Budget 2026 ISA data, PR B of #1418: ISA subscriptions per adult, the Lifetime ISA anchor, additive ISA income, policyengine-uk 2.164.2

changelog.d/1360-uk-additive-isa-income.changed.md

@@ -0,0 +1 @@
+The UK producer now stores interest outside ISAs in `savings_interest_income` and ISA interest apart in `individual_savings_account_interest_income`, and declares policyengine-uk's additive ISA income contract (`{"version": 1, "representation": "additive", "projection": "additive"}`) on every frame and H5 it writes; `tax_free_savings_income` is left to the engine, which adds the ISA parts. Frames carry the contract they were loaded with, so the enhanced-FRS reference and older Microcosm files keep their legacy gross reading, and a build refuses them as inputs. Base-year cash and tax are unchanged: predictors that read gross interest read the two parts together and the SPI stage-2 draws keep their salts. From the base year the engine grows each part by its own index instead of growing gross interest and taking off an unchanged ISA part, which removes a small overstatement of taxable interest in projected years (microcosm#1360).

changelog.d/1415-uk-isa-statistics-facts.added.md

@@ -0,0 +1 @@
+A new vendored resource, `hmrc_isa_statistics_facts.json`, carries the 1,074 rows of HMRC's 2026 annual savings statistics the ISA subscriptions stage and the Lifetime ISA anchor read (microcosm#1415): Table 9.7 2023-24 subscribers and holders by income and size of subscription for each type paid into, Tables 9.8 and 9.9 by age, sex and region, Table 9.4 accounts and amounts for 2023-24 and 2024-25, Table 9.6 market values, the cash ISA limit reduction impact note's shares and Lifetime ISA Table 1a. They come from the pinned `53815b2` feed, which carries PolicyEngine/chronicle#392's ISA statistics; the pin and the other vendored resources are unchanged. None is a calibration target.

changelog.d/1415-uk-isa-subscriptions.added.md

@@ -0,0 +1 @@
+The UK spine imputes ISA subscriptions per adult (microcosm#1415). A new last stage, `isa_subscriptions`, splits each household's cash and stocks-and-shares ISA balances across its adults (`cash_isa_balance`, `stocks_and_shares_isa_balance`) and draws who paid into which type in 2024-25 and how much (`cash_isa_subscriptions`, `stocks_and_shares_isa_subscriptions`): subscribers and their amounts follow HMRC's 2023-24 Table 9.7 by income, size of subscription and type paid into, raked to Tables 9.8 and 9.9, and grow to 2024-25: people by Table 9.4's account growth, and amounts to its 2024-25 amounts, holding subscribers already at the £20,000 overall limit and tapering amounts above £18,000 up to it. Every Lifetime ISA holder aged 18 to 47, the ages that can hold one in 2024-25, with room under the limit pays in (`lifetime_isa_payments`), drawn to Table 9.4's 2024-25 Lifetime ISA amount, so policyengine-uk's Lifetime ISA bonus is no longer zero. policyengine-uk 2.164.2 reports the subscriptions an ISA limit leaves out but taxes none of them yet, so a limit reform costs nothing until it does. No ISA fact is a calibration target; the national build reports them as checks. Two gates check the stage: `uk_stage_isa_subscriptions_invariants` holds the rules any population keeps (limits, ages, holdings, finite and non-negative amounts, household sums), and the synthetic smoke build blocks on it; `uk_stage_isa_subscriptions` holds Table 9.4's totals. See `docs/uk-isa-subscriptions-1415.md`.

changelog.d/1415-uk-lisa-anchor.changed.md

@@ -0,0 +1 @@
+The `was_lisa` stage anchors Lifetime ISA holders to HMRC's 2024-25 payers (microcosm#1415). The WAS round-8 model implied about 660,000 holders on the spine against HMRC's 1.14 million Lifetime ISA accounts subscribed to in 2024-25; every payer holds an account and pays into one a year, so the ownership logit now shifts up on ages 18 to 47, the ages that can hold and pay into one in 2024-25, until the holders at those ages who survive the financial-wealth cap equal that count. The same keyed uniforms decide ownership, so the anchor only adds holders. A new gate, `uk_stage_was_lisa_anchor`, checks the count.

changelog.d/uk-policyengine-uk-2-164-2.changed.md

@@ -0,0 +1 @@
+The UK lock moves from policyengine-uk 2.162.14 to 2.164.2 (microcosm#1415). 2.164.0 (PolicyEngine/policyengine-uk#2712) adds the ISA annual subscription limit, a cash ISA limit that current law leaves unset, per-person cash and stocks-and-shares ISA subscription and balance inputs, and the 20% Lifetime ISA withdrawal charge on a first home above the £450,000 cap; its household `cash_isa` and `stocks_and_shares_isa` sum the person balances. The release therefore stops storing those household totals, and the engine reads the `isa_subscriptions` stage's person columns. 2.164.1 moves the economic assumptions to the OBR's October 2026 Economic and fiscal outlook, and the vendored engine pins move with it: average earnings at 2025 fall 1.1% and household interest income 6.1%, while GDP per head grows by the same 8.8% from 2023 to 2025. On the national surface the eight SPI Table 3.10 self-employment bands, which mixed income per head takes to 2025, compile 0.55% lower, and the dividend bands move by under 0.001% (the compile-parity receipt is re-signed); the savings interest bands grow by the OBR's October savings income growth since microcosm#1420, so the engine pins no longer move them. The annual-income earnings index, the weight-year population path and the bonus-sacrifice anchor's earnings and population growth (factor 1.1149 to 1.1123) move too, and the CGT projection gate's pinned growth path is re-pinned to the October path (its horizon stays 2030, though the path now runs to 2031). 2.162.15 to 2.164.2 also change Housing Benefit, Income Support and Pension Credit rules, project the Scottish Carer Supplement and the Carer's Allowance earnings limit, and run the property allowance election on only the households with a choice. Concept coverage drops the two household ISA totals as inputs and gains the four person ISA columns and Housing Benefit's pre-assessed permitted-work disregard, which the build leaves at its default.

docs/uk-chronicle-feed-repin.md

@@ -345,3 +345,9 @@ The national register version moves with the new references, so the uk-candidate
 register is re-frozen before the next national solve. The high-value home stock still names
 `735820c`: its inputs are the same on `53815b2` (the feed-gated fingerprint check passes), so it
 was not rebuilt.
+
+The ISA subscriptions (microcosm#1415, PolicyEngine/microcosm#1421) then vendored a new resource
+from the same `53815b2` artifact: `hmrc_isa_statistics_facts.json`, the 1,074 rows of #392's ISA
+statistics the ISA subscriptions stage and the Lifetime ISA anchor read. None is a calibration
+target, so no reference moves for it. That pull request first pinned #392's head (`e56af9a`) and
+then its merge (`0b721a4`), and took `53815b2` when PR C merged first.

docs/uk-isa-subscriptions-1415.md

@@ -0,0 +1,206 @@
+# ISA subscriptions per adult (microcosm#1415)
+
+The `isa_subscriptions` stage gives every adult of the UK spine what they paid
+into ISAs in the year, by type, and the per-person ISA balances behind them. It
+runs last, on final incomes, and nothing in the spine reads its outputs.
+
+| Column | Meaning |
+|---|---|
+| `person.cash_isa_balance` | The person's share of the household's cash ISA balance (`was_wealth`), in pounds. |
+| `person.stocks_and_shares_isa_balance` | The person's share of the household's stocks-and-shares ISA balance. |
+| `person.cash_isa_subscriptions` | Paid into cash ISAs in 2024-25. |
+| `person.stocks_and_shares_isa_subscriptions` | Paid into stocks-and-shares ISAs in 2024-25. |
+| `person.lifetime_isa_payments` | Paid into the person's Lifetime ISA in 2024-25 (at most £4,000). |
+
+The `was_lisa` stage also anchors Lifetime ISA holders to HMRC's payers (below,
+and [the LISA notes](uk-lisa-1003.md)).
+
+The pinned policyengine-uk 2.164.2 reads all five:
+
+- `lifetime_isa_payments`, whose 25% bonus is `lifetime_isa_bonus`;
+- the subscriptions, against the limits of policyengine-uk#2712
+  (`isa_subscriptions_above_limit` and `cash_isa_subscriptions_above_limit` are
+  what a limit leaves out);
+- the person balances, which its household `cash_isa` and
+  `stocks_and_shares_isa` sum.
+
+The release does not store those household totals. A stored total would
+override the sum, so the means tests, which count the household `cash_isa`,
+read the sum of the person balances.
+
+## Sources
+
+- HMRC, [Annual savings statistics 2026](https://www.gov.uk/government/statistics/annual-savings-statistics-2026)
+  (September 2026), through the pinned Chronicle feed (`53815b2`, which carries
+  PolicyEngine/chronicle#392's ISA statistics):
+  - Table 9.7: individuals subscribing in 2023-24 by total income and size of
+    subscription, for each type paid into (cash only, stocks and shares only,
+    both). The tables head their columns "(thousands)" but count people.
+  - Table 9.8 by age and sex, and Table 9.9 by region.
+  - Table 9.4: accounts and amounts subscribed by type to 2024-25 (the latest
+    year provisional), including Lifetime ISAs.
+  - Table 9.6: market values, and Tables 9.10 to 9.12: holders by market value,
+    both checks only.
+- HMRC, the cash ISA limit reduction tax information and impact note: 22% of
+  cash subscribers under 65 paid in more than £12,000 in 2022-23 (a check).
+- HMRC, Lifetime ISA statistics, Table 1a: withdrawals (a check).
+- ONS Wealth and Assets Survey round 8 (UKDS SN 7215), the person tab's ISA
+  values, for the split of household balances across adults.
+
+How the tables relate, on HMRC's own figures:
+
+- Table 9.7's type blocks are the types paid into: cash only plus both
+  reproduces Table 9.4's 2023-24 cash accounts, and stocks and shares only plus
+  both its stocks-and-shares accounts. Lifetime ISAs, Innovative Finance ISAs
+  and Junior ISAs are outside Table 9.7; Help to Buy ISAs are inside cash.
+- Table 9.7's bands imply more money than Table 9.4 records. Valued at their
+  floors they give at least £114.3bn (cash only alone £78.1bn), against Table
+  9.4's £100.6bn of cash plus stocks and shares, so the two cannot both bind.
+- From 6 April 2024 (SI 2024/350) a person may pay into more than one ISA of a
+  type in a year, but still only one Lifetime ISA, so Table 9.4's 2024-25
+  accounts are an upper bound on people.
+
+## Method
+
+1. **Balances per adult.** Each household's cash and stocks-and-shares ISA
+   totals are split across its adults by the holding pattern of a WAS round-8
+   donor household with the same number of adults (up to four), drawn at
+   household-keyed uniforms and conditioned so every subscriber holds the type
+   they pay into. Person values sum exactly to the household totals, which are
+   unchanged.
+2. **Who subscribed in 2023-24.** Adults aged 18 or over in households holding
+   the type are eligible: cash only needs a cash balance, stocks and shares only
+   a stocks-and-shares balance, and both needs both. For each class, a small
+   iterative proportional fit rakes the eligible weighted mass to Table 9.7's
+   income bands (on the CGT stage's income proxy), Table 9.8's age and sex and
+   Table 9.9's regions (overseas dropped, each margin on Table 9.7's total).
+   Keyed walks select subscribers cell by cell, both first, then stocks and
+   shares only, then cash only. A cell without enough eligible mass is reported
+   (`shortfall_weight_unmet`); no holder is invented.
+3. **How much.** Within each income band and class, amounts follow Table 9.7's
+   subscription bands: uniform inside a band, exactly £20,000 at the top. The
+   rank blends the person's balance rank with a keyed uniform.
+4. **The both class.** Each person's amount is split by their balance shares,
+   shifted on the logit scale so the class's stocks-and-shares total equals
+   Table 9.4's 2023-24 stocks-and-shares amount less the stocks-and-shares-only
+   payments (about 27% stocks and shares). A shift the floor stops is named.
+5. **Growth to 2024-25.** Each class's people grow by Table 9.4's account
+   growth (cash accounts for cash only, stocks-and-shares accounts for stocks and
+   shares only, both together for both), added from eligible holders not yet
+   subscribing. Amounts then scale, one factor per type, to Table 9.4's 2024-25
+   amounts (£95.6bn of cash and £37.2bn of stocks and shares). Subscribers
+   already at the £20,000 overall limit stay there. A person whose combined
+   amount lies between £18,000 (Table 9.7's last band floor below the limit) and
+   the limit moves on a line from their scaled £18,000 up to it, so the range just
+   below the limit stays filled; the rest are capped at it.
+6. **Lifetime ISA payments.** Every holder aged 18 to 47 with room under the
+   overall limit pays in; a holder with none pays nothing. 18 to 47 are the ages
+   that can hold one in 2024-25, since accounts opened at 18 to 39 from April
+   2017. Amounts lie in (0, £4,000], drawn by a truncated-exponential tilt so
+   the total equals Table 9.4's 2024-25 Lifetime ISA amount (£2,818m from
+   1,136,000 accounts), so the baseline leaves nothing above the overall limit.
+   The rank blends the balance and income ranks with a keyed uniform.
+
+## Decisions (28 October 2026)
+
+- **Measure, bind none.** No ISA fact enters the weight solve: binding the cells
+  the stage draws to would be circular. A reviewed fence declares every vendored
+  ISA fact measured only; the checks below are reported and never block a build.
+- **Table 9.7's people, the stocks-and-shares residual.** Table 9.7 keeps its
+  people in every band; the both class's split takes its stocks-and-shares
+  total from Table 9.4's residual.
+- **People by accounts.** Growth uses Table 9.4's account growth, the upper
+  bound on people; the flat-people bound is reported beside it.
+- **Lifetime ISAs here.** Holders are re-anchored to HMRC's payers and payments
+  drawn in this change. The evidence is the next pair build.
+
+## Decisions (29 October 2026, on the review of microcosm#1421)
+
+- **Cash grows to Table 9.4's level, holding and tapering to the limit.**
+  Growing by Table 9.4's ratio carried Table 9.7's cash excess into 2024-25 and,
+  with the subscribers at £20,000 unable to grow, scaled everyone else by about
+  2.05. On populations built from HMRC's own tables that gave £127bn of cash
+  against Table 9.4's £95.6bn, 51% of cash subscribers above £12,000 and 40% of
+  all subscribers at the limit. Scaling to Table 9.4's 2024-25 amounts holds the
+  subscribers at the limit, since a plain factor below one would have moved every
+  one of them to about £19,100. One factor of about 0.89 below the limit still
+  left nobody between about £17,750 and £20,000, so amounts above £18,000 now
+  taper up to the limit. That gives a cash factor of about 0.878, 36.9% of cash
+  subscribers above £12,000, 28.8% at the limit (28.8% in 2023-24) and amounts
+  up to about £19,950 below it.
+- **The Lifetime ISA anchor's reach is counted on the pair.** Three later stages
+  read the whole frame (`uc_reporter_redraw`, `uc_take_up` and
+  `pension_credit_take_up`), and a Lifetime ISA counts as Universal Credit and
+  Pension Credit capital, so the added holders can move their draws. The pair
+  build's national counts, at its final weights, the benefit units whose
+  entitlement changes without the added holders' balances, and the units that
+  move in or out of the take-up pools. Where the financial-wealth cap binds, an
+  added holder also scales the household's existing balances down pro rata.
+
+## What the tables leave open
+
+- **The cash excess.** Table 9.7 at band midpoints implies about £92.5bn of
+  cash (£84.5bn cash only and £8.1bn from the both class) against Table 9.4's
+  £69.5bn, about £23bn above it. The stage keeps Table 9.7's people and bands,
+  so its 2023-24 cash total sits above Table 9.4's, and the receipt reports by
+  how much; the 2024-25 total meets Table 9.4.
+- **The impact note's year.** On HMRC's tables 36.9% of 2024-25 cash
+  subscribers pay in more than £12,000, against the impact note's 22% of under-65
+  cash subscribers in 2022-23. Age explains little of that: Table 9.8 puts about a
+  quarter of cash-only subscribers at 66 or over, who would need about 80% above
+  £12,000 to bring the under-65s to 22%. Most of it is the year: Table 9.4's
+  cash subscriptions were £41.6bn in 2022-23, £69.5bn in 2023-24 and £95.6bn in
+  2024-25, so the 22% does not validate 2024-25 either way. Amounts follow Table
+  9.7's bands within income band and class, not age.
+- **The both class's residual share** depends on where the stocks-and-shares-only
+  payments sit inside their bands: about 27% stocks and shares at band
+  midpoints, 50% at band floors and 7% at band tops.
+- **One year of growth.** The individual tables are 2023-24, grown one year
+  across the April 2024 rule change.
+- **Balances.** WAS balances are April 2020 to March 2022 pounds from a Great
+  Britain donor, not uprated; Northern Ireland adults take the Great Britain
+  patterns.
+
+## Gates
+
+`uk_stage_isa_subscriptions_invariants` holds the rules any population keeps:
+the limits, ages, holding what is paid into, finite and non-negative amounts and
+person balances that sum to the household totals. It is no population fact
+check, so the synthetic smoke build blocks on it as every licensed build does.
+`uk_stage_isa_subscriptions` holds the 2024-25 totals, the both class's residual
+and the Lifetime ISA amount to Table 9.4. A synthetic population cannot reach
+HMRC's national totals, so it is a population fact check.
+
+## Checks (reported on the pair build's national file)
+
+- Subscription cells before and after calibration, by class and band, and above
+  £12,000.
+- Holders against Tables 9.10 to 9.12, and market values against Table 9.6.
+- The impact note's 22% of under-65 cash subscribers above £12,000.
+- Lifetime ISA payers and amounts against Table 9.4, and `lifetime_isa_bonus`
+  against 25% of payments; the 2025-26 house-purchase withdrawals (99,750) as a
+  loose check on holders.
+- The flat-people growth bound.
+- The Lifetime ISA anchor's reach: benefit units whose Universal Credit or
+  Pension Credit entitlement changes without the added holders' balances, at
+  the final weights, and the units that move in or out of the take-up pools.
+
+## What the fields support
+
+The fields support the subscriptions a limit leaves out: the overall £20,000
+limit and a separate cash limit (the £12,000 under-65 setting is a reform in
+policyengine-uk), through `isa_subscriptions_above_limit` and
+`cash_isa_subscriptions_above_limit`. policyengine-uk 2.164.2 reads neither in
+any tax or spending variable, so a limit reform changes those amounts but costs
+nothing until the engine models where the excess goes. The subscriptions are
+flows the engine does not uprate, so a later year reads 2024-25 amounts. In
+particular the cash subscriptions above £12,000 in 2027-28, which a £12,000 cash
+limit from April 2027 would leave out, are 2024-25 levels here, not a 2027-28
+base: the dashboard's cash ISA card should not read them as one (Vahid's review
+of microcosm#1421). policyengine-uk's own docstring notes a producer could grow
+them and cap each person at the limit in force; this stage does not, since
+HMRC's tables stop at 2024-25. The
+Lifetime ISA bonus does reach government spending, so its rate and payment
+limit score. The fields also support analyses of who subscribes how much. They
+do not support withdrawals or first-home purchase rules, Innovative Finance or
+Junior ISAs, or behavioural responses to a limit.