Earnings Revisions and the Equity Cycle: What Consensus Signals, and What It Manufactures

Analysts revise their earnings estimates every quarter, and those revisions move share prices company by company. Added up across an index, they follow a ritual that makes them misleading when read as a market signal.

Consensus almost always drifts lower ahead of reporting season, then companies beat it. Reading a revision starts with knowing that slope.

What a revision measures

An earnings revision is an analyst’s change to an earnings-per-share estimate. FactSet turns them into a weekly measure for the whole index: the “bottom-up” estimate, built from the median estimates for all the companies in the S&P 500 and used as a proxy for index earnings. Its change between the start and end of a quarter sums up what analysts learned during that quarter.

The common reading treats it as a leading indicator for the market. When analysts raise their numbers, prices are supposed to follow, and when they cut them, the index is supposed to be setting up for a fall. The intuition works at company level. Applied to the index, it runs into two obstacles: a systematic slope in estimates, and the role of the discount rate.

Consensus forecasts earnings; it does not set their price.

Consensus walks down

The first obstacle is visible in every edition of FactSet’s Earnings Insight. In a typical quarter, analysts lower their estimates as the reporting date approaches. According to the October 2, 2026 edition, the bottom-up estimate for a quarter fell an average 2.2% during the quarter over the past five years, and more over ten, fifteen and twenty years.

Accounting research calls this the walk-down. In Contemporary Accounting Research (2004), Scott Richardson, Siew Hong Teoh and Peter Wysocki describe a “game” in which analysts first publish optimistic forecasts and then bring them down to a level the company can beat. They find the walk-down most pronounced when the firm or its insiders are net sellers of stock after the announcement.

Louis Chan, Jason Karceski and Josef Lakonishok (Journal of Financial and Quantitative Analysis, 2007, based on a 2003 NBER working paper) tie the pattern to analysts’ conflicts of interest during the 1990s bull market. Adjusting forecasts to avoid disappointments, they argue, shifted the distribution of surprises towards the positive.

The outcome is built in.

According to FactSet, a large majority of S&P 500 companies reported earnings above the mean estimate over the past five years (August 28, 2026 edition). In the second quarter of 2026 the share was the highest since the second quarter of 2021. A high “beat rate” is the norm the process produces. Reading surprises themselves as a cycle indicator is the subject of what earnings surprises say about the cycle.

Read a revision against the slope

If estimates fall 2.2% in an average quarter, a cut only carries information through its distance from that norm. The second quarter of 2025 is a case in point: FactSet’s bottom-up estimate fell by more than its average over the preceding five years. The revision was negative, and steeper than usual, in the quarter of the April 2025 US tariff announcements.

An increase, by contrast, is the exception. In the fourth quarter of 2025, the same estimate rose slightly between September 30 and December 19. In the third quarter of 2026 it gained again between June 30 and September 30, according to the Earnings Insight of October 2, 2026. In the same quarter, the number of companies issuing guidance above consensus reached a record according to FactSet.

The useful reading rests on two gaps: the quarter’s revision against its historical average, and the share of positive company guidance against its own. The size of the revision on its own mixes information with ritual.

At index level, the link fades

The second obstacle runs deeper. For a single company, good earnings news lifts the share price, and research has shown that prices take several quarters to absorb it. For the market as a whole, S. P. Kothari, Jonathan Lewellen and Jerold Warner (Journal of Financial Economics, March 2006) find a different pattern: aggregate returns are unrelated to past earnings, and they are negatively correlated with earnings in the same quarter.

Their explanation is the discount rate. Aggregate earnings and discount rates move together over the cycle: when profits accelerate, rates often rise too, and the higher discount rate can outweigh the profit gain in the index price. Their February 2003 MIT working paper concludes that discount-rate shocks explain a significant fraction of market returns.

Nineteen quarters of evidence

That result covers roughly 1970 to 2000. To test it on recent data, we matched, quarter by quarter, the revision to the bottom-up estimate published by FactSet after each quarter closed with the total return of the US market over the same quarter, computed from Kenneth French’s data. Nineteen of twenty-two quarters between early 2021 and mid-2026 could be rebuilt. The test covers revisions, not the reported earnings the three authors studied.

The correlation between the two is zero over the period. The negative sign of the 2006 study does not show up, but neither does any positive link. Of the four deepest cuts to consensus, only one came with a falling market: the third quarter of 2022 (−3.9% for the market). The other three (the fourth quarter of 2022, the first quarter of 2023 and the fourth quarter of 2023) came with market gains of 6.6%, 7.5% and 12.0%.

Conversely, the worst quarter of the period for the market, the second of 2022 (−17.2%), saw the estimate fall only modestly. Over that quarter, the 10-year US Treasury real yield went from −0.52% to 0.65% (DFII10 series): the discount rate moved, expected earnings barely did.

Nineteen quarters do not make a statistical proof. They are enough to rule out a mechanical link between the direction of revisions and the direction of the index over the same period.

Two recent quarters show the same decoupling. In the second quarter of 2025 the estimate was cut more than usual, while the US market returned 11.4% including dividends between the end of March and the end of June 2025 (total-return index from Kenneth French’s data library). In the third quarter of 2026, expected earnings over the next twelve months, a measure that rolls forward a quarter and so takes in a later period, rose much faster than the S&P 500, according to FactSet. The forward price-to-earnings ratio fell, as the 10-year real yield rose from 2.20% to 2.93%.

The market is not ignoring analysts. It combines their information with a discount rate whose real component went from −1.04% at the end of 2021 to 2.93% at the end of September 2026, the channel set out in how equity markets anticipate the economic cycle.

What revisions still tell you

These limits do not make revisions useless. They change the question to ask of them. An abnormal revision, in either direction, says something about the path of profits before the accounts confirm it. The second quarter of 2026 showed it: on June 30, the index reported year-over-year earnings growth far above what consensus expected, the strongest since the second quarter of 2021, according to FactSet.

Dispersion across sectors is the second contribution. A positive revision for the index can come from a handful of sectors while the rest decline. For the third quarter of 2026, FactSet expects double-digit growth in several sectors, led by Energy, Information Technology, Communication Services and Materials. The index figure does not say whether the improvement is broad or narrow, and that is often where the cycle reading is decided.

Long-horizon optimism, finally, remains well documented. Somnath Das, Carolyn Levine and K. Sivaramakrishnan (The Accounting Review, 1998) show that analysts issue more optimistic forecasts for companies whose earnings are hard to predict. Consensus for a distant year therefore tends to start high, before the walk-down brings it back towards what gets reported.

What weakens this reading

FactSet’s averages shift from one edition to the next: the five-year average decline changed noticeably between June 2025, December 2025 and October 2026 (2.2%). The norm depends on the window, and a run of unusual quarters moves it. A careful reading compares a quarter’s revision with several horizons at once.

The Kothari, Lewellen and Warner result has its own scope. It concerns earnings and returns in the same quarter, measured over three decades that end in the early 2000s. It does not prove that aggregate revisions carry no value, only that no simple, stable link ties earnings consensus to the level of the index.

What to watch

Three public markers are enough to read an earnings season without being taken in by the ritual. The change in the bottom-up estimate during the quarter, set against its five- and ten-year averages. The share of companies raising guidance, set against its own. And the forward price-to-earnings ratio, which shows whether the market is paying more or less for the same profits; how it differs from the CAPE is covered in three equity valuation measures compared.

When the first two markers rise and the third falls, as in the third quarter of 2026, earnings are improving and the market is paying less for each dollar of them. That is neither a buy signal nor a sell signal. The simultaneous rise in real yields is the simplest explanation, without ruling out a higher risk premium or doubts about estimates seen as too high.

Last updated — 8 October 2026