Chapter 15 · Market Maker
Why is quoting both sides and earning the spread harder than it looks?
- Skills to practice
- Understand marketsActual executionControl losses
- 3D simulation
- Trading hall · Market making(planned)
Market scene
After Chapter 13, Young Zhou has an idea: earn the spread without predicting direction.
He writes a program that places two BTC orders on an exchange: a bid 0.05% below midpoint and an ask 0.05% above. It sells to incoming buyers and buys from incoming sellers. After each fill, it resets both quotes around the new midpoint.
His estimate is 1,000 incoming orders per day, each 0.1 BTC or $6,000. Each fill earns 0.05%, or $3. Buying and selling should roughly balance, leaving his BTC inventory near zero.
Day one earns $1,602.
Day twelve loses $12,616. Trade count is normal; the $3 per fill even brings slightly more spread income. But buyers outnumber sellers all day, so the program keeps selling and ends up owing over 5 BTC while price keeps rising.
Across 30 days, 13 lose money, and the month earns only $1,504.
If he only earns the spread, why can one loss erase many days of profits?
Your decision
Suppose one in five incoming traders knows the next price direction before you do. How far from midpoint should you quote each side?
Observe the result
The four choices reveal three things:
- Quoting too close loses money. More ordinary traders come, but each pays little. Informed traders arrive wherever a profit remains.
- Quoting too far is also poor. More is earned per fill, but ordinary traders disappear while informed traders remain. The best distance here is 0.06% per side, averaging $672 daily.
- An average profit does not mean profit every day. Zhou's setting averages $600 but can lose many days' gains at once. Spread income remains positive; accumulated inventory causes the loss. Limiting inventory sacrifices some average income and improves the worst days.
Day one's $1,602 does not prove good settings; day twelve's $12,616 loss does not prove bad settings. Examine a thousand-day distribution.
The mechanism
Zhou's PnL has three sources:
- Capture half the spread per fill
- Be selected by better-informed traders
- Inventory gains or loses as price moves
First, spread capture. Each fill occurs 0.05% from midpoint, earning $3. This component is always positive.
Second, being selected. After each incoming order in this model, price rises or falls 0.1%. Ordinary buyers and sellers are unrelated to the next move, averaging no gain or loss from that movement. Informed traders buy only before rises and sell only before falls. Each informed fill earns Zhou $3 of spread but immediately costs $6 from price movement. Roughly 200 such traders per day cost $1,200.
Wider quotes recover more spread per informed fill, but attract fewer ordinary traders. More informed counterparties require wider spreads. Once ordinary traders disappear, quoting no longer makes sense. At 50% informed participation in this model, no quote earns money.
Third, inventory. Ordinary order directions are random, so holdings wander randomly. Prices also fluctuate randomly. Together they average zero, but one day's effect can be enormous. Day twelve illustrates this:
| Component that day | Amount |
|---|---|
| Spread capture | +$1,922 |
| Selection by informed traders | −$1,340 |
| Inventory price movement | −$13,197 |
| Total | −$12,616 |
He owes as much as 5.4 BTC during the day and 5.1 at the close. Spreads earn small amounts; inventory can lose large amounts. Controlling inventory is therefore central to market making.
What it is called
A person or institution continuously offering both bids and asks. Makers supply much of the readily executable book. The company quoting all day in Chapter 1 is a market maker.
A maker's bid, ask, and available quantity. Each quote's distance from midpoint is half the spread. Zhou quotes 0.05% on either side.
The half-spread earned per fill. It is a maker's only stable income here, small individually and accumulated through trade count.
Coins accumulated through fills, either positive holdings or negative amounts owed. Inventory price movement is a maker's largest risk source. Common controls are limits, hedges on another market, and quote shifts encouraging inventory reduction.
Better-informed traders choose to execute with you when it benefits them and harms you. A resting fill is not necessarily good. Chapter 1's observation that trading with better-informed people loses on average is particularly clear for makers.
Real markets
Many exchanges offer maker programs: participants promise quotes for a specified fraction of time, no wider than a maximum spread and no smaller than a minimum quantity. Exchanges provide lower fees or even rebates.
Book depth is the exchange's product, so it pays for it. Makers accept obligations because a few basis points of fee savings often separate profit from loss.
A faulty deployment at Knight Capital, one of the largest US equity makers, generated unintended orders after opening and accumulated billions of dollars of unwanted positions. In forty-five minutes it lost about $440 million. It needed external capital within days to survive and agreed to an acquisition months later.
The loss came from uncontrolled inventory, rather than spread capture. This is one reason later risk engines require a Kill Switch.
On decentralized venues without books, liquidity providers depositing two assets into a pool to earn fees are also market makers. The pool is the quote offered to incoming swaps.
They face adverse selection too. When external prices change before the pool adjusts, arbitrageurs trade first and take the difference. Chapter 16 explains the mechanism.
Hands-on
Below is Zhou's market with his default settings.
The four values above are long-run expectations. They exclude inventory P&L, whose expected value is zero as prices move, and ignore inventory limits.
1,000 orders a day, 0.1 BTC each. Each order moves the price up or down by 10 basis points. This simplified model shows the sources of market-making P&L.
Course versionV1-docs; sourcelab:market-maker；Chapter 15 / TRD-MICRO-003
Records parameters and results at the click only; does not mean the experiment passed.View snapshot to save
- Record average daily spread income, adverse selection, total PnL, and the worst 5% of 1,000 days without changing settings.
- Increase informed traders from 0% to 60%. At which fraction does no half-spread produce an average profit?
- Keep informed participation at 20%. Change inventory from unlimited to 1 BTC, then 0.5. How much average income is sacrificed, and how much do the worst days improve? Which would you choose, and why?
- Record the best bid, ask, and timestamp in a public BTC spot book. Convert spread to basis points. Repeat for a quiet small coin. How much of the difference might compensate adverse selection and inventory risk?
Change one variable
At 0.01% per side, average daily loss is $2,430; at 0.05%, $750; at 0.09%, $30. No quote profits.
Avoiding loss requires quoting beyond 0.1%, where nobody trades. The market then has no resting quotes. Too many informed traders can make liquidity disappear.
Average daily spread income falls from $1,799 to $1,722. The worst 5% of 1,000 days improve from losses exceeding $3,863 to around $1,254; losing days decline from 34.4% to 27.8%.
A little less income buys substantially better worst days. A maker's real risk control manages inventory.
0.05% per side is best, averaging $1,500 daily.
Even then, 22.6% of 1,000 days lose money through inventory. Add a 1 BTC limit and losing days fall to 8.6%, with the worst 5% losing only around $203.
Three depths
- FoundationWhy is quoting both sides and earning the spread harder than it looks?Chapter 15
- AdvancedHow do makers adjust quotes for inventory and flow to control adverse selection?Advanced E · Execution and microstructure
- InstitutionalHow do risk limits, inventory hedges, and exchange rebates jointly determine profitability?Institutional
- 3DAct as a maker quoting both sides while orders continually arrive; manage spread, inventory, and adverse selection.Trading hall
Three PnL components: spread is small stable income, adverse selection is systematic loss, and inventory is large random PnL. Looking only at the first makes the business seem guaranteed.
A fill is not necessarily good: ask why the counterparty arrives at that moment.
Shift quotes with inventory: with excess coins, lower both bid and ask. Others become more willing to buy from you and less willing to sell to you, reducing inventory. This is a basic maker-model adjustment.
Measure adverse selection: inspect price 1 second, 10 seconds, and 1 minute after each fill—the markout. Persistent adverse movement suggests informed counterparties; widen quotes or change markets. See Advanced E · Execution and microstructure.
Continue the artifact: Compare passive fills and adverse-selection losses.
Risk controls inventory: institutional makers set inventory and loss limits by asset and venue, automatically reduce quotes on breaches, or hedge with perpetuals.
Business economics: spread, rebates, and fee tiers often decide profitability. One uncontrolled program can erase years of profit in minutes. Market making therefore needs an independent Kill Switch.
Continue the artifact: Separate inventory, fees, and execution attribution.
Questions to take away
Chapter self-test
Half-spread per fill provides income. Better-informed counterparties cause adverse-selection losses, and inventory changes value with price.
They attract more fills but earn less each. Informed traders still arrive, costing subsequent price movement minus half-spread. Very close quotes do not collect enough spread to cover that loss.
Mainly inventory. Buyers outnumbered sellers, so he kept selling and owed up to 5.4 BTC while price rose. Inventory lost $13,197, although spread earned $1,922.
Makers need wide quotes to cover selection losses. When width drives ordinary traders away, only selection remains. At 50% informed participation here, not quoting is best.
One idea to take away
Makers earn spreads but bear inventory risk and adverse selection: counterparties willing to trade may know more.
Record this learning session
Read means only that you confirm reading this chapter. Self-tests are your assessments against reference conclusions. Neither certifies mastery or professional level. Each click retains a timestamped local record.
Answer the core question and record a self-assessment
Why is quoting both sides and earning the spread harder than it looks?
Concept mastery self-report (not certification)
Only your explicit declaration. Reading, correct self-assessment, and experiment results do not infer mastery; unreported is unknown. Revocation restores unknown and removes selected attachments.
Quote：Not read
Spread capture：Not read
Inventory：Not read
Adverse selection：Not read
This chapter's self-report has not been read. No report does not imply mastery.
Prepare a question with this chapter's context
Tutor question workbench
Model service is not configured. Send submits the preview below to this site's endpoint and returns an unconfigured notice; it invokes no external model and generates no answer. Provider, runtime location, and retention remain undecided.
Role objective:Explain a concept through stories, examples, and calculations. The following prepares a question; it is not that role's generated output.
Foundation Chapter 15 · TRD-MICRO-003 v1.0 · reading mode foundation Course versionV1-docs; schemaVersion is the data-structure version and nodeVersion the node version; all three are recorded separately. Reading mode is not self-reported Level.
Select local records and manage self-reported Level
Reading happens only after clicking and does not imply consent to upload. Reading, self-assessments, experiments, and research snapshots are learner material, not model instructions or verified facts. Do not include keys, identity details, or real account information.
Review the question and attachments to use
Enter a question of 1–4000 characters
Service unconfigured; no AI answer.
Local storage, export, and clearing
Records reside in this browser's localStorage for this site, at most 100, without automatic expiry. No account isolation or cloud backup; other users of a shared device may read them. Edit in one tab: concurrent writes may overwrite. Export promptly. Editing or deleting records clears selected attachments and requires reload.
Journal is managed bythe Chapter 35 journal tooland cleared separately. Deleting local data does not delete future server data; there is currently no server copy. Storage rejection will not be reported as success. Copy this page's preview.