Skip to main content

Earnings Surprises & Revisions

Updated Jun 24, 2026 at 2:35pm

Research Draft High 1,077 words

An earnings surprise is the gap between a company's reported result (usually EPS, sometimes revenue) and the Wall Street consensus estimate that preceded it; an estimate revision is the change in analysts' forward forecasts, typically triggered by that report or by interim news. The two are tightly linked — a surprise often sets off a wave of revisions — and both matter because markets price the expectation, not the absolute number. The core tension of the topic is that a result can be objectively good and still crush the stock, because what trades is the surprise relative to a moving, partly hidden bar. Among fundamental-analysis concepts, this is unusual: it has a genuinely well-documented return anomaly attached to it (post-earnings-announcement drift), making it one of the few places where fundamentals and price behavior provably intersect.

How it's calculated / formed

Raw surprise. Surprise = Actual EPS − Consensus EPS, often expressed as a percent of the estimate. Consensus is the mean (sometimes median) of sell-side EPS forecasts compiled by data vendors — Bloomberg, FactSet, LSEG (formerly Refinitiv), and S&P Capital IQ are the dominant aggregators (Investopedia; AAII).

Standardized Unexpected Earnings (SUE). The academic workhorse. SUE scales the surprise by its own variability so a 2-cent beat at a stable utility isn't equated with a 2-cent beat at a volatile growth name. Two common forms: divide the surprise by the standard deviation of analyst forecasts (cross-sectional disagreement), or — the original Foster/Bernard-Thomas version — by the standard deviation of past seasonal-random-walk forecast errors (time-series). Higher absolute SUE = larger, more statistically meaningful surprise (WestGA/B>Quest; Wikipedia: Earnings surprise).

Revisions. Tracked as the direction, magnitude, and breadth (count of up vs. down) of analyst EPS changes, and as revision momentum (the trend over recent weeks). Vendors also publish a whisper number — an unofficial buy-side expectation that frequently sits above published consensus.

How it's used in practice

Three distinct uses dominate:

1. Surprise as a signal of mispriced information. Large positive SUE flags that the market may be underreacting; this is the basis of earnings-momentum strategies that buy high-SUE deciles. 2. Revisions as a leading indicator. Because analysts adjust forecasts incrementally and with herding, the trend of revisions tends to continue. Rising consensus is itself a bullish signal independent of the last print — this is the logic behind systems like Zacks' rank and many quant "estimate-revision" factors. 3. Surprise-vs.-expectation as event interpretation. Practitioners decode why a stock beat consensus and still fell: it missed the higher whisper number, or — most commonly — management's forward guidance disappointed. The forward outlook routinely moves the stock more than the backward-looking beat (heygotrade). The headline beat is necessary but rarely sufficient.

Adoption, debate & evidence

This is one of the most-studied effects in finance. Ball and Brown (1968) first documented drift; Bernard and Thomas (1989, 1990) sharpened it, finding SUE-decile long/short portfolios earned roughly ~8–9% per quarter pre-cost in their sample, with the spread positive in 41 of 48 quarters (1974–1985). They attributed it to investor underreaction — prices behaving as if earnings followed a naive seasonal random walk and failing to recognize earnings autocorrelation. Notably, they found ~25–30% of the drift clusters in the 3-day windows around the next earnings date, strong evidence the effect is delayed reaction, not risk (Wikipedia: PEAD; Caltech/PEAD review).

On revisions, Chan, Jegadeesh & Lakonishok (1996, Journal of Finance) reported a spread of roughly ~7.5% over the following six months between the extreme standardized-earnings-surprise portfolios (1973–1993 sample), and linked momentum partly to analysts' sluggish, herding reaction to earnings news (Alpha Architect summary; CJL 1996 paper).

The honest caveat: the effect has decayed. Research finds classic numerical PEAD largely disappeared in non-microcaps after ~2001 and was near zero for large caps by ~2006, attributed to decimalization, Reg NMS / electronic arbitrage, and faster announcement-day adjustment (Wikipedia: PEAD). Kettell, McInnis & Zhao (2022) tie the decline to falling persistence of SUE itself (Columbia working paper). And studies estimate transaction costs consume 70–100% of paper PEAD profits, with the residual drift concentrated in illiquid, low-institutional-ownership names — exactly where it's hardest to harvest (Liquidity and PEAD, Financial Analysts Journal 2009).

Folklore vs. measured: the beat-rate is theater. By FactSet's tally the long-run share of S&P 500 companies reporting EPS above consensus runs roughly the mid-70s to low-80s percent (its 5- and 10-year averages have sat near 76–78%) (FactSet Earnings Insight) — because companies and analysts cooperate to set a beatable bar ("sandbagging"/guidance walk-down). A beat is therefore the baseline, not an edge; the size and surprise relative to the whisper/guidance is what carries information.

Strengths & limitations

When it works: SUE and revision momentum carry real predictive content, best in smaller, less-liquid, less-followed names where information diffuses slowly. Revision breadth and trend remain a respected quant input. Surprise is also a clean, objective, frequently-updated data point.

When it fails: In large, liquid, heavily-arbitraged stocks the drift is largely priced out in seconds. Naive "buy the beat" rules ignore that beats are ubiquitous and that guidance dominates. The biggest misuse is treating a headline beat as bullish without checking the bar — the whisper number, the magnitude vs. SUE, and especially forward guidance. A secondary trap is acting on stale consensus near the print, when the buy-side has already moved the effective bar higher.

Sources

Disputes flagged: (1) PEAD's cause — underreaction (Bernard-Thomas) vs. risk/limits-to-arbitrage — remains debated; the weight of evidence favors underreaction. (2) Whether any exploitable drift survives in liquid large-caps today: most recent work says no, net of costs.