02 / 97
Adverse Selection
Working definition
The systematic tendency to transact precisely when the counterparty knows something you do not — filled when the market is about to move against you, missed when it would have moved in your favour.
Adverse selection is the oldest tax in market microstructure. A resting quote is a free option granted to everyone who can see it: the better-informed and the merely faster will exercise it exactly when it is mispriced, and leave it alone when it is not. Classical market-making theory — Glosten and Milgrom’s sequential trade model is the canonical treatment — shows that part of the bid-ask spread exists precisely as compensation for this: the market maker loses to informed flow and recoups it from uninformed flow, and the spread is where the arithmetic balances.
The concept cuts in both directions. From the maker’s side, adverse selection is being picked off on stale quotes by counterparties with a faster view of the same market — the risk that motivates defensive machinery such as quote staleness tolerance, spread widening into events, and hold windows like last look. From the taker’s side, the same logic inverts: when a counterparty holds discretion over whether a trade completes — a hold window, a requote, an internalised book — the trader’s fills are no longer a random sample of the trader’s orders. The orders that complete are disproportionately the ones the counterparty was happy to be on the other side of.
That skew has a measurable signature: the markout. Track the market’s drift in the seconds and minutes after each fill; under symmetric execution, post-fill drift should average near zero net of spread. A fill population that systematically drifts against the trader immediately after execution is the fingerprint of being the adversely selected party — selected into bad trades, filtered out of good ones. Measuring it requires no theory of intent, only timestamps: fills, quotes, and a reference feed. The Latency Arb Tracker I am building is intended to compute exactly those adverse-selection signatures from a trader’s own execution records.
Commonly confused with
Neighbouring concepts that get used interchangeably, and the distinction that actually separates them.
- Slippage
Slippage is a price difference on the trades you got. Adverse selection is about which trades you got at all — a filtering of your order population rather than a repricing of it. A trader can show unremarkable slippage while being adversely selected, because the worst outcomes were never fills.
- Market impact
Impact is the price moving because your order consumed liquidity — your own footprint. Adverse selection is the price moving because the other side knew something. One is caused by you and the other is done to you, and they call for opposite remedies.
- Last look
Last look is one mechanism by which a counterparty exercises discretion over which of your orders complete. Adverse selection is the resulting skew in your fill population. Naming only the mechanism misses that requotes, internalisation and asymmetric rejection produce the same signature.
- A run of bad luck
Both feel identical from inside a losing week, and they separate cleanly in aggregate. Bad luck is symmetric — post-fill drift as often helps as hurts. Adverse selection is a population-level skew that persists across enough fills that chance stops being a plausible account of it.
How to measure it in your own data
A definition you cannot test is a definition you have to take on trust. This is the shortest honest route from the concept to a number you computed yourself.
- Records you need
Fills with timestamps, the quotes standing at those moments, and a reference feed for the same windows. No theory of anyone's intent is required and none should be assumed — the measurement is arithmetic on timestamps.
- What you compute
The markout: track the market's drift in the seconds and minutes after each fill, and take the distribution across the whole fill population rather than any individual trade.
- What the answer tells you
Under symmetric execution, post-fill drift should average near zero net of spread — prices move both ways after you trade. A fill population that systematically drifts against you immediately after execution is the fingerprint of being the adversely selected party: selected into the bad trades and filtered out of the good ones. The size and consistency of that drift is the measurement; the reason for it is a separate conversation your counterparty may have an answer to.
If this has already cost you
Post-fill drift across your whole fill population is the measurement that separates being unlucky from being selected.
- Execution-Chain Attribution“Was it the EA, the connection, or the broker?”Will not establish: Fault. The chain attribution shows where the divergence entered, not who owes you for it — a wide fill can be an honest market, a slow hop can be your own VPS, and the report distinguishes what the logs prove from what they merely suggest.
Intake is not open yet, so none of these can be commissioned today. They are listed here so you know the measurement exists and what it would and would not settle — the launch list hears first.
Questions and answers
What is a markout?
The market's drift measured over a short horizon after each fill — seconds to minutes. It is the standard instrument for detecting adverse selection because it asks the one question that separates being unlucky from being chosen: does the market reliably move against you immediately after you are filled.
Is adverse selection something my broker does to me?
The concept cuts both ways, which is why it is worth understanding rather than resenting. From a market maker's side, adverse selection is being picked off on stale quotes by faster counterparties — the risk that motivates spread widening, staleness tolerance and hold windows in the first place. From your side, the same logic inverts wherever a counterparty holds discretion over which of your orders complete.
Why is part of the bid-ask spread compensation for adverse selection?
Because a resting quote is a free option granted to everyone who can see it, and the better-informed exercise it exactly when it is mispriced. Classical market-making theory — Glosten and Milgrom's sequential trade model is the canonical treatment — shows the maker loses to informed flow and recoups it from uninformed flow, with the spread as the point where the arithmetic balances.
Can I measure adverse selection from ordinary retail records?
Yes, provided your records keep fill timestamps and you can obtain a reference feed for the same windows. That is the appealing part: the signature is computable from data you already have, and it requires no access to your counterparty's systems and no assumption about what anyone intended.
Related terms
Derived from the links this entry makes and the entries that link back to it.
In the research
Adverse Selection comes up in three research notes on this site.