How Quantitative Market Makers Work: Strategy Explained

What Is a Market-Making Strategy?

A market maker provides immediacy. When a trader wants to buy or sell without waiting for a matching counterparty, the market maker takes the other side. In return, it seeks to earn the difference between its buying and selling prices.

That description is accurate but incomplete. The spread is compensation for risk, not free income. A market maker may buy just before a price decline, sell just before a rally, accumulate an unwanted position, or lose queue priority before its order executes. Quantitative market makers manage these problems through automated pricing and risk controls. Their systems estimate fair value. They predict short-term order flow. They place and cancel limit orders, manage inventory, and hedge exposures across correlated instruments.

A market-making strategy continuously offers to buy at a bid price and sell at a higher ask price. The difference is the bid-ask spread:

Spread(t) = Ask(t) − Bid(t)

Example. A market maker quotes a stock at $99.99 bid and $100.01 ask. One trader sells to the bid. Another later buys from the ask. The maker buys at $99.99 and sells at $100.01. The gross spread capture is $0.02 per share.

The outcome changes if only one side trades. If the bid fills and the market falls to $99.90, the maker holds a long position with an unrealized loss. Earlier spread income may be too small to offset that inventory loss.

Regulatory context. Under Exchange Act §3(a)(38), a market maker is a dealer who holds itself out as willing to buy and sell a security for its own account on a regular or continuous basis. For Nasdaq-registered market makers, Rule 4613(a) requires a continuous two-sided quotation within a “Designated Percentage” of the National Best Bid and Offer (NBBO) — 8% for Tier 1 LULD-covered stocks, 28% or 30% for Tier 2 stocks, and 20% for Tier 1 stocks outside the 9:30–9:45 a.m. and 3:35 p.m.–close windows. A market maker whose quote drifts beyond the “Defined Limit” must re-quote or be excused from the obligation under Rule 4619.

How Quantitative Market Makers Make Money

Bid-ask spread capture. This is the most visible revenue source. The maker buys at its bid and sells at its ask, ideally before the reference price moves materially. The gross figure must be reduced by fees, hedging costs, adverse selection, and inventory price changes.

Exchange rebates and fee schedules. Some venues use maker-taker pricing: a resting order that adds liquidity may receive a rebate, while an aggressive order that removes liquidity pays a fee. Under Regulation NMS Rule 610(c), the SEC reduced the access-fee cap for protected quotations from $0.003 to $0.001 per share (for stocks priced at or above $1.00). Compliance has been repeatedly deferred — most recently to the first business day of November 2027 — so current fee schedules may not yet reflect the lower cap. Rebates can improve marginal economics, but they do not rescue a strategy with consistently negative post-fill markouts.

Cross-venue matching and short-horizon signals. Large firms offset client or exchange flow across several venues. Most competitive systems also estimate whether fair value is likely to change over the next few milliseconds or seconds. The objective is usually to avoid stale quotes, not to build a large directional position.

The Quantitative Market-Making Decision Loop

Step 1 — Estimate fair value. The midpoint is the simplest reference price:

mid = (best bid + best ask) / 2

A stronger estimator may include order-book imbalance, recent trades, cross-venue prices, futures, ETFs, correlated securities, volatility, and scheduled events. A microprice weights the best prices by opposite-side depth. When displayed bid depth exceeds ask depth, the microprice moves closer to the ask, signaling short-lived upward pressure. This relationship must be re-estimated for each market state.

Step 2 — Calculate a defensible spread. The required spread should cover explicit fees, expected adverse selection, inventory cost, operational risk, and required margin. A wider quote earns more per round trip but executes less often. A narrower quote trades more frequently but offers less protection against informed flow.

Step 3 — Skew quotes according to inventory. Inventory is the net position created by executions. When inventory is long, the strategy lowers its reservation price. Its ask becomes more competitive, its bid less competitive, and bid size may fall. These changes encourage offsetting sell-side executions while discouraging additional purchases. A short position produces the opposite response.

Step 4 — Estimate fill probabilities. A quote has no value if it never executes. Fill probability depends on:

● distance from the best price

● quantity ahead in the queue

● trade-arrival intensity

● cancellation rate

● order size

● volatility

● venue latency

● quote age

A common classroom model assumes execution intensity — the expected number of fills per unit time, not a probability — declines exponentially with quote distance:

λ(δ) = A · exp(−k · δ)

To convert intensity into a usable per-interval fill probability, apply p = 1 − e^−λΔt, where Δt is the simulation time step. Treating λ directly as a probability, without this conversion, overstates or understates fills depending on the size of Δt and is a common implementation error.

Step 5 — Apply independent risk limits.

● Maximum position per instrument and aggregate exposure

● Delta, gamma, and vega limits for options

● Maximum order size, message rate, and daily loss

● Stale-data protection, price collars, and self-trade prevention

● Venue kill switches and global strategy shutdown

Step 6 — Hedge residual exposure. A maker can hedge an individual stock with index futures or a sector ETF, an option book with the underlying, an ETF with its basket, or crypto spot inventory with perpetual futures. Hedging reduces the targeted risk but adds transaction costs and basis risk (the risk that the hedge and the hedged position do not move together).

Inventory Risk and Adverse Selection

Marked-to-market wealth equals cash plus inventory multiplied by the current reference price. Spread capture increases cash, but inventory price changes can dominate it. A strategy can record many favorable round trips and still lose money when one-sided flow leaves it holding a large position.

Adverse selection occurs when the maker trades against a counterparty with better short-term information. A practical measure is markout — the signed price movement after a fill, defined as:

Markout(h) = side · ( Mid(t + h) − Fill )

where side = +1 for a maker buy and −1 for a maker sell, measured at fixed horizons after the fill (for example, 100 ms, 1 s, 10 s, and 60 s). Persistent negative markouts across these horizons indicate that executions arrive just before unfavorable price changes — a sign of “toxic” flow. When toxicity rises, the system can widen its spread, reduce size, cancel one side, hedge sooner, or stop quoting. A common trigger rule: widen quotes when the 1-second markout falls below −0.3 times the current half-spread over a rolling 500-fill window.

Queue Position and Latency

A limit order at the best bid does not necessarily execute when a sell order arrives — other orders may have time priority. A backtest that ignores quantity ahead will overstate fills. Latency adds a further risk: a faster participant may trade against a stale quote before its cancellation reaches the venue.

Production teams measure market-data latency, decision latency, order-entry latency, cancellation-acknowledgment latency, and fill probability conditional on queue position. Faster infrastructure does not, by itself, create a profitable signal. It prevents a valid pricing model from being undermined by stale execution.

The Avellaneda–Stoikov Model

Marco Avellaneda and Sasha Stoikov formalized a widely taught inventory-aware model in 2008. It chooses quotes that balance spread revenue against the risk of holding inventory. The model uses a reference price S(t), inventory q(t), risk aversion γ, variance σ², and remaining time T − t.

Reservation price:

r(t) = S(t) − q(t)γσ²(T − t)

If inventory is zero, the reservation price equals the reference price. A long position moves the reservation price lower; a short position moves it higher. The adjustment grows with inventory, risk aversion, variance, and remaining exposure time.

Optimal half-spread:

δ   = (1 / γ) · ln(1 + γ / k)  +  ½ · γ · σ² · (T − t)

bid = r − δ
ask = r + δ

Because the bid is offset below the reservation price by δ and the ask is offset above it by the same amount, the total quoted spread equals 2δ, not δ. Each of the two terms in δ plays a distinct role: the first term reflects execution-intensity sensitivity (governed by κ), and the second reflects inventory-holding risk (governed by γ and σ²). The table below shows how the balance shifts as k changes, holding γ = 0.1, σ² = 4 fixed:

k	T−t = 1.0	T−t = 0.5	T−t = 0.1
1.5	Intensity term dominates	Intensity term dominates	Intensity term dominates
15	Balanced	Balanced	Intensity term shrinks
150	Risk term dominates; fills become rare	Risk term dominates	Risk term dominates

The main contribution of the model is not a permanent parameter set — it is the principle that quotes should respond systematically to inventory, volatility, time, and execution intensity. Later work adds hard inventory constraints, multiple assets, directional signals, and more flexible price processes.

Worked example. With σ² = 4 (annualized), γ = 0.1, k = 1.5, and T − t = 1.0, the inventory skew for one unit is q · γ · σ² · (T − t) = 1 × 0.1 × 4 × 1.0 = 0.40, and the half-spread is (1 / γ) · ln(1 + γ / k) + ½ · γ · σ² · (T − t) = (1 / 0.1) · ln(1 + 0.1 / 1.5) + 0.5(0.1)(4)(1) ≈ 0.645 + 0.20 = 0.845, giving a total quoted spread of 2δ ≈ 1.69. As inventory grows or time-to-close shortens, both the skew and the spread widen further — this is the core mechanism the model exists to capture.

Why Textbook Models Fail in Production

● Reference prices contain jumps, discrete ticks, volatility clustering, and cross-asset dependencies.

● Orders cluster and respond to market state rather than arriving independently.

● Fill probability depends on queue position, not only distance from mid-price.

● Volatility and arrival parameters change across instruments, venues, and time.

● Firms optimize drawdown, inventory use, markout, and capital consumption alongside P&L — not P&L alone.

How to Backtest a Market-Making Strategy

OHLCV candles are rarely sufficient. A credible test needs timestamped order-book updates, trades with aggressor-side classification, venue fees, tick rules, queue estimates, realistic latency, partial fills, and inventory constraints. Use walk-forward evaluation and include volatile sessions, scheduled announcements, exchange interruptions, and thin trading periods.

Metric	Purpose	Formula / Threshold
Net P&L	Profit after fees, rebates, and hedging	Cash + inventory × price, net of costs; target positive over out-of-sample window
Maximum drawdown	Largest decline from a prior P&L peak	max(running peak − current P&L); set a hard stop-loss threshold
Maximum inventory	Worst position exposure	max(|inventory|); should stay within pre-set risk limit
Fill ratio	Executions relative to submitted orders	fills ÷ quote-time-weighted exposure (not raw order count)
Markout	Post-fill adverse selection	side × (mid − fill) at 100ms/1s/10s/60s; flag if persistently negative
Hedge cost	Cost of reducing exposure	transaction cost + basis risk from the hedge instrument

Market Making Across Asset Classes

Equities. Liquid equities often trade at narrow spreads. Competition centers on queue position, cross-venue routing, fee schedules, and fair-value signals from futures, ETFs, and related stocks.

Options. Options market makers manage delta, gamma, vega, and volatility-surface risk across a portfolio of contracts; inventory measured only in contract count is insufficient. Options market makers also carry formal quoting obligations distinct from equities — for example, Lead/Primary Market Makers are commonly required to provide two-sided quotations in roughly 90% of a class’s open trading seconds, and standard Competitive Market Makers in roughly 60%.

Futures and foreign exchange. These markets have strong links across maturities and venues. Market makers monitor calendar spreads, spot-futures relationships, funding, and macroeconomic releases.

Cryptocurrency. Crypto runs continuously across fragmented exchanges. Market makers face venue credit risk, unstable fee schedules, perpetual-futures funding, token inventory, and transfer delays. Unlike equities and options, crypto currently carries no equivalent regulatory two-sided-quoting obligation — participation is purely economic, not mandated. Rolling risk horizons replace the basic model’s fixed daily close.

Frequently Asked Questions

Conclusion

Quantitative market making does more than place a bid below the market and an ask above it. It estimates fair value, prices execution risk, skews quotes against inventory, monitors adverse selection, and hedges exposures before small spread gains become large directional losses. The Avellaneda–Stoikov model provides a clear mathematical starting point. Production performance depends on data quality, queue-aware fill modeling, latency, and independent risk controls. The most useful next step is an event-driven simulator that reports spread capture, markouts, inventory, hedging costs, and drawdown separately, alongside the feasibility inputs below.