It Is Not the Trades, It Is the Imbalance: Cont, Kukanov and Stoikov

If you ask a finance person what moves prices in the short run, they will say trades. Buys push the price up, sells push it down. The entire market impact literature is built on this, and it is not wrong.

But it is incomplete, and Rama Cont, Arseniy Kukanov and Sasha Stoikov showed exactly how incomplete. A trade is only one of three things that can happen at the top of an order book, and the other two matter just as much.

The problem: trades are the tip of the iceberg

Watch the best bid and best ask of a stock over a single second. Here is everything that can change the picture.

1. A market order arrives and consumes some of the resting size. This is a trade, and it is the only event most models look at.
2. A limit order arrives and adds new size to the bid or ask. Nothing traded. But the book just got deeper on one side.
3. A limit order is cancelled and size disappears from the bid or ask. Nothing traded. But the book just got thinner on one side.

In a modern electronic market, events two and three are vastly more common than event one. The overwhelming majority of orders are cancelled, not executed. A model that only looks at trades is watching a small minority of what is actually going on.

And crucially, all three events do the same fundamental thing: they change the balance of supply and demand at the top of the book. A big buy trade removes sell-side depth. A wave of cancellations on the ask also removes sell-side depth. From the perspective of the pressure on the price, these two are doing something similar, and neither the trade-only models nor common sense obviously distinguishes them.

The key idea via analogy: the tug of war rope

Picture a tug of war. The price is the marker in the middle of the rope, and it moves toward whichever side is pulling harder.

Now, the naive model says the marker only moves when someone yanks the rope, a yank being a trade. But that ignores two other things happening constantly on both sides.

• People are joining the team, adding their weight to one side. That is a limit order.
• People are letting go of the rope and walking away. That is a cancellation.

If ten people quietly let go of the rope on the sell side, the marker moves toward the buyers just as surely as if the buyers had given a hard yank. Nobody traded. The balance simply shifted.

Cont, Kukanov and Stoikov define a single quantity that captures the total pull on the rope. They call it order flow imbalance, usually shortened to OFI. It aggregates all three types of event, additions, cancellations, and trades, on both sides of the book, into one signed number: net buying pressure at the top of the book.

And their headline empirical result, from a study of fifty US stocks, is beautifully simple. The relationship between order flow imbalance and price change is linear.

This is a striking finding, because market impact measured against trade volume is famously and stubbornly non-linear, the whole concave, square-root business that has consumed the field for two decades. Yet measured against order flow imbalance, over short intervals, impact turns out to be a straight line. Twice the imbalance, twice the price move.

The suggestion is that the concavity everyone has been wrestling with may partly be an artefact of looking at the wrong variable. Look at the whole book instead of just the trades, and the relationship straightens out.

There is a second result, and it is the one practitioners tend to seize on. The slope of that linear relationship, how much the price moves per unit of imbalance, is inversely proportional to the depth of the book. In other words, the same imbalance moves a thin book much more than a thick one. Depth is the shock absorber. This gives you a single, computable coefficient that tells you how sensitive a given stock is to order flow right now, and it explains a great deal of the variation in impact across stocks and across the trading day.

Why it mattered

• It is one of the most useful signals in high-frequency trading. OFI is now a standard feature in short-horizon prediction models everywhere. If you want to know where the price will be in the next few seconds, the imbalance of order book events is one of the best things you can look at. This paper is a large part of why.
• It shifted attention from trades to the full event stream. In a market where the vast majority of order activity never results in a trade, focusing only on trades was leaving most of the information on the floor. The paper made that case decisively.
• The linearity result is genuinely surprising and clarifying. Recovering a clean linear relationship in a field defined by its non-linearities suggested that some of the field's central difficulties came from measuring the wrong thing.
• The depth-scaled slope gives a practical, computable impact coefficient. It is exactly the sort of result that makes it straight from a paper into a production system.

The honest limitations

• It is a short-horizon story. The linear relationship holds over short time intervals. Over hours and days, where metaorder impact lives, the concavity comes roaring back. OFI is a wonderful tool for predicting the next few seconds and a poor tool for estimating what a two-day execution will cost.
• It only looks at the top of the book. Order flow imbalance as defined here concerns the best bid and best ask. Activity deeper in the book, and the hidden liquidity that never appears at all, is invisible to it.
• Correlation, not causation. Does imbalance cause the price to move, or does an impending price move cause traders to adjust their orders, producing the imbalance? The paper documents a very strong relationship. Untangling the direction of causality in a market where everything happens in microseconds is genuinely hard.
• Everyone knows about it now. OFI was a strong, relatively underexploited signal when this was published. It is now in every high-frequency shop's feature set, and the profitability of naively trading it has been competed away substantially.
• The data predates a lot of market evolution. The empirical work uses US equity data from a specific period. Tick sizes, venue fragmentation and the behaviour of high-frequency market makers have all moved since.

The one-line takeaway

Cont, Kukanov and Stoikov showed that short-term price moves are driven not by trades alone but by order flow imbalance, the combined effect of trades, new limit orders and cancellations at the top of the book, and that once you measure things that way the relationship with price becomes cleanly linear, with a slope set by how deep the book is.