The Order Book as a Queue: Cont, Stoikov and Talreja

Most microstructure theory is about why people trade. Informed traders, uninformed traders, strategic behaviour, beliefs, equilibrium. It is economics, and it is deep.

Cont, Stoikov and Talreja's 2010 paper does something almost provocatively different. It throws all of that away.

No informed traders. No beliefs. No equilibrium. Just a limit order book, treated as a mechanical system in which orders arrive at random, get cancelled at random, and get executed when they meet. It is engineering, not economics.

And it works beautifully, because it answers a question the economics could not touch.

The problem: economists could not tell you what happens next

Suppose you are a trader with a very concrete question. The order book right now has 5,000 shares on the bid and 500 on the ask.

What is the probability the price ticks up before it ticks down?

This is not an abstract question. It is the question if you are deciding whether to join the queue, cancel your order, or cross the spread. Everyone who has ever traded on an order book has an intuition about it: a thin ask and a fat bid means the ask is more likely to get eaten first, so the price probably goes up.

Now try to get that answer out of Kyle's model, or Glosten and Milgrom. You cannot. Those models are about how information gets into the price over the long run. They have no order book in them at all, no queues, no depth, no notion of "what happens in the next thirty seconds given the current shape of the book."

There was a large gap between the theory of price formation and the practical reality of a trader staring at a book. This paper filled it.

The key idea via analogy: the supermarket checkout

Think of a limit order book as a bank of checkout queues.

Each price level is a queue. Orders join the back of a queue (a new limit order arrives). Orders leave the queue voluntarily (a cancellation). And orders get served from the front (a market order arrives and executes against the best price).

That is it. That is the whole model. Three kinds of events, each arriving randomly at some average rate:

• Limit orders arrive, adding depth at a price level
• Cancellations happen, removing depth
• Market orders arrive, consuming depth at the best bid or ask

When the queue at the best ask is completely emptied, the best ask moves up a tick. The price has moved. Price movement is not a decision anybody makes. It is what happens when a queue runs out.

This is a genuinely liberating reframing. It says: you do not need to know why anyone is trading in order to compute what the book will do. You just need to know the rates at which orders arrive, get cancelled, and execute, and those rates are directly and easily estimable from data. There is nothing to infer, no hidden types, no unobservable beliefs. Count the events, get the rates, done.

And because the model is built from the standard tools of queueing theory, it is analytically tractable. That is the paper's real gift. You can compute, with formulas rather than simulation:

• The probability the mid price goes up before it goes down, given the current book
• The probability your limit order gets filled before the price moves away from you
• The probability the spread widens before the next trade
• The distribution of how long you will wait in the queue

These are the actual questions of practical trading, and this paper gave the profession a way to answer them from the state of the book.

The authors estimated the model on data from the Tokyo Stock Exchange and showed it reproduces the empirical behaviour of real order books.

Why it mattered

• It made the order book a computable object. Before this, the book was something you looked at. After this, it was something you could compute conditional probabilities from. That is the foundation of essentially every modern short-horizon order book signal.
• It is the theoretical ancestor of queue-position modelling. Every high frequency trader who cares about where they sit in the queue, and they all care enormously, is working with the intuitions this paper formalized. Whether your order fills before the level gets swept is a queueing question, and this is the paper that framed it as one.
• It provided the foundation for order flow imbalance signals. The empirical fact that a lopsided book predicts the next price move, which is one of the most robust and widely used short-horizon signals in existence, follows naturally from this framework. The follow-up work by Cont, Kukanov and Stoikov on the price impact of order book events built directly on it.
• It legitimized a mechanistic approach. The paper is a strong argument that a great deal of short-horizon market behaviour can be explained without any reference to information or strategy at all, purely from the mechanics of a queueing system. That is a genuinely important scientific claim, and it aligns microstructure with the econophysics tradition of Bouchaud, Farmer and others.

The honest limitations

The paper's assumptions are its strength (they make it solvable) and its weakness (they are wrong).

• Order flow is not random, and it is not memoryless. The model assumes events arrive independently at constant rates, with no memory of what came before. Real order flow is famously persistent: buys follow buys, activity clusters violently, and rates spike and collapse. The mathematical convenience that makes the model tractable is precisely the assumption real markets violate most flagrantly.
• There is no strategy. Real traders cancel orders because the book moved against them, place orders because they see an imbalance, and race to the front of a queue. In this model, all of that is random noise. The feedback loops that dominate real order book dynamics are simply absent.
• All price levels are treated alike. The model typically assumes the same arrival rates independent of distance from the mid, which is not remotely true: activity is enormously concentrated at the best levels.
• No hidden liquidity, no fragmentation. Iceberg orders, dark pools, and the same asset trading on ten venues at once are all outside the model.
• It predicts short horizons only. The model is excellent at "what happens in the next few seconds." It has essentially nothing to say about where the price will be tomorrow, and it should not be pushed there. Fundamentals do not appear anywhere in it.

The honest summary is that this is a useful caricature. It gets the mechanics right and the behaviour wrong, and for short-horizon questions the mechanics dominate.

The one-line takeaway

Cont, Stoikov and Talreja modelled the limit order book as a set of queues where orders arrive, cancel, and get served at random, which strips out all the economics but makes the book computable: given the depth on each side right now, you can calculate the odds that the price ticks up before it ticks down, and whether your order fills before it does.