Consensus Estimates: How the Number Is Built and Why It Moves Before a Report

A consensus estimate is the average of the forecasts published by the analysts who cover a company, for one figure and one period: earnings per share for the current quarter, revenue for the full year, and so on. It is the number a result is measured against when a headline says a company beat or missed. It is also a moving target. The consensus on report day is rarely the consensus from three months earlier, and the difference between the two is often the more interesting number.

What goes into it

Each analyst covering a company maintains a model and publishes forecasts from it. Data providers collect those forecasts and publish summary statistics: the mean, the median, the high, the low and the number of estimates. Different providers collect from slightly different sets of analysts and treat unusual items differently, which is why two sites can quote consensus figures a cent or two apart for the same quarter.

The figure being forecast matters as much as the forecasts. Most earnings consensus is for adjusted earnings, the company's own measure that leaves out items it calls one-off, rather than the figure under standard accounting rules. A company can beat the adjusted consensus and report a loss under the standard figure in the same release. Which figure a headline is quoting is worth checking before reading anything into it. What earnings per share is made of in the first place is in earnings per share.

A hypothetical quarter

To show how the number moves, here is a made-up company, covered by twelve analysts. All figures are hypothetical.

Three months before the report, the twelve estimates for quarterly earnings per share run from $1.02 to $1.22. The mean is $1.12.

Over those three months, a larger competitor reports a soft quarter and the company's own management sounds cautious at an industry conference. Five analysts cut their estimates, by six to eleven cents each. The other seven leave theirs alone. On report day the estimates run from $1.02 to $1.22 again, because neither the lowest nor the highest analyst moved, but the mean is now $1.08.

The company reports $1.11.

Measured against the consensus on the day, that is a beat by three cents. Measured against the consensus three months earlier, it is a cent short. Both statements are true. The headline will carry the first one.

The estimate nobody updated

Look again at the high end of that range. The $1.22 estimate did not move in three months, while five colleagues cut theirs. It may be a considered view. It may also be an estimate nobody has revisited, which happens: analysts update when they have a reason to, and some coverage is maintained less actively than the rest.

Leave that one estimate out and the hypothetical consensus falls from $1.08 to $1.07. One stale forecast in twelve moved the average by a cent, which is a third of the reported beat. This is why the median, the number of estimates and the date of the most recent revisions are worth a glance alongside the mean. A mean with a wide range and a few old estimates is a looser number than the same mean from a tight, recently updated group.

Why estimates tend to drift lower into a report

The pattern in the example is not unusual. For many years, estimates for large US companies have tended to be revised down over the course of a quarter, and a clear majority of those companies have then reported above the lowered figure. Firms that track this have put the share of companies beating the earnings consensus at around three quarters in a typical recent quarter.

The usual explanation is that the estimate is partly negotiated. Companies give guidance, analysts build near it, and management teams have an interest in a figure they can clear. None of that is hidden, and it means a beat of a few cents against a consensus that was walked down during the quarter carries less information than the word beat suggests. It is one reason the market reaction often seems unrelated to the headline, a subject covered in why a stock can fall on an earnings beat.

More than one consensus

There is a consensus for every figure and every period analysts forecast. For a single company that usually means at least:

• Earnings per share and revenue for the current quarter, which the report is judged against
• The same figures for the next quarter and the full year, which the new guidance is judged against
• Earnings for the next full year, which is what forward valuation measures such as the forward P/E ratio are built on

The current-quarter number gets the attention on report day. The full-year and next-year numbers are the ones that move after it, as analysts rebuild their models from what the company said. A quarter can beat while the full-year consensus falls, and in the weeks after a report it is the second movement that shows how the result was actually read.

Revisions as their own reading

Because the consensus moves, its direction over a month or a quarter is information in itself: estimates being raised across several analysts, or cut across several. The consensus price target works the same way, and it moves far more often than the consensus rating, because an analyst can nudge a number without changing a label. The caution that applies to targets applies here too, and it is set out in what price targets are worth: a published analyst figure is a model output, updated when there is a reason.

What revisions do not do is settle anything. Rising estimates describe what analysts are publishing, not what the company will report, and they can be revised straight back down. They are a description of the conversation about a company, and a useful one, because they change far more often than ratings do.

Reading it in practice

Before a report on a stock you hold, four things together say more than the mean alone: the mean, the range, how it moved over the past month, and which figure it is for. We put those into a short pre-report routine in before a stock you own reports. For what the word consensus covers beyond estimates, and why an average of disagreeing views can represent nobody, see what market consensus actually means.

EquityBias is a market data research tool. Nothing here is financial advice. Consensus figures describe published third-party estimates, and the example company is hypothetical.