Withdrawn: are dividend cuts at the UK's biggest companies shallower?

Withdrawn 10 October 2026. The first version of this note said that when a UK company worth $50bn or more cut its dividend, the median cut was -16.5%, about half the -32.9% median for companies worth $2bn to $10bn. We rechecked it. The gap is smaller (-17.8% against -29.5%), it moves between 5.9 and 13.8 points depending on how one-off specials are treated, and once you allow for which companies happen to be in each group it cannot be told apart from zero. Many of the shallow cuts at the top are currency movements: dividends declared in dollars or euros, recorded in pence. The original figures are kept below for the record. What we found.

What the first version claimed

The note took every UK-listed share from 2015 to 2024, split by market cap at each semiannual observation date (mid $2bn to $10bn, large $10bn to $50bn, mega $50bn and over), and measured the size of each 12-month dividend cut. It reported that mega-cap cuts were 16.4 points shallower than mid-cap cuts, that the gap held between 15.8 and 17.6 points at every specials threshold, and that the same pattern held in the US.

What we found

The note's analysis script was never committed, so we rebuilt its method from the note's own description. The rebuild gives a mega-cap median of -17.0% from 69 cuts and a mid-cap median of -30.2%, close to the published figures, which confirms the reading. We then re-ran it on the dataset's own cut label: each company's full payment record, dated by payment date, in split-adjusted amounts, with cut sizes taken from the same 12-month sums the label uses. That label matches a fresh recomputation on all 23,056 rows. Pandemic observation dates (July 2019 to July 2020) are excluded, as in the first version.

UK, size of the median cut, by company size

Share of UK cuts deeper than each threshold, rechecked

The first version reported 15% of mega-cap cuts at 50% or more. Rechecked it is 22.2%, higher than the large-cap tier, so the pattern does not fall steadily with size.

The specials threshold

The gap between the mega and mid medians is 5.9 points with no specials filter, 13.8 at 2.0x, 11.7 at the 2.5x headline setting, 11.3 at 3.0x and 6.8 at 5.0x. The first version said it held between 15.8 and 17.6 points.

Chance

The 72 mega-cap cuts come from 19 companies. Resampling companies rather than cuts, the 90% range on the 11.7-point gap runs from -2.6 to +18.1 points, so it includes zero.

Currency

AstraZeneca, Shell, Unilever and HSBC declare their dividends in dollars or euros. Our data records the London line in pence, so a fall in the exchange rate of more than 2% reads as a cut even when the dividend in the company's own currency did not change. Those four supply 14 of the 72 mega-cap cuts, with medians between -2.9% and -13.3%, the shallow end that the first version's headline rested on. AstraZeneca held its dividend flat in dollars for most of the period. National Grid supplies 7 more from a dividend record that, in our data, mixes adjusted and unadjusted amounts.

The US

Rechecked, the US medians are -31.2% (mid), -25.0% (large) and -25.0% (mega), against -47.3%, -25.0% and -31.3% in the first version. The mega-mid gap is 6.2 points, with a company-resampled range of -0.2 to +10.0. Of 972 US mid-cap cuts, 95 sit within a point of -25%. That is the pattern a quarterly payer produces when one payment lands just outside the 12-month window, with no change in the dividend.

What this does not show

It does not show that cuts at large and small UK companies are the same size. In both halves of the decade the median mega-cap cut was shallower than the median mid-cap cut (-15.4% against -28.9% before 2020, -19.0% against -31.8% after). But with 19 companies at the top, a gap that moves with the specials threshold, and currency movements counted as cuts, the data cannot say how large the difference is or whether it is there at all.

Reproduce it

cd dividendmapper && npx tsx scripts/research/ttm-cut-sector-and-magnitude-label.ts

It rebuilds the first version's method, prints the rechecked tables, the specials sensitivity, the company-resampled intervals, the split halves and the per-company cut counts, and fails if its recomputed outcome disagrees with the frozen label on any row.