Sector Earnings Trends: How to Read Results Season

In the same quarter, on the same line item, three large American banks told investors three different things. Bank of America raised its 2026 net interest income growth guidance to 6%–8%. Truist cut its own to 2%–3%, from 3%–4%. Huntington said it now expected to land "at the low end of our guided range." Same three months, same rate environment, same metric, and one raise against two downgrades.

Any sector summary published that quarter would have reported bank earnings as a single direction, because that is what a summary is for. The direction was not single. And the most useful fact of the three sits buried inside the raise: Bank of America's finance chief attributed it to the interest-rate curve shifting "from 2 rate cuts expected to having none currently." The bank had not discovered new lending. Its forecast improved because its assumption about the Federal Reserve changed.

That distinction — a business change against an assumption change — is most of what sector analysis actually is, and it is completely invisible at the aggregate level. What follows is a method for getting at it, illustrated with real companies, and it is a lesson in reading rather than a view on any of them.

Why the sector aggregate hides the trend

"Sector earnings grew 8%" is a fact about the past, assembled from companies that reported on different dates with different year-ends, different accounting and different business mixes. It arrives after the market has priced it, and in the assembling it averages away the one thing you actually wanted: which companies inside the group are describing a changing world, and in which direction.

The forward view is not in the aggregate at all. It sits in the guidance, said out loud, company by company, inside a three-week window every quarter, scattered across forty transcripts that nobody bothers to assemble. Which is precisely why assembling it is worth something.

Step one: build a cohort, not a sector

A sector is too coarse a unit to compare across. Banks and insurers are both financials and share almost no operating metric worth the name. What you want is the narrowest group whose managements are answering the same questions.

The test is simple enough to apply in your head: a cohort is a set of companies for which the same three KPIs matter. For US banks that means net interest income, net interest margin and provisions, with the CET1 ratio sitting behind them. For enterprise software it means annual recurring revenue, net dollar retention and remaining performance obligations. For industrials it means book-to-bill, backlog and price against cost. If two companies do not share those, they are not a cohort, whatever the sector label on a screener says. Six to twelve names is enough — beyond that you gain very little and, more to the point, you stop reading properly.

Step two: track one shared line across the whole cohort

Pick the single metric the cohort competes on, and read only that, across every company, in the same quarter. For the banks above, that line is the full-year net interest income guide.

Three data points, and the sector picture is already sharper than any summary of it: net interest income expectations are diverging rather than moving together. That distinction carries real weight. When a cohort moves as one, the driver is the environment and there is not much to do about it. When it splits, the driver is company-specific — balance-sheet mix, deposit costs, loan growth — and the split is telling you which managements are exposed to what.

Step three: separate the business change from the assumption change

This is the step that separates a sector read from a sector headline, and it is the one most often skipped.

Guidance rests on assumptions the company does not control: interest rates, currency, commodity prices, tariffs. So when a guide moves, the first question is whether the business changed or the assumption did. Bank of America's raise came with its own answer attached — the rate curve moved from two expected cuts to none, and net interest income improves mechanically when rates stay higher for longer. Truist's cut, in the same quarter, was attributed to deposit mix and client balances, which is the business.

Both are legitimate disclosures and neither is a criticism. They are simply not equally informative. An assumption-driven raise will reverse the moment the assumption does, and it tells you nothing whatever about how well the company is being run. A business-driven cut, in that same quarter, tells you a great deal. The habit worth building is small: next to every revision in the cohort, write one word — business or assumption — and then only compare like with like.

Step four: normalise for the sector's house style

Here is the part almost nobody does, and it changes what the word "raised" is worth.

Sectors have distinct guidance cultures, and the differences are large. Sorting the tracked US commitments into cohorts by the operating metrics their managements actually discuss:

A software company raising guidance is doing the ordinary thing: the cohort raises nine times for every cut and clears what it guides better than nine times in ten. A pharma company raising guidance is doing something comparatively unusual. The same headline therefore carries something like four times the information in one cohort as in the other, and reading them identically is how a sector rotation ends up narrated backwards. The corollary matters more than the observation: a cut in enterprise software, where cuts are rare, is a far louder signal than a cut in a cohort that revises in both directions as a matter of routine. Judge a revision against its cohort's habit, never against zero.

Step five: read the divergence, not the average

With the cohort built, the shared line tracked, each revision tagged and the house style understood, the question to ask is the one an averaging summary can never answer: who is diverging, and why?

A single company moving against its cohort is the most information-dense event of the whole results season. It is either a company-specific problem the sector does not share, or an early read on something the others have not yet admitted, and both are worth an afternoon. The company moving with the cohort has told you about the weather, which you could have got from the aggregate anyway.

A results-season routine

Before the season starts, fix your cohorts and the one shared metric for each, and do not change them mid-season — the temptation to redefine a group around whatever looks interesting is exactly how you end up with a finding that is really an artefact. As calls land, log that metric's guide and its direction for every company in the cohort, tagging each revision business or assumption. At the end of the window, look only at the divergences, the names that moved against the group. Then, for each one, read the Q&A of that call properly, because that is where the reason will be if there is one.

Assembling the same line across a dozen transcripts inside a three-week window is the part that fails by hand; the US earnings call archive is organised for exactly this, one company's guidance quarter by quarter, in a form you can put side by side.

How Inve makes a cohort readable in one sitting

The method above has one practical enemy: results season is a three-week window, and a cohort of ten companies reporting inside it is ten transcripts you have to read while they are still comparable. Miss the window and you are reading history.

Inve removes that constraint. Every US call is parsed on arrival into the same shape — guidance with the figure, period and speaker; the analyst exchanges and how directly each was answered; and a set of sector KPIs keyed to what that industry actually competes on, so a bank's call surfaces net interest income, net interest margin, CET1 and provisions while a software company's surfaces ARR, net dollar retention and remaining performance obligations. Cohort-building stops being a judgement call and becomes the shape the data already has, and pulling one line across a peer group is reading a column rather than opening ten documents. The Bank of America, Truist and Huntington net-interest-income guides in this article sit three clicks apart, and the reason Bank of America gave for its raise — a rate curve that moved from two expected cuts to none — is on the record next to the number, which is what lets you separate an assumption change from a business one without re-reading the call.

Track the same cohort over several quarters and the house style in the table above stops being a general statistic and becomes your peer group's own baseline.

Where this approach can mislead you

The strongest counter-case is that cohorts are constructed, and construction is something you can get wrong. Group by the metrics management discusses and you may end up grouping companies with genuinely different economics — a regional bank funding itself with retail deposits and a wholesale-funded peer share every KPI name and almost none of the sensitivity. If the cohort is wrong, the divergence you find is an artefact of your own grouping rather than a fact about the world. The sanity check is to ask whether these companies would recognise each other as competitors.

The cohort figures carry their own caveat. Those groupings are derived from which operating metrics each company's calls actually discuss, rather than from a standard industry classification — deliberate, but it is not GICS, and a company can land in the wrong bucket. The raise-to-cut ratios also reflect a roughly two-year window that has been broadly favourable for software, so read them as this period's house style rather than as a law.

And a raise is not good news, nor a cut bad news. The most valuable thing in the bank example was not the direction of any of the three guides. It was that one of them was driven by a Federal Reserve assumption that none of the three controls and all of them are exposed to.

Frequently asked questions

The owner's question

Results season rewards speed, and speed is precisely the wrong instinct here. The question worth carrying through a cohort is a slower one: if this whole industry is describing the same change, what does that make the industry worth in five years — and if only one company is describing it, which one is early? Neither answer is in the sector aggregate. Both are in the calls, spread across a fortnight, waiting for somebody to line them up.

Related reading

• Earnings Surprise: Why Beating Estimates Tells You Little
• How to Track Management Guidance Across Quarters