Market Structure Is Policy Written in Microseconds

An electronic market is a rule system before it is a stream of prices. The matching engine defines priority; the tick constrains admissible quotes; transparency rules determine what information is public; fees alter the net price of execution; and venue categories determine where orders may interact. Trading algorithms do not encounter these choices as legal prose. They encounter them as state variables and constraints evaluated in microseconds.

The research claim is not that regulation determines every market outcome. It is that regulation defines the feasible set within which private optimisation occurs. A smaller tick changes the return to queue priority; a transparency waiver changes the information cost of displaying size; continuous matching preserves the value of arriving first. Observed spreads and routing are equilibrium responses to those settings, not natural constants.

Section 01Rules become incentives

The cleanest demonstration of "policy compiled to microseconds" is the speed race itself. Budish, Cramton and Shim argued that the high-frequency trading arms race is not a quirk of greedy engineering but a predictable consequence of one specific design choice: matching orders in continuous time. Using high-resolution data, they showed that correlations which hold over normal horizons break down at the millisecond scale, opening mechanical, recurring arbitrage opportunities available only to whoever is fastest. The race to be that participant, to shave nanoseconds off a path, is the rational response to the rule. Their proposed remedy is itself a policy: replace continuous matching with frequent batch auctions, clearing orders in discrete intervals so that speed advantages measured in microseconds stop being decisive. Same traders, same technology, different rule, and the incentive evaporates.

The general lesson is to separate conduct from mechanism. Participant intent matters for enforcement, but recurring races can persist even when each firm follows the rules because the allocation rule rewards marginal speed. Market design asks which private responses a rule makes profitable.

Section 02The map of venues

Figure 4 sketches the journey of an order through a fragmented market. A signal becomes a parent order; a smart order router splits it into children and decides, continuously, where each child should go. The destinations are not interchangeable. A lit order book displays depth and contributes to public price formation. A dark venue lets large orders rest unseen, trading the benefit of reduced market impact against the cost of opacity. A periodic auction batches interest into discrete crossings. A systematic internaliser fills client flow against a firm's own book. Each exists because a rule permits it, often with conditions (transparency waivers, size thresholds, tick-size regimes) attached.

Fragmentation across these venue types is the defining feature of modern equity markets, and it is double-edged. Competition among venues can tighten spreads and lower explicit fees; it also disperses liquidity, complicates price formation, and turns "best execution" from a slogan into a genuine engineering and compliance problem. The European framework built around MiFID II has spent years adjusting exactly these dials (how much trading may occur in the dark, under what waivers, with what transparency) precisely because each setting changes the microstructure that results. The router in the diagram is where all of that policy is finally adjudicated, thousands of times a second, on behalf of a single order.

Market design does not remove agency. It determines which forms of agency are rewarded often enough to become the market's visible behaviour.

On market design

Section 03When the design fails visibly

Stress episodes make the design legible. On 6 May 2010, US equity markets fell and rebounded violently within minutes. The official SEC and CFTC report identified a large automated sell programme in E-Mini S&P 500 futures, executed according to a volume-based rule without regard to price or elapsed time, as an important initiating event. High-frequency participants initially absorbed inventory and then reduced it rapidly as conditions deteriorated, while effects propagated across linked venues. The episode does not prove that one design caused the event; it shows how execution rules, inventory behaviour, and fragmentation can interact non-linearly.

The point is not that fragmentation or automation is bad. It is that a market's behaviour under stress is a property of its design, and that design is a policy artefact. A different set of rules (different circuit breakers, different obligations on liquidity providers, different matching cadence) would have produced a different event, better or worse. The microstructure is where the policy is tested, and the test only runs when something breaks.

continuous matching  → first message to arrive wins
correlated prices    → a move on venue A predicts venue B
microsecond horizon  → whoever reacts first captures the gap

Section 04Policy catching up

If microstructure is policy, then supervision is the slow human loop that watches the fast machine loop and adjusts the rules. That loop is visibly active. In February 2026, ESMA published a supervisory briefing on algorithmic trading in the EU intended to align how national regulators oversee algorithmic and high-frequency activity under MiFID II. Its themes are precisely the levers in Figure 4: governance of trading algorithms, testing and stress-testing requirements, outsourcing arrangements, pre-trade controls, and the growing use of AI within algorithmic-trading workflows. The briefing is not new primary law; it is an attempt to make existing rules bite consistently, to ensure that the policy already written into the microstructure is actually supervised at the speed the microstructure runs.

The behaviour you see at the touch is the policy you wrote, executed at machine speed.

The operational implication is model risk. A strategy calibrated to queue length, venue fill probability, fee schedules, or transparency waivers contains regulatory parameters even if the code labels them as market data. When those rules change, historical execution estimates can lose transportability without any change in the predictive signal. Microstructure research should therefore version the rulebook alongside the dataset.

This research is analysis and commentary for general information. It is not investment advice, legal advice, an offer, or a solicitation, and it contains no price forecasts. Regulatory descriptions are summaries of cited sources and should not be relied on for compliance.