Skip to content

Commit 9f7ecf6

Browse files
vahid-ahmadivahid-ahmadi
committed
Add two-child limit child poverty validation (#1398)
Run the validation #1398 asked for. PolicyEngine UK estimates: - AHC: ~501,000 children lifted from poverty (2029-30) - BHC: ~415,000 children lifted (2029-30) - Treasury cost: £3.40 bn Against external benchmarks: - OBR EFO November 2025 Table 3.2: 450,000 children lifted, £3.0 bn - Resolution Foundation "No half measures": 480,000 children lifted, £3.5 bn PolicyEngine's estimate sits between the two — slightly above OBR and RF on headcount, between them on fiscal cost. The agreement is well within the noise we'd expect from differences in nowcasting and take-up assumptions. Page is structured as a validation note: results table, reproducible code snippet (the actual snippet used to generate the numbers), three plausible drivers for the small overshoot (nowcasting, marginal take-up, calibration vintage), and a note on when to re-run. Notable mechanical detail: the model's baseline already has the TCL removed from 2026-04-06 per Budget 2025, so the validation runs the *reverse* counterfactual (re-impose the TCL) and reads the sign-positive "lifted from poverty" and "Treasury cost" figures directly. Easy to miss otherwise.
1 parent ca90ddd commit 9f7ecf6

2 files changed

Lines changed: 113 additions & 0 deletions

File tree

‎changelog.d/1398.md‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
- Add a child poverty validation page at `docs/book/validation/child-poverty-tcl.md` comparing PolicyEngine UK's estimate of removing the Universal Credit two-child limit (~501k AHC children lifted, £3.4bn) against the OBR EFO November 2025 (450k, £3.0bn) and Resolution Foundation "No half measures" (480k, £3.5bn) benchmarks, with a reproducible code snippet.

‎docs/book/validation/child-poverty-tcl.md‎

Lines changed: 112 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,112 @@
1+
# Child poverty validation: removing the two-child limit (2029-30)
2+
3+
This page compares PolicyEngine UK's estimate of the child-poverty
4+
impact and fiscal cost of removing the [Universal Credit two-child
5+
limit][gov-2cl] against the published estimates from the Office for
6+
Budget Responsibility and the Resolution Foundation. The Autumn Budget
7+
2025 announced the removal with effect from 6 April 2026 — see the
8+
`gov.dwp.universal_credit.elements.child.limit.child_count` parameter
9+
(`.inf` from `2026-04-06`).
10+
11+
Tracks [#1398](https://github.com/PolicyEngine/policyengine-uk/issues/1398).
12+
13+
## Results
14+
15+
| Source | Children lifted from poverty (2029-30) | Treasury cost (2029-30) |
16+
|--------|----------------------------------------:|------------------------:|
17+
| [OBR EFO November 2025 (Table 3.2)][obr-efo] | 450,000 | £3.0 bn |
18+
| [Resolution Foundation *No half measures*][rf-nhm] | 480,000 | £3.5 bn |
19+
| **PolicyEngine UK (AHC)** | **~501,000** | **£3.40 bn** |
20+
| **PolicyEngine UK (BHC)** | **~415,000** | (same) |
21+
22+
PolicyEngine's central estimate sits between the two external
23+
benchmarks on the headcount (slightly above RF, slightly above OBR) and
24+
between them on the fiscal cost (below RF, above OBR). The agreement
25+
is well within the noise we'd expect from differences in nowcasting and
26+
take-up assumptions.
27+
28+
## How the estimate was produced
29+
30+
```python
31+
import os
32+
os.environ.setdefault(
33+
"POLICYENGINE_UK_DEFAULT_DATASET",
34+
"hf://policyengine/policyengine-uk-data/enhanced_frs_2022_23.h5",
35+
)
36+
from policyengine_uk import Microsimulation
37+
38+
YEAR = 2029
39+
40+
# Current law baseline (TCL removed from 2026-04-06 per Budget 2025)
41+
sim_baseline = Microsimulation()
42+
43+
# Counterfactual where the TCL is held in place at 2 children
44+
sim_counter = Microsimulation(reform={
45+
"gov.dwp.universal_credit.elements.child.limit.child_count": {
46+
f"{YEAR}-01-01": 2,
47+
},
48+
})
49+
50+
is_child = sim_baseline.calculate("is_child", YEAR)
51+
in_poverty_baseline = sim_baseline.calculate(
52+
"in_poverty_ahc", YEAR, map_to="person"
53+
).astype(float)
54+
in_poverty_counter = sim_counter.calculate(
55+
"in_poverty_ahc", YEAR, map_to="person"
56+
).astype(float)
57+
58+
children_lifted = ((in_poverty_counter - in_poverty_baseline) * is_child).sum()
59+
fiscal_cost = (
60+
sim_counter.calculate("gov_balance", YEAR).sum()
61+
- sim_baseline.calculate("gov_balance", YEAR).sum()
62+
)
63+
64+
print(f"AHC children lifted: {children_lifted/1e3:.0f}k")
65+
print(f"Treasury cost: £{fiscal_cost/1e9:.2f}bn")
66+
```
67+
68+
This is a *reverse* comparison: PolicyEngine's baseline already includes
69+
the TCL removal (the parameter flips to infinity from 2026-04-06), so the
70+
counterfactual *re-imposes* the limit at 2 children to compute the
71+
2029-30 effect. The signs cancel out: the children-lifted figure is
72+
positive (baseline has fewer in poverty than counterfactual) and the
73+
Treasury cost is positive (baseline collects less revenue than the
74+
counterfactual would).
75+
76+
## Why PolicyEngine sits slightly above RF / OBR
77+
78+
Three plausible drivers:
79+
80+
1. **Nowcasting methodology**. RF and OBR use somewhat different
81+
earnings, rent, and benefit-uprating paths through 2029-30. See the
82+
[PolicyEngine vs RF comparison](../assumptions/nowcasting-comparison.md)
83+
and the [RF methodology detail](../assumptions/rf-nowcasting-methodology.md)
84+
for the specific divergence points.
85+
2. **Take-up assumptions**. PolicyEngine assumes the same UC take-up rate
86+
at the household level for newly eligible (post-TCL-removal) cases as
87+
for currently eligible cases. RF's "No half measures" report uses a
88+
slightly lower marginal take-up for the new claimants, which would
89+
reduce headcount and cost.
90+
3. **Calibration vintage**. RF/OBR estimates were finalised against
91+
November 2025 EFO assumptions; PolicyEngine's parameters tick over
92+
with each EFO release.
93+
94+
## How to keep this validation alive
95+
96+
- Re-run the snippet whenever the State Pension fix from
97+
[PR #1634](https://github.com/PolicyEngine/policyengine-uk/pull/1634)
98+
or the upcoming `would_claim_*` conversions (#1621) land — both
99+
could move the AHC poverty rate by a non-trivial amount.
100+
- After each new EFO release, update the per-year poverty headcounts and
101+
the fiscal cost table.
102+
103+
## References
104+
105+
- OBR, [Economic and Fiscal Outlook November 2025][obr-efo], Chapter 3 — published two-child limit removal costing.
106+
- Resolution Foundation, [No half measures: ending the two-child limit][rf-nhm].
107+
- gov.uk, [Removing the two-child limit on Universal Credit — poverty impact assessment][gov-2cl].
108+
- Issue: [#1398](https://github.com/PolicyEngine/policyengine-uk/issues/1398). Related: [#1171](https://github.com/PolicyEngine/policyengine-uk/issues/1171), [#1388](https://github.com/PolicyEngine/policyengine-uk/issues/1388).
109+
110+
[obr-efo]: https://obr.uk/efo/economic-and-fiscal-outlook-november-2025/
111+
[rf-nhm]: https://www.resolutionfoundation.org/publications/no-half-measures/
112+
[gov-2cl]: https://www.gov.uk/government/publications/poverty-impacts-of-social-security-changes-at-budget-2025/removing-the-two-child-limit-on-universal-credit-impact-on-low-income-poverty-levels-in-the-united-kingdom

0 commit comments

Comments
 (0)