We test claims about markets.
There is a large published literature on what moves prices, and a much smaller set of it that still works once you take it out of the sample it was fitted on. Sorting one from the other is most of what we do.
Microstructure, flow, and the effects people still trade on.
Our work sits close to the tape. How order flow arrives and what it does to price. Where liquidity actually is, as opposed to where the book says it is. How volatility behaves around known events. And the long tail of published effects, drift after earnings, opening ranges, seasonality, positioning, that get repeated for years after the edge in them has gone.
We are not trying to discover a new anomaly. We are trying to find out which of the old ones are still true.
The same four steps, every time.
We take the paper at its word and write down exactly what it claims, in a form precise enough that it can fail. We rebuild the result from the data and the rules described, without shortcuts. Then we run it on periods the authors never saw. Whatever does not survive gets thrown away, and the fact that it did not survive gets written down.
The method in full
What we publish.
Short notes with the data, the rule and the source stated plainly enough that someone else can check them. Where a number comes from a simulation we say so, and we say what was simulated.
The rules we hold ourselves to.
- No performance claimsWe do not publish returns, and we do not describe any result as an edge you can expect to keep.
- Simulated is simulatedIf a number came from a backtest, it is labelled as simulated and out of sample, in the same sentence as the number.
- No recommendationsNothing we publish is advice, and nothing in it is a suggestion to buy or sell anything.
- Negative results countA study that found nothing is still a study, and it is the one most likely to save someone else the month.