Add PolicyEngine vs RF nowcasting comparison docs page (#1171) (#1706)

#1171 asked for a docs page describing how PolicyEngine's nowcast differs
from the Resolution Foundation's (per their Living Standards Outlook
appendix). Add it under docs/book/assumptions/nowcasting-comparison.md.

The page is high-level rather than a parameter audit, covering the five
axes where the two methodologies typically diverge:

- wage-floor / NLW handling and spillover effects (RF models the floor
  explicitly; PolicyEngine uprates earnings by the OBR index),
- benefit uprating and the triple lock,
- take-up handling,
- calibration targets (PolicyEngine adds LA-level rent and council-tax
  targets that the public RF nowcast does not),
- horizon (the long-run divergence beyond 2030).

It also gives a short "when results disagree, look here first" guide so
users can interpret discrepancies between the two nowcasts.

Author: vahid-ahmadi

changelog.d/1171.md

@@ -0,0 +1 @@
+- Add a docs page comparing PolicyEngine UK's nowcasting approach to the Resolution Foundation's Living Standards Outlook methodology — wage-floor handling, benefit uprating, take-up, calibration targets, and horizon — under `docs/book/assumptions/nowcasting-comparison.md`.

docs/book/assumptions/nowcasting-comparison.md

@@ -0,0 +1,102 @@
+# PolicyEngine vs Resolution Foundation nowcasting
+
+The Resolution Foundation (RF) publishes a UK microsimulation nowcast as
+part of its [Living Standards Outlook][lso] and pre-budget previews. RF
+documents its methodology in the LSO appendix; PolicyEngine's approach is
+spread across this repo, `policyengine-uk-data`, and the parameters under
+[`gov/economic_assumptions/`](./growthfactors.md). This page summarises
+how the two approaches differ at a high level so users can interpret
+discrepancies between the two nowcasts.
+
+For a focused write-up of how PolicyEngine generates the economic growth
+factors, see [`growthfactors.md`](./growthfactors.md).
+
+## Common ground
+
+Both PolicyEngine and RF start from the same survey backbone — the **Family
+Resources Survey** (FRS) plus its companion datasets (HBAI) — and project
+it forward using a mix of:
+
+- macroeconomic year-on-year growth rates (CPI, average earnings, RPI,
+  CPIH),
+- benefit and tax parameter uprating,
+- a calibration step that reweights households to match published
+  aggregates (caseloads, fiscal totals).
+
+In broad strokes both nowcasts produce gross-to-net household incomes and
+participation-weighted programme spending consistent with OBR-published
+totals over the next two-to-three fiscal years.
+
+## Where the methodologies diverge
+
+### Earnings growth
+
+| Aspect | Resolution Foundation | PolicyEngine UK |
+|-------|------------------------|-----------------|
+| Wage floor | Models the National Living Wage / NMW explicitly each year, lifting anyone below their age-appropriate minimum to the floor. | Earnings are uprated by the OBR average-earnings index without an explicit floor; minimum wage uprating is held in `gov/hmrc/minimum_wage/` and only enters when a reform invokes it. |
+| Spillovers | Includes a spillover effect for workers just above the wage floor (compression of the lower wage distribution). | No automatic spillover; earnings shift uniformly with the OBR index. |
+| NLW extensions | Models the 2024-25 extension of the NLW to 21- and 22-year-olds, and provisionally the 2029-30 extension to 18-20-year-olds. | NLW age bands are encoded in parameters (`gov/hmrc/minimum_wage`) but extensions only flow into outputs if a household's wage actually falls below the modelled floor — there is no explicit re-anchoring of earnings. |
+| Beyond 2025-26 | Assumes the wage floor grows in line with average earnings. | Same effective behaviour, but driven by uprating rather than a wage-floor rule. |
+
+**Practical implication.** Low-paid workers' incomes can grow faster under
+the RF nowcast than under PolicyEngine because the wage floor binds and
+ripples up; PolicyEngine treats them like everyone else under the OBR
+earnings index.
+
+### Benefit uprating
+
+| Aspect | Resolution Foundation | PolicyEngine UK |
+|-------|------------------------|-----------------|
+| Working-age benefits | Uses statutory uprating with explicit overrides for announced policy (e.g. CoL Payments, benefit freezes). | Same approach. Parameters under `gov/dwp/` and `gov/hmrc/child_benefit/` track legislated rates; ad hoc payments live under `gov/treasury/cost_of_living_support`. |
+| State Pension | Models the triple lock explicitly, using its own internal earnings/CPI forecasts. | Models the triple lock via `gov/dwp/state_pension/triple_lock/*.yaml`; the uprating value tracks announced DWP rates with a fallback to the maximum of earnings, CPI and the 2.5% floor. |
+
+### Take-up
+
+| Aspect | Resolution Foundation | PolicyEngine UK |
+|-------|------------------------|-----------------|
+| Approach | Uses HMRC/DWP outturn take-up rates and aligns simulated caseloads to published totals via calibration. | Same: each means-tested benefit has a `*_takeup_rate` parameter populated stochastically into `would_claim_*` flags in `policyengine-uk-data`, and the reweighting step in `policyengine-uk-data` enforces caseload alignment. |
+
+### Calibration / reweighting
+
+| Aspect | Resolution Foundation | PolicyEngine UK |
+|-------|------------------------|-----------------|
+| Targets | Aligns to a mix of HMRC, DWP and ONS aggregates. | Aligns to the same family of targets, but is publicly documented in `policyengine-uk-data` (`targets/`) and includes finer-grained local-authority targets for council tax and rents (see `policyengine-uk-data/datasets/local_areas/`). |
+
+### Horizon
+
+- **RF**: typically 4-5 years ahead in detail; long-run sensitivities run
+  separately.
+- **PolicyEngine**: 5+ years ahead using the OBR EFO supplemented by
+  long-run equilibrium assumptions in `yoy_growth.yaml` (see
+  [`growthfactors.md`](./growthfactors.md) and the
+  `gov/economic_assumptions/README.md` for the 2031+ construction).
+
+## Interpreting discrepancies
+
+When PolicyEngine and RF disagree on a published statistic (e.g. relative
+poverty, fiscal cost of a reform), the main sources of difference tend to
+be:
+
+1. **Wage-floor handling** — concentrated in low-pay groups (young
+   workers, hospitality, retail). RF nowcasts higher real earnings growth
+   here.
+2. **Take-up calibration vintage** — both models drift between official
+   take-up statistics releases; align to the same vintage before comparing.
+3. **Local-area targets** — PolicyEngine's calibration includes
+   LA-level rent and council tax targets that the public RF nowcast does
+   not, which can shift housing-cost and disposable-income distributions.
+4. **Long-run horizon** — beyond 2030 the divergence widens because the
+   two models use different long-run rules; treat post-2030 figures as
+   illustrative rather than authoritative.
+
+## References
+
+- Resolution Foundation, [Living Standards Outlook 2025][lso] — appendix
+  details the RF nowcast methodology.
+- Resolution Foundation, [budget preview reports](https://www.resolutionfoundation.org/publications/topic/budget-2025/) — operational pre-budget application.
+- PolicyEngine, [growthfactors.md](./growthfactors.md) — economic
+  assumption growth rates.
+- PolicyEngine, [`policyengine_uk/parameters/gov/economic_assumptions/README.md`](https://github.com/PolicyEngine/policyengine-uk/blob/main/policyengine_uk/parameters/gov/economic_assumptions/README.md) — three-horizon construction of growth series.
+- `policyengine-uk-data` — calibration targets and reweighting code.
+
+[lso]: https://www.resolutionfoundation.org/publications/living-standards-outlook-2025/