Adverse Selection: The Maker's Nightmare

We’ve named adverse selection three times now — as a slice of the spread, as the leakage between effective and realized spread, and as the reason retail flow is prized. It’s time to make it the main character, because it’s the deepest reason markets have spreads at all. Strip away processing costs and inventory risk entirely, give a maker infinite capital and a perfect hedge, and it would still need a spread — purely to survive trading against counterparties who know more than it does. This lesson is about that irreducible danger and the elegant logic it forces onto every quote.

The lemons problem, applied to quotes

Analogy. The classic “market for lemons”: you sell used cars sight-unseen at one fixed price. Buyers who know their car is a gem (“a peach”) won’t sell to you cheap — they hold. Buyers who know their car is a dud (“a lemon”) happily dump it on you. So the cars you actually buy are disproportionately lemons — you’re adversely selected. To survive, you must lower your buy price to account for the lemon-heavy mix, which makes peaches even less likely to sell to you, and the market can collapse.

Definition. Adverse selection in markets is the maker’s version of the lemons problem: because the maker posts a fixed quote that anyone can hit, the counterparties who choose to trade against it are disproportionately the ones who know the quote is mispriced in their favour. When someone lifts your ask, it’s slightly more likely they know the price is about to rise; when someone hits your bid, slightly more likely they know it’s about to fall. The maker can’t tell informed from uninformed in the moment — it just sees an order. So every fill carries the risk of being on the wrong side of an imminent move.

Informed vs uninformed flow

Definition. Order flow splits into two idealized types:

• Informed flow — traders acting on an edge: real information, a better model, faster data, or a correct read of where the price is going. They trade selectively, only when the odds favour them. Trading against informed flow is, on average, a loss for the maker — the price tends to move against the maker right after.
• Uninformed (noise) flow — traders with no edge on direction: index funds rebalancing, retirees, people raising cash for a house, hedgers, the impatient. They trade for reasons unrelated to short-term price prediction. The maker profits from them, earning the spread with no systematic move against it.

The maker’s whole P&L is a tug-of-war: it earns the half-spread from uninformed flow and loses (typically more than the half-spread) to informed flow. The spread must be wide enough that the gains from noise traders cover the losses to informed ones. Raise the informed share and the spread must widen to compensate — or the maker bleeds out.

The simulator below makes this tug-of-war live. Drag the share of flow that is informed and the quoted half-spread, and watch the maker flip between profitable and bleeding. The grid shows incoming orders coloured by type; the readouts compute the profit per uninformed fill, the loss per informed fill, the blended net edge, and the break-even informed share at which the maker stops making money.

Two lessons jump out. First, a little toxic flow is survivable; a lot is fatal — past the break-even informed share, the net edge goes negative no matter how clean your operation. Second, the maker’s only defence is to widen the spread, which raises the profit per uninformed fill and the loss-cushion against informed ones. Everyone else then pays that wider spread — the informed minority taxes the uninformed majority through the maker.

The Glosten–Milgrom intuition: the spread IS adverse selection

Definition. A landmark result (the Glosten–Milgrom model, 1985) showed something startling: even a market maker that is risk-neutral, has zero inventory cost, zero processing cost, and earns zero profit (perfect competition) will still post a positive spread — entirely to break even against adverse selection. The logic:

• The maker sets its ask so that conditional on someone buying (which makes that buyer slightly more likely to be informed), the ask equals the expected value of the asset given that a buy arrived.
• It sets its bid symmetrically: the expected value given that a sell arrived.
• Because a buy is bad news for the seller (the maker) and a sell is bad news for the buyer (the maker), these two conditional expectations differ — and that difference is the spread.

In other words: the spread exists because the act of trading with you is itself information, and the maker must price that information into every quote. Strip away every other cost and the adverse-selection spread remains. It’s the floor under all spreads.

Worked example — break-even spread arithmetic

Suppose 80% of flow is uninformed (noise) and 20% is informed. When the maker trades with an informed trader, the price moves 8¢ against it on average. The maker quotes a half-spread h (so it earns h from each uninformed fill and loses 8 − h from each informed fill, since the informed trader still pays the half-spread before the 8¢ move). Break-even requires expected profit = 0:

• Expected profit per fill = 0.80 × h − 0.20 × (8 − h).
• Set to zero: 0.80h − 1.6 + 0.20h = 0 → h − 1.6 = 0 → h = 1.6¢.
• So the break-even quoted spread is 2 × 1.6 = 3.2¢, purely to offset adverse selection — before adding a cent for processing or inventory.

Now spike informed flow to 40%: 0.60h − 0.40 × (8 − h) = 0 → 0.60h − 3.2 + 0.40h = 0 → h = 3.2¢*, a 6.4¢ quoted spread. Double the toxic share and the adverse-selection spread doubles — even with everything else held fixed. That’s the whole story of why spreads explode around news: pending information spikes the informed share.

Toxic flow, the winner’s curse, and why news blows out spreads

Definition. Practitioners call informed, maker-hostile order flow toxic flow, and measure order-flow toxicity (e.g. the “VPIN” metric) as the imbalance of buys vs sells that signals informed trading. When toxicity rises, makers widen spreads, reduce size, or pull quotes — because each fill is more likely to be a loss. A spike in toxicity often precedes volatility, which is why toxicity metrics are watched as early-warning signals.

There’s also a winner’s curse flavour. If a maker is the one whose quote gets hit, that very fact is bad news: among all the makers competing, you were the one offering the price the informed trader wanted — i.e., the most mispriced quote. Winning the trade means you were the most wrong. Makers must price this curse into their quotes, quoting a touch more conservatively than naive expected value would suggest.

This is the deepest reason spreads blow out around scheduled news (earnings, central-bank decisions, economic releases). Before the announcement, the share of informed traders — and the size of the potential move they could exploit — both spike. The break-even spread balloons accordingly, and many makers simply stop quoting until the dust settles. The “no liquidity right before earnings” experience is adverse selection, quantified.

Tying it back: PFOF, and why retail is the dream counterparty

Now the PFOF story from earlier snaps fully into focus. Wholesalers pay brokers for retail order flow because retail flow is overwhelmingly uninformed — the cleanest, least-toxic flow there is. Filling it earns the spread with almost no adverse selection, so wholesalers can even offer price improvement (filling you slightly inside the public quote) and still profit handsomely. Meanwhile, informed flow gets routed to lit exchanges where it faces wider, more defensive quotes. The market quietly segments flow by toxicity: harmless flow gets pampered with tight fills, toxic flow gets the wide public spread. Adverse selection is the invisible hand sorting everyone.

Putting it together

Adverse selection is the irreducible reason spreads exist: because a maker posts a fixed, hittable quote, the counterparties who choose to trade against it skew toward those who know it’s mispriced in their favour — the lemons problem in market form. Flow splits into informed (the maker loses to it) and uninformed noise (the maker profits from it), and the spread must be wide enough that noise profits cover informed losses. The Glosten–Milgrom result proves the spread survives on adverse selection alone — zero inventory or processing cost needed — because the act of trading with you is itself information. When the informed share (toxicity) rises, the break-even spread balloons, which is why spreads blow out around news and why makers face a winner’s curse. And it all ties back to PFOF: harmless retail flow is the dream counterparty, so the market segments flow by toxicity, pampering the uninformed and charging the rest.

Key Takeaways