An earnings surprise needs an unrevised expectation

The expectation must precede the event

An earnings surprise compares a reported outcome with an expectation that existed before the release. Using the latest consensus downloaded after the announcement can erase or reverse the surprise. Record the estimate snapshot's timestamp, contributing-analyst coverage, fiscal period and measurement basis. An accurate actual result paired with the wrong expectation is still an invalid signal.

LSEG's I/B/E/S Actuals documentation distinguishes comparable actuals, restatement actuals and go-forward actuals, and describes timing and normalization fields. That distinction matters because analyst estimates may use an operating basis different from reported GAAP earnings. The documentation is evidence of field concepts, not evidence that a researcher possesses the necessary historical feed or has reconstructed its vintages correctly.

Calculate the same event three ways

Suppose fictional adjusted EPS is $1.21 and the last eligible pre-release consensus is $1.10. The difference is $0.11, or 10% of the absolute consensus. If a later consensus snapshot reads $1.18, the same actual produces only about 2.54%. That smaller number answers a different, contaminated question because the expectation has changed after the event.

If the company's GAAP EPS is $0.95, comparing it with the adjusted $1.10 expectation produces approximately -13.64%. The arithmetic is correct but the measurement bases are mismatched. The SEC's non-GAAP guidance emphasizes reconciliation with comparable GAAP measures; for research, retain both reported and adjusted definitions rather than treating whichever number is easiest to retrieve as interchangeable.

Choose a scaling rule with known failure cases

A percentage surprise divided by absolute consensus becomes unstable when consensus approaches zero. An absolute EPS difference avoids that denominator but is not directly comparable across share prices or split histories. A standardized surprise might divide by historical forecast-error variability, which creates another estimated input. Choose the scale in advance and document what happens with zero denominators, sparse histories and missing forecasts.

Match actual and expected figures on fiscal period, currency, share basis and dilution convention. Do not pair a next-quarter forecast with the just-reported quarter because both share a convenient calendar label. Track stock splits affecting per-share values. If the consensus excludes a category of expense, verify that the actual adjustment policy is genuinely comparable rather than relying on an adjusted field name.

Separate an event study from a tradable strategy

A post-announcement return study must define the release timestamp and the earliest eligible trading decision. A close-to-close event return may include a gap that an after-release strategy could never capture. Analyst coverage also selects the sample: firms with usable estimates may differ systematically from those without them. Show coverage and exclusion counts alongside surprise distributions.

Historical findings about earnings effects are hypotheses to investigate, not guarantees of a repeatable opportunity. A complete evaluation would need original consensus vintages, comparable actuals, executable timing and costs. This guide uses invented EPS inputs solely to expose measurement sensitivity. No licensed estimates were downloaded, no actual earnings events were analyzed, and no predictive return claim follows from the calculations.

Inspect the example

Sources & reproduction

Primary references support the definitions and method. Numerical examples are illustrative unless a source record is explicitly identified.

• LSEG: Actuals Data for I/B/E/S Estimates
• SEC: Non-GAAP Financial Measures

Continue the review