Repository navigation
UK: the £1,000 personal allowance costing sits 13–14% above HMRC's ready reckoner while 1p on the basic rate is within 1% #1370
Description
Follow-up from Vahid's review of #1344 (item 27).
What we see
On the pre-Budget arm at e399d21ef (policyengine-uk 2.147.2, signed with a throwaway key, not attested), the static 2025 costings compare with HMRC's 2025-26 ready reckoner (January 2025) like this:
| Change | Microcosm | HMRC | Gap |
|---|---|---|---|
| +1p on the basic rate | £6.51bn | £6.55bn | −0.6% |
| +£1,000 personal allowance | £8.81bn | about £7.7–7.8bn (from £0.78bn for £100 and £9.65bn for 10%) | +13–14% |
#1344 moved the allowance cost the right way: it was 15–17% above HMRC before. The gap was there before that PR, so it is not a regression.
Why it points away from the basic-rate base
A 1p change reads the basic-rate base. An allowance rise is worth 20p, 40p or 45p per pound to each taxpayer, by their marginal rate, and it moves people in and out of tax and the £100,000 taper. With the base within 1%, the gap more probably comes from the number of taxpayers at each marginal rate, or from the taper zone.
Checks
- Taxpayers by marginal rate. Compare the model's 2025 taxpayer counts by marginal rate with HMRC's Income Tax liabilities statistics. The
hmrc_itl_marginal_raterows fit these counts on annual income (tax_band_annual), while the costing reads the engine's tax on current income, so compare both measures. - The taper zone. Compare the number of people with adjusted net income between £100,000 and £125,140 with HMRC's, and their cost share in the reform.
- The reform definition. Confirm the model's reform holds what HMRC's holds: the basic-rate limit (so the higher-rate threshold moves with the allowance) and the £100,000 income limit.
- Scope. Note 1 of the January 2025 bulletin says its figures are the exchequer impact on the UK government. They include block-grant effects and exclude the devolved elements for Scotland and Wales. The model's cost counts every UK taxpayer's tax, including Scottish rates. Pin down a like-for-like scope before the gap is read as a model error.
Done when
The gap is split into scope, reform definition and model parts, with each part either explained or filed against the calibration rows or the engine. Use the first certified national build from main for the figures, not the throwaway-key arm.
🤖 Generated with Claude Code
Activity
juaristi22 commentedon Oct 23, 2026
The gap, split
Arm at
e399d21ef(policyengine-uk 2.147.2, throwaway key) at 2025, static, on the engine's current income. These are weighted totals; HMRC's figures are from its January 2025 ready reckoner and Income Tax liabilities statistics. Most of the 13–14% is not the model's. HMRC's 2025-26 column is a first-year receipts figure on the UK government's scope; the model's is a full-year liability of every UK taxpayer.£1,000 on the personal allowance
Basis. The reckoner's 2025-26, 2026-27 and 2027-28 columns follow the Self Assessment payment cycle (balancing payment plus first payment on account in the second year). Fitted to it, a full year is 9.0% (1% row) and 9.8% (10% row) above the first-year figure. That gives £8.40–8.46bn for £1,000. The same pattern shows on 1p on the basic rate (11%) and on the higher-rate and basic-rate-limit rows (about 20%), where Self Assessment is denser.
Scope. HMRC's figure is E&NI income tax plus the change in Scotland's and Wales's block grants, without the devolved elements. In the model:
Counts.
Rest. The same 3% sits between HMRC's own counts priced (£0.878bn per £100) and its full-year reckoner (£0.850bn). Its source is the reckoner's vintage, the basis fit or behaviour, not Microcosm's counts.
On the annual-income measure that the calibration fits (39.5m taxpayers), the cost is £8.98bn (£8.71bn at HMRC's scope).
1p on the basic rate
rates.uk[0]only)On annual income the model like for like is £7.25bn (−0.4%). Moving the savings basic rate as well would add £0.04bn, if HMRC's row moves it; the row does not say. HMRC's figure is net of behaviour, so its static equivalent is higher.
So the "within 1%" was two offsetting differences: the narrower reform and missing block-grant effect (−£0.56bn) against HMRC's receipts basis (−£0.73bn).
Like for like, #1344 against #1338's build
829c1bc26)Both at HMRC's scope, in calendar 2025, against HMRC's 2025-26 full year. Moved to 2025-26, the arm costs about 1% more on both, which gives the +1.9% and −1.9% above.
What follows
3ea2fd822).🤖 Generated with Claude Code
juaristi22 commentedon Oct 23, 2026
Decision: accept the definitions, no model change
The two large parts of the gap are definitions, not model errors. HMRC's 2025-26 column is a first-year receipts figure on the UK government's scope; Microcosm's costing is a full-year liability of every UK taxpayer. Both are right for what they measure. Calibrating to the reckoner would make every full-year UK costing about 14% too low.
The rest of about 2% (+1.9% on the allowance, −1.9% on 1p) is inside the uncertainty of the Self Assessment fit. HMRC's own counts show a similar gap against HMRC's own figure. Closing.
Two checks, neither needing a build:
Reopen if the rest moves above about 3–4%. The lead then is the income distribution just above the allowance: Microcosm's cost per pound falls faster than HMRC's as the allowance rises.