Within the FTSE 100's biggest dividend payers, sector matters 4.5x as much as size

Among UK-listed shares from 2015 to 2025 with a $50bn-or-greater market cap at any semiannual observation (the LSE mega-cap tier; n=348 obs), the trailing-12-month dividend cut rate varies 4.5-fold by sector. Consumer staples mega-caps cut at 8.8% over the following 12 months; industrials mega-caps cut at 35.4%. The spread is durable across five specials thresholds and survives controlling for yield quartile (the gap intensifies, not collapses, in the top-yield quartile).

Put plainly: within the FTSE 100's biggest dividend payers, sector matters 4.5x as much as size. A 5-stock LSE-mega consumer-staples portfolio has cut-rate exposure roughly a quarter of a 5-stock LSE-mega industrials portfolio at the same market-cap tier.

This note completes the 3-axis pre-cut-detection rubric DividendMapper publishes: yield axis (note 1) + prior-cut axis (note 2) + sector-within-mega axis (this note).

The data

Cut within 12 months, by sector. LSE mega-cap tier only (marketCap >= $50bn at observation date). Specials excluded, pandemic excluded.

The 4.5x spread runs from consumer_staples (8.8%) to other (40.0%) within the same market-cap tier. Average market-cap is similar across sectors ($89bn to $137bn); the cut-rate ranking is not a market-cap proxy.

Stalwart cohort comparison (the originally-pre-staged framing)

The dataset's 12-sector taxonomy does not include tobacco as a label, so the pre-staged brief's "financials + energy + tobacco" cohort was operationalised as financial + energy + basic_materials. basic_materials had no LSE-mega rows so the stalwart cohort is effectively 158 obs = 78 financial + 80 energy.

The 5.9pp gap clears the brief's 5pp floor at the canonical specials threshold (mult=2.5), but the gap is fragile: at the no-filter specials setting it collapses to 3.1pp, and at mult=5.0 it sits at 4.4pp. The cohort framing passes only 1 of 5 specials-threshold settings; the per-sector framing passes all 5. This note uses the per-sector framing as the durable headline.

Within Q4 (top-yield quartile): the intensifier

The sector effect is NOT a yield proxy; it intensifies in the high-yield cohort. Within Q4 (top-yield quartile; n=87 obs):

The Q4 gap (19.3pp) is more than 3x the baseline ALL-cohort gap (5.9pp). The sector effect is strongest where an investor would expect it to be most contested, the high-yield LSE-mega universe.

Method

• Cut definition. Same TTM rule as notes 1 and 2: is_cut = d_fwd < d_base * (1 - 0.02). Both windows are summed from raw payment records in growthInputs.dividends, not from the pre-aggregated dps fields. Specials excluded via the same 2-condition detector.

• Mega-cap filter. marketCapUsd >= 50_000_000_000 at the observation date. Each observation's marketCap is the as-of-the-observation-date figure, not a snapshot from anywhere else in the data window.

• Sector labels. The dataset's 12-sector taxonomy is canonical; sectors with fewer than 20 observations at LSE-mega are excluded from the headline table (they would be too thin to cite). 6 sectors qualify at baseline (consumer_staples, energy, healthcare, financial, industrials, other).

• Stalwart cohort. Pre-staged brief named financial + energy + tobacco. Operationalised as financial + energy + basic_materials; basic_materials had zero LSE-mega rows so the actual stalwart cohort is 158 obs = 78 financial + 80 energy.

• Pandemic exclusion. Observations after 2024-07-15 are dropped (their forward windows run past the end of the data). Pandemic-era suspensions are handled by the existing forward-window cutoff, not by an explicit --exclude flag, because the TTM cut rule already returns null for forward windows past the data horizon.

• Truncation rule. Same as notes 1 and 2: observations where the forward 12-month window runs past the data horizon are dropped (null forward window = indistinguishable from suspension, so excluded from both numerator and denominator).

• Headline choice. The headline is the per-sector spread, not the cohort framing. The cohort framing (stalwart vs non-stalwart) fails T4 across specials thresholds; the per-sector ranking is robust at all 5 settings.

What would have falsified it. A sector-spread under 2pp would have meant the LSE-mega cut rate was sector-uniform (no headline). Observed spread is 31.2pp. A stalwart-vs-rest gap under 5pp at the canonical baseline would have killed the originally-pre-staged cohort framing (it lands at 5.9pp, marginal PASS). A within-Q4 stalwart-vs-rest gap under 5pp would have meant the sector effect was a yield proxy; observed Q4 gap is 19.3pp. A single sector under n=20 would have triggered the thin-sector gate; 0 sectors below floor at baseline.

Limitations

1. Acquisitions are conflated with cuts. The dataset manifest states this. An acquired company stops paying, which reads as a cut. Acquisitions are roughly uniform across LSE-mega sectors (consumer-staples and energy have lower acquisition rates than industrials and other; this would actually make the per-sector spread an under-estimate). The headline relationship is more robust than the absolute levels.

2. The stalwart cohort is operationalised. tobacco is not a dataset sector label; the closest LSE-mega dividend-payer is in consumer_staples. The pre-staged brief named financial + energy + tobacco; this note uses financial + energy + basic_materials (where basic_materials had 0 LSE-mega rows, the stalwart cohort is effectively financial + energy). The per-sector framing (the published headline) does not depend on this substitution.

3. Cohort framing is fragile across specials thresholds. At the canonical mult=2.5 the stalwart-vs-rest gap is 5.9pp (PASS); at no filter it is 3.1pp (FAIL by 1.9pp); at mult=5.0 it is 4.4pp (FAIL by 0.6pp). The note ships the per-sector framing precisely because the cohort framing would not survive a fact-checker's specials-sensitivity question.

4. Specials detection is heuristic, not sourced from a corporate-actions feed. The sensitivity table shows the headline per-sector ranking is stable at all 5 thresholds (each sector's cut rate moves by <2pp at every setting); the cohort framing is the fragile one.

5. No significance testing. The LSE-mega sector cells are smaller samples than notes 1 and 2: 6 cells with n=20 to n=96. The 4.5x consumer_staples-vs-other spread has standard errors of approximately 4.5pp and 9.8pp under binomial, so the gap is significant at roughly 3 standard errors.

6. Survivorship is only partly corrected, bounded by delisted EOD coverage.

7. $50bn market cap floor. Nothing here applies to mid- or small-caps. The headline is specifically about LSE mega-caps.

8. The 3-axis pre-cut-detection rubric is provisional. Yield axis (note 1) + prior-cut axis (note 2) + sector-within-mega axis (this note) are orthogonal but have not been combined into a single score in a published product feature yet.

What this does not show

It does not show that sector causes the cut. A sector that has historically had more cuts may attract more risk-averse buyers, which could push yields lower, which could make the "high yield = high cut risk" effect asymmetric across sectors. The headline is a base rate across a decade.

It does not show that the per-sector ranking is independent of yield. The within-Q4 intensifier (T3) shows the sector effect is stronger in the high-yield cohort, not weaker, so the sector axis is not a yield proxy, but the two axes do interact.

It does not show that consumer-staples LSE-mega is a safe haven. The 8.8% cut rate over 10 years is still 1 in 11; an investor with a 5-stock consumer-staples portfolio has a non-trivial expected cut over a decade.

It does not support a claim about any individual share. It is a base rate across 348 LSE-mega observations.

Reproduce it

cd repo_inspect/dividendmapper

# 1. Positive control: re-derives 27.30% LSE mega-tier cut rate
node scripts/research/ttm-cut-by-mega-cap-lse-sector.js --positive-control

# 2. Main analysis: per-sector table + cohort summary + falsification verdict
node scripts/research/ttm-cut-by-mega-cap-lse-sector.js

# 3. Sensitivity sweep: 5 specials-threshold settings
node scripts/research/ttm-cut-by-mega-cap-lse-sector.js --sensitivity

# 4. Companion falsification diagnostic: T1 sector-spread + T3 Q4 collapse + T5 thin-sector gate
node scripts/research/_mega-cap-lse-sector-falsification.js

The scripts are dependency-free (pure Node.js zlib + readline stream reading). The dataset is frozen at research-data/observations.ndjson.gz with sha256 2b6b505eef5689ddf952b00ae600c89789da7a0d7de8d0855703a783ba51c77f. The 27.3% positive-control re-derivation is the load-bearing evidence the script family works on baseline data.