UK companies that cut their dividend once are 2.5x more likely to cut again within a year

Among UK-listed shares from 2015 to 2025, those whose trailing five-year payment record contains at least one rebuilt TTM cut cut their dividend again within the following 12 months 28.9% of the time. UK companies without a prior cut cut 11.5% of the time. The gap is +17.4 percentage points, a 2.5x lift.

The US pattern is the same direction with a larger lift: 33.2% vs 6.3%, a 5.3x multiple.

The gradient is monotonic across all four yield quartiles. Within the top yield quartile on LSE, prior-cut names cut 42.9% of the time, against 17.4% for clean histories (n=296 vs n=276, gap +25.5pp).

The data

Cut within 12 months, by rebuilt prior-cut history. Specials excluded, pandemic excluded.

Within yield quartile. The gradient survives controlling for yield.

The prior-cut flag is not a yield proxy. Within any single yield quartile the gap is large and positive; the cut-rate ranking (prior-cut > no-prior-cut) holds at every band.

Robustness: specials threshold

The headline is stable across every specials setting. ALL markets, gap between prior-cut and no-prior-cut:

The headline number sits within 1pp of every sensitivity cell on ALL markets. The finding does not depend on the specials threshold chosen.

Method

• Cut definition. Same TTM rule as the first research note. is_cut = d_fwd < d_base * (1 - 0.02). Both windows are summed from raw payment records in growthInputs.dividends, not from the precomputed dps fields. Specials excluded via the two-condition detector (mult 2.5 against trailing 24-month median AND no comparable counterpart within 45 days of one year earlier).

• Prior-cut flag. For each observation at t0, scan the ticker's prior semiannual observations in (t0 - 5y, t0). If any of those observations had a rebuilt TTM cut (per the rule above), the obs is flagged priorCut = true.

• The precomputed flag (buyInputs.dividendCutInLast5Years) was used only as a reconciliation check and is not used in any reported number. Per the rebuild-from-primitives rule, derived fields must be rebuilt. The reconciliation finds 21.2% row-level disagreement, large enough that quoting the rebuilt label is the only honest choice.

• Quartiling is on the as-reported trailing yield assigned within each observation date and then pooled. Same convention as the first research note. As-reported yield is what an investor sees on a screener.

• Truncation. Observations after 2024-07-15 are dropped: their forward window runs past the end of the data, where an absent total is indistinguishable from a suspension.

• Specials and the prior-cut loop. The prior-cut flag uses the SAME TTM-cut rule as the forward label, which uses the SAME specials detector as the trailing baseline. A loop is avoided because the prior-cut flag looks at historical obs dates, which are themselves past the current t0.

What would have falsified it. A gap that vanished within any single yield quartile would have meant the prior-cut flag was a yield proxy. Every quartile holds the gap. A gap that collapsed when specials detection moved would have meant the result was an artifact of the specials heuristic. It holds at every threshold. A sector where the gap goes negative would have meant the lift was sector-specific, not general. Every sector with n>=30 on both sides holds a positive gap.

Limitations

1. Acquisitions are conflated with cuts. The dataset manifest states this. An acquired company stops paying, which reads as a cut. This inflates all rates. It should affect prior-cut and no-prior-cut rows similarly.
2. A prior cut may not mean the same thing in every sector. For REITs, structural distribution changes can produce a false-positive prior-cut flag. The ALL-markets direction is robust; the LSE-sector levels should be read with sector context.
3. Specials detection is heuristic, not sourced from a corporate-actions feed. The sensitivity table shows the headline moves within 1pp at every threshold on ALL markets.
4. No significance testing. Sample sizes are large (n>=1,076 per LSE cell). The LSE ALL-cell gap of 17.4pp on n=2,245 has a standard error of approximately 1.6pp under the binomial, so the LSE lift is significant at roughly 11 standard errors.
5. Survivorship is only partly corrected, bounded by delisted EOD coverage.
6. $2bn market cap floor. Nothing here applies to small caps.
7. 5-year lookback is arbitrary. A 3-year or 7-year lookback would shift the headline levels but not the qualitative finding.

What this does not show

It does not show that a prior cut causes the next cut. It does not support a claim about any individual share. It is a base rate across a decade. It also says nothing about total return: a company that cuts may still outperform.

It does not show that prior-cut is independent of yield. The headline ALL-markets gap of +25.6pp is a marginal effect over the prior-cut flag alone; within-quartile gaps remain large (+10.9 to +31.8pp) but smaller, so a citable pitch line should report the marginal, not the within-quartile.

Reproduce it

node scripts/research/ttm-cut-by-rebuilt-prior-cut.js --exclude 2019-07-15,2020-01-15,2020-07-15
node scripts/research/ttm-cut-by-rebuilt-prior-cut.js --sensitivity
node scripts/research/ttm-cut-by-rebuilt-prior-cut.js --sector --exclude 2019-07-15,2020-01-15,2020-07-15

The precomputed flag was reconciled against the rebuild with the one-off probe scripts/research/_probe-prior-cut-flag.js, retained because the disagreement itself is the strongest argument for rebuilding.