Quantitative trading research
Edge isn't designed.
It's selected.
Evolution has no opinion - it has results. We apply the mechanism behind natural selection to find trading edge that survives out of sample. Sometimes it doesn't. You'll know either way.
Our process
Define the question.
Evolutionary search finds what survives.
Every feature in your data can be transformed, lagged, and combined with every other. You tested the ones that made sense to you. The algorithm tests the rest.
The mechanism is borrowed from biology. Features compete like traits in a population - the fittest survive out of sample, pass their characteristics to the next generation, and mutate. The weak ones die off. No hypothesis. No researcher picking winners. Just the data.
What survives isn't what you'd choose by hand. It's what the data couldn't kill.
Domain expertise
You know how markets work - but you've never been able to test it
Skew reads, fair value surfaces, vol regime timing, spread relationships - translated into testable hypotheses and validated out of sample.
Target selection
You've optimized parameters - but never tested the question
Sizing, Sharpe, trade frequency, book width - alone or combined. Different targets explore different edges. Most people optimize one and never question it.
Production integration
You've found edge and need it running live
Data pipelines, historical-to-live matching, real-time execution. From data ingestion to deployed system - not a PDF.
If this sounds familiar - here's what the engagement looks like:
a research sprint that tests your edge, and if it survives,
a production system that trades it. See services
Why the results hold
Decision-time data
Features are stamped to what you'd actually know at decision time. Fills are modeled at realistic execution windows - not the close that generated the signal.
Execution realism
Fill simulation models adverse selection - when your signal is strongest, liquidity is thinnest. Costs scale with conviction, not just volume.
Holdout validation
Walk-forward holdout. The search never touches OOS data - not for feature selection, not for parameter tuning, not for stopping rules. One pass. No reruns.
Most strategies don't survive the process.
The ones that do are worth trading.
Ready to test the question?
Start a conversation