UK calibration has no capital gains targets; CGT distribution is ~3.9x HMRC taxpayer count

[vahid-ahmadi]

Summary

The published populace-uk calibration (release populace-uk-2023-dd68c73-4aa4b14-20260619T023711Z) contains 149 targets, none of which constrain capital gains. As a result the calibrated weights leave the CGT distribution far from HMRC administrative data — in fact further away than the stock Enhanced FRS on two of the three key statistics.

Measurements

Read directly from populace_uk_2023.h5 (capital_gains weighted by household_weight), compared with the stock Enhanced FRS and HMRC's published CGT statistics:

Statistic	populace-uk 2023	Enhanced FRS (2026)	HMRC 2023-24
CGT taxpayers (gains > AEA)	1.47m	1.29m	~378k
Gains of those taxpayers	£96.0bn	£112.0bn	~£65.9bn
Mean gain per taxpayer	£65,374	£87,158	~£174,000

So the calibrated dataset has roughly 3.9x HMRC's taxpayer count, with a mean gain about 62% below the administrative figure. The direction is consistent with an unconstrained imputation: gains are spread across too many households in amounts that are individually too small.

Why it matters

Capital gains are one of the places the FRS imputation is furthest from administrative data, and the error is distributional rather than just a level shift. Any analysis of a CGT reform on these weights will overstate how widely the reform is felt — in our case the share of people losing income came out roughly 3x too high (5.0% vs 1.65% once CGT targets were added), and mean gain per payer was off by a factor of two.

The only capital-gains-adjacent item in the target surface is the column household.household_is_capital_gains_clone, which is a feature flag from the imputation rather than a target.

Suggestion

Add HMRC CGT targets to the UK target registry, e.g.:

• total taxable capital gains (HMRC CGT statistics, table 1)
• CGT taxpayer count
• ideally gains by size band, since the concentration matters more than the total for distributional work

For reference, calibrating with just the first two (£70bn gains, 400k taxpayers) alongside held targets for income tax, net income, population and household counts brought all six targets within 0.13% at an ESS cost of only 981 -> 905, so the CGT targets do not appear to be in serious tension with the existing surface.

Happy to share the target definitions we used if helpful.

[vahid-ahmadi]

Update: tracing this further, part of the cause sits upstream in policyengine-uk-data rather than here.

Its OBR EFO parser reads the cash-basis receipts rows from a hard-coded sheet "3.9", but in the March 2026 tables (the configured vintage) current receipts are on 3.8 and 3.9 is the APD forecast. Every lookup raises, is caught, and logs a warning — so the OBR capital gains tax target, along with NICs, VAT, fuel duties and SDLT, never reaches the calibration surface at all. Fix in PolicyEngine/policyengine-uk-data#439.

That restores an OBR revenue target for CGT (£20.80bn for 2026-27). It does not close this issue: gains totals and taxpayer counts are still unconstrained, which is what drives the 3.9x taxpayer-count gap reported above. The two are complementary — #439 constrains what CGT raises, this issue is about how gains are distributed across households.