The cost that is in no fee table
Every maker knows the fee schedule by heart. Almost none can tell you what their fills cost them in the ten seconds after they happen, and that number is larger.
What it is, in one sentence
Adverse selection is being filled by a counterparty who knows more than you about the next few seconds. Your resting order is taken, and the price immediately continues in the direction that hurts you.
It is not bad luck and it does not average out. Informed flow selects your order precisely because it is mispriced relative to what is about to happen, so the losses cluster instead of cancelling.
- It is systematic, not random: the fills you get are the ones someone wanted.
- It shows up as a price move after the fill, called markout.
- It never appears on a fee statement, which is why it is underestimated.
The measurement that changed how we quote
In August 2026 we tested quoting further from the touch on a metals book, expecting more fills to mean more spread captured. The opposite happened, and the mechanism was unambiguous: cost per million of turnover went from 143 to 350, and the maker share collapsed from 99.8% to 81.9%.
Quoting deeper does not get you filled more often in calm markets. It gets you filled when the market is moving hard, because that is the only time the price reaches your level. The position then builds to twice the intended size, and the emergency exit crosses the spread at taker rates. We rolled the change back within twenty minutes.
- Deeper quotes fill at the worst moment, not more often.
- Cost per million rose 2.4x; the maker share fell 18 points.
- The conclusion was not 'quote tighter'. It was 'fix the exit first'.
Three things that reduce it
Nothing removes adverse selection; it is the price of providing liquidity. But three levers move it measurably, and none of them is about being cleverer at pricing.
- Shorten hold time. The damage is done in the seconds after the fill, so the fastest exit wins.
- Cap inventory. Once a position exists, stop the side that would grow it; only the reducing side keeps quoting.
- Requote often. Following the mid at 0.2 bp instead of 1.0 keeps you from being the stalest order in the book.
How to tell whether it is hurting you
Measure the price move in the seconds after each fill, signed by your side, and average it. If that markout is consistently negative and large relative to the spread you captured, adverse selection is your problem and no fee tier will fix it.
One warning from our own mistakes: an unsigned metric is blind to selection. Averaging the size of the move without its direction hides the whole effect, because informed and uninformed fills look identical by magnitude.
- Sign the markout by your side. Unsigned averages hide the effect entirely.
- Compare it against the spread you earned, not the quoted spread.
- Segment by hold time; the tail is where the damage concentrates.
Questions
What is markout?
The price move after your fill, measured at a fixed horizon such as ten seconds and signed by the side you took. Negative markout means the market kept moving against you after you were filled, which is adverse selection expressed as a number.
Can adverse selection be eliminated?
No. It is the cost of standing in the book with a resting order. It can be reduced by exiting faster, capping inventory and requoting more often, and it can be made survivable by trading a market where the spread is worth more than the damage.
Does quoting tighter make it worse?
Usually the opposite. Quoting deeper is what fills you during fast moves. In our own test, moving away from the touch raised cost per million from 143 to 350 while cutting the maker share from 99.8% to 81.9%.
Is it worse on DEX perpetuals than on centralised venues?
It differs by book rather than by venue type. What matters is how much informed flow the market carries relative to its spread. A thin book with a wide spread can be safer than a deep book with a one-tick spread.