UK WAS stage: split private pension wealth out of corporate_wealth (uk-data#452 M2)

[MaxGhenis]

What

The UK WAS wealth stage folds WAS private pension wealth less the current-employment defined-benefit component (totalpenr8_aggr − dvvaldbt_scaper8_aggr) into corporate_wealth, and policyengine-uk counts corporate_wealth as capital in every means test (UC means_test/capital/sources.yaml and the HB/JSA/ESA/IS/PC equivalents). The value of a right to receive a pension under an occupational or personal pension scheme is disregarded capital (UC Regs 2013 Sch 10 para 10; HB Regs 2006 Sch 6 paras 31–32 and HB(SPC) Regs 2006 Sch 6 para 24; JSA Regs 1996 Sch 8 paras 28–29; ESA Regs 2008 Sch 9 paras 28–29; IS Regs 1987 Sch 10 paras 23–23A; SPC Regs 2002 Sch V paras 22–23). On the incumbent enhanced_frs_2024_25.h5 (uk-data 1.56.14) at 2025, 46% of benefit-unit records that report UC fail the £16,000 test, 96% of them on corporate_wealth alone.

Diagnosed as mechanism M2 of PolicyEngine/policyengine-uk-data#452 (the UC caseload shortfall); machinery lives here per the 2026-08-22 routing (uk-data gets at most one last mirrored release, then freezes).

Measured on the donor (WAS round 8, aggregates)

The folded remainder is £3,659bn of the £4,318bn old corporate_wealth mass (84.7%): current occupational DC 314 + AVCs 12 + retained DC 550 + current personal pensions 204 + retained DB 864 + pensions in payment 1,692 + pension-sharing rights 23 — so it is not DB-free and is named private_pension_wealth (WAS/ONS private pension wealth less current-employment DB), not non_db_pension_wealth.

Change

• WAS stage emits private_pension_wealth as its own household output (drawn first in chain segment 2; share-like holdings conditional on it; segment 3 conditional on both) and keeps corporate_wealth to employee shares/options, UK shares, unit and investment trusts and the stocks-and-shares ISA fold. Row identity old corporate_wealth == corporate_wealth + private_pension_wealth on the donor.
• Lockstep spec, regenerated support bounds, UK_ALLOWED_EXTRA_EXPORT_COLUMNS / uk_export_surface allow-list entry, coverage manifest and digest re-cuts; measurement receipt in experiments/452-uk-pension-wealth-split.md.
• Model side (policyengine-uk, separate PR): private_pension_wealth input as an explicit capital disregard; total_wealth re-sums it; corporate_sector_wealth = corporate_wealth + private_pension_wealth keeps the shareholding / corporate land value / employer-NI keys whole.

Effect (microcosm machinery at a fixed seed, weights unchanged, 2025)

UC benefit units 5.19m → 6.27m (+1.07m, +21%); eligible benefit units 9.8m → 16.3m; reporter records over £16k 3,165 → 1,757; GB elements: housing +0.51m, LCWRA +0.36m, carer +0.08m, childcare +0.14m, child +0.52m. Full table and caveats in the experiments receipt.

Follow-ups registered here

• Current-employment DB entitlements (DVValDBT_SCAPE, £1,149bn on the donor) are not carried; a future current_db_pension_wealth would complete WAS private pension wealth (it would change total_wealth).
• Cash ISAs (cash_isa, WAS DVCISAVR8) are capital for the means tests but the model's savings maps savings accounts only — the opposite-direction gap.
• The microcosm re-draw of the whole wealth block sits ~1.15m UC benefit units below the incumbent artifact on the same households even with the old arithmetic (continuous-quantile full-leaf draws vs microimpute's 10-point grid; predictor surface) — worth its own look alongside microcosm#145's predictor-quality revisit.
• Ordering: microcosm split → uk-data mirror (held for Max) → re-pin the efrs-post-calibration input-mass reference; re-pinning first would breach the input-mass gate on corporate_wealth.
• Engine coupling: bump the uk extra lock to the policyengine-uk release carrying private_pension_wealth before any national build/calibration (the UK targets ons.land.corporate_land_value / land_value are materialised through the engine's keys).

Refs: PolicyEngine/policyengine-uk-data#452, microcosm#701, microcosm#145, microcosm#736.