Market Design: Ticks, Venues, and the Speed Race

The mechanism built in the first lesson looks inevitable once you have seen it, and none of it is.

Price-time priority is a choice; some markets use price-size priority instead. Continuous matching is a choice; auctions match at intervals. The minimum price increment is set by rule. Whether orders are visible, whether they must be routed to the best price available anywhere, whether trading can happen away from an exchange at all: all decisions, all different across jurisdictions and asset classes.

These choices are collectively market design, and they are not neutral plumbing. Each one determines which behaviours are profitable, and therefore which participants appear and what they spend their money on.

The useful habit is to stop asking whether a market outcome is good and start asking which rule produced it. Most of what looks pathological in modern markets is a rational response to a rule, which means it is fixed by changing the rule rather than by disapproving of the response.

The minimum price increment is the cleanest example, because its effects are entirely mechanical.

If the tick is large relative to the natural spread, quotes cannot compete on price. Everyone who wants to be at the front posts the same price, and the only remaining way to gain priority is to arrive earlier. Competition moves from price to speed, and long queues form at each level.

If the tick is very small, quotes compete on price freely. Undercutting by a negligible amount takes priority from everyone ahead, so queues never build, and the top of book becomes unstable and thin.

Neither extreme is desirable, and that is the point: the tick is a dial that trades off price competition against queue stability and depth. Set it too fine and displayed liquidity evaporates because posting is pointless; set it too coarse and traders pay a spread wider than the market would otherwise support.

This is also why tick size regimes are usually tiered by price and liquidity rather than set as one number.

The single book of lesson one is also a simplification. In most developed equity markets the same instrument trades on many venues at once, each with its own order book.

Fragmentation has a genuine benefit: venues compete on fees, speed and features, and that competition has driven costs down substantially. It also imposes a cost that falls on every participant. There is no longer one book to read. Assembling a view of total available liquidity means consuming data from every venue and combining it, and acting on that view means sending orders to several places at once.

The consolidated picture is a construction, not an observation, and it is only as good as the slowest feed in it.

That gap between the consolidated view and the venues themselves is where a large amount of modern trading activity lives. Anyone whose picture updates faster than the common one knows something about the present rather than the future, which is a much easier thing to know.

Now the central argument. Eric Budish, Peter Cramton and John Shim set it out in "The High-Frequency Trading Arms Race: Frequent Batch Auctions as a Market Design Response", Quarterly Journal of Economics, volume 130, issue 4, 2015, pages 1547 to 1621.

The observation is about what a continuous limit order book means. Orders are processed one at a time, in arrival order, in continuous time. So when new public information appears, whoever reacts first gets to trade against quotes that have not been updated yet.

That is a race, and its prize does not depend on being right. It depends on being first, over public information everyone can see. Two participants with identical models and identical information still have a strict incentive to outspend each other on latency, because the faster one takes the whole prize.

The authors' framing is that this is a mechanical consequence of the design. Continuous serial processing turns public information into a winner-takes-all contest, and rents from that contest are competed away into spending on speed rather than into better prices.

The proposed alternative is a frequent batch auction. Instead of matching continuously, collect orders over a short interval, then clear them all at once at a single price, treating everything in the batch as having arrived at the same time.

The change is small and its consequence is not. Being a microsecond earlier no longer wins, because within a batch there is no earlier. Competition is forced back onto price, which is the dimension that benefits the person on the other side of the trade.

The left branch is worth reading carefully, because it explains a cost that is invisible. A market maker who gets picked off on stale quotes must recover that loss somewhere, and the only place available is a wider spread charged to everyone else. The race is not free even for participants who never enter it.

Batch auctions are not hypothetical. Most equity markets already run them at the moments that matter most, and the reasons are instructive.

A market typically opens with an auction and closes with one. Orders accumulate, and a single clearing price is chosen to maximise the volume that can be matched. Everything executes at that one price, so no participant gains from being marginally earlier within the window.

The closing auction is often among the largest concentrations of volume in the day, because index funds and anyone benchmarked to a closing price need to trade at that price rather than near it.

So the design question is not whether auctions work. It is where to draw the line. Continuous trading offers immediacy, which is genuinely valuable to someone who needs to trade now. Batching offers protection from the speed race. Markets currently resolve this by running continuously most of the time and batching at the open and close, and the debate is about whether that boundary is in the right place.

A further axis is whether an order is displayed at all, and it exists because of the inference problem from the previous lesson.

A large displayed order announces intent, and the market prices that intent before the order completes. So venues offer ways to reduce what is revealed. Hidden and iceberg orders show none or only part of their size while still resting in the book. Venues that display nothing pre-trade let large orders meet without either side broadcasting.

The trade-off is exact and unavoidable. Concealment reduces information leakage, and it also reduces the chance of being found by a counterparty who would have traded with you. You cannot be simultaneously invisible and easy to find.

Design then determines how the two coexist. Rules that give displayed orders priority over hidden ones at the same price are a deliberate subsidy to visibility, on the argument that a market where nobody displays anything has no prices for the others to reference.

Read down the second column and the participant mix of any market becomes predictable. Speed firms concentrate where the rules reward speed. That is not a claim about their character; it is where the rules put the money.

The transferable skill is inverting this table. Given an outcome that seems strange, find the rule that pays for it. Given a proposed rule change, ask which column-two behaviour it stops rewarding and what it starts rewarding instead, because it will always be something.

Four lessons, one idea, stated four ways.

A market has no privileged observer. There is no participant who knows what an asset is worth, and no mechanism that reveals it. There are only orders, and the inferences everyone draws from them.

The spread exists because a trade is evidence. Impact exists because a sequence of trades is stronger evidence. Market design matters because it determines who gets to act on that evidence first, and therefore what participants invest in.

The practical residue is a way of reading any market question. Ask what each participant can observe, what they can infer from it, and what the rules pay them for acting on it. That framing survives changes in technology, asset class and regulation, because it is about information rather than about any particular market's plumbing.

It also sets up everything downstream: execution is the problem of trading without revealing, and risk management is the problem of surviving the times your inference was wrong.