Does Copy Trading Work on Polymarket? What 54,000 Copies Showed

Copy trading is sold as the shortcut: find a winning wallet, mirror it, share the profits. Since every Polymarket trade settles publicly on-chain, it is technically easy to do - so we tested whether it actually pays. We ran a shadow experiment across roughly 54,000 simulated copies of leaderboard traders and measured the result. The findings were not what we hoped, and they are more useful than a success story.

The headline result: broad copying is flat

Copying leaderboard traders as a group produced an average edge of about +0.03 percentage points, with an uncertainty band far wider than the result itself. In plain terms: statistically indistinguishable from zero. And zero is not neutral - after fees it is a loss. A trader-quality filter we applied on top made no significant difference either, which was the first uncomfortable finding: the thing everyone does - scroll the leaderboard, pick the impressive names, mirror them - has no measurable edge behind it.

The ranking methods that failed

We then tested the obvious ways to pick better candidates. Every one of them died on a large sample. Ranking by past ROI did not predict future results. Neither did trade frequency, entry price bands, market category, or copy lag within the ranges we could achieve. We also tested exiting when the leader exits - the intuitive move - and simply holding to resolution beat it.

The common thread is hindsight. A rule fitted to past data will always look good on that data; the question is whether it survives contact with the future. Almost none did. This is the same sample-size discipline that runs through reading smart money and the time-of-day study - and it is easy to preach and painful to apply to your own ideas.

Two methodology lessons worth more than the result

Measure edge in probability points, not ROI. Edge - your win rate minus the price you paid - turned out roughly three times less noisy than return, which means you need far fewer observations to see whether something is real. If you evaluate anything on these markets, evaluate it this way.

Verify forward, not backward. The only method that survived is unglamorous: fix your candidate list, publish it to yourself, then judge each candidate purely on what happens afterwards. It is slow - weeks per wave - and it kills most candidates. That is the point. Every shortcut we tried was a way of asking the past a question it always answers yes to.

The bar you have to clear

Before any of this matters, the arithmetic sets a floor. With a taker fee around 1.25 percentage points at mid-range prices, plus any additional builder fee, a copied strategy needs roughly 1.5 points of measured edge just to break even. That is a high bar, and it explains why so many copy-trading records look profitable in screenshots and lose money in practice - the screenshots rarely subtract costs. The full cost breakdown is in fees, deposits and withdrawals, and why cost moves your breakeven is in the guide to edge.

So should you copy anyone?

Possibly - but only as the output of a process, never as a reaction to a leaderboard position. If you want to try, the honest procedure is: vet a wallet properly using the wallet check method, write down your candidates before you copy anything, judge them only on what happens next, measure in edge points, and size small enough to survive being wrong - see position sizing. The mechanics of what changes between their result and yours are covered in copy trading.

We are publishing this because the null result is the useful part. Anyone can show you a wallet that went up; almost nobody shows you the test that failed. If you want to run your own version, the raw material is public - start from the leaderboard, the base rates on records, and the measurement approach in Polymarket analytics.