What Are the Odds the Next Trader Knows Something? The PIN Model

Every theory of the bid-ask spread since Copeland and Galai says the same thing: market makers widen their quotes because they are afraid of trading against someone who knows more. The more informed traders there are lurking, the wider the spread must be.

That is a lovely theory with one gaping practical problem. Informed traders do not wear badges. You cannot count them. You cannot survey them. They have every incentive to look exactly like everyone else. So how do you ever test the theory?

Easley, Kiefer, O'Hara and Paperman's answer, in 1996, is one of the genuinely clever pieces of inference in finance. You cannot see the informed traders, but you can see the footprints they leave in the daily balance of buys and sells.

The problem: spreads differ, and nobody knew why

The specific puzzle that motivated the paper: infrequently traded stocks have wider spreads than actively traded ones. Everyone knew this. Nobody agreed on the cause.

One story is inventory: a dealer stuck with a thinly traded stock waits a long time to unload, so bears more risk, so charges more. That is Stoll.

The other story is information: thinly traded stocks are less covered, less scrutinized, and so a larger fraction of the trades in them come from people who genuinely know something. That is adverse selection.

To settle it, you need to actually measure the fraction of informed trading, stock by stock. That is what PIN does.

The key idea via analogy: reading the tide from the beach

Imagine you cannot see the ocean, but every evening someone hands you a note saying how much seaweed washed up on the left side of the beach and how much on the right.

On a normal day, roughly equal amounts wash up on both sides. Wind and waves are random and there is no reason for a lopsided result.

But some days, a strong current is running. On those days the seaweed piles up dramatically on one side. You never see the current. But you know, from the lopsidedness of the piles, that it must have been there.

That is the entire PIN idea, and it is beautiful in its economy.

• Uninformed traders are the random waves. They buy and sell for their own reasons, and their buys and sells roughly balance out over a day. They generate a base level of activity on both sides.
• Informed traders are the current. Crucially, they only push one way. If someone knows the stock is worth more, they buy, and they do not sell. If they know it is worth less, they sell, and they do not buy. Information is directional by nature.

So the signature of a day with informed trading is an abnormally lopsided day: unusually many buys relative to sells, or vice versa. And the signature of a stock with lots of informed trading is that it has lots of lopsided days.

The model formalizes this with a simple story about how each trading day begins. First, nature flips a coin: is there private information today at all? If yes, another flip: is the news good or bad? Then, depending on that state, buys and sells arrive at rates that either balance out (no information) or tilt heavily one way (information). The researcher never observes the coin flips. They only observe the daily counts of buys and sells over many days.

From that stream of daily counts, the model works backwards to estimate the underlying rates: how often information events occur, how many uninformed traders show up on a typical day, and how many informed traders pile in when there is news. Combine those, and you get PIN, the probability that any given trade is coming from someone who is informed.

Their headline empirical finding is the one the theory demanded: PIN is lower for high-volume stocks. The heavily traded names have proportionally less informed trading, which is precisely why their spreads are tighter. Adverse selection, not just inventory, explains the spread pattern across stocks.

Why it mattered

• It made an invisible quantity into a number. This is the achievement. "How much informed trading is in this stock" went from a philosophical question to a parameter you can estimate from public data. Whether the number is exactly right is a separate argument, but the leap from unmeasurable to measurable is what makes a field progress.
• It confirmed adverse selection empirically. The 1996 paper is one of the strongest pieces of evidence that the information story of the spread, not just the inventory story, is doing real work in explaining why some stocks cost more to trade.
• PIN escaped microstructure entirely. It became a standard control variable across corporate finance and accounting. Researchers use it to study whether insider trading rises before takeovers, whether better disclosure reduces information asymmetry, whether a firm's cost of capital depends on how much private information trades in its stock. A microstructure parameter became a general-purpose measure of "how asymmetric is the information around this firm."
• It set up its own successor. When markets went high frequency, PIN's daily-count framework became too slow. Easley, Lopez de Prado and O'Hara's VPIN was the direct response, adapting the same core idea, order imbalance reveals information, to a world measured in volume rather than days.

The honest limitations

PIN has been vigorously attacked, and some of the attacks land.

• It depends on trade classification, and errors bias it. You have to infer which trades were buys and which were sells, usually with a rule like Lee and Ready. Those rules misclassify, and misclassification is not innocuous: it mechanically makes days look less lopsided than they were, which biases PIN in ways that are hard to sign cleanly.
• The estimation is numerically fragile. Fitting the model involves maximizing a likelihood that is notoriously badly behaved. For actively traded stocks the calculation overflows, and different researchers using different numerical tricks and starting values can get materially different PIN estimates from the same data. This is not a minor computational footnote, it has produced published disagreements.
• Order imbalance has other causes. The model's entire identification rests on the claim that lopsided order flow means information. But order flow can be lopsided because of index rebalancing, because a pension fund is executing a large uninformed order over several days, because of retail sentiment, or because of momentum chasing. PIN cannot tell an informed trader from a large, patient, utterly uninformed one. It measures imbalance, and calls it information.
• The two-state world is a caricature. Information arrives as a discrete daily event that is either good, bad or absent. Real information dribbles in continuously, with varying precision, and is partially known to many people to varying degrees.
• Poor performance in high-frequency markets. The model assumes a day is a meaningful unit and that trades are chunky, discrete events. In a market with millions of small algorithmic prints per day, both assumptions strain badly. This is the failure mode VPIN was built to address.

The one-line takeaway

Easley, Kiefer, O'Hara and Paperman showed that you can count informed traders you cannot see, because information only pushes one way: informed trading reveals itself as abnormally lopsided days of buys versus sells, and the frequency of those lopsided days tells you the probability that the person on the other side of your trade knows something.