How we measure guidance reliability

One question, asked of every company we cover: when this management team forecasts its own results, how close does it come?

This is not “did the company beat consensus”. Consensus is a survey of analysts, it is available from any data vendor, and beating it says as much about the analysts as about the company. This measures the company’s own published forecast against the company’s own published outcome, over as many periods as we can resolve.

The S&P 100 — 100 companies, membership taken from the published index and each company’s identity resolved against the SEC’s own ticker-to-CIK register.

A bounded universe is a deliberate choice rather than a limitation. Guidance is voluntary and practice varies enormously: many companies issue none at all, and without a fixed membership list there is no way for a reader to tell “this company does not guide” apart from “these people failed to find it”. Fixing the universe turns every gap into a statement about the company rather than about us. It also makes the leaderboard a real claim — “the most reliable guider in the S&P 100” means something; “the most reliable guider” does not.

Source of constituents: published index membership. Expanding the universe later is a configuration change, not a rebuild.

Every figure on these pages — both halves of every comparison — comes from a company’s own filing with the U.S. Securities and Exchange Commission: Form 8-K, Exhibit 99.1, the quarterly earnings press release. Each row links to the exact document for the forecast and the exact document for the outcome.

No paid or licensed data source is used anywhere in this feature. No analyst estimate, consensus figure or third-party summary enters the record. That is partly principle and mostly practical: a record whose only asset is credibility cannot afford a link a reader is unable to follow.

A forecast cannot enter this dataset without a declared rule for how it will be judged, stated before it is stored. That constraint is enforced by the database schema rather than by process, because process is what gets skipped during a hurried backfill.

An outcome anywhere inside the stated range, inclusive of both bounds, resolves in line. Outside it, the deviation is measured from the nearest bound. Point guidance carries a tolerance band — 0.5% of the guided figure for revenue, 1% or one cent (whichever is larger) for earnings per share — because a company that guides to $2.00 and reports $2.00 has not missed.

Because landing inside a range scores zero deviation, a company could in principle look flawless by guiding wide. So we publish two more numbers beside the hit rate: the average position of the outcome within the range, and the average width of the range itself. A company that lands inside its own range every quarter, always in the top third, of a range four percent wide, is doing something specific and legible — and it is not “forecasting accurately”.

Percentages are expressed against the midpoint of the guided range, floored at $1m for revenue and $0.25 for earnings per share so that a company guiding two cents of EPS and reporting four cannot post a +100% deviation that dominates every average on the site.

Periods are matched on the issuer’s own fiscal labelling — “the first quarter of fiscal 2025” — not on calendar dates. A company with a September year end calls the quarter ending in December Q1, and so does its guidance three months earlier. Matching the label rather than the date needs no fiscal calendar and cannot silently pair a forecast with the wrong three months.

Guidance and outcome must also share an accounting basis. A forecast given on an adjusted or non-GAAP basis is resolved only against the company’s own adjusted figure, never against a GAAP result. Comparing the two would not be imprecise; it would be wrong.

Below 6 resolved forecasts, a company’s record is published in full and no verdict is stated at all. Those companies appear in no ranking on this site.

Five observations cannot separate a pattern from a coin. “Beats its own guidance 100% of the time” from a sample of three is the kind of claim that loses precisely the reader this record is built for, and it is not worth the traffic.

When a company withdraws or suspends guidance, every forecast it had outstanding is marked withdrawn — a state of its own, in the data and on the page, never counted as a miss.

Companies withdraw guidance for real reasons: a pandemic, a large acquisition, a regime change in currency or tariffs. Scoring that as a failure would be both wrong and the easiest thing in this dataset to attack. Withdrawals are shown in the distribution in their own colour so they can be seen rather than quietly dropped.

• Guidance given only as a growth rate (“organic revenue growth of 5 to 6 percent”), a margin, or a constant-currency figure. Resolving those requires importing a base figure and an assumption, and the assumption would then be doing the measuring. Companies that guide exclusively this way appear with no captured guidance, which is an honest description of what we can and cannot check.
• Open-ended guidance (“at least $5.00”). A bound with nothing on the other side of it cannot be missed.
• Guidance given on the earnings call but not in the release. It is real, and it is out of scope for now: the call transcript is not a filing, is not uniformly available, and cannot be linked as a primary source for every company.
• Segment and regional forecasts, and metrics other than total revenue and diluted earnings per share.
• Any view on the company. We do not rank companies as investments, do not derive a recommendation, and do not suggest what a reader should do with any of this.

If a row on this site misstates what a company said or reported, we want to hear about it and will correct it with the change recorded in the log below. Send the company, the fiscal period and the filing you believe we misread. We correct errors of fact; we do not remove accurate rows on request, including from the companies they describe.

Unflattering results are published, including for companies that would prefer otherwise. Contact details are in the impressum.

100 companies · 1473 forecasts on record · 677 resolved · 0 withdrawn · 23 companies past the threshold · earliest forecast 16 Dec 2021, most recent 01 Sept 2026