Where Does the Price Really Get Made? Hasbrouck's Information Share

The same stock trades in many places at once. Apple trades on Nasdaq, on the NYSE, on a dozen alternative venues, in dark pools, and as an option, a future, and an ETF constituent. Bitcoin trades on scores of exchanges. Gold trades in London and Chicago.

All these prices track each other closely, because arbitrage is fast and merciless. But "they track each other" leaves a much more interesting question wide open:

Which one moves first?

Somewhere, some venue is where new information actually enters the price. Everywhere else is, to some degree, following along. Hasbrouck's 1995 paper is about how to tell them apart, and it produced the metric the field has argued about ever since.

The problem: everyone is looking at everyone else

You might think this is easy. Just check which venue's price changes first.

It is not easy, for a reason that is worth sitting with. All the venues are watching each other. They quote nearly identical prices, they update almost simultaneously, and the correlation between their price changes is enormous. When two series move together at essentially the same time, disentangling who is leading and who is following is a genuinely hard statistical problem.

Worse, the answer you want is not really about prices. Every venue's price is the true value plus some local junk: its own bid-ask bounce, its own inventory pressures, its own microstructure noise. What you actually want to know is: whose contribution moves the underlying truth, as opposed to just jiggling their own local noise?

The key idea via analogy: the invisible conductor

Imagine an orchestra where no one can see the conductor. Each musician plays, listens to the others, and adjusts. From the outside, all you hear are the individual instruments, each slightly out of tune, each with its own scratches and squeaks.

But there is an underlying piece of music being played. Hasbrouck calls it the efficient price: the unobservable true value that all the venues are noisily tracking. Every venue's quoted price is that efficient price plus its own local static.

Here is the key structural insight, and it is what makes the whole thing work. Because all the venues are pricing the same asset, their prices cannot wander apart forever. Arbitrage tethers them. In econometric language, they are cointegrated: they share one common permanent component, the efficient price, and everything else about their differences is temporary and mean-reverting.

That gives Hasbrouck exactly the leverage he needs. If there is one shared permanent trend and a bunch of temporary local noise, then you can ask a sharp question: when the efficient price makes a permanent move, whose price innovation was responsible?

His answer, the information share, is the proportion of the variance in the permanent, efficient-price innovations that can be attributed to each venue. If a venue's own surprises are the ones that end up permanently moving the shared truth, that venue is discovering the price. If a venue's surprises are all temporary local noise that gets corrected away, that venue is a price taker, following along.

He ran it on the thirty Dow stocks in the early 1990s, when they traded on the NYSE and on various regional exchanges, and the answer was emphatic: the overwhelming majority of price discovery was happening at the NYSE, with the regionals contributing very little. The NYSE was the conductor. The regionals were reading the sheet music off the NYSE's stand.

Why it mattered

• It gave fragmentation a scoreboard. As markets shattered into dozens of competing venues, the crucial policy and business question became: is fragmentation splintering price discovery, or does the price still get made in one place while everyone else just executes? The information share is how that question gets answered, and it has been applied to essentially every fragmented market: equities across venues, spot versus futures, stocks versus their ETFs, options versus underlying, onshore versus offshore, spot crypto versus perpetual futures.
• It sharpened the concept of "price discovery" itself. Before this paper, "price discovery" was a hand-wave. After it, it was a number between zero and one that you could compute. That is what makes a concept scientific.
• It found real, actionable structure. A recurring finding across markets, following Hasbrouck's method, is that price discovery often happens in the derivative rather than the underlying: index futures lead the cash index, ETFs can lead their own baskets, and perpetual futures often lead spot crypto. That is enormously useful to anyone building a signal, because it tells you which screen to watch.
• It handed regulators a tool. Debates about whether a new venue, a dark pool, or a payment-for-order-flow arrangement is "free-riding" on someone else's price discovery are conducted, quite literally, in the currency of information shares.

The honest limitations

This is a case where the limitations are unusually famous, because they generated a decades-long methodological fight.

• The measure is not unique when venues move together. This is the big one. If two venues' price innovations are contemporaneously correlated, and in fast markets they always are, then the method does not deliver one number. It delivers a range, an upper and a lower bound, depending on the arbitrary order in which you list the venues. Researchers typically report the midpoint of the bounds, which is a convention, not a result. When venues update almost simultaneously, the bounds can be embarrassingly wide.
• A rival measure exists and disagrees. Gonzalo and Granger proposed a different decomposition (the "component share") which asks about permanent error-correction weights rather than variance contributions. The two measures answer subtly different questions and can rank venues differently, and there is a substantial literature arguing about which one you should want. This is not settled.
• Sampling frequency changes the answer. Aggregate to coarser time intervals and simultaneity increases, the bounds widen, and the venue that looks dominant can shift. The result is not invariant to a choice the researcher makes.
• Speed is not the same as information. A venue can have a high information share simply because it is fastest, not because it hosts the smartest traders. In a world of latency arbitrage, the venue that updates first may just be the one with the shortest wire, mechanically front-running everyone else's quote updates. The measure cannot tell "genuinely discovering the price" apart from "reacting to a common signal a microsecond sooner."
• The efficient price is never observed. The whole edifice rests on an unobservable common factor that the model assumes into existence. It is a good assumption, but it is an assumption.

The one-line takeaway

Hasbrouck showed that when one asset trades in many venues, they all noisily track a single invisible true price, and you can score which venue's surprises actually move that true price permanently, versus which are just making local noise, though the score is a range rather than a single number whenever venues move at the same time.