Research
We publish what we measure, including the results that do not flatter us. Everything below is a negative result: the validation signal our own pipeline selects on does not predict forward return, and neither does the obvious replacement for it. These are research papers, not advice, and none of them is a reason to trade.
Publications
Preprint · 10 August 2026
Selection, Not Strategy
Meet Zoro Research · 1,971 blind out-of-sample tests · 8,046 walk-forward observations · 59 crypto and US equity instruments
Fourteen strategy families were tested on data their parameters never saw, across two asset classes. Crypto fell 63.1% and the strategies returned −28.2%; US equities rose 22.1% and the strategies returned −22.1% — the same loss in opposite markets, which identifies apparent defensiveness as low market exposure rather than skill. A profitable configuration existed for 58 of 59 instruments, and the selection protocol almost never chose it.
Study I
Blind out-of-sample matrix
1,971 tests · 14 strategy families · 59 instruments · 12 unseen months
Selection was pinned to a cutoff twelve months in the past, then every family's champion was frozen and replayed over the year that followed. In-sample metrics barely rank the future, and the embargoed holdout — the stage that exists to correct overfitting — ranks it worse than the training return it corrects. Bar size mattered more than the choice of strategy.
Read the resultsStudy II
52-week walk-forward
8,046 picks · 52 weekly anchors · 53 instruments · one week ahead each time
If fitted metrics do not predict, ranking candidates by their realised out-of-sample return might. It does not. The correlation between a configuration's ranking-window return and the week that follows is −0.059, mean return is not monotone in rank, and an equal-weight portfolio of every pick compounds to −7.69% over the year.
Read the results