The Spread You Can See Without Seeing the Quotes: Roll's Implied Bid-Ask Estimator

Every trade you make costs something before the market even moves. You buy at the ask, you sell at the bid, and the gap between the two is a toll you pay for the privilege of trading right now. That gap is the bid-ask spread, and it is one of the most important numbers in finance.

There is just one problem. For most of financial history, nobody recorded the bid and the ask. Old databases stored the closing price and the volume, and that was it. If you wanted to study trading costs across decades of stock history, or across a market that never published quotes, you were stuck.

In 1984, Richard Roll pointed out something that sounds almost like a magic trick: the spread leaves a fingerprint in the sequence of trade prices themselves. You do not need the quotes. You can read the toll off the receipts.

The problem: costs you cannot observe

Imagine you want to answer a simple question. Are small companies more expensive to trade than large ones? To answer it you need the spread for thousands of stocks. But if your data has only daily closing prices, you have no spread column, no bid, no ask. You have nothing to measure.

Researchers faced this everywhere: historical equity data, thinly traded stocks, foreign markets, bonds, anything before high-quality intraday feeds existed. The cost of trading was a first-order fact about a market, and it was invisible.

The key idea via analogy: the bouncing ball

Here is the intuition. Picture the stock's "true" value as a ball drifting randomly through the air. Nobody can trade at the true value. Instead, there are two invisible walls, one just below the true value (the bid) and one just above it (the ask).

Every actual trade is a print on one of those two walls. A buyer hits the ceiling. A seller hits the floor. Now watch the sequence of prints. If the true value is not moving at all, and trades arrive as a random mix of buyers and sellers, the printed price will bounce back and forth between floor and ceiling: up, down, down, up, up, down. That is the famous bid-ask bounce.

And bouncing has a statistical signature. It means an up-tick tends to be followed by a down-tick, purely as an artifact of which wall got hit, not because of any real news. In plain terms: price changes look negatively correlated with themselves, and the wider the walls are apart, the stronger that negative correlation is.

Roll's insight is that you can run this logic backwards. Measure how strongly consecutive price changes tend to reverse each other. The stronger the reversal, the wider the spread must be. From that single statistic, and nothing else, you get an estimate of the effective spread. All you need is a history of prices.

The beautiful part is the assumption structure. Roll assumed the market is efficient, so the true value moves as a random walk and carries no predictable pattern. That means any systematic tendency for prices to zig-zag cannot be coming from the true value, because the true value is unpredictable by assumption. It has to be coming from the trading mechanism. The spread is the only suspect left.

Why it mattered

Roll's estimator did three big things.

• It made trading costs measurable across all of history. Suddenly you could estimate spreads for any stock, any era, any market, as long as you had a price series. Roll himself found that his implied spread lined up with firm size, which was exactly the kind of empirical fact that had been hard to establish before.
• It gave microstructure its first bridge to asset pricing. Once you could put a number on liquidity for thousands of stocks, you could ask whether illiquid stocks earn higher returns. That question, taken up shortly afterwards by Amihud and Mendelson and many others, turned liquidity into a genuine pricing factor.
• It founded an entire genre. The idea that the microstructure of trading contaminates observed prices in a measurable way became the seed of what we now call market microstructure noise. Every later low-frequency spread estimator, and a large slice of the high-frequency volatility literature, descends from Roll's basic move: separate the true value from the trading artifact by looking at the correlation structure.

The honest limitations

Roll was refreshingly candid, and the follow-up literature was blunter still.

• Sometimes the answer is imaginary. The estimator only works if consecutive price changes really do tend to reverse. In practice, for a substantial fraction of real stocks in real samples, they do not: the measured relationship comes out with the wrong sign, and the formula asks you to take the square root of a negative number. Researchers usually respond by throwing those observations away or setting them to zero, which is exactly the sort of quiet fudge that should make you nervous.
• The model assumes trades are a coin flip. Roll's setup treats buy and sell orders as arriving independently, like flipping a fair coin at each moment. Real order flow is nothing like that. Buys tend to follow buys, because large traders slice big orders into small pieces. That persistence in order flow works against the bounce and biases the estimate.
• It assumes the spread never changes and the market maker never learns. In reality spreads widen and narrow all day, and market makers adjust their quoted midpoint after seeing an informed trade. Both effects break the clean floor-and-ceiling picture.
• It measures the whole spread, not its parts. Roll tells you how wide the walls are. He cannot tell you why they are wide: whether the market maker is charging for inventory risk, for order processing, or for the danger of trading against someone who knows more. Later work, notably Glosten and Harris, took on that decomposition.

None of this made the estimator useless. It made it a first, cheap, back-of-the-envelope read that you should sanity-check against something better when better data exists.

The one-line takeaway

Roll showed that the cost of trading is written into the wiggle of the price series itself: because trades ping-pong between the bid and the ask, prices reverse more often than an efficient market alone would explain, and the strength of that reversal tells you how wide the spread is, using nothing but prices.