Repository navigation
UC-eligible units with six-figure dividends: dividends treated as yield of a few thousand pounds of imputed capital #1948
Description
Problem
On the Enhanced FRS, benefit units with six-figure dividend income receive Universal Credit. Their dividends are treated as the yield of a few thousand pounds of imputed capital, so they are dropped from unearned income, and that capital stays under the £16,000 limit.
Enhanced FRS 2026-27, real run of main 44240bd (a copy of enhanced_frs_2024_25.h5, sha256 e433e532…), UC recipients with dividend income above £50,000:
| Benefit units | 30.9k (19 survey records) |
| UC paid to them | £0.517bn |
| Their dividend income | £4.70bn |
Their imputed corporate_wealth |
£0.268bn |
| Dividends counted as UC unearned income | £0.022bn |
Every one of these units has uc_tariff_income > 0 and uc_assessable_capital < £16,000. Above £20,000 of dividends it is 50.3k units (31 records), £0.738bn of UC and £5.41bn of dividends.
Mechanism (read in the code)
uc_unearned_income.pysubtractsdividend_incomefrom unearned income when tariff income applies and the household hascorporate_wealth> 0 (or reported capital). This treats the dividends as the actual yield of capital that the tariff income replaces.- The imputed
corporate_wealthbehind £4.70bn of dividends is £0.268bn, an implied yield of about 1,750%. It keeps assessable capital between the lower and upper limits, so the tariff income is small and the unit stays eligible. - Until UC earned income deducts the whole benefit unit's tax, including tax on pensions and other unearned income #1942 is fixed,
uc_earned_incomealso deducts the tax on those dividends from the unit's earnings, often down to zero. That fix removes £0.39bn of the UC these units get (UC to units with over £50,000 of dividends falls from £0.517bn to £0.123bn). The eligibility question here is separate.
Questions
- Should dividends be excluded as capital yield when the dividend income is many times the imputed capital? The dividends and the wealth come from different sources (the FRS and the WAS imputation), so they need not agree for a household.
- Many of these are probably owner-managers paid in dividends. UC has specific rules for a person who owns a company (UC Regs 2013 reg. 77). The model does not implement them, and they have not been checked for this issue.
- How does Exclude corporate_wealth from Universal Credit countable capital pending a pension split #1837, which takes
corporate_wealthout of UC capital, change this? Without that capital the tariff income may not apply, and the dividends would count in full.
Found while measuring #1942, where 17 of these records carry £0.394bn of that fix's £0.505bn.
Activity
MaxGhenis commentedon Sep 30, 2026
Update: #1950 answers question 1. Under UC Regs 2013 reg 66(1), which is a closed list, and reg 72(3), dividends are never UC unearned income. So the defect this issue describes is the imputed capital behind them, not the dividend rule. That data side is now PolicyEngine/policyengine-uk-data#495, and reg 77 (owner-managed companies) has its own follow-up. #1950 does not by itself change these units' eligibility: on main their dividends are already dropped as capital yield. Whichever of #1949 and #1950 merges second will re-run the combined Enhanced FRS impact.
MaxGhenis commentedon Sep 30, 2026
Findings
Edited 2026-10-13 after review: finding 2 is qualified, the #1837 section separates the issue's cohort from each rule set's own population, and totals for small cohorts (fewer than ten FRS households) are removed.
All figures are aggregates from real runs on the issue's dataset (sha256
e433e532…), 2026, linked to the raw FRS 2024-25 where noted.1. The six-figure dividends are an imputation artefact, not survey data.
impute_incomereplaces every FRS respondent'sdividend_incomewith a draw from an SPI-trained model that predicted from age, gender and region (policyengine-uk-data#529 has since added earnings groups), not from anything about investments or benefits.frs.pysums the survey's own investment-account dividends on each person's row position, not theirperson_id. Almost no dividends survive: well under £0.1bn in 2024-25 against about £8bn keyed correctly.The issue's cohort is mostly FRS-half rows, with some SPI donors. Linked to the raw FRS, the cohort's households report a small fraction of a percent of the dividends the model gives them. Across all FRS-reported UC units it is £2.91bn in the model against £0.011bn in the raw FRS.
2. The dividends are not owner-manager income the survey recorded. No director in the cohort's FRS households reports dividends from their company. That does not show that no one in the cohort is an owner-manager under reg 77: the comparison is at household level, and the FRS records neither a company's trading status nor its profits, so a director who takes no dividends would still be within reg 77. UC's rule for a person who owns a company (reg 77) was missing from the model. It is added in #1965, which is inert on this dataset, because the data has no owner-manager inputs.
3. Dividends are never UC income. Reg 66(1) is a closed list with no head for dividends, and reg 72(3) treats them as capital. #1950 implements that. Shares count through their value, as capital (regs 46, 49, 72).
4. How #1837 changes the picture. On the same dataset, 2026:
With dividends treated as the law requires, the rules cannot stop these units. The data must.
5. The fix is in the data: PolicyEngine/policyengine-uk-data#498.
person_id.Production-setting rebuilds of policyengine-uk-data main (
b45c373, before the 10/8 batch) and the branch, simulated with this repo's main for 2026, show the effect. The UK hub is re-measuring on post-batch main.would_claim_uc. That is capital coherence for the donor half (policyengine-uk-data#495).microcosm has the same FRS-half redraw and needs the same change; that is tracked separately.