Repository navigation
UK spine: housing benefit £2.7bn against OBR £12.1bn — pensioner and supported-accommodation HB missing after the UC migration #867
Description
Found while measuring the first calibrated rowwise UK local candidate (#762, receipts R8–R15 on uk-rowwise-candidate-762-b). The row is a reviewed measure exclusion under #762 A16, expiring 2026-10-03, tracked on #736.
Measured on spine-m (FRS 2024-25, policyengine-uk 2.92.1, calibration year 2025), single-block engine resolution: obr.housing_benefit — spine £2.67bn against the OBR EFO £12.10bn (−78 % at initialization). Block-insensitive (identical under per-block and single-block resolution), so the miss is real.
What binding it does: under the ruled grain_equal weighting the solver chases the aggregate by multiplying the few HB households — £9.4bn at 512 epochs, £11.4bn at 1500 — which halves the effective sample (ESS fraction 0.161 → 0.105 at K=15) and doubles Glasgow's social-rent count. Excluded, the concentration returns (max/median weight 2,857 → 398, R13).
Likely cause: housing benefit after the UC migration. OBR's HB line is dominated by pensioner HB and supported / temporary accommodation, which the FRS either does not capture as HB or which the spine's take-up / migration stages route into the UC housing element. The FRS-reported HB caseload should be checked against DWP Stat-Xplore HB caseload by age band and accommodation type.
Ask: reconcile the spine's HB caseload and spend against the DWP HB caseload (pensioner / working-age / supported accommodation) and the OBR spend line; add an imputation or reclassification stage for the missing pensioner and supported-accommodation HB; and, if the OBR line legitimately includes spend the household frame cannot carry, declare that reconciliation on the target.
Acceptance: the row sits within 25 % at initialization on the next spine and the A16 exclusion retires.
Activity
juaristi22 commentedon Sep 16, 2026
Re-adjudicated to 2026-12-08 in #937 on the mechanism receipts (
tools/diagnose_uk_legacy_benefits.py,experiments/882-uc-repairs-receipts.md). The source is not the gap: FRS 2024-25 reports 1,559,991 Housing Benefit benefit units at design weights, 99.9 percent renters, against a DWP GB caseload of 1.94m in January 2025 falling to 1.46m by December. The spine carries 1.21m GB reporters because the SPI support channel re-imputes benefit receipt onto its synthetic households without tenure coherence: 363k of its 451k Housing Benefit reporters sit in owner-occupied households and can never be paid (thefrschannel's 758k are all renters). The engine then pays only reporting units with a rented tenure, not drawn into Universal Credit, under the capital limit and carrying the household's rent (rent is the household head's benefit unit's;claims_all_entitled_benefitsis a scalar on microdata, so there is no take-up lever). The 55 percentwould_claim_ucdraw had marked 3.75m all-pension-age units as claimants, inert for UC but zeroing Housing Benefit for 377k reporters on Q50f; #937 gates the draw to units with an adult under State Pension age (rate unchanged, U8 population re-signed): on the licensed twins at design weights the engine's Housing Benefit rises from 401,489 to 644,183 units (£2.61bn to £4.11bn), almost all pension-age. Q50f pays 235k units and £2.13bn against the OBR £12.10bn. Bound as diagnostics on the calendar-2025 window:dwp.hb.households,_social_rented,_private_rentedfrom the Stat-Xplore client-type-by-tenure cube (chronicle#259). Remaining repair: the SPI channel's benefit-receipt fill (a #840/#145 item); the OBR line also carries supported and temporary accommodation the frame has no concept for.