Genetic Programming for Intraday Equity Alpha Discovery

Summary

The document describes using genetic programming to discover stock-selection factors from intraday price and volume data. Factor expressions are represented as trees; successive candidate populations are produced through selection, crossover, and mutation. The input consists of 30-minute bars, and candidate factors are evaluated using information coefficient, long-side excess return, and the monotonicity of returns across factor-ranked groups.

The reported research uses data from 2017–2018 to search for factors and 2019 as an out-of-sample period. It says that 100 factors were identified, gives several example expressions, and reports that their selection performance persisted out of sample, with some decay from in-sample results. The factors are also described as having relatively low pairwise correlations. These results are a summary of the source study, not an independently verified replication. The document provides limited detail on implementation, transaction costs, and robustness beyond the stated sample, so the reported factor performance should not be treated as a guarantee of future returns.

Key ideas

• The method searches for intraday equity factors by evolving tree-structured expressions with genetic programming.
• It uses 30-minute price and volume data as model inputs.
• Factor fitness combines information coefficient, long-side excess return, and monotonicity across ranked groups.
• The study reports 100 factors discovered on 2017–2018 data and evaluated out of sample on 2019 data.
• Reported factor performance decayed out of sample, and the summary does not establish future profitability or account for all implementation costs.

Tags

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