Extending OpenFE with Rolling Futures Features and XGBoost Ranking

Summary

This article describes modifying OpenFE’s automated feature engineering workflow for futures research. It adds rolling mean, standard deviation, and sum operators across instrument groups, with preset window lengths, and explains how these operators are incorporated into candidate feature enumeration. The article also reviews OpenFE’s expand-and-reduce process: generate candidate features, screen them in stages, and retain useful additions. Its first-stage screening uses successive halving with incremental feature evaluation; the second stage considers interactions and feature importance in a LightGBM model. The author also discusses using XGBoost for feature importance assessment.

The evidence is a procedural walkthrough and code examples, rather than reported trading results or a measured comparison. The author says the second-stage evaluation is not being reworked, focusing customization on feature generation and first-stage evaluation. The proposed rolling calculations depend on grouping by instrument and on the chosen windows; the article does not discuss safeguards against look-ahead bias, out-of-sample validation, or the effects of data ordering. The method therefore outlines a research tool, not a validated trading strategy.

Key ideas

• OpenFE generates candidate features by combining base variables with operators, then reduces the candidate set through staged screening.
• The article adds rolling means, standard deviations, and sums with preset windows to the operator library.
• Rolling features are grouped by instrument and included in the feature enumeration process.
• Successive halving screens candidate features using incremental evaluation before a second stage considers interactions.
• The article provides implementation details but reports no trading results or validation evidence.

Tags

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