How to Check a Dividend Figure Before You Rely on It
Equinor has grown its dividend by 15.51% a year over five years. Over ten years it has shrunk it by 14.75% a year. Both figures are ours, both are arithmetically correct, and they describe the same payment record.
That is the awkward truth about dividend data. Most of the numbers you will read are not facts so much as the output of decisions somebody made, and the decisions are rarely shown. Here are five questions that separate a figure you can rely on from one you cannot.
Which years is it actually measuring?
A growth rate compares two years. Which two is a choice, and it changes the answer more than anything else in this article.
Equinor ASA (EQNR.OL)Energy · Oil & Gas IntegratedData as of 08 Sept 2026high confidence
Equinor's two rates disagree because the ten-year window contains a deep cut and the five-year window starts after it. Measured from the bottom, the recovery is spectacular. Measured across the whole period, the company pays less than it did. Neither number is a trick. They answer different questions, and only one of them is usually printed.
We found the same pattern across companies with long payment records, running in both directions. Some read as shrinking over five years and growing over twenty. Others read as growing over five years and shrinking over ten, because a cut had aged out of the shorter window.
Two habits protect you. Ask what window a growth figure covers before comparing it to another one, and treat a rate quoted without its window as unusable rather than as approximate. Our piece on why dividend growth figures disagree between sites works through the four decisions that produce different answers from identical data. And the aristocrats that are quietly shrinking shows what happens when a payment streak and a payment amount are allowed to tell different stories.
Which currency was it paid in?
A company can declare its dividend in one currency and be listed in another. When the exchange rate moves, the same payment rises in one currency and falls in the other.
Equinor is the clearest example we have found, and it is the same company again for a different reason: a raise in dollars and a cut in kroner, from one unchanged decision. Fifteen instruments we cover declare in a currency they are not listed in, and the data feed delivered almost all of them in the wrong one before we corrected them. A growth rate built on the uncorrected series measures the currency market rather than the dividend.
Ask which currency a dividend figure is denominated in, and be especially careful with any company whose listing currency differs from its reporting currency. The detail is in companies that pay dividends in a currency they are not listed in.
What is being counted as a payment?
"Total dividends for the year" sounds unambiguous and is not.
A special dividend can double a year's total and then not repeat, which turns an ordinary year into an apparent collapse. A company moving from two payments a year to four changes every per-payment comparison while the annual total is unchanged. A European company paying an interim and a final is not paying two equal halves, and treating them as such misreads the year badly.
Cognex is the case that taught us to check. Its dividend has grown 7.63% a year over five years on the regular payments alone. Include a large one-off payment sitting at the start of that window and the same five years read as a decline, because the comparison is now against a year that contained something which was never going to repeat. It carries a score of 87.
Cognex Corporation (CGNX.US)Technology · Scientific & Technical InstrumentsData as of 08 Sept 2026high confidence
The question to ask is whether a total includes one-off payments, and whether a per-payment comparison is being made across a change in schedule. Interim and final: how European dividends actually pay covers the cadence half of this, which matters more in Europe than in America, where 97% of the companies we cover simply pay four times a year.
Is the number missing, or is it zero?
A blank and a bad score are different things, and conflating them is the error we see most often.
When a figure is absent from our own pages it is usually because an input did not resolve, not because the company failed a test. Of the instruments we cover, 78 carry no safety score. Roughly half are funds, which we will never score because a fund has no board deciding a distribution and no balance sheet behind it. Most of the rest are companies missing one specific input, and several have payment records running past forty years.
A blank of that kind also fills in. A company can acquire a score after a later filing without anything about the business having changed, which is worth knowing before you read an absence as a judgement.
Treat an absent figure as an absent figure. If a screen filters on a metric, the companies lacking that metric silently leave the screen, which is a selection you did not intend to make and will not see. Why some stocks have no dividend safety score sets out what actually produces a blank, and why publishing nothing is better than publishing a number nobody can stand behind.
Has anyone checked whether it predicts anything?
The previous four questions are about whether a number means what it says. This one is about whether it is worth anything.
Most dividend metrics are published without evidence that they work. A payout ratio, a growth streak, a safety rating: each is presented as informative, and almost none have been tested against what happened next. It is a fair question to ask of any of them, including ours.
We tested our own and published the result, misses included. Companies rated very safe cut their dividend within eighteen months 1.2% of the time, against 20.0% for those rated very risky. The ordering held after removing the pandemic years, removing property companies and controlling for yield. It also got things wrong, which the article names.
The harder half of that exercise was not the scoring. It was deciding what counted as a cut, which needs all four of the questions above at once. Which window, which currency, what counts as a payment, and what to do about a company that stopped paying rather than reducing. A first attempt flagged a company that has never cut its dividend in its life, because it had changed its payment schedule. If a test of a dividend metric does not describe how it defined the outcome, that is the part to ask about. Does a dividend safety score actually predict cuts is the whole test.
If a metric you rely on has never been checked this way, that does not make it useless. It makes its usefulness an assumption rather than a finding.
Why this happens at all
None of this is anyone behaving badly. Dividend data is genuinely difficult in ways that price data is not.
A price is a single observed number. A dividend figure is an aggregate over irregular events in variable currencies with inconsistent labelling, and every step from the raw payment to the published rate needs a decision. Vendors make those decisions differently, most do not publish which ones they made, and the resulting figures disagree in ways that look like errors and are usually not.
What separates a trustworthy source from an untrustworthy one is therefore not accuracy in the abstract. It is whether the decisions are visible and consistent, and whether the source says so when it does not know.
Consistency matters more than it sounds. A site that always measures growth over completed fiscal years will disagree with one that uses calendar years, and both can be useful, because within either you can compare two companies fairly. A site that switches between them, or that silently changes when a company's fiscal year does not start in January, produces figures that cannot be compared with anything, including its own.
The same applies to us. Where our numbers differ from another source, the useful question is not who is right but which decision differs, and that is answerable only if both sides publish theirs.
What to do with all this
You are not going to audit every figure you read, and you do not need to. The practical version is short.
- Check the window on any growth rate before comparing it with another.
- Check the currency on anything involving a company listed outside its home market.
- Check whether specials are included in an annual total that looks surprising.
- Treat a blank as missing information, not as a bad result.
- Ask what evidence exists that a metric predicts anything, before weighting it.
Every figure on our stock pages carries its basis, and the Dividend Health Score publishes what it looks at rather than only what it concludes. Where we cannot stand behind a number we leave it blank, which is less satisfying than a figure and more honest than a guess. The screener applies the same rules across the whole catalogue, so a company missing an input is visibly missing it rather than quietly scoring zero.
Frequently asked questions
- Why do two websites show different dividend growth rates for one company?
- Because a growth rate is a comparison between two years, and the choice of which two is not obvious. Change the window, include or exclude a special dividend, or bucket by fiscal rather than calendar year, and the same payment record produces a different rate. All of them can be arithmetically correct.
- Which dividend figure should I trust?
- The one whose definition you can see. A figure published with its window, its currency and its treatment of special payments is checkable. A figure published on its own is a claim you cannot audit, whoever produced it.
- Can a company increase its dividend and cut it at the same time?
- Yes, if it declares in one currency and is listed in another. The shareholder receives more of the declared currency and fewer of the listed one. Both descriptions are true and they point in opposite directions.
- Is a missing dividend figure the same as a bad one?
- No, and treating them alike is the more common error. A blank usually means an input did not resolve rather than that the company failed a test. A long payment record is real whether or not anyone has scored it.
- How would I know if a dividend metric actually works?
- Somebody has to test it against what happened next and publish the result, including the misses. Very few dividend metrics have been checked that way, which is worth knowing before you weight one heavily.
DividendAtlas provides data and research for informational purposes only. Nothing here is investment advice or a recommendation to buy or sell any security. Always do your own research.