The Old Rules Do Not Apply: O'Hara on High Frequency Market Microstructure

Maureen O'Hara helped build market microstructure. She co-wrote the PIN model, she wrote the textbook that trained a generation, and she spent decades developing the theory of how informed traders and market makers interact.

Which is what makes her 2015 paper so striking. It is, in substantial part, an argument that a lot of the framework she helped build needs rethinking, because the market it was designed to describe no longer exists.

The problem: the concepts are older than the market

Classical microstructure was built around a picture. There is an asset. There are traders, some informed, some not. There is a market maker. Orders arrive, the market maker quotes, prices form. The unit of analysis is a trade: a discrete, deliberate decision by a person to buy or sell a meaningful quantity.

Every model in the field rests on that picture. Kyle, Glosten and Milgrom, Ho and Stoll, PIN: all of them assume that a trade is a decision by an agent with a reason.

Now look at a modern equity market. A single institutional decision to buy a million shares becomes thousands of tiny child orders, sprayed across a dozen venues by an execution algorithm, most of them cancelled before they ever execute, sized in odd lots, timed by a scheduler nobody is watching. The "trades" you see on the tape are not decisions. They are debris, the visible residue of algorithms interacting with algorithms.

O'Hara's point is that this is not a change of degree. It is a change of kind, and it breaks the concepts.

The key idea via analogy: the map no longer matches the terrain

She walks through what has actually changed, and each item quietly invalidates something the old models assumed.

A trade is no longer a decision. Trade size is now meaningless as a signal of conviction. The old intuition, that a big trade means someone knows something, is dead: an informed trader now hides in a thousand small orders, and a large print might be a clumsy algorithm. Every model that keys off trade size, including much of the adverse selection literature, is working with a corrupted variable.

The order, not the trade, is the unit of information. Most orders in a modern market are cancelled. They never trade. But they are not nothing. The pattern of placement and cancellation is the strategic behaviour, and it contains most of the information. Yet classical microstructure studies trades, because trades are what the old data recorded. We have been studying the shadow and ignoring the object.

Time is the wrong ruler. Machine markets are wildly uneven in activity. A minute at the open and a minute at lunch are not comparable units. O'Hara pushes the case for event time or volume time, sampling by market activity rather than by the clock, which is the same argument that drives VPIN and later the machine learning literature on information-driven bars.

Speed itself is a strategy, and asymmetric information now includes speed. In the old models, the informed trader knows something about value. In the new one, the fast trader may know nothing about value and everything about what is about to happen in the order book. This is a genuinely different kind of informational advantage, and the old models do not have a slot for it.

Liquidity is no longer a stock, it is a flow. Old models treat the market maker as standing ready. New liquidity providers post and cancel thousands of times a second. The depth you see on the screen may not be there when you reach for it. "How much liquidity is in this market" is no longer a well-posed question at a point in time.

Fragmentation and the design of venues matter enormously. Maker-taker fees, order types, colocation, priority rules. These used to be plumbing. Now they are strategic variables that determine outcomes, and O'Hara insists they belong inside microstructure theory, not outside it.

Why it mattered

• It is a rigorous statement of the research agenda. The paper functions as a to-do list for a whole field: stop studying trades and start studying orders, stop measuring in clock time, put speed into your models of information, take venue design seriously as an economic variable. Much of the microstructure research of the last decade is working through this list.
• It gives practitioners a warning label. If you are computing PIN, or Roll's spread, or a Lee and Ready classification on modern data, this paper explains exactly why the number you get may be a fiction. The tools were built for a world with human decisions behind each print, and applying them to algorithmic debris produces numbers that look fine and mean nothing.
• It came from the right person. A critique of classical microstructure from someone who built classical microstructure carries a weight that the same critique from an outsider would not. It is an unusually candid act of intellectual honesty.
• It connected microstructure to modern data practice. The insistence on event-based sampling, on studying the full order book rather than executions, and on treating cancellations as data, is now standard in quantitative trading. This paper is one of the clearest academic statements of why.

The honest limitations

• It is a diagnosis, not a cure. O'Hara is far more convincing about what is broken than about what should replace it. The paper identifies the problems and gestures at directions. It does not deliver a new model of price formation for machine markets, and, candidly, the field still does not have one.
• It may overstate the discontinuity. Menkveld's HFT market maker turned out to behave exactly as Ho and Stoll would have predicted: earn the spread, lose to adverse selection, manage inventory to flat. The fundamental economics of liquidity provision seem to have survived the transition to machines rather well. What broke is the measurement apparatus, not necessarily the theory underneath.
• A survey is a snapshot. Written in 2015, it does not cover subsequent developments in speed bumps, batch auction implementations, retail order flow internalization at its current scale, or the crypto market structure experiments that have since become a live laboratory for exactly these questions.
• The evidence is mostly US equities. The claims are pitched generally, but the empirical world described is largely the American stock market, which has an unusual and idiosyncratic structure.

The one-line takeaway

O'Hara argues that high frequency markets broke the field's basic vocabulary: a trade is no longer a decision, the clock is the wrong ruler, cancelled orders carry more information than executed ones, and speed is now a form of private information, so microstructure needs to be rebuilt around orders and events rather than trades and time.