor templates. Submitted code must place its work inside the designated main function and return a correctly formatted datetime field for evaluation.

The guide describes evaluation outputs: long-short single-factor Sharpe, average excess return across 16 cutoff times, and the Sharpe of those excess returns, with portfolio and trade details available by cutoff. These are evaluation measures, not reported results or evidence that any factor is profitable. Scores may take hours to arrive, and failed submissions receive no score; the guide notes that the specific error is withheld to reduce data leakage. Its advice is specific to this contest and platform.

Key ideas

• The contest asks participants to build 15-minute stock factors from three-second snapshot data.
• The stated rules prohibit creating factors through model synthesis and require the submitted logic to run inside the main function.
• Factor output must include a datetime field with the required intraday time information.
• Evaluation includes long-short Sharpe, average excess return across cutoff times, and the Sharpe of those excess returns.
• Backtest details expose portfolio and trade information for each cutoff time, but the guide reports no factor performance results.

Tags

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