A market maker earns compensation by supplying executable bids and asks, but every fill also creates inventory exposure and the possibility that the counterparty traded against a stale or unfavorable quote. That makes algorithmic market making less a machine for collecting the bid-ask spread than a continuous balancing problem: quote competitively enough to trade, but not so aggressively that inventory, adverse selection, costs, or venue risk overwhelm the economics.
The quoted spread is only the starting point. Sustainable performance depends on realized spread, post-fill price movement, inventory mark-to-market, fees or rebates, hedging, funding, latency, margin, operational reliability, and the ability to stop when conditions move outside the strategy’s control envelope. Avellaneda and Stoikov’s foundational market-making framework formalizes this tradeoff by connecting inventory risk with quote placement and execution probability.
Algorithmic Market Making Is Not an Automated Market Maker
Algorithmic market making refers here to electronic liquidity provision through explicit bid and ask orders on an order book or comparable quote-driven venue. The strategy continuously observes market conditions, estimates a reference value, sets prices and sizes, manages fills, and adjusts inventory.
An automated market maker, conventionally abbreviated AMM, is different. Liquidity sits in a pool, and protocol logic determines the terms at which trades occur. Kenson’s DEX and automated market maker research covers that pool-based model separately. CPS 191 is about order-book market making, not liquidity-pool strategy.
How the Quote Cycle Works
A market-making cycle begins with data, not an order. The system ingests the order book, trades, and any approved reference markets, then checks freshness and data integrity. A pricing model estimates a reference value and evaluates volatility, depth, and current inventory. It then selects a spread, skews the bid and ask if inventory is off target, and chooses quote sizes.
Before submission, hard controls should test order size, price, notional exposure, venue eligibility, inventory, margin, and data freshness. The model can want to quote, but a separate risk layer should decide whether the quote is permitted. After submission, the system monitors queue position, order state, fills, cancellation acknowledgements, and partial executions. Each fill updates inventory, which can trigger new skew, size reduction, hedging, or a stop condition.
Post-trade analysis then asks whether the fill was economically useful. Markouts, realized spread, hedge cost, fees, and inventory movement feed back into strategy calibration. This sits alongside the broader order flow and market depth questions that determine how much executable liquidity the market actually has.
Reference Price, Spread, and Inventory Skew
Midpoint pricing is one possible reference, not a universal fair value. A market maker may also use weighted order-book information, recent trades, cross-venue prices, index inputs, or spot-derivative relationships. The important distinction is between a reference value and the maker’s inventory-adjusted reservation value.
If the strategy is close to its target inventory, its reservation value may sit near the reference. If repeated bid fills leave it long, the strategy typically becomes less willing to buy more. It can lower bid aggressiveness, reduce bid size, make the ask more competitive, increase ask size, hedge elsewhere, or stop the bid side near a hard limit. A short inventory produces the reverse incentives.
Quote width is also conditional. Higher volatility, stronger adverse-selection risk, slower cancellation, expensive hedging, weaker fill quality, or a stressed inventory position can justify wider quotes or smaller size. Competition and queue value can push in the opposite direction. A narrow spread that fails to compensate for these risks may look liquid while generating poor realized economics.
Queue Position, Fill Probability, and Adverse Selection
Being at the best bid or ask does not guarantee execution. On venues using price-time or related priority rules, older orders at the same price can remain ahead of a new quote. Queue position therefore affects fill probability and the value of maintaining an existing order.
That creates a cancel-and-replace tradeoff. Cancel too slowly, and a changing reference price can leave the maker exposed on a stale quote. Cancel too aggressively, and the strategy may surrender queue priority, reduce fill probability, and consume message or API capacity. Research on queue-position valuation shows why passive-order value depends on both execution priority and the adverse-selection risk associated with being filled.
Adverse selection explains why a fill is not automatically a success. A passive ask may execute just before fair value rises, or a passive bid may fill just before fair value falls. A post-fill markout compares the execution with a later reference value to estimate what happened after the trade. The relevant horizon depends on strategy speed and market structure, so there is no universal markout interval.
The Real Economics of Market Making
Quoted spread, realized spread, and net market-making P&L are different measures. The quoted spread describes displayed prices. Realized-spread analysis asks how much spread remained after subsequent price movement. Net P&L must also reflect inventory mark-to-market, explicit fees, maker rebates where offered, taker fees, slippage on hedges, funding or borrow cost, margin cost, infrastructure, custody, and settlement expenses.
Fee structures matter because a quote that appears viable before costs may be uneconomic after them. Maker rebates are not universal, and fee schedules often depend on venue, account type, and activity tier. Likewise, hedging can consume the apparent spread if the maker must cross another venue aggressively to reduce inventory.
This is why fill rate alone is a poor success metric. A strategy can fill frequently while producing consistently negative post-fill markouts. Conversely, a wider, lower-fill strategy may create better risk-adjusted economics. Measurement needs to explain where P&L came from rather than treating volume as proof of quality.
Latency, Venue Risk, and Capital Constraints
Latency risk is not one number. It can arise in market-data delivery, computation, order entry, acknowledgement, cancellation, fill reporting, or the hedge path. Dropped messages, sequence gaps, API throttling, or a delayed cancel acknowledgement can leave a quote live after the pricing model believes it should have disappeared.
Controls can include feed-age limits, heartbeat and sequence monitoring, quote withdrawal, cancel-on-disconnect where supported, reference-price divergence checks, backup feeds, and venue disablement. Lower latency can reduce stale-quote exposure, but its value depends on volatility, competition, queue rules, and strategy horizon.
Digital-asset venues add another layer of exposure. A maker can depend on exchange availability, API stability, custody, collateral accessibility, withdrawal access, margin rules, and counterparty solvency. BIS/FSI’s 2026 work on cryptoasset intermediaries highlights credit, liquidity, market, and interconnectedness risks associated with multifunction cryptoasset intermediaries.
Capital and margin therefore constrain liquidity provision. Leverage can expand quoting capacity, but it also amplifies inventory losses and liquidation risk. Venue concentration and limited hedge capacity can turn a normal market-making position into digital asset liquidity risk and digital asset operational risk.
Pre-Trade and Runtime Controls
A production system should separate pricing logic from hard permissions. Pre-trade digital asset internal controls can include maximum order size, maximum inventory, gross-notional limits, venue exposure limits, price collars, minimum or maximum quote-width rules, stale-data blocks, order-rate limits, approved asset and venue lists, self-trade prevention, and margin thresholds.
The principle is familiar from regulated electronic market access. SEC guidance on Rule 15c3-5 requires covered U.S. broker-dealers to maintain controls such as credit or capital thresholds, price and size checks, authorization restrictions, and controls for erroneous orders. That rule is not a universal crypto requirement, but it is a useful control-design precedent.
Runtime responses should be graduated. The system can widen spreads, reduce quote size, stop one side, hedge inventory, disable a venue, or cancel all orders. A kill switch, cancel-all function, venue-health monitoring, abnormal-fill detection, inventory-breach response, and manual escalation path are more important than preserving uninterrupted quote uptime.
Two Illustrative Market-Making Scenarios
Inventory skew
Assume a reference value of $100.00 and an initially neutral quote of $99.95 bid and $100.05 ask. Several bid orders fill, leaving the maker long. An illustrative response might move the bid farther away, reduce bid size, and make the ask more competitive so that subsequent flow is more likely to reduce inventory. The maker may also hedge on an approved venue. The exact prices depend on the model, market state, and limits; the example shows the direction of the control, not a profitable formula.
Volatility and stale quotes
Suppose the reference value moves rapidly from $100.00 toward $101.00 while an old ask remains at $100.05. If cancellation is delayed, the maker can be filled at a price that is already stale. A volatility or freshness control may instead cancel the quote, widen the replacement spread, reduce size, or suspend quoting until the data and venue state are acceptable. The control reduces exposure but cannot guarantee that every adverse fill is avoided.
Where Algorithmic Market Making Fails
- One-sided fills create excessive inventory. Position and margin monitoring should trigger skew, size reduction, hedging, or a side-specific stop; hedge execution can still fail.
- A market-data feed becomes stale. Feed-age controls should withdraw quotes or switch to an approved fallback; correlated backup feeds remain a residual risk.
- Cancel acknowledgements slow during volatility. Venue-health logic can stop new orders and escalate to cancel-all or manual action; already-live orders may remain exposed briefly.
- The maker fills often, but post-fill markouts are consistently negative. Adverse-selection monitoring should widen or reduce size; the flow regime may change again.
- A hedge venue goes offline after inventory has accumulated. The strategy should reduce new inventory acquisition or stop quoting; alternative venues may have poorer depth.
- Fee, funding, or margin economics deteriorate. Net-cost monitoring should reprice or stop the strategy rather than maintain volume at uneconomic terms.
Metrics for Measuring Liquidity Provision
Performance needs several metric families. Liquidity measures can include quoted spread, quoted depth, quote uptime, and time at the best bid or ask. Execution measures include fill rate, partial-fill rate, rejection rate, cancel latency, and queue position where observable.
Inventory measures include average and maximum inventory, turnover, position age, and time outside target bands. Economic measures include gross spread capture, realized spread, explicit fees, rebates, hedge cost, funding or borrow cost, and net P&L. Adverse-selection analysis should include post-fill markouts by side, venue, and market regime.
Risk and operations complete the picture: limit breaches, kill-switch activations, margin utilization, venue concentration, feed gaps, API failures, stale-quote cancellations, unmatched fills, and reconciliation breaks. There is no universal target for these indicators. The threshold that matters is the one connected to a widening, reduction, hedge, escalation, or stop decision.
Market Integrity and On-Chain Differences
Fast order updates and high cancellation rates are not automatically manipulative. Spoofing, layering, wash trading, and misleading quote behavior depend on conduct, intent, market context, venue rules, and applicable law. CFTC guidance on disruptive trading practices distinguishes bona fide cancellation and modification from spoofing in its covered markets based on intent and factual circumstances. IOSCO’s crypto-asset market recommendations likewise emphasize surveillance and controls addressing manipulation, fraud, conflicts, and operational risks.
On blockchain-connected order books, market makers may also face network fees, ordering uncertainty, confirmation or finality considerations, and failed or partial executions. Those specialist topics are covered separately in Kenson’s research on transaction ordering and MEV and failed and partial on-chain executions. Multi-venue hedge and execution choices are treated in more depth in institutional liquidity routing and Kenson’s work on slippage in low-liquidity token trades.
Implementation and Risk Checklist
- Pricing: Which inputs define the reference value, and what makes that reference invalid or stale?
- Inventory: What are the target, soft limits, hard limits, hedge triggers, and one-sided stop conditions?
- Execution: How are queue position, partial fills, rejections, acknowledgements, and cancellations monitored?
- Economics: Is performance measured after fees, rebates, markouts, hedging, funding, margin, and infrastructure costs?
- Controls: Which hard limits sit outside the quoting model, and who can change or override them?
- Venue risk: What happens if the quoting venue, market-data feed, custody path, or hedge venue becomes unavailable?
- Monitoring: Which metrics trigger widening, size reduction, hedging, venue disablement, or cancel-all?
A Controlled Market Maker Knows When Not to Quote
Algorithmic market making can support executable liquidity when pricing, inventory, costs, data, venue conditions, and controls remain within acceptable bounds. The strongest operating model is not the one that quotes continuously at all costs. It is the one that can identify when liquidity provision no longer compensates for the risk and can reduce, widen, hedge, or stop in a controlled way.
For further educational information on market structure, digital-asset liquidity, and execution controls, contact us to connect with Kenson Investments and explore related Knowledge Centre research.
