The Hidden Hand of the Market: Understanding Market Makers and Inventory Risk

The last part followed derivatives as the shadow engine of markets, their flows reaching the visible price through funding, liquidations and creations. This part steps closer to the screen, to the participant who stands between buyers and sellers. The market maker is usually cast as a referee, neutral and impartial. Look behind that neutrality and you find someone with a stake in every trade they make.

This page is meant to be come back to. It runs from the referee story, through what the spread pays for (its three components, each with its own cause), how inventory pushes a dealer's quotes around, why liquidity disappears exactly when it's wanted, what happens off the exchange, and how a dealer's hedging carries a trade from one market into another. Then it works through 6 May 2010 as a case study, and ends with where these readings fail and a checklist for reading a market through its dealers.

#The Illusion of the Referee

We're told the market maker is a silent guardian of liquidity: the neutral hand that keeps the game flowing, so that buyers always find sellers and sellers always find buyers. It's an appealing story, because a referee means rules and rules mean fairness.

A market maker stands inside the game. It quotes both sides, and it profits from the gap between them on every trade, whatever the asset is worth. It adjusts those quotes to its own position. A dealer that fills client orders directly sees that flow before the rest of the market does. Liquidity is a service it sells, at a price it sets, and it stops selling it when the price of providing it rises too far.

None of that is corruption. Markets need someone willing to stand in the middle, and nobody does it for free. The point of this part is to see the price, and who pays it.

#The Spread as Price of Admission

Every market has a toll, and in trading the toll is the spread, the gap between the best bid and the best ask that Part 3 defined. Cross it with a market order and you pay it, to whoever was resting on the other side.

The spread bundles several costs. Market-microstructure research splits it into three, and the split explains how spreads behave:

• Order processing. The fixed cost of being a market maker: technology, exchange fees, staff, capital. It's roughly the same in calm and stress.
• Inventory. Every fill leaves the dealer holding something it didn't choose, exposed to the price moving before it can lay the position off. The more volatile the market, the more that holding costs, and the wider the spread has to be to pay for it.
• Adverse selection. Some of the flow crossing the spread knows more than the dealer does. A dealer that sells to someone who knows the price is about to rise loses on the trade. Glosten and Milgrom showed in 1985 that this alone is enough to create a spread, even with no other costs: the dealer widens the spread so the profit from uninformed flow covers the losses to informed flow.

How the three divide depends on the market. Huang and Stoll's 1997 study of large US stocks found a large order-processing component and smaller, but significant, inventory and adverse-selection components. That's the calm-market picture. The stress picture is different, because the two components that grow with volatility and uncertainty are inventory and adverse selection, and adverse selection grows fastest when everyone suspects someone else knows something.

So the spread is a measure of risk perception. In calm markets it narrows to a tick, the illusion of a frictionless market. When storms gather it widens, because what the dealer believes it must charge to protect itself has gone up, whatever the asset is worth. The toll at the bridge is set by whoever controls the bridge.

#Inventory and the Skew

A dealer's job doesn't end with a quote. The moment it buys from a seller or sells to a buyer, it inherits risk, and that risk sits on its book as inventory. Buy too much and it's heavy with exposure; sell too much and it's dangerously short. So it leans.

The standard model of this, Avellaneda and Stoikov's 2008 paper building on Ho and Stoll's work from 1981, gives the dealer a reservation price, the price at which it's indifferent to its current inventory, shifted away from the market's mid in proportion to how much it holds:

where is the mid price, the inventory (positive when long), the dealer's aversion to risk, the volatility and the time left in its horizon. The dealer quotes around rather than around . A long dealer's reservation price sits below the mid, so its whole quote moves down: its offer becomes the cheapest on the screen, its bid the least attractive, and the flow it attracts takes the inventory off its hands.

This is a hidden push on price. A dealer that's overexposed in one direction and quietly skewing to unwind moves the quotes everyone else sees, and what looks like momentum can be a market maker cleaning its book. Where a few firms supply much of a venue's resting liquidity, their skews add up. The chart shows price and volume; the inventory behind them sits off its edge.

#Adverse Selection and Withdrawal

Every market maker lives with a quiet fear: that the person trading against it knows more than it does. Most of the time the flow is noise: retail orders, funds rebalancing, hedges. Now and then it's a trader who knows something, or a fund moving ahead of news. Sell too cheaply to that trader, or buy too dearly from them, and the dealer inherits the loss.

The spread is the first defence. The second is to stop quoting. When volatility spikes and rumours swirl, the fear of being picked off can become larger than anything the spread can pay for, and market makers pull their quotes and step back. Buyers and sellers still exist. What disappears is the willingness of anyone to stand between them, and the continuous market everyone assumed turns out to have been conditional all along. That's why depth thins in crashes faster than any single seller could explain, and it's the subject of the case study below.

On 11 October 2025, the morning after crypto's largest recorded liquidation day, I put the same mechanism in a short post on X: as forced sells push price into more stops, "liquidity thins as market makers pull inventory", and arbitrage carries the wave between perpetuals and spot. CoinShares' own account of that day also put the violence of the move down to market makers' behaviour. The liquidation side of that day is in Part 5.

#Maker Rebates

Some venues pay market makers to rest orders: a maker rebate, a small credit per trade for providing liquidity, funded by the fee takers pay. It's a subsidy for exactly the behaviour this part describes, and in quiet conditions a disciplined maker can live on it. I did, briefly, years ago, turning about $30 into about $2,000 by posting passive liquidity on BitMEX while it paid rebates; the arithmetic and the kill-switches are in A Small Edge, Carefully Kept. What that taught me is the same thing the spread decomposition says: the rebate pays you for the fills you want and does nothing about the ones you don't. Inventory and adverse selection are still yours.

#OTC Desks: The Shadow Extension

Plenty of rebalancing happens out of sight. A block too large for the visible book goes to an over-the-counter (OTC) desk, which negotiates a price with the client privately. There, size can move without disturbing the surface, and the desk prices the risk it's taking on, with its own interest in the price as much as on the exchange.

The trade doesn't stay off the book for ever. The desk now holds the client's position, and it hedges or unwinds it on public venues, usually in pieces and over time, in spot, futures or perpetuals. So an OTC trade reaches the public market later and in disguise, as a run of ordinary-looking orders that nobody outside the desk can connect back to the block.

How much of the market trades this way, and what that does to the price you see, is Part 7's subject.

#Hedging: The Dealer's Side

A market maker rarely leaves itself exposed for long. Every position it takes on has to be offset somewhere, and that offsetting is how a trade in one market becomes a trade in another. A dealer long spot after a client sale may short a perpetual to neutralise it. A dealer that's sold a client exposure through a derivative buys the underlying to cover it. A fund's authorised participant creating shares buys the underlying, which is a dealer's hedge by another name.

Part 5 describes these channels from the instruments' side. From the dealer's side, they share one property: the hedge is compelled by the dealer's balance sheet, whatever the dealer thinks of the market. When the flows are small, that's background noise. When they're large and one-sided, the hedging becomes a force of its own, and what looks like conviction in the price is a book being balanced.

#Reference

#Case Study: 6 May 2010

The flash crash of 6 May 2010 is the cleanest public record of what happens when market makers step back. In about twenty minutes that afternoon, US share prices fell sharply and recovered, briefly erasing $862 billion in value. Part 3 covered the spoofing case tied to that day. The dealers' side of it is in the joint SEC and CFTC staff report.

What the regulators found was a liquidity event. As prices fell and automated systems paused, some market makers withdrew from the market completely and left behind only automatically generated stub quotes: placeholder bids of a penny and offers as high as $100,000, posted to meet the obligation to quote both sides without any intention of trading there. With the real bids gone, market sell orders went looking for the best bid left in the book and found the stub. Shares of Accenture, a large and liquid company, traded at a penny.

The SEC's staff wrote that the largest and most erratic price moves that day were caused by withdrawals of liquidity and the execution of trades at stub quotes, and executions against stub quotes were a significant proportion of the trades later broken. In November 2010 the SEC banned stub quotes, from December requiring market makers' quotes to sit within 8% of the national best bid or offer for stocks covered by the new circuit breakers, and within 30% for the rest.

That case says something the rest of this part only argues. A market maker's quote is a promise to trade only as long as trading suits it. On an ordinary day the promise holds and the referee story looks true. On 6 May 2010 the promise was withdrawn in minutes, and what was left in the book was a set of placeholder quotes that were never meant to be filled.

#Where It Fails

• "The dealer moved the price" is usually unprovable. Inventory skew, hedging and withdrawal are real mechanisms, and from outside you see only their footprint. Attributing a specific move to a specific dealer needs data you don't have.
• Spread decompositions vary. The split between order processing, inventory and adverse selection depends on the market, the period and the model. The shape of how they change with stress is robust; the percentages aren't.
• Not every wide spread is fear. Thin assets, odd hours and tick sizes widen spreads with no informed flow in sight.
• Competition narrows the toll. Where many makers compete, spreads compress towards cost. The tilt is largest where dealers are few and can step away together.
• Withdrawal can be prudence. A dealer that keeps quoting into a crash can fail, and a failed dealer provides no liquidity at all. Stepping back is a choice the rules allow, and sometimes the right one.

#Checklist

1. What's the spread now, and how does it compare with its usual width in calm conditions?
2. Is the depth near the top of the book thinning while volume rises? That fits dealers stepping back.
3. Which way are quotes leaning relative to recent flow? A persistent skew is consistent with inventory being shed.
4. Who are the market makers on this venue, and how many of them are there?
5. Does the venue pay maker rebates, and does it impose quoting obligations?
6. Is there a large trade likely to have gone through an OTC desk, and where would its hedge land?
7. If the market moves violently, which venue's book will empty first?

#Conclusion: The Hidden Hand

If you look only at price, you'll miss the hand that shaped it. The market maker plays the game it appears to referee. Its spread is a toll priced for its own risk. Its inventory leans on every quote. Its hedges carry trades from one market into the next, and its OTC desk keeps the largest trades off the screen until they leak back in pieces. And its liquidity, which everyone treats as a property of the market, is a service that's withdrawn when it stops paying.

None of that makes the market maker a villain. Markets need intermediaries, and intermediaries charge. Seeing the hand is what lets you price it: knowing when the toll is cheap and when it's about to rise, and knowing that the depth you're leaning on belongs to someone who can take it away.

Series note: Part 6 of Understanding Market Mechanics. Previous: Part 5 · The Shadow Engine of Markets. Next: Part 7 · The Market's Blindfold. The practical companion is Applied Market Mechanics.

---

Understanding Market Mechanics describes how markets work, published purely as education. Nothing in it is financial advice or a recommendation to buy or sell anything, and I'm not a financial adviser. Every case study is drawn from the public record and has already resolved, and nothing in a part should be read as a statement of what I hold.