Dividend growth streak: 9 years or 55

Ask this screener on 2026-09-13 for a listing whose dividend has been held or raised for ten years running and it returns 376 of them. The longest run in the pool is 55 years.

Ask the same question of the 1,034 listings whose streak is read out of filed cash flow statements and the answer is none, because the highest value any of them can take is nine.

One field, two sources, two ceilings. The ceiling is a property of the record a number was read from, and no coverage count will ever show it to you.

Where each ceiling comes from

The archive holds at most ten filed fiscal years per company. A streak read from those statements counts the consecutive years in which the dividend did not fall, so ten statements yield nine comparisons and nine is the ceiling. That is arithmetic, not policy.

Since 11 September 2026 the figure is measured per share instead, from the record of declared dividends, and that record reaches back much further than ten years. Where it is used, the ceiling is wherever the record starts.

The metric registry states the arrangement on the figure itself:

Per share, so buybacks do not break it and new shares do not hide a cut.

Beyond the filed years, the record is the data vendor's.

When that record begins inside the run, the count is a floor and the source trail says so.

The third caveat is the honest one. On 2026-09-13, 212 of the streaks in the pool are marked as floors because the record begins inside the run rather than before it. The earliest record behind any of them starts in 1970.

The two populations, measured

Counts over the 5,518 active equity listings in the pool on 2026-09-13.

Every listing at ten years or more is in the first row. Not one is in the second, and that is not a fact about those companies.

Why the old answer was wrong in both directions

Until 11 September this page reported that nine was the highest value anywhere in the pool, and that the field was held for 5,126 listings. Both were true readings of what was stored. Both were measuring a defect.

A total is not a per-share figure. A company buying back its shares can pay less cash in a year it raised the dividend per share: Apple paid 14,081m in fiscal 2020 against 14,119m the year before, so it read a streak of five.

Worse, a year whose common dividend could not be read out of the filing ended the run as though it were a zero. A zero is a measurement, meaning the company paid its common holders nothing. A year we cannot read is not a measurement at all, and a 0 on the page does not read as "unknown", it reads as "never paid one". Johnson & Johnson carried a 0 on that basis, and so did Procter & Gamble. Both have raised their dividend every year for more than half a century, and both now read 55.

The figure is n/a when nothing was measured and 0 only for a break that was measured. That is why coverage fell from 5,126 to 1,839: the missing 3,287 were never streaks, they were blanks wearing a number.

The rest of the dividend family, re-counted

Coverage over the same 5,518-listing pool, after the recompute that finished at 03:03 UTC on 2026-09-13.

The payout ratio row moved the furthest and it moved the other way. It is now filed common dividends over filed net income, both off the same fiscal year's statements, and the concepts the ingest reads were widened on 11 September, so a figure that existed for 83 listings exists for 1,572. Of those, 242 are above 100 percent, which is a real reading and is published as measured rather than clamped.

That ambiguity is the general case rather than a quirk of one field. A zero standing in for a value nobody measured is the one failure a coverage count cannot warn you about, because the arithmetic did run and the cell is full.

Two questions before you filter on any of this

How many companies is it held for. The coverage count answers this, and the table above is why the answer is worth re-reading rather than remembered.

What values can it take at all, and from which record. No coverage count answers that. A bounded metric quietly turns a demanding threshold into an empty result, and an empty result reads like a discovery. "At least ten consecutive years" returned nothing on this site for months, and the reason was the source, not the market.

What this leaves out

Every count is one reading of the production database on 2026-09-13, after the recompute that finished at 03:03 UTC. They change nightly and /coverage carries the live figures. The 2026-09-07 column is the reading this page carried before it was rewritten, kept so the change is visible.

The per-share record beyond the filed years is the data vendor's, not a filing this product holds. For every filed fiscal year it is checked against the cash flow statement and a year that disagrees by more than 15 percent is not used, but a run reaching back to 1970 rests on a record we did not file and cannot show you page by page. That is a weaker claim than a figure read from a statement, and it is why the source trail names the basis on every row.

Where the record begins inside the run, the count is a floor. 212 streaks are in that state on this reading.

165 listings have dividends on file whose common share cannot be separated from the preferred, and 3 have a gap in the record that the weekend backfill has not yet closed. All 168 read n/a with the reason attached, not 0.

The streak does not distinguish a dividend held flat from one that rose, and it says nothing about whether the next one will be paid.

Nothing here is a view on any company, and nothing here says a long streak or a short one is a reason to do anything.