Who market makers are, and when they step away

A market maker is a firm that posts both a bid and an offer continuously, earning the spread and fee rebates in exchange for standing ready to take the other side. Nothing obliges them to keep doing it. Understanding the two risks that govern the business, inventory and adverse selection, explains most of what looks arbitrary about liquidity vanishing.

The economics of quoting

The naive description is that a maker buys at the bid, sells at the offer and pockets the difference. In practice the flow does not arrive symmetrically. When one side is being hit repeatedly, the maker accumulates inventory in a falling asset, and the mark-to-market loss on that inventory can exceed a great many captured spreads. This is inventory risk, and it is managed by skewing quotes, by reducing size, and by hedging.

Adverse selection is the sharper problem. A resting quote is an option that the market has been granted for free, and it is exercised most often by whoever knows something first. Every quote loses to informed flow and profits from uninformed flow, so the width of a spread is essentially a forecast of how toxic the arriving flow will be. This is why spreads widen ahead of scheduled events and during periods when information is arriving quickly, without any change in sentiment about the asset.

Venues subsidize the activity through maker and taker fees, paying or discounting the passive side and charging the aggressor. Tier structures reward volume, which concentrates the business in a small number of firms able to hit the top tiers across many venues.

The arrangements nobody sees

Two distinct businesses share the name. Proprietary firms quote for their own account on major pairs and are compensated purely by the spread and by fee tiers. Designated makers are engaged by a token issuer or a venue to support a specific market, typically under an agreement that specifies minimum quoted size, maximum spread and uptime.

The compensation structure in that second arrangement deserves attention. A common form is a loan of tokens from the issuer to the maker, combined with call options on those tokens at set strike prices. The maker is then quoting a market in an asset it has borrowed from the party that benefits from the market looking active, while holding options whose value depends on the price. These agreements are private, and the borrowed inventory is generally not disclosed as part of circulating supply, so measured float can be understated. When such an agreement ends, quoting can stop abruptly and borrowed tokens are returned or sold, which is a supply event unrelated to anything about the project.

Hedging is what makes continuous quoting possible

A maker holding unwanted inventory needs to neutralize it quickly, and the instrument of choice is usually a perpetual future on a liquid venue. That creates dependencies most observers never see. The maker needs margin available at the hedging venue, needs the funding rate to be tolerable, needs to be able to move collateral between venues, and needs the correlation between the hedged and quoted asset to hold. Capital is fragmented across venues that do not net against each other, and there is no central clearing and rarely a prime broker spanning them, so a firm's usable capital at any single venue is a fraction of its balance sheet.

When a hedge becomes unavailable or expensive, the quote goes with it. That is the mechanism connecting a stress event in one corner of the market to spreads widening in an asset with no obvious link to it.

The recognizable withdrawal triggers

• A volatility jump. Risk limits reduce size automatically, and a wider spread is the correct response to less certainty about fair value, not a judgment about the asset.
• A credit or custody event. If a venue's solvency is questioned, firms pull balances first and quote later. The 2022 failures produced exactly this sequence across venues that were themselves solvent.
• A stablecoin dislocation. Most quoting is against stablecoin pairs, so a depeg makes the unit of account itself uncertain and pricing anything in it temporarily meaningless.
• An outage or a withdrawal suspension. Inability to move collateral is functionally identical to not having it.
• A regulatory or listing shock. A firm that cannot be certain it may legally hold an asset stops holding it, and quoting requires holding it.
• The end of a designated agreement. Unlike the others this is scheduled, invisible from outside, and permanent.

The algorithmic alternative and its limits

On-chain, the equivalent role is played by a liquidity provider in an automated market maker, and the comparison is instructive because the two fail differently. A provider deposits inventory into a formula that quotes continuously and cannot be persuaded to stop, so on-chain quotes do not vanish during a volatility spike the way book quotes do. What happens instead is that the pool keeps offering a stale price and is systematically traded against until an arbitrageur brings it back, which transfers value from providers to informed traders rather than removing the quote. Providers can withdraw, but only by sending a transaction that competes for blockspace at exactly the moment blockspace is most contested. The two mechanisms therefore fail in opposite ways: one stops quoting, the other keeps quoting badly.

What withdrawal looks like in the data

The first sign is spread widening with unchanged price. The second is depth falling at the touch while the far levels stay populated, since firms cancel their tightest quotes first. The third is quote flicker, where prices update rapidly with almost no size behind them. Trade-by-trade the market takes on a gappy quality: prices jump between levels rather than grinding, and slippage on ordinary size rises even though reported volume may look normal or elevated, because forced and urgent trades still occur. A reader watching only 24-hour volume will miss all of this; volume can rise while liquidity collapses, and the two frequently move together in the wrong direction.

The general lesson is that measured depth describes willingness at a moment, not capital committed to a market. Liquidity is supplied by firms whose risk systems are correlated with each other, which is why it tends to be present in all venues at once or absent in all venues at once. The risk pages carry the liquidity measures per asset, on-chain venue activity shows where quoting happens algorithmically instead, and incidents catalogs episodes in which quoting stopped and what followed.