Zero fees don't make alpha: a cost-floor experiment on a 0% perp DEX
If your edges die net of fees, a zero-fee venue should resurrect them. We ran that test on a 0% maker/taker DEX. Nothing came back — because the fee was never what killed them.
Run enough relative-value census on crypto and you accumulate a stack of verdicts that all rhyme: positive gross, negative net. The signal makes money in the abstract and loses it once you subtract the fee and the spread. Stare at that stack long enough and a tempting inference forms — the fee is the marginal killer, the thing standing between the strategy and a profit. If that were true it would be testable, because there are venues now that charge zero maker and zero taker fee. Move the same signals to a zero-fee perp DEX, set cost to the spread alone, and the ones that were dying by a fee's width should cross into the black.
We ran it. Nothing crossed.
The experiment
Four families — cross-sectional cointegration, momentum, funding-carry, and an intraday session reversion — on a universe of crypto majors, identical engines and identical universe to the fee-venue runs, with the fee term zeroed so the only cost charged was a spread proxy. A clean A/B: same signals, same data shape, fee removed. If the fee was the binding constraint, this is where the edges reappear.
What came back: nothing, and why
Each family failed on the zero-fee venue, and each failed for the same reason it had failed everywhere else — a reason the fee had nothing to do with:
- The cointegration line that looked like a 3.0 Sharpe at hourly bars was a stale-close artifact; it collapsed the moment the bars were coarsened, fee or no fee.
- Momentum was sign-unstable across sampling frequency — strongly negative at one horizon, positive at another — the signature of a best-of-grid fit and regime beta, not a factor.
- The session reversion was real in direction but far too small to clear even the spread, and did not survive multiple-testing control.
- Carry's best line was a single-name selection on a handful of trades — an anecdote dressed as a result.
Removing the fee changed none of those diagnoses. It could not, because none of them were about the fee.
The error in the reasoning
The inference “negative net of fee, therefore fee-limited” quietly assumes that a backtest's gross is a real edge and the fee is what eats it. Usually that is the wrong decomposition. An apparent gross that goes negative once realistic costs are applied is, more often than not, not edge minus fee — it is an artifact whose fake gross simply happens to be smaller than the costs you finally charged it. Delete the fee and you are left with the artifact, exactly as untradeable as before, now flattered by a smaller cost line.
Put differently: a cheaper venue lowers the cost of a trade that is already good. It does not manufacture a good trade. The intuition “cheaper venue, more edges clear” is only true on the thin margin where a genuine edge sits just under the fee — and in our census that margin was empty. Every candidate that was negative net of fee was negative for a structural or measurement reason that survived the fee going to zero.
Where zero fee does matter
This is not an argument that zero-fee venues are pointless. Fees are decisive — for strategies whose binding constraint actually is the fee. A high-turnover passive book that round-trips the same edge thousands of times a day lives or dies on the per-trade fee and rebate, and a zero-fee or rebate-positive venue changes that economics completely. But that is a maker-economics question, not a relative-value-signal question; the signal's problem is whether there is an edge net of spread, and on klines-level relative value the fee was a rounding error next to the spread and the artifact. Zero fee is a genuine advantage stacked on top of a real edge. It is not a source of one.
What this is not
It is not a claim that no crypto relative-value edge exists — only a claim about a specific and common reasoning error. And it is not a claim that the zero-fee venue is otherwise uninteresting: removing the fee is exactly the right thing to do once you hold an edge that clears the spread, and the spread and depth on those venues are their own worthwhile measurement. The narrow point is this — if your edges only exist net of fees, the most likely explanation is not that the fee is in your way. It is that the edge is the fee: a number small enough to vanish the instant you price it honestly, and small for reasons a zero-fee venue leaves completely intact.