Symbolic-regression equity alpha

In plain terms

Instead of hand-writing a stock-ranking formula, this lets a genetic-programming search evolve one — then puts it through the hardest test the firm runs.

How it works

A symbolic-regression engine searches a large space of candidate formulas on a survivorship-correct panel; the best is then taken to a pre-registered, one-shot blind holdout it has never seen.

What it’s tested against

Strong in two samples — in-sample Sharpe 0.50 rising to 0.71 out-of-sample, with a Fama–French six-factor alpha of 7.2% a year — and then put through a blind holdout, because a one-shot holdout is the only honest verdict on a search this wide.

Data

Survivorship-correct, point-in-time WRDS US-equity panel.

Funding-grade — the engineering and the discipline are the asset; the blind holdout is the point.