What the spread pays for
Buying at the ask and immediately selling at the bid loses money with certainty. There is no market condition in which that round trip is free, and the amount lost is the spread.
That looks like a toll, which invites the question of who collects it and for what. The answer is specific: the spread is the price of immediacy, paid by whoever demands it to whoever supplies it, and it covers three distinct costs borne by the supplier. Each of the three moves independently, and the spread you see is their sum.
Immediacy is the thing being sold
A resting order is an option granted to everyone else. By posting a bid you have committed to buy at that price if someone chooses to sell to you, at a moment of their choosing rather than yours, and you cannot withdraw the commitment for orders already matched against it.
The person crossing the book gets certainty about time. The person resting gets a worse price in exchange for accepting uncertainty about it. That transfer is the economic content of the spread, and it is why the roles are priced differently by fee schedules as well.
The conventional decomposition of what the resting side is being compensated for has three components.
Order-processing cost
The mechanical cost of being in the market at all: connectivity, systems, the operational overhead of maintaining quotes, and any per-message or per-fill charge the venue applies.
This component is largely fixed per unit of activity, which has one structural consequence worth noting. Fixed costs spread over more volume are smaller per unit, so this part of the spread tends to be smaller where activity is high — not because anyone is being generous but because the arithmetic is different.
It is also the smallest of the three in most conditions, and the least interesting, because it does not respond to what is happening in the market.
Inventory risk
Someone quoting both sides accumulates a position whenever the two sides do not fill evenly, and they did not choose that position — it is whatever the flow handed them.
Holding it has a cost, because the price can move while it is held. So the compensation demanded for resting an order rises with anything that increases the size or duration of unwanted inventory: a wider expected price range over the holding period, thinner conditions that make offsetting harder, and larger typical order sizes relative to what can be absorbed.
This is the component with the clearest mechanical signature. When prices are moving further per unit of time, the cost of being caught holding something rises, and the spread widens in step. Nothing about that requires anyone to have a view on direction — the exposure is undirected, and the cost is symmetric.
Adverse selection
The subtlest of the three, and the one that explains most of the spread’s odd behaviour.
A resting order is filled by whoever chose to trade against it. That selection is not random. Among all the participants who might cross your bid, the ones most likely to do so are the ones who think your bid is generous — and some fraction of them will be right for reasons you do not have access to.
Adverse selection is the cost of being systematically on the wrong side of that filter. It is not a claim that anyone is being cheated; it is arithmetic. A passive order transacts when someone else finds transacting attractive, so the population of your counterparties is biased by construction.
The compensation for it is a wider spread. And because the bias is worse when information is arriving quickly — when the price a moment from now is more likely to differ from the price now — this component widens sharply in exactly the conditions where the other two do.
The mechanism
THE MECHANISM — what the gap is charging for
· Crossing the book
→ you pay roughly half the spread
against the mid, and receive
certainty about timing.
· Resting in the book
→ you earn roughly half the spread,
and accept uncertainty about
whether and when.
· Fixed operating cost per unit
→ order-processing component. Falls
per unit as activity rises.
· Unwanted position from uneven fills
→ inventory component. Rises with
how far prices are travelling per
unit of time.
· Being filled by whoever chose to trade
→ adverse-selection component. The
counterparty population is biased
BY CONSTRUCTION.
· Resting to capture the spread
→ NO GUARANTEE of capture. The fill
may be one-sided, and the
position is the cost.
· Whether the venue also charges or
rebates on top, and any quoting
obligations
→ VENUE-SPECIFIC. Fee models change
the net of every row above.Worked example
Illustrative figures, synthetic throughout. Suppose a best bid of 40,000 and a best ask of 40,004 — a spread of 4, a mid of 40,002.
A taker buying 1 unit pays 40,004, which is 2 above the mid. A taker selling 1 unit receives 40,000, which is 2 below. Each crossing costs half the spread against the mid, and the round trip costs the whole spread: 4, before any fee.
Now trace the other side across two cases with the same quotes.
The even case. Someone rests both a bid at 40,000 and an ask at 40,004. One taker buys from the ask and another sells into the bid. The resting participant bought at 40,000, sold at 40,004, holds nothing, and is 4 better off. That is the spread being earned as advertised.
The one-sided case. Only the bid is hit. The resting participant now holds 1 unit bought at 40,000 and no offsetting sale, and the quotes have moved to 39,980 / 39,984 because the same flow that hit the bid moved the market. Marked at the new mid of 39,982, the position is 18 worse than the purchase price. The 2 of edge against the old mid has been swamped by an 18 move on a position that was not chosen.
Both outcomes come from the same quotes. The spread is the compensation for the second case occurring some fraction of the time, and a spread of 4 is only adequate if the one-sided case is rare enough and small enough. When either changes, the quote that makes sense changes with it.
Why spreads widen exactly when it is least convenient
Two of the three components — inventory risk and adverse selection — rise with the rate at which prices are moving. Both therefore rise together, and both rise fastest when depth is thinnest, because thin conditions make an unwanted position harder to offset and make each individual order more informative.
Which produces the pattern everyone notices and few people connect to a cause: the visible cost of trading rises at the moment the book is least able to absorb size, and it rises for whoever is crossing, not for whoever is quoting. There is no version of the mechanism where the price of immediacy stays constant while the cost of supplying it rises. Quotes widen, or they are withdrawn.
Withdrawal is the same phenomenon at its limit. A participant who cannot price the risk stops quoting, and displayed depth disappears rather than being traded against.
The failure mode
The spread is frequently treated as a fixed characteristic of a market — a number you can look up and budget for. It is a price, it is set by participants who can change or remove it, and it responds to conditions rather than describing them.
The specific way that bites: a cost estimate built on the spread you have usually seen is an estimate of the cost of trading in the conditions you have usually traded in. It has no validity in other conditions, and it is least valid precisely when the difference matters most.
And the other side of it is worth stating plainly, because the arithmetic above makes resting look like the profitable seat. Earning the spread requires being filled on both sides at prices that have not moved through you in between, and there is no guarantee of either. The compensation exists because the risk exists. It is not a free half-spread; it is a payment for taking something on, and this site has nothing to say about whether anyone should.