Quant Analytics: from scratch to pro · Microstructure, execution and costs · Advanced
Measuring liquidity from prices: the Roll spread and Amihud illiquidity
When quotes and order-book data are missing, liquidity can still be estimated from prices alone. Roll's spread estimator reads the bid-ask bounce in price changes, and Amihud's ratio measures how far the price moves per unit of money traded.
In this lesson
- Estimate the effective spread from the covariance of consecutive price changes (Roll)
- Compute the Amihud illiquidity ratio from daily returns and traded value
- State the assumptions behind each measure and when each fails
- Compare liquidity across assets without mixing units
Read first: Order books, spreads and depth, Time series basics: random walks, autocorrelation and stationarity
The best data on liquidity is the order book and the record of quotes. It is also the data most often missing for long histories, for many markets, and for many assets. Two classic measures need only prices and volume, and they are the base of much academic work on liquidity.
Roll: the spread hidden in price changes
Roll (1984) started from a simple picture. The true value of an asset is taken as steady over the sample, and trades alternate at random between the bid and the ask. The transaction price then bounces by the spread s between two levels. A move up is likely to be followed by a move down, so consecutive price changes have negative serial covariance. With a bounce of size s, the covariance is -s^2 / 4, and so:
s = 2 x sqrt(-cov(dP_t, dP_t-1))
where dP_t is the price change from one trade (or period) to the next. This gives an estimate of the effective spread using nothing but a price series.
Worked example: Roll
The sample covariance of consecutive price changes of a stock at about 50 is -0.0004. Then s = 2 x sqrt(0.0004), which is 0.04. On a price of 50 that is 8.0 basis points. The chart shows the formula. The estimate is steep near zero: a small error in a small covariance changes the spread a lot.
Assumptions matter. The formula expects no trend, no information in trades, and a constant spread. It breaks when the covariance estimate is positive, which happens often in short samples, and one common remedy is to treat that case as missing. At daily frequency the bounce is small compared with the daily noise, so the estimate is least reliable there.
Amihud: price response per unit traded
Amihud (2002) proposed a ratio that needs only daily returns and daily traded value:
ILLIQ = average over days of |return| / traded value
The idea is a rough price impact. If a stock moves a lot on little volume, it is illiquid. A higher value means less liquid. See Amihud illiquidity.
Worked example: Amihud
Five days. Absolute returns are 1.0%, 0.5%, 2.0%, 0.8% and 1.2%. Traded value is 50, 40, 20, 60 and 30 (millions). With returns in percentage points, the daily ratios are:
| Day | Ratio |
|---|---|
| 1 | 0.0200 |
| 2 | 0.0125 |
| 3 | 0.1000 |
| 4 | 0.0133 |
| 5 | 0.0400 |
Their mean is 0.0372 percentage points of price move per million traded. If the same returns came with ten times the traded value, the ratio would be 0.00372, so the asset would be ten times more liquid by this measure. Day 3 dominates: a 2.0% move on 20 million drives most of the average, which shows how single days can distort the ratio.
What each measure tells you
| Roll | Amihud | |
|---|---|---|
| Needs | prices | returns and traded value |
| Estimates | effective spread | price response to volume |
| Unit | price or basis points | return per currency unit |
| Weakness | undefined when covariance is positive | news days look like illiquidity |
Neither is a cost. A cost for a given order needs the size, as in slippage and market impact.
Common mistakes
- Comparing Amihud values computed with different units for volume or different currencies.
- Using a Roll estimate when the covariance is positive and reporting a spread of zero.
- Treating an Amihud ratio as the cost of a trade.
- Using a very short sample. Both measures are noisy.
- Mixing periods of different volatility in one average.
In practice
Use these measures to rank and screen assets, to spot a change in liquidity over time, and as a check where real spread data is not available. Where quotes are available, prefer them. Data on FinUties provides price series to practise these checks on; the data APIs are free after registration.
Roll spread against the size of the negative covariance
Worked numbers
Each result is recomputed from the formula shown.
| What | Formula | Result |
|---|---|---|
| covariance of consecutive price changes | -0.0004 | -0.0004 |
| Roll spread | 2 * sqrt(-cov) | 0.04 |
| Roll spread in basis points of a price of 50 | roll / 50 * 10000 | 8.0 basis points |
| day 1 ratio | 1.0 / 50 | 0.0200 |
| day 2 ratio | 0.5 / 40 | 0.0125 |
| day 3 ratio | 2.0 / 20 | 0.1000 |
| day 4 ratio | 0.8 / 60 | 0.0133 |
| day 5 ratio | 1.2 / 30 | 0.0400 |
| Amihud ratio, mean of five days | (a1 + a2 + a3 + a4 + a5) / 5 | 0.0372 |
| same returns, ten times the traded value | amihud / 10 | 0.00372 |
Check yourself
Consecutive price changes have a covariance of -0.0004. What is the Roll spread?
Spread = 2 x sqrt(-cov) = 2 x sqrt(0.0004) = 2 x 0.02 = 0.04.
The covariance in a sample comes out positive. What does Roll's formula give?
Nothing usable: the square root of a negative number is undefined. A positive covariance means trends or noise dominate the bounce. One common practice is to treat the estimate as missing, not as zero.
Stock A and stock B have the same absolute returns on the same days, but B trades ten times the money. How do their Amihud ratios compare?
B's ratio is one tenth of A's, since volume is in the denominator. B is more liquid by this measure.
Why does Amihud's measure not give a cost in basis points?
It is a ratio of price response to money traded, with units of return per currency amount. It ranks assets by price sensitivity to volume, but a cost needs a specific order size and a model such as the square-root rule.
A stock that is 'illiquid' by Amihud has a very large return on a day of tiny volume caused by news. What goes wrong?
The ratio treats the whole move as the price of low volume. News moves prices on any volume, so single days like this inflate the average. Averaging over many days, or using the median, reduces the effect.
Key terms
Sources
- A Simple Implicit Measure of the Effective Bid-Ask Spread in an Efficient Market. The Journal of Finance 39(4), 1127-1139 (Richard Roll), 1984. Accessed 9 Nov 2026
- Illiquidity and stock returns: cross-section and time-series effects. Journal of Financial Markets 5(1), 31-56 (Yakov Amihud), 2002. Accessed 9 Nov 2026
- Trades, Quotes and Prices: Financial Markets Under the Microscope. Cambridge University Press (Jean-Philippe Bouchaud, Julius Bonart, Jonathan Donier, Martin Gould), 2018. Accessed 9 Nov 2026