Research
The method is public. The questions are not.
This page sets out how work is done here and the standard it is held to. What is currently being investigated stays private until it is finished, published, or abandoned — and a good deal of it ends in the third category.
Scope
Why there is no list of projects on this page
A line of enquiry is worth most before anyone else is walking down it, and in this field an idea can be handed over simply by being named. A published roster of open questions gives that away for nothing, to everyone, permanently.
So EdgeQuant does not disclose its active research directions: not the questions being asked, the data being assembled, the techniques under evaluation, nor the instruments, venues and time horizons in scope. Not in summary, not as a teaser, and not as a “coming soon” list. This is a considered policy rather than an oversight, and it is not going to change.
What does get published is the finished article — a written result with its method, its null model, its cost assumptions and its failure modes attached, at the point where it is complete enough to be argued with. Before that point there is nothing useful to say, and saying it anyway would cost more than it is worth.
At the broadest level the work sits in market microstructure and price formation: how prices come to be what they are, and how much of the conventional account of that survives being checked. That is as specific as this page gets, and enquiries asking it to be more specific are politely declined.
Standards
Methods, and why they are strict
| Practice | What it means here |
|---|---|
| Pre-stated null | The null model is chosen and written down before the result is looked at, and is constructed to preserve the uninteresting structure already present in the data while destroying only the effect being claimed. |
| Out-of-sample holdout | A period is set aside untouched and reported separately. If a result only exists in-sample, that is the finding. |
| Multiple-comparison honesty | The number of variants tried is reported, not just the one that worked. Testing forty things and reporting the best is not a result. |
| Realistic cost model | Spread, slippage, financing, borrow and tax applied before anything is described as an edge. Idealised fills are never used. |
| Reproducibility | Data-processing steps and parameters published with the write-up so a reader can disagree with the specific choices rather than the summary. |
| Negative results published | Ideas that failed appear alongside those that did not, with the same amount of detail. |
Publication policy
What gets published, and what deliberately does not
Findings are published as written notes on the blog, with methods and figures attached. Where a study was commissioned, only the parts the client has agreed to release are published, and the commissioning relationship is disclosed on the note itself.
Two things are never published: anything that would function as a trade recommendation, and any parameterisation packaged so that it could be run directly as a trading system. Research is published to be argued with, not deployed. This is a deliberate constraint on the work, and it rules out some otherwise interesting write-ups.
Corrections. Errors are fixed in place, with a dated note recording what changed and why. If you find one, say so — it will be credited unless you would rather it were not.