Capital gains tax reform scoring (mode 2): the Budget's flagship measure, and the one where the behavioural response IS the disagreement

[vahid-ahmadi]

Capital gains tax is the single most-reported broad-based option for the 28 October Budget, with the "big three" (income tax, NI, VAT rates) ruled out by the manifesto pledge. Aligning CGT rates with income tax bands is reported at around £14bn.

PolicyEngine-UK can express it cleanly, verified against an installed 2.89.2:

parameter	today (2026)	aligned
gov.hmrc.cgt.basic_rate	0.18	0.20
gov.hmrc.cgt.higher_rate	0.24	0.40
gov.hmrc.cgt.additional_rate	0.24	0.45

The scorecard already carries HMRC capital_gains_tax level claims, but no CGT reform scoring at all.

Why this one is worth doing properly rather than quickly

The behavioural response IS the disagreement. Published CGT estimates differ mostly about how far realisations fall when the rate rises, not about the arithmetic — reported receipts have ranged £12–22bn historically on a broadly stable rate structure. PolicyEngine-UK carries a real response module (capital_gains_behavioural_response, relative_capital_gains_mtr_change, elasticity at gov.simulation.capital_gains_responses.elasticity), so this lane can do something the reckoner lanes cannot: score the same measure static AND post-behavioural, and publish the wedge between them.

That wedge is the interesting number. A static-only PE figure is not comparable to a behavioural OBR costing, and #67 already found that pretending otherwise produces an undecomposed divergence nobody can attribute.

Steps

• Extend the divergence axes registry with a behavioural_elasticity axis for this family, sized by the paired static/behavioural runs rather than left as a recipe (OBR costings divergence decomposition: axes registry + identity-checked pipeline (#59) #67's rule: recipes are executed, not asserted)
• Compute both legs on the certified bundle via UK counterpart compute pipeline (managed sims, pe_gap-degrading registry) #51's machinery, on the announcement-consistent baseline
• Stage through the campaign contract, claim_id-direct, with the elasticity and the static/behavioural basis on every row
• Harvest the OBR costing when the EFO lands on 28 October and compare like-for-like (post-behavioural to post-behavioural)
• Record the elasticity's provenance in the assumptions registry (Assumptions registry: catalog every source model's methodology for apples-to-apples comparison #10) — it is the parameter the whole comparison turns on

Blocked on #51 merging and #77-style run capacity. Registry entry already exists: ab2026__cgt_align_with_income_tax_rates in #96.