How we think about the search.
We are not a product. We are a quantitative research firm that invests its own capital, alongside a small number of partners, in the strategies our pipeline produces. These are the principles that govern what we publish, what we ship, and what we discard.
Scale is the honest answer.
Most ideas do not survive contact with the market. The only intellectually honest response is to test enough of them that the survivors are signal rather than story. Working at this scale is not a flex. It is the price of admission.
Decay is the default.
Every edge erodes. The only question is how fast. We assume any strategy we deploy today will be retired, and we plan our research pipeline around the steady cost of replacing it. A book of permanent winners is a story other people tell.
Validation is the hard part.
It is easy to fit a model. It is easy to deploy one. The discipline, and most of the work, is in killing the overwhelming majority that look promising in-sample and have nothing real underneath. We discard more than we keep, by a wide margin, on purpose.
Machines search, humans choose.
Compute can reach hypothesis spaces a desk could not explore by hand. It cannot decide what is worth searching for, what is interesting about a result, or when to walk away. We pair both deliberately. Each, alone, produces noise.
Out-of-sample is the only judge.
In-sample performance is a courtesy of the optimiser. We treat it as a starting point, never a verdict. A strategy that has not survived data it has never seen has not earned a position. Most do not earn one.
We say what we know.
We publish what we are learning, not what we are doing. The methods stay in-house; the conclusions do not have to. When we are uncertain, we say so. When we are wrong, we revise. There is no other way to run a research firm honestly.
Compute is the substrate.
The search we run is not possible without the machines it runs on. We measure working memory in terabytes, historical data in petabytes, and our compute is the most powerful available for AI inference and training. What we can look for is bounded by what we can run, we have chosen to invest accordingly.
OnlyAlpha does not sell signals, manage retail accounts, or solicit speculation. Capital is committed alongside a small number of partners. Everything you find on this site is what we have chosen to publish, never what we have.
The pipeline that does not stop.
Continuous hypothesis generation, statistical validation, out-of-sample testing, regime gating, and post-deploy monitoring – all running in parallel across markets, timeframes and instruments. Strategies survive on stability and retention, not in-sample flash.
Coming soon →