Where VWAP Came From: Berkowitz, Logue and Noser Measure the Real Cost of Trading

If you have ever heard a trader say "we'll work it VWAP," you have heard the echo of this paper. The volume-weighted average price is the most widely used execution benchmark in the world, embedded in broker algorithms, in institutional mandates, and in the daily language of every equity desk. It came from here.

What is nice about the paper is that VWAP was not the point. It was a tool the authors needed to answer a different and more basic question, and it happened to be so useful that the industry ran off with it.

The problem: how do you tell a good fill from a bad one?

Suppose a pension fund buys a large block of shares and the average price paid is 50.20. Was that good or bad?

You cannot answer without something to compare it to, and every obvious comparison has a fatal flaw.

• Compare to the price before you started. Fine, but the market moves on its own. If the stock was going up all day for reasons unrelated to you, you look terrible even if your trader did a superb job.
• Compare to the closing price. Same problem, in reverse, and it lets a trader look like a hero by pure luck.
• Compare to the best price of the day. Unfair to the point of absurdity. Nobody trades at the low.

What the authors wanted was a fair yardstick: a price that represents what an ordinary, unremarkable participant would have paid for that stock on that day, with no skill and no cheating.

The key idea via analogy: the price everybody else got

Think about buying petrol on a road trip. The price at the pump moves around all day and varies station to station. To judge whether you got a good deal, you would not compare against the cheapest station in the country. You would ask: what did the average driver pay for a litre today?

And crucially, you would weight that average by how much petrol was actually sold at each price, not just by how many price signs existed. A station that sold one litre at a silly price should barely count. A station that sold ten thousand litres should count a lot.

That is exactly VWAP. Take every trade in the stock during the day, weight each price by the number of shares traded at it, and average. The result is, in a very real sense, the price the market as a whole paid.

Now the comparison becomes meaningful. If you bought at 50.20 and the day's VWAP was 50.35, you did better than the crowd. If VWAP was 50.05, you paid up. And because VWAP moves with the market, it automatically strips out the "the stock just went up today" problem that ruins the naive benchmarks.

Armed with this yardstick, the authors went and measured the actual cost of institutional trading on the New York Stock Exchange across a large sample of real institutional trades. The headline finding was one that surprised people at the time: the market impact component of trading cost was small on average, and considerably smaller than the commissions being paid. In other words, the cost institutions worried about least, brokerage commissions, was the bigger line item, while the mysterious and much-feared "impact" was, on average, modest.

That average, though, hides everything interesting. The finding is about typical trades. A genuinely large, urgent order in an illiquid name is a different animal entirely, and later research spent decades mapping exactly how impact scales with size.

Why it mattered

• It made execution quality measurable at all. Before a credible benchmark existed, arguments about whether a broker was doing a good job were pure assertion. VWAP turned it into arithmetic. The entire discipline now called transaction cost analysis grew from this root.
• It became the target, not just the measure. This is the great irony of the paper. Once funds started grading their traders against VWAP, traders started explicitly trying to beat VWAP, and brokers started selling algorithms designed to track it. A VWAP algorithm slices your order in proportion to expected volume through the day, guaranteeing you land close to the benchmark. Billions of shares are traded this way daily, all because of a measurement tool.
• It quantified something the industry was arguing about blind. Putting real numbers on institutional trading costs, and showing commissions were the larger piece for typical trades, changed how funds negotiated with brokers.

The honest limitations

• VWAP is easy to game, and everyone knows it. If you are being graded against the day's VWAP, the safest thing you can do is trade in proportion to volume all day. You will land almost exactly on the benchmark and score a perfect grade. But that is not the same as doing a good job for the investor, who might have been far better served by finishing early. VWAP rewards conformity, not quality.
• It ignores the decision price entirely. This is the deep flaw, and it is precisely the hole Perold's implementation shortfall was designed to fill. If the portfolio manager decided to buy at 9:30 and the trader dawdled until the stock had risen two percent, the trader can still hit VWAP beautifully and the fund has still lost two percent. VWAP is blind to delay cost and blind to orders you never filled.
• You are part of the benchmark. If your order is a big chunk of the day's volume, your own trades are inside the VWAP calculation. Beating a benchmark you are helping to set is a strange exercise, and for very large orders the measure becomes close to meaningless.
• It suits patient, uninformed flow only. If you are trading because you have genuine short-lived information, spreading your order across the whole day to match VWAP is the worst thing you could possibly do. Your edge decays while you are busy tracking a benchmark.

The one-line takeaway

Berkowitz, Logue and Noser needed a fair way to judge institutional fills, so they proposed comparing them to the volume-weighted average price, the price the rest of the market paid that day, and in doing so they accidentally created both the industry's favourite execution benchmark and the algorithm designed to chase it.