Market Making and Liquidity Provision — Deep Reference

Market making is the business of standing ready to buy and sell a financial instrument at posted bid and ask prices and earning the spread between them, net of inventory risk and adverse-selection losses. The discipline begins with Hans Stoll’s 1978 “Supply of Dealer Services” framework, extends through Glosten-Milgrom 1985 and Kyle 1985 on adverse selection, reached its modern stochastic-control formulation with Avellaneda-Stoikov 2008, and operates today across centralized limit order books, over-the-counter dealer markets, and on-chain automated market makers (AMMs). This note covers the modern stack: inventory models (Stoll, Ho-Stoll, Garman, Amihud-Mendelson), information models (Glosten-Milgrom, Kyle, Easley-O’Hara, PIN, VPIN), P&L decomposition, optimal spread setting (Avellaneda-Stoikov, Cartea-Jaimungal-Penalva, Gueant-Lehalle-Fernandez-Tapia, Bergault-Drissi-Gueant), limit order book dynamics (Cont-Stoikov-Talreja, queue position theory, hawks-and-doves), strategy taxonomy (passive vs aggressive MM, cross-venue arbitrage, latency arbitrage, statistical-arbitrage MM), the maker-taker fee structure post-Reg NMS, the major participant landscape (Citadel Securities, Susquehanna, Jane Street, Jump, Virtu, HRT, DRW, Tower, Two Sigma Securities, IMC, Optiver, XTX, GTS), crypto market making (Wintermute, Cumberland DRW, GSR, Auros, Amber, B2C2), AMM market making (Uniswap V2/V3/V4, Curve, Balancer, Trader Joe Liquidity Book, Maverick, Bunni/Arrakis), ETF authorized participants, basket and index arbitrage, and the regulatory regime (Reg SCI, MAR, MiFID II RTS, FINRA CAT, LEI).

See also

• market-microstructure-and-hft
• options-pricing-deep
• derivatives-and-quant-finance
• portfolio-construction-and-risk-deep
• fixed-income-deep
• structured-products-deep

1. Stoll 1978 — the inventory model

Hans Stoll’s seminal 1978 Journal of Finance 33:1133 “The Supply of Dealer Services in Securities Markets” laid the foundation for inventory-based bid-ask spread theory. The dealer maintains a target inventory $I^{\ast }$ and adjusts bid and ask prices to attract trades that move actual inventory $I_t$ toward the target. The half-spread compensates the dealer for the holding cost of an unwanted inventory position:

$\text{half-spread}=\frac{1}{2}(\text{ask}-\text{bid})\approx \alpha {\sigma }^2|I-I^{\ast }|+c_0$

where ${\sigma }^2$ is the instrument’s return variance, $\alpha$ is the dealer’s coefficient of inventory aversion (related to risk tolerance and time-to-unwind), and $c_0$ is a fixed cost reflecting bookkeeping and information processing. Stoll’s empirical contribution: spreads should be increasing in volatility (verified) and decreasing in trading volume (also verified — high-volume names have tighter spreads because inventory is mean-reverting faster).

Ho-Stoll 1981 Journal of Financial Economics 9:47 “Optimal Dealer Pricing Under Transactions and Return Uncertainty” extended this to dynamic optimization. The dealer solves a Hamilton-Jacobi-Bellman PDE for value $V(I,t)$ subject to:

$\frac{\partial V}{\partial t}+\mu I\frac{\partial V}{\partial I}+\frac{1}{2}{\sigma }^2{I}^2\frac{{\partial }^{2}V}{\partial I^2}+{\sup }_{b,a}\left[{\lambda }_b(b)(V(I+1)-V(I)+b)+{\lambda }_a(a)(V(I-1)-V(I)-a)\right]=0$

with ${\lambda }_b,{\lambda }_a$ the trade-arrival intensities at the chosen bid and ask (decreasing in distance from mid). The optimal bid and ask are symmetric around an inventory-skewed reservation price:

$r(I,t)=S_t-\gamma {\sigma }^{2}I(T-t)$

where $\gamma$ is the dealer’s risk-aversion. When long ($I>0$) the dealer shifts both bid and ask downward to incentivize sell trades.

Garman 1976 Journal of Financial Economics 3:257 modeled the market-maker as a monopolist setting fees on Poisson buy and sell flows. Less influential operationally than Stoll/Ho-Stoll, but foundational to subsequent work.

Amihud-Mendelson 1980 Journal of Financial Economics 8:31 introduced the dealer-optimal inventory view — the dealer’s exposure is bounded, and exceeding the bound triggers either spread-widening or aggressive position-reduction trades.

2. Information models — Glosten-Milgrom and Kyle

The inventory paradigm captures the dealer’s perspective on holding-cost risk but misses adverse selection — the risk that the counterparty has private information about the instrument’s value. Two foundational papers:

Glosten-Milgrom 1985 Journal of Financial Economics 14:71 “Bid, Ask and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders” — sequential-trade model with informed traders and noise traders. The market maker quotes a bid $B_t$ and ask $A_t$ such that conditional expected payoff equals zero given the trade direction:

$A_t=\mathbb{E}[V\mid \text{buy at }t]$ $B_t=\mathbb{E}[V\mid \text{sell at }t]$

The spread $A_t-B_t>0$ even with zero inventory cost because trade direction itself is information. The market maker loses money on every trade with an informed trader (who knows $V>A$ when she buys) and recoups via trades with noise traders. The Bayesian update on $V$ from each trade drives price discovery.

Kyle 1985 Econometrica 53:1315 “Continuous Auctions and Insider Trading” — strategic single informed trader trading against a competitive market maker through noise traders. Equilibrium price impact is linear:

$\Delta p_t=\lambda \cdot \Delta q_t$

where $\lambda ={\sigma }_V/(2{\sigma }_u)$ is Kyle’s lambda — informed-trader information variance divided by twice the noise-trader volume volatility. Kyle’s lambda is the canonical measure of permanent market impact, used extensively in execution-algorithm design (Almgren-Chriss bake it into the linear-impact term).

Easley-O’Hara 1987 Journal of Financial Economics 19:69 extended Glosten-Milgrom to simultaneous-trade settings and trade-size signaling. Larger trades are more likely informed → market makers quote wider for larger size.

Easley-Kiefer-O’Hara-Paperman 1996 Journal of Finance 51:1405 introduced PIN (Probability of Informed Trading) as a single empirical statistic estimated from the daily buy-sell trade imbalance. The model: each day is either an “information day” (probability $\alpha$, with informed traders present at intensity $\mu$) or a “no-information day”; both regimes feature uninformed traders arriving at intensity ${ϵ}_b,{ϵ}_s$. ML estimation of $(\alpha ,\mu ,{ϵ}_b,{ϵ}_s)$ from daily buy-sell counts yields:

$\text{PIN}=\frac{\alpha \mu }{\alpha \mu +{ϵ}_b+{ϵ}_s}$

PIN became the workhorse adverse-selection metric in academic microstructure literature 1996-2010.

Easley-Lopez de Prado-O’Hara 2012 Review of Financial Studies 25:1457 introduced VPIN (Volume-synchronized PIN), replacing clock-time bars with volume-time bars to handle the dramatic intraday volume variation in modern HFT markets. VPIN can be computed near-real-time and is used as a “flash crash early warning” signal at some prop desks. The authors argue VPIN spiked above 90th percentile in the hour before the May 6, 2010 Flash Crash.

3. Market-maker P&L decomposition

A market maker’s daily P&L splits cleanly into three components:

$\text{P\&L}=\underset{\text{passive earnings}}{\underbrace{\text{spread capture}}}-\underset{\text{losses to informed flow}}{\underbrace{\text{adverse selection cost}}}-\underset{\text{P\&L impact of holding risk}}{\underbrace{\text{inventory cost}}}$

• Spread capture = $\frac{1}{2}(\text{ask}-\text{bid})\times \text{volume}$ — half-spread earned on every fill (both sides), summed across the day. For tight US large-cap names this is single-digit basis points per share traded.
• Adverse selection cost = the price drift against the position after each fill. Measured as the realized post-trade price move (e.g., the 30-second drift away from the trade price). For passive market makers in informed-flow venues this can erase the entire spread capture.
• Inventory cost = the P&L impact of holding non-zero inventory through random price moves. Captures both the realized variance experienced by held inventory and the slippage of unwinding it. For high-volatility instruments this dominates.

A well-run market-making operation runs spread capture at ~2x the sum of adverse selection plus inventory cost. The Tier-1 firms (Citadel Securities, Jane Street, Virtu) report spread-capture-to-cost ratios in their disclosures; Virtu’s 10-K disclosures (VIRT, listed Apr 2015) show market-making revenue ~$1.5-2.0 billion/year against principal-trading risk that has historically produced single-digit losing days per year.

4. Avellaneda-Stoikov 2008 — the workhorse optimal MM model

Marco Avellaneda and Sasha Stoikov’s 2008 Quantitative Finance 8:217 “High-Frequency Trading in a Limit Order Book” is the most-implemented quote-setting model in HFT and on-chain market making. The framework:

• Asset mid-price follows $d{S}_t=\sigma d{W}_t$ (Brownian, zero drift).
• The market maker quotes bid $S_t-{\delta }^b$ and ask $S_t+{\delta }^a$.
• Buy orders arrive at the bid at Poisson rate ${\lambda }^b({\delta }^b)=A\exp (-k{\delta }^b)$; sell orders at the ask analogously.
• Holding inventory $q$ at horizon $T$ exposes the MM to terminal P&L variance ${\sigma }^2(T-t)q^2$.

The MM maximizes exponential utility of terminal wealth. Solving the resulting HJB gives the reservation price (the price at which the MM is indifferent between holding current inventory and not):

$r(s,q,t)=s-q\gamma {\sigma }^2(T-t)$

and the optimal bid-ask spread:

${\delta }^a+{\delta }^b=\gamma {\sigma }^2(T-t)+\frac{2}{\gamma }\ln \left(1+\frac{\gamma }{k}\right)$

with the bid and ask placed symmetrically around the reservation price. The model produces an explicit prescription: when long, skew quotes downward (lower the ask to attract sells, lower the bid to discourage further buys); when short, skew quotes upward. The framework’s elegance and analytical tractability made it the standard reference for inventory-aware MM across both TradFi HFT desks and DeFi liquidity-management vaults.

Cartea-Jaimungal-Penalva 2015 Algorithmic and High-Frequency Trading (Cambridge University Press) is the canonical textbook treatment, generalizing Avellaneda-Stoikov to: multi-asset MM with cross-asset inventory penalties, ambiguity-averse robust MM (Cartea-Donnelly-Jaimungal 2017), informed-trader-aware MM, and MM with model uncertainty.

Gueant-Lehalle-Fernandez-Tapia 2013 Mathematics and Financial Economics 7:477 derived an asymptotic dealer-optimal bid-ask formula for inventory-constrained dealers, widely cited in the dealer-RFQ literature.

Bergault-Drissi-Gueant 2022 generalized to multi-asset MM with quadratic cross-inventory cost — the standard formulation for ETF basket market making and for multi-pool AMM liquidity management.

5. Limit order book dynamics

A modern limit order book (LOB) is the price-time priority queue of resting buy and sell orders at each price level. Modeling it requires both the steady-state distribution of liquidity and the dynamics of order arrival, cancellation, and execution.

Cont-Stoikov-Talreja 2010 Operations Research 58:549 “A Stochastic Model for Order Book Dynamics” modeled the LOB as a multi-class queueing system. At each price level $p$ orders arrive (limit + market) and leave (cancellation + execution). The stationary distribution of book shape follows from balance equations on bid/ask volumes at each level. Predictions: bid-ask spread distribution, queue lengths, intraday-volume patterns.

Maglaras-Moallemi-Zheng 2015 Operations Research 63:1056 (workshop version, refined in 2022 Quantitative Finance publication) introduced queue position theory — the value of a passive limit order depends critically on its position in the price-level queue, since orders ahead of it absorb adverse selection. The expected fill time and the expected post-fill price drift both depend on queue rank. Practical implication: HFT market makers rank-jump by canceling and reposting; passive institutional orders (which won’t cancel-and-replace) are systematically disadvantaged unless using midpoint or dark venues.

Avellaneda-Lipkin 2003 “A Market-Induced Mechanism for Stock Pinning” introduced the “hawks-and-doves” intuition — aggressive HFT participants (hawks) take liquidity when they spot mispricings; passive participants (doves) post resting orders and earn the spread. The equilibrium proportion of each type and the steady-state spread emerge endogenously.

Foucault-Pagano-Roell 2013 Market Liquidity: Theory, Evidence, and Policy (Oxford University Press) is the standard graduate-level treatment of market vs limit order choice, venue selection, and market-design implications.

Order-book imbalance is the most-exploited LOB signal in practice. Define:

${\text{OBI}}_k=\frac{{\sum }_{i=1}^k{V}_i^b-{\sum }_{i=1}^k{V}_i^a}{{\sum }_{i=1}^k{V}_i^b+{\sum }_{i=1}^k{V}_i^a}$

where $V_i^b,V_i^a$ are bid and ask volumes at level $i$. Cartea-Jaimungal 2016 formalize that ${\text{OBI}}_1$ has predictive power for next-tick mid-price moves on the order of 10-60 seconds in liquid futures; the effect decays rapidly at longer horizons.

6. Strategy taxonomy — what market makers actually do

Real-world market-making operations run a portfolio of strategies, not a single textbook quoting algorithm:

• Passive market making — post resting limit orders on both sides of the book, earn the spread (plus rebates where applicable), manage inventory via skew. Dominant strategy on liquid US equities, Treasuries, listed options. The “deploy LOB queue, manage inventory” mode of Avellaneda-Stoikov.

• Aggressive market making (taker-side MM) — take liquidity (lift offers, hit bids) when inventory deviates from target, supplementing passive quoting with active inventory management. Common in fast-moving futures markets and in crypto where venue fragmentation creates frequent cross-venue mispricings.

• Cross-venue arbitrage market making — sit on the bid in one venue and the offer in another, profiting from the inter-venue spread. Standard in equity wholesaling (post on a maker venue, offload on a different venue at a tighter inside) and in crypto (Coinbase ↔ Binance ↔ OKX).

• Latency arbitrage — pick off stale resting quotes after a price-update event. Controversial: SIP-vs-direct-feed latency arbitrage in US equities (chronicled by Michael Lewis Flash Boys 2014) motivated IEX’s 350-microsecond speed bump. Now also a standard concern on AMMs where slow LP price updates leave “stale liquidity” exploitable by MEV searchers.

• Statistical arbitrage market making — pair trades, basket trades, index-arbitrage flow combined with market making in the constituent legs. The MM provides liquidity to one leg while opportunistically initiating the spread trade. Citadel Securities, Two Sigma Securities, and Jane Street all run this hybrid.

• Options-implied market making — equity MM that uses options-market signals (skew, IV term structure, dealer-gamma positioning) to adjust quote intensity. Common at SIG, Optiver, IMC where options and equities desks are tightly integrated.

7. Maker-taker fees and the access-fee economy

Reg NMS Rule 610 caps exchange access fees at 30 mils ($0.003/share) for stocks ≥ $1.00. Within that cap, US exchanges run two opposed fee models:

• Maker-taker: rebate to liquidity providers (typically -0.20 to -0.25 mils), charge to liquidity takers (+0.28 to +0.30 mils). Used by NYSE, Nasdaq main book, Cboe BZX, EDGX. Attracts displayed liquidity by paying rebates to passive orders.

• Taker-maker (inverted): charge to liquidity providers, rebate to takers. Used by Nasdaq BX, Cboe BYX, EDGA, NYSE National, NYSE American, NYSE Chicago, MEMX Members Exchange (some routes). Attracts marketable order flow by paying rebates to takers; designed for liquidity-seeking algorithms and for situations where capturing the trade is worth more than the rebate.

• No-rebate / fee-pro-rata / membership fee: IEX Investors Exchange (since 2016) — no rebates, all participants pay flat exchange fees; the 350-microsecond speed bump (delivered by 38 miles of coiled fiber in Mahwah NJ) prevents latency arbitrage by making the SIP catch up to direct feeds for the IEX matching engine.

The economic effect of the rebate-fee structure has been a long-running policy debate. Battalio-Corwin-Jennings 2016 Journal of Finance 71:2193 showed that retail broker order-routing decisions are distorted by rebate capture in ways that may not align with best execution.

MEMX (Members Exchange) launched September 2020, backed by Bank of America, BlackRock, Charles Schwab, Citadel Securities, Citigroup, Fidelity, Goldman Sachs, JPMorgan, Morgan Stanley, UBS, Virtu, and Wells Fargo. MEMX positioned itself as a low-cost alternative with both maker-taker and inverted-fee routes.

The SEC’s Tick Size and Access Fee Rule (adopted September 2024) reduces the access-fee cap to 10 mils for tick-constrained stocks (those quoted in mostly 1-cent ticks) and introduces variable sub-penny tick sizes. The rule rebalances the rebate-fee economy for the first time since 2005.

8. Major participants — the firms

The US equity market-making landscape consolidated dramatically through the 2010s. The dominant participants today:

• Citadel Securities (founded 2002 as Citadel’s market-making arm; CEO Peng Zhao since 2017; founder Ken Griffin). Executes ~35-40% of US listed equity retail order flow as a wholesaler. Major options market maker (50%+ of US listed options retail). Crypto MM via Citadel Securities Digital Assets launched 2022. Disclosed revenue $7.5B in 2022; $9B+ in 2024 per press reporting.

• Susquehanna International Group (SIG) (founded 1987 Bala Cynwyd PA by Jeff Yass) — one of the largest options market makers globally; major presence in ETF and equity MM. Internal hedge fund SIG Capital. Notable for “poker culture” recruiting and training; Yass is a major political donor.

• Jane Street Capital (founded 2000 NYC by Tim Reynolds, Marc Gerstein, others ex-Susquehanna). Dominant ETF market maker globally — Jane Street executes a substantial fraction of all institutional ETF block trades. Net revenue $14B+ 2023 (Bloomberg, FT reporting). Crypto MM through Jane Street Crypto. Listed reorg path debated.

• Jump Trading (founded 1999 Chicago by Bill DiSomma, Paul Gurinas) — multi-strat HFT plus Jump Crypto (founded 2021). Crypto-side losses during the May 2022 Terra UST collapse and the November 2022 FTX collapse were material; Jump Crypto led the recovery of Wormhole bridge exploit $320M in February 2022 by replenishing from corporate balance sheet.

• Virtu Financial (founded 2008 by Vincent Viola; NYSE: VIRT since April 2015 IPO at $19/share). Acquired KCG Holdings July 2017 ($1.4B), ITG July 2019 ($1.0B). Multi-asset multi-venue electronic market maker.

• Hudson River Trading (HRT) (founded 2002 NYC) — quantitative MM across equities, futures, options, fixed income, crypto. Notable for academic-research-style culture.

• DRW Holdings (Don R. Wilson, founded 1992 Chicago) — futures, options, FX, crypto. DRW Cumberland is the crypto-trading arm — one of the original institutional Bitcoin MMs. Reported losses in the early CME Bitcoin futures launch (Dec 2017); Wilson sued the CFTC over a 2013 swaps-mispricing prosecution and won in 2018, a notable industry precedent.

• Tower Research Capital (founded 1998 by Mark Gorton) — HFT in equities and futures.

• Two Sigma Securities (founded 2009 as MM arm of the Two Sigma hedge fund parent) — wholesaling, listed-options MM.

• IMC Trading (Amsterdam, founded 1989) — global options MM, especially Asian + EU options. Major ETF market maker.

• Optiver (Amsterdam, founded 1986 by Johann Kaemingk) — global options + ETF MM. ~2,000 employees globally; offices in Amsterdam, Chicago, Sydney, Shanghai, Singapore, Mumbai.

• Maven Securities (London, founded 2011) — derivatives MM; quietly substantial in European options.

• Akuna Capital (Chicago, founded 2011) — derivatives MM with strong futures and options presence.

• Wolverine Trading (Chicago, founded 1994) — derivatives MM specialist.

• Belvedere Trading (Chicago, founded 2002) — derivatives MM.

• GTS (Global Trading Systems) (founded 2006 NYC) — NYSE Designated Market Maker (DMM) since November 2016 for >100 of the most-active NYSE issues, succeeding Goldman as a major DMM presence.

• Quantlab Financial (Houston, founded 1998) — secretive quant MM.

• XTX Markets (London, founded 2015 by Alex Gerko ex-Deutsche Bank) — ML-driven FX, equity, and fixed-income MM. Annual revenue exceeded $1B by 2023. Built a deliberately low-headcount, high-research-density operation.

• Mercury One — a newer entrant (2024-2025) focused on quantitative MM in options.

9. Crypto market making

Crypto MM is structurally distinct from equity MM: 24/7 markets, 100+ active venues, severe venue fragmentation, on-chain and off-chain settlement co-existing, large basis spreads between centralized and decentralized prices.

• Wintermute (London, founded 2017 by Evgeny Gaevoy, ex-Optiver) — largest non-bank crypto MM. Reported daily volume $5B+ in 2023. Suffered a $160M DeFi exploit September 2022 (Profanity address-vulnerability); recovered operationally. OTC desk + venue MM + on-chain MM.

• Cumberland DRW — DRW Holdings’ crypto arm; one of the earliest institutional Bitcoin OTC desks.

• GSR (founded 2013 Hong Kong by Rich Rosenblum, Cristian Gil) — algorithmic OTC + venue MM in crypto. Operates in ~10 jurisdictions; subject to several enforcement reviews 2023-2024 over alleged stablecoin and token-issuance involvements.

• Auros (Hong Kong, founded 2018) — algorithmic crypto MM; recovered from a 2022 customer-fund-loss event tied to FTX exposure.

• Amber Group (Hong Kong, founded 2017) — crypto MM + asset management; reduced staffing 2023 amid bear market.

• B2C2 (London, founded 2015 by Max Boonen, acquired by SBI Group 2020) — OTC crypto MM, particularly strong in BTC, ETH, and stablecoins.

• Caladan (founded 2017 by John Lo, ex-Wintermute) — crypto MM with strong derivatives presence.

• Symbolic Capital Partners — crypto MM and derivatives.

• Flow Traders (Amsterdam, listed AMS:FLOW) — ETF MM that expanded into crypto MM 2017-2022; pulled back somewhat post-FTX.

• Genesis Trading (filed bankruptcy January 2023 amid the broader Digital Currency Group/Genesis lending crisis; restructured 2024).

Crypto MMs face unique structural challenges: counterparty risk in non-prime venue trading (the FTX collapse wiped out $2B+ of MM-held collateral); regulatory uncertainty across jurisdictions; on-chain MEV (maximal extractable value) pressure on LP positions; sandwich attacks against AMM LPs.

10. Automated Market Makers — DeFi’s quote-engineering

AMMs are smart-contract-based liquidity pools that quote prices algorithmically from pool inventory rather than from an off-chain quoter. The single most important innovation in market-making since 2010.

Uniswap V2 (deployed May 2020) — constant-product AMM:

$x\cdot y=k$

where $x$ and $y$ are the pool reserves of two tokens and $k$ is constant. The marginal price is $p=y/x$; trades shift the price along the constant-product curve. LPs provide both tokens symmetrically and earn 30 bps fee per trade. The model is permissionless, capital-inefficient (most liquidity is far from the active price), and exhibits impermanent loss — the LP’s terminal portfolio underperforms a 50/50 HODL portfolio when prices drift.

Uniswap V3 (deployed May 2021) — concentrated liquidity. LPs choose a price range $[p_a,p_b]$ within which to deploy capital. Inside the range the position behaves like a Uniswap-V2 position scaled by leverage; outside the range it sits entirely in one of the two assets. Capital efficiency increases 100-4000x for tight ranges. Fee tiers: 1, 5, 30, 100 bps. The math: each position has a “virtual reserve” pair $(L/p_b-L/p_a$ relationships); total pool reserves are summed across all active positions. V3 LPs face active-management risk — out-of-range positions earn nothing — leading to a market of “LP managers” (Arrakis, Gamma, Bunni) that rebalance ranges algorithmically.

Uniswap V4 (deployed January 2025) — adds hooks: contract callbacks at swap and LP-event lifecycle points enabling custom fee curves, dynamic fee tiers, MEV-protection logic, just-in-time liquidity, on-chain limit orders. V4 also moves from per-pool contracts to a singleton architecture (one contract holds all pools), reducing deployment gas by 99%+ for new pool deployments.

Curve Finance — stableswap AMM. Curve’s invariant blends constant-product and constant-sum:

$A{n}^n\sum x_i+D=AD{n}^n+\frac{D^{n+1}}{n^n\prod x_i}$

with amplification coefficient $A$. For correlated assets (USDC/USDT/DAI, ETH/stETH, BTC/WBTC) the constant-sum behavior at low slippage near peg dominates; for unbalanced pools the constant-product behavior kicks in. Curve dominates stablecoin AMM TVL. The CRV token (launched August 2020) governs fee distribution; the veCRV (vote-escrowed CRV) model — boost rewards by locking CRV for up to 4 years — became the canonical DeFi tokenomics structure adapted by dozens of subsequent protocols.

Balancer — generalized N-token weighted pools with arbitrary weights. The invariant:

${\prod }_i{x}_i^{w_i}=k$

where $\sum w_i=1$. Allows 80/20 LP positions (LP exposure dominated by one asset) and pool-based index funds. Balancer Boosted Pools layer Aave-yield-earning wrapped tokens for capital-efficient pool yields.

Trader Joe Liquidity Book (Avalanche, deployed November 2022) — bin-based AMM with discrete price ranges (bins) that hold both tokens proportionally. Combines the capital efficiency of Uniswap V3 with the no-impermanent-loss-within-bin property of bin pricing.

Maverick Protocol (Ethereum + zkSync, 2023) — directional LP — LP positions can be configured to track the price as it moves up (right mode), down (left mode), or both (both mode). Reduces impermanent loss for directional LP views.

Cetus (Aptos, also Sui) — concentrated liquidity AMM on Move-based L1s.

LP-management overlays: Arrakis Finance, Gamma Strategies, Bunni, Steer Protocol — automate range-rebalancing for Uniswap V3 LP positions.

TVL evolution: AMM total value locked peaked at ~$200B in late 2021, contracted to ~$30B by mid-2023, recovered to $100B+ by late 2024 as DeFi summer 2.0 emerged with restaking and the BTC ETF flows.

11. ETF authorized participants and basket trading

The ETF arbitrage mechanism keeps fund prices close to NAV via continuous creation and redemption. Authorized Participants (APs) are designated institutions that can create or redeem ETF shares directly with the ETF sponsor in large block sizes (typically 25,000-100,000 shares per creation unit). When the ETF trades at a premium to NAV: the AP buys the underlying basket, delivers it to the sponsor, receives ETF shares, sells them on-exchange, earns the basis. Reverse mechanism for discount.

The major US-equity ETF APs:

• Goldman Sachs Securities — AP for hundreds of ETFs; major basket trading desk.
• JPMorgan Securities — major AP, particularly in fixed-income ETFs.
• Bank of America Securities — AP for many BlackRock iShares funds.
• Citigroup Global Markets — major AP.
• Susquehanna SIG — particularly active in options-rich ETFs.
• Virtu Financial — substantial market-making + AP presence.
• Citadel Securities — AP for hundreds of ETFs.
• DRW — particularly in commodity and crypto-linked ETFs.
• Jane Street Capital — the dominant institutional ETF AP, particularly in international and corporate-bond ETFs where basket-pricing skill is critical.
• Optiver, IMC, Flow Traders — European-headquartered, very active in EU-listed ETFs.

ETF basket arbitrage is a multi-step trade: estimate the basket fair value from underlying-stock real-time prices, compare to the ETF on-exchange price, execute the basket leg via a portfolio-trading algo while simultaneously executing the ETF leg, deliver the basket to the sponsor for creation (T+1) or sell the redeemed basket (T+1). The realized basis is a few basis points on a creation-unit-sized trade; cumulative APP revenue across the major APs runs to billions of dollars annually.

Bond ETF arbitrage is particularly challenging because the underlying basket is illiquid OTC corporate bonds. APs hold inventory in cash bonds, may substitute portions of the basket with cash or with custom baskets per the sponsor’s rules, and exploit the basket-trading skill that distinguishes Jane Street from generalist arbitrageurs.

12. HFT regulation

The major regulatory regimes shaping market making:

• Reg SCI (Systems Compliance and Integrity) (SEC, adopted November 2014, compliance November 2015) — applies to SCI entities (exchanges, clearing agencies, plan processors) and requires comprehensive systems-management programs: business continuity, disaster recovery, capacity planning, incident reporting. Triggered by the May 2010 Flash Crash and the August 2012 Knight Capital event.

• MAR (Market Abuse Regulation) (EU, applied July 2016) — supersedes the Market Abuse Directive (2003). Defines and prohibits insider dealing, unlawful disclosure of inside information, and market manipulation (including spoofing, layering, wash trading). Direct effect across all EU markets and most UK/EEA markets.

• MiFID II RTS 5, 6, 7, 8 — Regulatory Technical Standards on:

  • RTS 5 (clock synchronization to UTC at microsecond granularity);
  • RTS 6 (organizational requirements for algorithmic trading);
  • RTS 7 (trading-venue requirements for technological systems);
  • RTS 8 (market-making schemes for trading venues — defines market-maker obligations and benefits).

• FINRA OATS (Order Audit Trail System) — retired September 1, 2021, replaced by the SEC’s Consolidated Audit Trail (CAT) under Rule 613. CAT captures every order, route, modification, cancel, and execution across US equities and options at the broker level; FINRA CAT LLC operates the repository. Funding disputes have continued through 2022-2025.

• Trader IDs and LEI — Legal Entity Identifier (ISO 17442) — mandatory for OTC derivatives reporting under Dodd-Frank, EMIR, MiFIR. Trading-firm identifiers within CAT and EU equivalent reporting.

• SEC Rule 15c3-5 (Market Access Rule) (adopted November 2010) — requires broker-dealers providing sponsored or direct market access to implement risk controls (pre-trade order size, daily exposure limits, prevention of erroneous orders). The “Knight Capital prevention rule.”

• CFTC Reg AT (Regulation Automated Trading, proposed November 2015; partially shelved 2017; some provisions adopted) — would have established Algorithmic Trading Person registration and pre-trade risk controls for futures. The current US futures regime applies many of these requirements via CFTC Part 1.73 and exchange rules.

13. Notable people

• Hans Stoll (Vanderbilt, foundational inventory MM theory, deceased 2019).
• Lawrence Glosten (Columbia, Glosten-Milgrom).
• Paul Milgrom (Stanford, Nobel 2020 for auction theory; Glosten-Milgrom co-author).
• Albert “Pete” Kyle (Maryland, Kyle 1985).
• Maureen O’Hara (Cornell, Market Microstructure Theory 1995).
• David Easley (Cornell, PIN + VPIN with O’Hara and López de Prado).
• Marcos López de Prado (Cornell ORIE; Abu Dhabi Investment Authority; AQR previously; VPIN co-author; Advances in Financial Machine Learning 2018).
• Marco Avellaneda (NYU Courant, deceased 2022; Avellaneda-Stoikov, Avellaneda-Lipkin).
• Sasha Stoikov (Cornell, Avellaneda-Stoikov).
• Álvaro Cartea (Oxford, Cartea-Jaimungal-Penalva).
• Sebastian Jaimungal (Toronto, Cartea-Jaimungal-Penalva).
• Olivier Guéant (Paris 1 Panthéon-Sorbonne, Guéant-Lehalle-Fernandez-Tapia).
• Charles-Albert Lehalle (Capital Fund Management; Guéant-Lehalle).
• Rama Cont (Oxford, Cont-Stoikov-Talreja).
• Costis Maglaras (Columbia, queue position theory).
• Ken Griffin (Citadel, Citadel Securities).
• Jeff Yass (SIG).
• Vincent Viola (Virtu).
• Alex Gerko (XTX Markets).

14. The 2024-2026 frontier

• Machine-learning-driven quoting — neural-network alphas trained on order-flow signals are standard at XTX, Citadel Securities, Jump, HRT, and the major DeFi MM operations. Reinforcement learning for quote-skewing and inventory management has migrated from research papers (Spooner et al. 2018) to production at multiple Tier-1 HFT firms.

• Cross-venue real-time risk consolidation — major HFT firms now run unified risk views across CME, ICE, Eurex, NYSE, Nasdaq, Cboe, IEX, MEMX, plus 20+ crypto venues, with latency below 100 microseconds across the global stack.

• AMM-CEX arbitrage — Wintermute and the major crypto MMs run continuous arbitrage between Binance/OKX/Coinbase order books and Uniswap V3/V4 LP positions. As V4 hooks introduce dynamic-fee curves, AMM quoting will resemble Avellaneda-Stoikov-style inventory-aware quoting at the contract level.

• On-chain MEV-aware market making — flashbots SUAVE, the various block-builder ecosystems on Ethereum, and the order-flow auction layer (CowSwap, UniswapX, 1inch Fusion, Bebop) restructure MEV from extractive sandwich attacks to MM-friendly auction-mechanism quoting.

• 0DTE options market making — 0DTE flow has restructured front-end SPX gamma profiles and changed the daily PnL distribution of options MMs (SIG, Citadel Securities, Optiver, IMC, Jane Street). Dealer gamma calculations now include intraday gamma waves.

• Regulatory pressure on retail wholesaling — the SEC’s Order Competition Rule (proposed December 2022) would require retail orders be auctioned among multiple liquidity providers rather than internalized by a single wholesaler. The proposal has been heavily contested and as of mid-2026 remains not finalized; the broader Reg Best Execution adopted in modified form in 2024.

15. Production examples

• Citadel Securities executing AAPL retail flow: order arrives from Robinhood at the Citadel SOR. Citadel quotes inside the NBBO (price improvement of $0.0005-$0.002 per share typical for liquid mega-caps). Citadel internalizes the trade — buys from the retail customer at $0.001 above the NBB or sells at $0.001 below the NBO, then hedges via its inventory book or via outbound venues.
• Jane Street creating SPY ETF units: Jane Street’s algo detects that SPY is trading $0.005 above the basket fair value. Jane Street buys the 500 underlying stocks in proportional weights via portfolio-trading algos, simultaneously sells SPY on-exchange. Delivers the basket to State Street (SPY sponsor) the next morning, receives SPY shares, closes the position.
• Susquehanna market-making SPX 0DTE: SIG quotes thousands of SPX strikes across each trading day, hedging delta intraday with E-mini S&P 500 futures, hedging gamma with vanilla options. Net 0DTE P&L is driven by realized-vs-implied volatility and by the dealer gamma profile.
• Wintermute providing two-sided BTC quotes on Binance: Wintermute’s quoter posts bid and ask near the BBO, sizing based on inventory (${\Delta }^a,{\Delta }^b$ via Avellaneda-Stoikov). Inventory mean-reverts via the natural buy-sell flow and via cross-venue rebalancing trades to/from Coinbase, OKX, and OTC counterparties.
• Uniswap V3 LP active-management via Arrakis: an institutional LP supplies USDC + WETH liquidity in a tight range around the current price. Arrakis’s manager rebalances the range every few hours as price drifts, with fees configured to net positive over the historical realized-volatility distribution.

16. Risk management for market makers

A market-making operation runs a risk framework that decomposes into:

• Position limits by instrument, by sector, by venue, by trader. Hard limits trigger flatten-orders or order-cancel signals; soft limits trigger alerts.
• Realized P&L attribution by strategy, by instrument, by hour. Anomalous attribution triggers review.
• Heartbeat monitoring — every algo is required to send a heartbeat to the central risk system at sub-second intervals; loss of heartbeat triggers kill-switch.
• Kill switch (post-Knight, mandatory under Reg 15c3-5 and Reg SCI) — operator can cancel all orders and disable trading within seconds. Multiple independent kill paths (manual button, automated thresholds, exchange-side).
• Pre-trade risk checks — every outgoing order passes through size, price, frequency, and gross-exposure checks at the network-interface layer (often FPGA-implemented for sub-microsecond evaluation).
• End-of-day reconciliation — every position must match the venue’s official record; discrepancies flagged for next-day investigation.

The Knight Capital event of August 1, 2012 — $440M loss in 45 minutes from an erroneous algorithm — remains the canonical “what could go wrong” case study. Every MM operation now runs the post-Knight playbook.

17. Cross-asset extensions

Market making generalizes across asset classes with similar but not identical structure:

• Treasury MM: BrokerTec (CME-owned IDB), Fenics USTreasuries, MarketAxess. Reg SBSR transaction-reporting layer. Treasury MM dominated by HFT firms (Citadel Securities, Jane Street, HRT, Jump, DRW) since the 2010s displacement of primary dealers from intraday liquidity provision. March 2020 Treasury dislocation + October 2014 Treasury flash rally + basis trade unwinds 2020-2024 all stress the Treasury MM ecosystem.

• FX MM: spot FX trades 24/5 across LMAX, FXall, Hotspot, Currenex, EBS, Reuters Matching. Major MMs: XTX Markets (~10-15% of spot G10 FX volume), HSBC, JPMorgan, Citi, Deutsche Bank, Goldman Sachs, UBS, Bank of America. Last-look quotation (the LP retains a brief option to reject the trade) is the central market-structure controversy.

• Listed options MM: SIG, Citadel Securities, Optiver, IMC, Wolverine, Belvedere, GTS, Akuna. The Cboe 1099-MM market making program rewards designated MMs with fee discounts in exchange for quoting obligations.

• Listed futures MM: CME Globex, ICE Futures US/Europe, Eurex. DRW, Jump, HRT, Tower, Optiver, IMC, Akuna, Wolverine, RSJ, Citadel Securities.

• Corporate bond MM: increasingly electronic via MarketAxess (~20%+ of US IG TRACE volume), Tradeweb. Citadel Securities, Jane Street, Millennium have built systematic corporate-bond MM capabilities competing with traditional dealer (JPMorgan, BofA, Goldman) market making.

• Crypto MM: see §9.

• DeFi AMM MM: see §10.

18. Liquidity-provision economics in summary

The market maker captures the spread, pays for inventory risk, and pays for adverse selection. The economic equilibrium of a competitive MM industry:

${\text{spread}}^{\ast }\approx \text{(adverse selection)}+\text{(inventory cost)}+\text{(infrastructure cost)}+\text{(competitive margin)}$

For a US large-cap with intraday volatility $0.50, daily volume 10M shares, and 5 active wholesalers: equilibrium spread is one cent (the minimum tick), spread capture is ~$0.005 per share, adverse-selection cost is ~$0.002, inventory cost is ~$0.001, infrastructure cost is ~$0.0005, residual margin is ~$0.0015. Aggregate across the market this is a multi-billion-dollar business.

For an illiquid small-cap with intraday volatility $0.30, daily volume 100K shares, and 1-2 active MMs: equilibrium spread is 3-5 cents, MMs earn proportionally more per share but trade much less volume.

For a Tier-1 crypto pair (BTC/USDT on Binance) with 24/7 volatility 50%+ annualized: spreads are ~1-2 bps in 100 BTC clip size during liquid hours, widening to 5-10 bps at low-liquidity overnight Asia.

For a deep Uniswap V3 LP position in USDC/WETH 5bps tier: effective spread is ~1 bp at the active price, widening rapidly out of the deployed range.

Further reading

• Maureen O’Hara, 1995, Market Microstructure Theory.
• Larry Harris, 2003, Trading and Exchanges: Market Microstructure for Practitioners.
• Álvaro Cartea, Sebastian Jaimungal, and José Penalva, 2015, Algorithmic and High-Frequency Trading.
• Olivier Guéant, 2016, The Financial Mathematics of Market Liquidity.
• Robert Almgren and Neil Chriss, 2001, “Optimal Execution of Portfolio Transactions,” Journal of Risk 3:5.
• Marcos López de Prado, 2018, Advances in Financial Machine Learning.
• Paul Wilmott, 2006, Paul Wilmott on Quantitative Finance.
• Mark Joshi, 2008, The Concepts and Practice of Mathematical Finance.
• Hans Stoll, 1978, “The Supply of Dealer Services in Securities Markets,” Journal of Finance 33:1133.
• Lawrence Glosten and Paul Milgrom, 1985, “Bid, Ask, and Transaction Prices in a Specialist Market,” Journal of Financial Economics 14:71.
• Albert Kyle, 1985, “Continuous Auctions and Insider Trading,” Econometrica 53:1315.
• Marco Avellaneda and Sasha Stoikov, 2008, “High-Frequency Trading in a Limit Order Book,” Quantitative Finance 8:217.
• Rama Cont, Sasha Stoikov, and Rishi Talreja, 2010, “A Stochastic Model for Order Book Dynamics,” Operations Research 58:549.
• Michael Lewis, 2014, Flash Boys: A Wall Street Revolt.
• Robin Wigglesworth, 2021, Trillions: How a Band of Wall Street Renegades Invented the Index Fund (for the AP / ETF arbitrage history).
• Paul Glasserman, 2003, Monte Carlo Methods in Financial Engineering.

Adjacent

• stochastic-calculus
• probability-distributions
• numerical-linear-algebra
• fpga-and-hardware-acceleration
• lock-free-data-structures-and-rdma
• game-theory
• securities-regulation-deep
• derivatives-and-instruments-catalog