Count statutory sick, maternity and paternity pay as Universal Credit earned income

[MaxGhenis]

Fixes #2337. Also fixes defect 3 of #1997.

What and why

Universal Credit treats statutory sick, maternity and paternity pay as employed earnings: UC Regs 2013 reg. 55(4)(a) to (c). Reg. 55(5) deducts income tax and primary Class 1 NI from "the amount of general earnings or benefits specified in paragraphs (2) to (4)". The regulations leave no second reading.

The model already charged NI on all three payments and income tax on sick and maternity pay, and deducted both from UC earned income. But uc_individual_earned_income_before_mif left the payments out of earned income, so the tax and NI on them came off other earnings. UC rose by 0.55 × (20% + 8%) = 15.4p per pound of sick or maternity pay above the thresholds, and by 4.4p per pound of paternity pay.

Changes:

• uc_individual_earned_income_before_mif adds statutory_sick_pay, statutory_maternity_pay and statutory_paternity_pay to earnings. Reg. 55(4) also lists adoption, shared parental, parental bereavement and neonatal care pay, but the model has no variables for those.
• statutory_paternity_pay is now in HOUSEHOLD_BENEFIT_VARIABLES, both HBAI lists and GOV_SPENDING_VARIABLES, next to statutory_maternity_pay. Sick and maternity pay were already in all four lists; paternity pay was in none. Once UC tapers paternity pay, leaving it out would make household net income fall as the pay rose: £400 of pay would cut UC by £220 and never count as income. uk-data does not fill statutory_paternity_pay, so this moves no survey aggregate.

Invariants (Hypothesis properties in tests/test_uc_statutory_pay_properties.py, plus deterministic paths)

1. Monotonic. Raising a claimant's statutory pay, with everything else fixed, never raises UC before the benefit cap. The final award rises only where the family becomes exempt from the cap. Three exceptions are lawful, each tested and labelled:
   • Childcare work condition. Reg. 32(2)(b) treats a person on statutory pay as in paid work, so the first pound can open the childcare element. The award rises by no more than the element.
   • Marriage Allowance cliff. ITA 2007 s. 55B(2)(b) withdraws Marriage Allowance in one step. The test shows the cliff happens and is identical for wages. Families claiming Marriage Allowance are left out of the monotonic property.
   • Benefit cap earnings exception. Under reg. 82(1)(a), a pound of pay can lift the cap. This can't arise in the model yet, because is_benefit_cap_exempt_earnings ignores statutory pay. Use each rule's own Universal Credit earned income in the benefit cap, targeted childcare and local CTR #1999 changes that by reading uc_individual_earned_income_before_mif.
2. Same as wages. Sick or maternity pay of s on top of wages of e gives the same income tax, NI, earned income and pre-cap award as wages of e + s. Paternity pay is left out until the model taxes it (Charge income tax on statutory paternity pay #2368). So are recipients whose pension relief is capped by earnings, because the model's relief excludes statutory pay.
3. Accounting. Earned income before the minimum income floor = wages + statutory pay + trading profit (a loss counts as nil) − relievable pension contributions − tax on earnings − NI on earnings, floored at nil.
4. Bounds. A pound more of pay adds between nothing and a pound to the recipient's earned income.
5. Claimants only. A dependant's statutory pay changes neither the claimants' earned income nor the award (reg. 22(1)(b)).
6. Counted once. Each payment counts once in household_benefits: Δhousehold_benefits − Δ(other listed benefits) = Δpay. Each payment appears exactly once in all four lists.

Power: on unfixed main, 6 of the 7 original YAML cases and 4 of the 6 original properties fail. The other 2 are controls: the dependant property and the childcare exception. With paternity pay removed from household_benefits, invariant 6 and the list check fail.

Impact

Measured at the pre-merge head 0dce50c against its base, main 3f174d0. Since then the only change is a merge of main (0654c4c), which brought in main's own changes, including #2327's switch to taxable_miscellaneous_income in the same list. Real run, 2026, one Microsimulation each. Both runs use the certified enhanced FRS: uk-data 1.58.0, enhanced_frs_2024_25.h5, HF revision 47bbb58d, sha256 5ba399ea…d960b6, verified at load. Main is 3f174d0; this branch is 0dce50c, a clean tree. The runs went through the heavy lock, one at a time. Script and outputs: ~/reviews/uk-statutory-pay-gov-spending-2026-10-12/ (impact.py, compare.py, runs/).

2026, £bn	Main	This branch	Change
universal_credit	83.589	83.365	−0.225
household_benefits	552.579	552.354	−0.224
household_net_income	1,710.908	1,710.684	−0.224
gov_spending	552.208	551.984	−0.224
gov_balance	469.236	469.460	+0.224
benefit_cap_reduction	1.171	1.155	−0.016
gov_tax	1,021.444	1,021.444	0

• Who loses UC. 73 sample households (67k weighted) lose UC: 40 receive SSP and 33 SMP. No household gains. Among them, UC falls by £0.39 per £1 of statutory pay; the median pay is £6,444 a year and the largest fall is £10,235 a year. Main's error was +£0.154 per £1 for pay above the thresholds, so the move is the taper less that error, smaller where pay is under the work allowance or the minimum income floor.
• Benefit cap. Capped families' pre-cap UC falls, so the cap takes £0.016bn less.
• Tax-Free Childcare. In 2 sample households (672 weighted), UC falls to nil and household benefits fall £619.74 less than UC. That fits Tax-Free Childcare becoming payable: the model withholds it while the benefit unit has any pre-cap UC (disqualifying_benefits.yaml, Childcare Payments Act 2014 s. 11). These runs did not record that variable.
• No paternity-pay effect. uk-data does not fill statutory_paternity_pay, so adding SPP to the household lists moves nothing here.
• Data caveat. uk-data sets SSP and SMP from the FRS weekly amounts × 52 (policyengine_uk_data/datasets/frs.py), so a payment current at the interview counts for the whole year. That is a data-side convention this PR does not change.

Tests run locally

At the merged head 0654c4c: test_uc_statutory_pay_properties.py and test_aggregation_lists.py, 15 passed. Sol's round-3 review also ran uc_statutory_pay.yaml there: 9 passed. The rest below ran at 0dce50c.

• uc_statutory_pay.yaml: 9 passed. Expected values are hand-computed from the 2026-27 rates in the file header, and every one was re-derived by the reviewer.
• test_uc_statutory_pay_properties.py: 7 passed in 38 s. A randomized stress run also passed at the original head 70851ab: 200 random populations for the monotonic property and 150 for wages and accounting. It wasn't repeated at 0dce50c.
• YAML folders: universal_credit 423, household 135, benefit_cap 8, council_tax_reduction 173, dfe 218, plus 140 cases in 11 other YAML files that read statutory pay or the lists. All passed.
• test_aggregation_lists.py, test_household_benefits_properties.py, test_uc_childcare_work_condition_properties.py, test_council_tax_reduction_pension_credit_properties.py: 27 passed.
• ruff format --check and ruff check pass on every changed Python file.

Review

GPT-6.1 Sol (subfleet run --task review --tier hard):

• Round 1 at 70851ab requested changes, with two blocking findings: paternity pay was missing from household income, and the final-UC monotonic property was false under reg. 82. Both are fixed in 0dce50c.
• Round 2 approved head 0dce50c.
• Round 3 reviewed the merge of main, whose one conflict was Relieve a miscellaneous loss only against miscellaneous income #2327's taxable_miscellaneous_income in the same list. It approved head 0654c4c: both changes are kept, and nothing main brought in changes an expected value or property.

Reviews are in ~/reviews/uk-statutory-pay-gov-spending-2026-10-12/review/ (out_2365.md, out_2365_r2.md, out_2365_r3.md). Its non-blocking suggestions:

• S4: SSP counted as government spending although employers fund it → Statutory sick pay is counted as government spending, but employers fund it #2369 and decision d1402.
• S5: HB/CTR leave out paternity pay → Charge income tax on statutory paternity pay #2368.
• S1/S2: docstring numbering and a duplicate test simulation → left, to keep the approved head.

Interactions

• Charge income tax on statutory paternity pay #2368 (income tax on paternity pay; also counts SPP in HB and CTR) should merge after this. It then adds SPP to TAXED_AS_WAGES in this PR's property test. No file overlap.
• Use each rule's own Universal Credit earned income in the benefit cap, targeted childcare and local CTR #1999 reads uc_individual_earned_income_before_mif for the benefit cap earnings exception, so the cap counts statutory pay once both land. Exception (c) above is written for that. Whichever lands second should rerun its tests.
• Count Scottish Child Payment, Healthy Start and the Two Child Limit Payment in government spending #2366 (Scottish Child Payment and Healthy Start are household benefits but missing from government spending #2338) also edits gov_spending.py, in a different hunk. Its code-health test (household benefits ⊆ government spending) passes with this PR, because SPP goes into both lists.
• Report each pull request's effect on national aggregates on the certified dataset #2335 (W3's fiscal partition) needs a statutory_paternity_pay spending head; see the comment on #2335.
• Count the Pension Age Winter Heating Payment in household and HBAI income #2047 (pawhp) edits other lines of the same lists. No conflict.

Ownership: no W1 file (simulation.py, tax_benefit_system.py, utils/scenario.py, utils/parameters.py) and no W3 file (tests/conftest.py, aggregate-impact.yaml, reforms_config.yaml) is touched.

axiom: TheAxiomFoundation/rulespec-uk#387 queued. Reg. 55 is not in the pinned corpus release, so the signed encoder can't cite it yet. The statutory-pay leaves and companion cases are added there.

🤖 Generated with Claude Code

Reproducer: policyengine_uk/tests/policy/baseline/finance/benefit/family/universal_credit/income/uc_statutory_pay.yaml, policyengine_uk/tests/test_uc_statutory_pay_properties.py::test_award_never_rises_with_a_claimants_statutory_pay, policyengine_uk/tests/test_uc_statutory_pay_properties.py::test_each_payment_is_household_income_and_government_spending, policyengine_uk/tests/test_uc_statutory_pay_properties.py::test_statutory_pay_counts_as_wages_and_earned_income_adds_up, policyengine_uk/tests/test_uc_statutory_pay_properties.py::test_a_dependants_statutory_pay_is_not_the_claimants_earned_income, policyengine_uk/tests/test_uc_statutory_pay_properties.py::test_award_falls_along_a_path_of_rising_statutory_pay

[MaxGhenis]

Hand-off to the UK hub from session local_615fdd79-8fa2-4544-b8ce-67188d179ad6 (started by the UK hub).

• Head: 0654c4c. This is 0dce50c plus a merge of main 0bc08f2, which had one conflict (Relieve a miscellaneous loss only against miscellaneous income #2327's taxable_miscellaneous_income). Both sides are kept.
• Review: GPT-6.1 Sol approved this exact head in round 3 (run 20261012-045417-review-pr2365-r3). Round 2 approved 0dce50c, and round 1 requested changes on 70851ab, which have since been fixed. All three reviews are in ~/reviews/uk-statutory-pay-gov-spending-2026-10-12/review/.
• Checks: Test (YAML) passes at this head; Lint, install and the three Python test shards were still pending at hand-off. The previous head 0dce50c passed all six checks of the old layout. Locally at this head: the property tests and test_aggregation_lists.py (15 passed); Sol ran the 9 YAML cases there and all passed. GitHub reports MERGEABLE.
• Impact: real 2026 runs on certified uk-data 1.58.0, at 0dce50c against main 3f174d0. UC −£0.225bn, household net income −£0.224bn, gov_balance +£0.224bn. 73 sample households lose UC (67k weighted) and none gain. Details are in the PR body.
• Depends on: nothing.
• Order notes:
  • Charge income tax on statutory paternity pay #2368 (income tax on paternity pay) should land after this. Its session is paused on a usage limit until 17 October.
  • Use each rule's own Universal Credit earned income in the benefit cap, targeted childcare and local CTR #1999 reads the same variable for the benefit cap: whichever lands second, rerun its tests.
  • It is compatible with Count Scottish Child Payment, Healthy Start and the Two Child Limit Payment in government spending #2366 in either order.
  • Report each pull request's effect on national aggregates on the certified dataset #2335 needs a statutory_paternity_pay head (comment).
• Decisions: d1402 (should gov_spending keep statutory sick pay? Statutory sick pay is counted as government spending, but employers fund it #2369). It doesn't block this PR.
• Axiom: Encode UC Regs 2013 reg 55 and 57 earnings deductions per person (PE parity: policyengine-uk#1942) TheAxiomFoundation/rulespec-uk#387, queued with companion cases.

[MaxGhenis]

UK hub (2026-10-12): this PR's head 0654c4c9b is merged unchanged into rollup B, #2380, with #2365, #2366 and #2375. Its one CI run tests all three together on current main. Merging #2380 with a merge commit marks this PR merged. The hub cancelled this PR's own queued runs, which the rollup replaces, to free slots in the starved Actions queue. Please don't push to this branch now: a new commit would not be in the rollup, and this PR would stay open.