What an earnings 'beat' actually measures

On July 31, 2026, Exxon reported adjusted earnings of $3.52 per share. BNN Bloomberg called it an 8-cent miss against an LSEG consensus of $3.60. Yahoo Finance called it a miss against $3.63, which makes it 11 cents.

Same company, same quarter, same $3.52. The shortfall was either 2.2% or 3.0% depending on which page you happened to open.

Neither outlet made a mistake. That three-cent gap is just what consensus is, and once you understand where the number comes from, a lot of earnings coverage starts reading differently.

How often do S&P 500 companies beat earnings estimates?

About four times in five: FactSet puts the 5-year average at 78%. Start with the base rate, because it reframes the whole vocabulary.

FactSet tracks this every quarter. In its S&P 500 earnings season update of July 24, 2026, with 27% of the index having reported Q2: “86% have reported actual EPS above estimates, which is above the 5-year average of 78% and above the 10-year average of 76%.”

Read that again. In a normal year, 78% of the largest companies in America beat expectations. Over ten years, 76%.

A thing that happens roughly four times out of five is not an achievement. It is the default. When a company beats, it has done what the overwhelming majority of companies do every quarter, and the headline “Company X beats estimates” carries about as much information as “Company X filed its 10-Q.”

The interesting cases are the other 22%.

Where does the consensus estimate come from?

From data vendors that average sell-side analysts’ forecasts. There is no official expectation. Nobody at the SEC or the exchange publishes one.

What exists is this: sell-side analysts at banks and brokerages each build a model of the company and publish an EPS forecast. Data vendors — LSEG (formerly Refinitiv), FactSet, Bloomberg, S&P Capital IQ, Zacks — collect those forecasts and average them. That average is “consensus.”

Every step introduces variation:

• Different vendors poll different analysts. One might have 28 estimates in its panel, another 24, and the overlap is not complete.
• Different cutoff times. An analyst who revises the morning before the print may or may not be in the average, depending on when the vendor froze it.
• Stale models. Some analysts do not update between quarters. Their number sits in the average anyway, anchored to information that is three months old.
• Different treatment of adjustments. More on this shortly, and it is the big one.

So when a headline says a company missed by 8 cents, the honest translation is: reported EPS came in below the average of the particular set of forecasts that one particular vendor happened to be holding at one particular moment. The direction is usually reliable. The precision is not. Anything inside a few cents is inside the noise of how the number was assembled.

Is a beat measured against GAAP or adjusted EPS?

Almost always adjusted EPS, a figure the company defines itself. Here is the part that surprises people most.

Beats and misses are almost always measured against adjusted EPS, not GAAP EPS. GAAP is the standardized accounting figure. Adjusted is the company’s own presentation of what it thinks its “real” earnings were, after excluding items management considers unrepresentative — restructuring charges, acquisition costs, litigation settlements, asset writedowns, sometimes stock-based compensation.

Some of those exclusions are perfectly reasonable. A genuine one-time legal settlement really does obscure the run rate. But the company decides what goes in the bucket, and the same company decides it every quarter. A “one-time” restructuring charge that appears for six consecutive quarters is not one-time, and it is still being excluded.

This is worth checking rather than assuming, and the direction is not always what you expect. In Chevron’s second quarter of 2026, adjusted earnings of $11.98 billion came in below GAAP net income of $12.07 billion, because special items happened to be net favorable. That is the unusual direction. Far more often adjusted sits above GAAP, and the gap is where the story is.

When the gap between GAAP and adjusted is small, adjusted is a fair yardstick. When it is large and persistent, the beat is being scored against a number the company constructed.

Who sets the expectations a company beats?

Largely the company, through its guidance. Management issues guidance. Analysts build their models substantially off that guidance. Consensus is then an average of models built off the company’s own forecast, and the company then reports against it.

If you were graded on a test where you got to tell the examiner in advance roughly what score to expect, you would learn to suggest a number you could clear. Companies have learned this. It is not fraud and it is not secret; it is simply how the incentive runs. A 78% long-run beat rate is what that incentive looks like in aggregate.

The practical consequence: a small beat against guided-down expectations can coexist with a business that is deteriorating. The beat tells you management calibrated well. It does not tell you the quarter was good.

Why can a company beat estimates and still get a negative write-up?

Because the beat scores the past quarter while the analysis turns on guidance and what changed. This is the question that brings most people here: the company beat, so why does the write-up sound negative?

Three reasons, roughly in order of frequency.

The market was expecting more than consensus. Published sell-side consensus is not the same as what large investors actually expect. The unpublished, higher bar — sometimes called the whisper number — is the bar some investors hold in mind. A company can clear the published estimate and miss the real one.

The guidance changed. A reported quarter is history. Guidance is the future, and it is usually where the analysis turns. Roblox’s Q2 2026 is the clean example: revenue grew 36%, but full-year bookings guidance came down by roughly $900 million at the midpoint. The beat was real and irrelevant.

One sentence changed the story. Reddit’s Q2 2026 beat on revenue, beat on EPS, and guided above consensus. The caveats were in the letter, which described search referrals as choppy, and in the US daily active uniques figure, which declined sequentially. Everything in the quarter was good except the part that described what happens next.

Notice what these have in common. In all three the market was pricing forward-looking information, and the beat-or-miss headline was scoring backward-looking information. They are answering different questions.

Can the aggregate earnings surprise mislead?

Yes — a few huge results can dominate it. Even the surprise magnitude can mislead when you zoom out. That same FactSet update reported companies beating “by 39.3% above estimates, which is above the 5-year average of 7.0%.”

A 39.3% aggregate surprise is not evidence that corporate America suddenly became five times better at its job. Q2 2026 contained a wartime energy windfall, and a handful of enormous energy results distort an aggregate like that badly. Averages across an index hide their own composition. If a statistic looks like a regime change, check which sector produced it before believing it.

What should you read instead of beat or miss?

The level and trend, the guidance change, cash flow, the GAAP-to-adjusted bridge and segment mix. None of this means ignore earnings. It means score them yourself. In rough order of how much they matter:

1. The level and the trend, not the delta. Is revenue growing, decelerating, or shrinking? Compare this quarter to the last four, not to a forecast.
2. The guidance change. Raised, held, or cut, and by how much at the midpoint. This is usually the most price-relevant sentence in the release.
3. Cash flow. Operating cash flow and free cash flow are far harder to present flatteringly than adjusted EPS. When earnings and cash flow diverge for more than a quarter or two, believe the cash.
4. The GAAP-to-adjusted bridge. Look at what was excluded and whether the same items keep reappearing.
5. Segment mix. A company can beat on the total while its most important business deteriorates. The segment table is where that shows up, and it is the part of the release almost nobody reads.
6. Where the beat came from. A beat driven by a lower tax rate, a buyback shrinking the share count, or a one-off gain is arithmetic. A beat driven by margin expansion is operations. They deserve different weight.
7. Share count. EPS can rise while total profit falls, if enough stock was retired.

What should you check before reacting to an earnings headline?

Before you react to any earnings headline:

• Was the beat against GAAP or adjusted? How wide was the gap?
• Did guidance go up, down, or nowhere?
• Did operating cash flow move in the same direction as earnings?
• Is the growth rate accelerating or decelerating against the last four quarters?
• Did the share count fall enough to flatter EPS on its own?
• Which segment actually produced the change?

If a headline gives you none of that, it has told you the company landed near a forecast. That is genuinely all it has told you.

What does an earnings beat measure, in short?

“Beat” measures the distance between a company’s own adjusted figure and an average of analyst models that were largely built from that company’s own guidance, collected by a vendor whose panel differs from the next vendor’s. Four out of five companies clear it in a typical year.

It is a fact about the forecast. It is not a fact about the business. Once you stop reading it as a grade, earnings season gets considerably easier to interpret.

---

Sources and method. Beat-rate and surprise-magnitude figures are quoted from FactSet’s S&P 500 earnings season update of July 24, 2026, linked above. The Exxon consensus figures are as reported by BNN Bloomberg (LSEG, $3.60) and Yahoo Finance ($3.63) on July 31, 2026; the Chevron, Roblox and Reddit figures are from those companies’ reported results and are sourced in full on their individual analysis pages. The description of how consensus is assembled, and the reading of the incentives around guidance, are our own analysis. Educational content, not investment advice.

Worked examples from the same week: Exxon Q2 2026 out-earned Chevron by $2.7 billion and was reported as a miss, while Chevron Q2 2026 earned less and was the day’s winner.