Making markets
Market making
A market maker stands ready to buy at one price (the bid) and sell at a slightly higher one (the ask). When both sides trade, it keeps the difference. The business is millions of tiny, repeated edges – and the craft is not losing them to people who know more than you do.
At a glance
At a glance
- Makes money from
- The bid–ask spread, plus rebates some exchanges pay for posting orders
- Holds positions for
- Seconds to minutes
- Trades
- Shares, futures, currencies, bonds, crypto – anything quoted electronically
- Biggest risk
- Adverse selection: being filled just before the price moves against you
- Who does it
- Electronic market makers such as Citadel Securities, Virtu, Optiver, IMC and Jane Street
- $0.0030 – US cap on the fee to take a share (falls to $0.0010 in Nov 2027)
- $8.6m – Virtu’s net trading income a day (2025, firm-reported)
- ~35% – Citadel Securities’ share of US retail volume (firm-stated)
- 51–52% – trades that make money (reportedly, Virtu, 2014)
- 1 in 1,238 – Virtu’s losing days, 2009–2013 (from its IPO filing)
- Seconds – typical holding period
The business
Investors who want to trade now – a pension fund rebalancing, a retail investor buying a few shares – would rather not wait for someone with the opposite wish to turn up. The market maker is always there, and charges for it through the spread. Quote 100.00 to buy and 100.02 to sell, trade with a buyer and a seller, and you have made 2 cents a share without any view on where the price is going.
Who is on the other side shapes the whole business. Exchange-listed market makers post quotes in public order books and compete on price and on their place in the queue; many sign up to exchange programmes that oblige them to quote, in return for better fees. In the US, most retail orders never reach an exchange: brokers route them to wholesalers – Citadel Securities, Virtu, G1 and a few others – who fill them from their own inventory and often pay the broker for the flow. The SEC estimated that more than 90% of marketable retail orders went to six wholesalers in early 2022.
Where the orders come from
The edge per trade is tiny, so the business is volume and consistency. Virtu, one of the few listed market makers, reported adjusted net trading income of about $8.6 million a trading day in 2025, and its 2014 listing documents showed one losing day in 1,238. Its chief executive was reported in 2014 as saying only 51–52% of its trades made money: the law of large numbers, not a high hit rate, is what makes the business steady.
How big the business is
| Firm | Figure | Year | Source |
|---|---|---|---|
| Virtu | Total revenue $3.63bn; net income $912m | 2025 | Annual report |
| Virtu | Adjusted net trading income about $8.6m a day | 2025 | Firm-reported |
| Citadel Securities | About 35% of US-listed retail share volume | Current | Firm website |
| Citadel Securities | Net trading revenue about $12bn | 2025 | Press reports |
| XTX Markets | Net revenue about £3.9bn across its UK entities | 2025 | Company filings |
The market maker’s quotes in the order book
| Bid size | Price | Offer size |
|---|---|---|
| 100.04 | 1,800 | |
| 100.03 | 1,200 | |
| 100.02 | 900 | |
| 100.01 (market maker’s quote) | 500 | |
| spread 0.02 · fair value 100.00 | ||
| 500 | 99.99 (market maker’s quote) | |
| 1,100 | 99.98 | |
| 1,400 | 99.97 | |
| 2,000 | 99.96 | |
A trade, step by step
The textbook picture – buy at the bid, sell at the ask, keep the spread – hides most of the work. A real round trip involves a fill you did not choose, a price that moves while you hold the shares, and quotes moved to get rid of them.
Worked example
One round trip in a $100 stock
Fair value is 100.000. You quote 500 shares bid at 99.99 and 500 offered at 100.01. The exchange pays a maker rebate of $0.0025 a share for resting orders that are traded against, and clearing costs $0.0004 a share. All figures are illustrative.
- A seller hits your bidBuy 500 at 99.99; you are now long 500 sharesrebate +$1.25
- One second later, the mid has slippedMarkout −$2.50
- Skew the quotes against the positionLower both quotes a cent: bid 99.98, offer 100.00still long
- A buyer lifts your offerSell 500 at 100.00; you are flatrebate +$1.25
- Trading profit+$5.00
- Rebates less clearing+$2.10
Net +$7.10. The one-second markout said the purchase was a bad one – the price fell – but skewing the offer down a cent got the shares out at a profit anyway. Had the price kept falling, the skew would have sold them at a loss rather than holding them longer, which is the point: a market maker earns the spread, not a view.
- Estimate fair value continuously from everything that leads the price: trades on other exchanges, the futures market, related ETFs, the imbalance between bids and offers in the book (the microprice).
- Quote around it – tighter when the market is calm and the flow is uninformed, wider when it is volatile or a news event is due, and not at all through scheduled releases like the 8:30 New York economic data.
- Skew for inventory: long shares you did not want, lower both quotes so the next trade is more likely a sale.
- Hedge what you cannot shed quickly, often with a future or ETF that trades more cheaply than the stock itself.
- Mind the queue. Under price-time priority the front of the queue fills first, from small and mostly uninformed orders; fills at the back of the queue come disproportionately from large orders sweeping the level, which are more often informed.
The maths that matters
Three models from the academic literature are the vocabulary of the trade, and interviewers use their logic even when they do not name them.
Adverse selection (Glosten and Milgrom, 1985). Even a dealer who expects to make no profit quotes a spread, because some traders know more than it does. The ask is what the asset is worth given that someone wants to buy, , and the bid what it is worth given that someone wants to sell. The more informed traders there are, the wider the spread.
Inventory (Avellaneda and Stoikov, 2008). A market maker who dislikes risk quotes around a reservation price that moves against its position: , where is the mid, the inventory, risk aversion and the variance. The total spread is , where measures how quickly fills fall off as quotes move away from the mid.
Quotes skew against inventory
Informed trading (Kyle, 1985). With one informed trader hiding among noise traders, prices move by per unit of net order flow, : the more noise trading there is, the deeper the market and the more the informed trader earns – at the noise traders’ expense.
Where the money comes from
Profit on 1,000 shares bought and sold
Market makers separate the spread from adverse selection with markouts: the profit on each fill measured against the mid a fixed time later – 100 milliseconds, a second, a minute. A fill that looks profitable at the moment it happens and loses money a second later was against someone who knew something.
Markouts: what a fill is worth, measured later
Rebates matter more than they look. US exchanges may charge at most $0.0030 a share to the side that takes liquidity, and many pay part of that to the side that provided it – a maker rebate of $0.0020–$0.0030 in 2025 schedules. On a stock with a one-cent spread that is a large share of the edge. The SEC voted in 2024 to cut the fee cap to $0.0010 and to allow half-cent ticks for the most liquid stocks; after delays, both are due in November 2027.
Payment for order flow runs the other way: it is a cost to the wholesaler, paid to the broker for routing it orders. An SEC committee memo in 2016 put typical equity payments at $0.0010–$0.0031 a share. Wholesalers pay it because retail flow is mostly uninformed – the cheapest flow there is to trade against.
Risk and what goes wrong
- Informed flow: a run of trades on one side, just before news, leaves you holding the wrong inventory.
- Gaps: the price jumps past your quotes before you can cancel them.
- Technology: a bad release or a runaway algorithm can do more damage in minutes than a year of spreads earns.
- Competition: as more firms compete for the same flow, spreads narrow and the edge per trade shrinks.
The Flash Crash · 6 May 2010
A fund’s algorithm sold 75,000 E-mini S&P contracts, about $4.1 billion, at 9% of volume with no price limit. High-frequency firms passed contracts among themselves, the E-mini fell more than 5% in four minutes, and the Dow was briefly down nearly 1,000 points. Some stocks traded at a penny against placeholder "stub" quotes, which were banned afterwards.
Knight Capital · 1 August 2012
A faulty software release sent more than 4 million orders in 45 minutes while trying to fill 212 customer orders, trading about 397 million shares. The loss was more than $460 million, and the firm was rescued and later merged. The SEC fined it $12 million under its rule on market access controls.
The Treasury flash rally · 15 October 2014
The ten-year yield fell 16 basis points and came back within twelve minutes, with no news. A joint official report found principal trading firms did more than half the volume and order books thinned, but no single cause.
GameStop and payment for order flow · January 2021
Retail trading in a few stocks surged; one large broker restricted buying after its clearing deposit jumped from $125 million to $1.4 billion. Congressional hearings and an SEC staff report put payment for order flow under scrutiny; reforms proposed in 2022 were withdrawn in 2025.
Technology and data
A market maker is a software business. Market data arrives from every venue as binary messages; the system rebuilds each order book, updates fair value, decides on quotes and sends them – thousands of times a second, with risk checks on every order that US rules require before it leaves the firm.
- Servers in the exchanges’ own data centres – CME at Aurora, Illinois; US equities at Mahwah, Carteret and Secaucus in New Jersey – so the signal travels metres, not miles.
- Network cards that bypass the operating system, pinned processor cores and code with no memory allocation or locks on the path from market data to order.
- Hardware (FPGA) for the most time-critical parts: decoding the feed and firing pre-set orders.
- Research on full message-level data, with simulations that reproduce queue position – a backtest that assumes you are always filled is the classic way to fool yourself.
Who works here
Trader
Owns a group of products: sets spreads, skew, size and limits, watches markouts and fill quality, and decides when to widen or pull quotes around events. Talks to researchers when profit drifts, and handles the auctions and the exceptions the system cannot.
Quant researcher
Builds the short-horizon fair-value and signal models – order-book imbalance, lead–lag from futures, trade flow – and the models of fills, markouts and inventory risk. Tests them on message-level data with realistic queue simulation.
Developer
Builds the feed handlers, order gateways, pre-trade risk checks, the simulation and backtest platform and, at the fastest firms, the hardware logic. Measures latency with hardware timestamps, and ships changes through staged roll-outs – the lesson of Knight.
A day on the desk
Before the open: overnight profit and markout reports, risk limits and any new code. The opening auction, then the session, with quotes widened or pulled around scheduled data. Research and code review in the quiet middle of the day; the closing auction; then a review that splits the day’s profit into spread, rebates, adverse selection and hedging cost.
In interviews for this desk
Expect to make markets on unknown quantities, adjust after being traded against, and reason about who is on the other side. Fast, accurate mental arithmetic and expected value come up in nearly every process.
Make me a market on the sum of two dice.
The mean is 7 and the variance , a standard deviation of about 2.4. Quote around the mean, say 6.5 bid, 7.5 offered. Then ask what matters: how big is the trade, and who is asking? If you are lifted repeatedly by someone who may have seen the dice, widen and move up.
An asset is worth 100 or 110 with equal chance. A fifth of traders know which; the rest buy or sell at random. Where do you quote, expecting no profit?
Given a buy, informed traders buy only if it is 110: and , so and the ask is . By symmetry the bid is 104. The spread, 2, is the informed share times the range – more informed traders, a wider spread.
With a mid of 100, inventory long 10, , , one period left and , where are your Avellaneda–Stoikov quotes?
The reservation price is . The total spread is , so quote about 99.858 bid and 100.062 offered. Being long moves both quotes down.
You buy 1,000 shares at 99.99 with the mid at 100.00. A second later the mid is 99.97. What is your one-second markout, and what does it say?
: $10 of spread captured at the fill, then $30 lost to the move. A pattern of fills like this means you are being selected against – widen, shade the bid or look at who is trading with you.
Each fill earns half a cent of spread plus a 0.2-cent rebate. A "toxic" fill then moves 2 cents against you; the rest do not move. What share of toxic fills can you afford?
Break-even is , so . If the fee cap falls to $0.0010 and the tick to half a cent, the edge per fill is at most about 0.35 cents and the break-even share halves to about 17.5%.
In Kyle’s model, what happens to price impact and the insider’s profit if noise trading doubles?
Price impact halves – the market is deeper – and the insider’s expected profit, , doubles. The extra profit is paid by the noise traders.
Why does your place in the queue matter, beyond how often you are filled?
The front of the queue is filled by small orders, which are mostly uninformed. The back is reached mainly when a large order sweeps the level, and large sweeps are more often informed – so back-of-queue fills have worse markouts. Queue position is as much about who you trade with as how often.
Firm guides: Citadel Securities · Optiver · IMC Trading · Jane Street · Flow Traders
Glossary
- Bid–ask spread
- The gap between the best price to sell at and the best price to buy at.
- NBBO
- The national best bid and offer across all US venues.
- Maker–taker
- Exchange pricing that pays a rebate to resting orders and charges the orders that take them.
- Access-fee cap
- The legal maximum fee to take liquidity in US shares: $0.0030 a share, falling to $0.0010 in November 2027.
- Tick size
- The smallest price step: a cent in US shares, half a cent for the most liquid from November 2027.
- Payment for order flow
- What a wholesaler pays a broker for sending it customer orders.
- Internalisation
- Filling customer orders from the firm’s own inventory rather than on an exchange.
- Adverse selection
- The expected loss from trading with people who know more than you do.
- Markout
- A fill’s profit measured against the mid a fixed time later.
- Inventory skew
- Moving both quotes to steer the position back towards flat.
- Queue position
- Your place in line at a price under price-time priority.
- Microprice
- A fair-value estimate that weights the bid and offer by the size waiting on each side.
Further reading
- Trading and Exchanges: Market Microstructure for Practitioners – Larry Harris, 2003. The standard introduction to who trades, why, and how markets are built
- Algorithmic and High-Frequency Trading – Álvaro Cartea, Sebastian Jaimungal and José Penalva, 2015. The mathematics of optimal quoting and execution
- High-frequency trading in a limit order book – Marco Avellaneda and Sasha Stoikov, 2008. The inventory model above
- High frequency trading and the new market makers – Albert Menkveld, 2013. One real high-frequency market maker’s trades, studied in detail
- Findings Regarding the Market Events of May 6, 2010 – CFTC and SEC staff, 2010