model catalog

One microsimulation, six models that extend it.

The microsimulation says what a family pays and what a reform costs. The six macro models supply what it cannot see — shocks, forecasts, long-run change — and a score can chain them.

the suite

The core, and six extensions.

Every extension names where it joins the microsimulation.

Taxes and benefits, for a household or a populationthe core · UK + US · pe-microsim

PolicyEngine’s own engine, and the one behind policyengine.org: the tax-benefit rules as code over enhanced survey microdata, for one household or the whole population. The only member covering both countries.

→£50,000 UK employment income → £39,520 net. Household answers are exact rule arithmetic

→Population answers inherit survey-data error and carry no interval

→Static by design — no forecast, no GDP, inflation or behavioural response. The six below add those

→Hosted — MCP tools and a CLI, no account needed

Open PolicyEngine microsim →

A selected UK fiscal or macro scenario over 3–5 yearsUK · obr-macro

A transpiled, solvable copy of the OBR’s published macroeconometric model, anchored to the latest EFO: shock a lever, re-solve, and read GDP, consumption and investment against the official path.

→Tracks the published forecast within 0.15% (GDP) / 0.25% (consumption) with the anchors held; free-running, 4.48%

→Household multiplier ~0.35 vs the OBR’s 0.3; the spending lever is 1.0000 by construction vs 0.6

→Joins the core: score_reform feeds the static costing in through HHDI_ADDFACTOR

Open the OBR emulator →

What drove UK GDP and inflation at the latest model-data vintageUK · boe-svar

An eight-variable Bayesian structural VAR: named shocks identified by sign and zero restrictions, a decomposition of recent GDP and inflation, and a forward projection with credible bands.

→Frozen-edge check: 14 of 14 outturns inside the 68% bands (bands run wide)

→Against a drifting random walk no skill survives 64 tests — smallest q is 0.36

→Joins the core: the CPI path drives benefit-uprating incidence (svar_inflation_incidence, experimental)

Open BoE SVAR →

A monetary or fiscal shock in the US over coming quartersUS · frb-us

The Federal Reserve’s FRB/US model — 284 equations parsed from the Board’s own model file — solved around the April 2026 LONGBASE baseline. Pick a lever and a policy rule, read the deviation path.

→Levers: federal funds rate, government spending, transfers, taxes

→Matches the Fed’s pyfrbus to ~1×10⁻⁸, its own noise floor. Reproducing LONGBASE is an identity, not evidence

→Joins one way: wage and hours drive employment-income incidence (frbus_shock_incidence, experimental). score_reform refuses it

Open FRB/US →

How US shock responses differ across the household wealth distributionUS · us-hank

A two-asset heterogeneous-agent New Keynesian model (an ABRS 2021 replication): one monetary, fiscal or productivity shock, solved for the whole distribution of households.

→Consumption responses by liquid-wealth quartile and the hand-to-mouth share, warm in seconds from a cached sequence-space Jacobian

→Stylized replication — not a forecaster, no reform scoring, and responses are first-order

→Joins the core one way: wage and labour IRFs drive employment-income incidence (hank_shock_incidence, experimental)

Open US HANK →

Work, saving, and capital effects over decadesUK · psl-og

80 overlapping cohorts choose work and saving over their lifetimes. Solve the steady state or the 60-period transition path and read where a reform takes GDP, investment and revenue in the long run.

→Worked example: +1pp basic rate → GDP −£5.0bn (−0.14%) and tax revenue +0.29pp of GDP by 2030

→Local-only and experimental: calibrated, with no published ground truth

→Joins the core both ways: statute in through PolicyEngine-estimated tax functions, long-run changes back out into a second microsim run (og+microsim)

Open OG-UK →

A UK climate-policy scenario, as deltas against a baselineUK · define-uk

The authors’ ecological stock-flow-consistent model of the UK economy and climate, run unmodified at a pinned commit: one published scenario, read as deltas over 2023–2037.

→Deltas only, never levels — the baseline sits far from outturns, so nothing here is a forecast

→Partial replication: the macro block replicates manual Table 4, and scenario deltas are design-gated because no numeric v1.1 scenario results are published

→Joins the core one way: scenario income deltas drive employment-income incidence (define_scenario_incidence, experimental)

Open DEFINE-UK →

Already have two candidates? Compare their assumptions and tradeoffs.

combine the models

Every join starts or ends at the microsimulation.

score_reform accepts microsim, obr, og and og+microsim. The other four members connect the opposite way.

What would a reform cost — and what would it do to the economy?score a reform

The pipeline chains two members: the microsimulation produces the static costing from the actual tax-benefit rules, then the OBR emulator returns the second-round effects.

Reform pe-microsim obr-macro Scored result

→Worked example: 1p on the basic rate — £6.46bn (2026) → £7.38bn (2030), GDP −0.02% on impact building to −0.06%

→The static costing runs below HMRC’s ready reckoner at every matched year — −6.4% in 2026–27; a benchmark, not a validation

→UK only: no US macro bridge exists, so a US reform is scored statically

pe-macro score --country uk --reform '{"gov.hmrc.income_tax.rates.uk[0].rate":0.21}' --model obr

Read the caveats before the headline: the costing inherits survey-microdata uncertainty, and the emulator’s household-route multiplier is ~0.35 vs the OBR’s published 0.3 for income tax, while its spending multiplier is 1.0000 by construction vs 0.6 — treat macro feedback as sign and order of magnitude. Mechanics on the microsimulation and OBR emulator pages; setup on the Use page.

What a reform does once wages and the capital stock adjustUK · og+microsim

OG-UK solves the reform’s long-run steady state, then its wage and interest-rate changes feed a second microsimulation run, so the distributional result carries general-equilibrium feedback.

→Local-only and experimental; the OG solve needs its own environment, so run the two-step --og-payload pipeline

→The steady-state factor is applied flat: no transition path, and no price-level effects, because the OG model is real

→Effective-labour incidence is reported, not allocated — and no scored result is published

Open OG-UK →

How a macro shock lands across householdsexperimental · four members

The four members that cannot score a reform run the arrow the other way instead. boe-svar, frb-us, us-hank and define-uk each take a macro path in and return distributional incidence through a microsimulation run — all four experimental, and none of them a reform bridge. The row-by-row mapping is in the comparison table.

The routes that run these chains end to end are listed on the Forecasts page.

compare the models

Seven models, side by side.

The seven models compared across question class, method, country, vintage, microsimulation connection and run surface
	PolicyEngine microsimulation pe-microsim	OBR macroeconometric obr-macro	Bank of England structural VAR boe-svar	Federal Reserve model frb-us	US heterogeneous-agent NK us-hank	Overlapping generations psl-og	Ecological stock-flow consistent define-uk
Question class	What a specific family pays and receives; distributional effects	Selected near-term UK fiscal and macro scenarios	What drove the UK economy at the latest vintage; short-run forecasts	US macro dynamics: monetary and fiscal shocks, quarter by quarter	Distributional transmission of stylized US shocks: who responds, by wealth	Long-run incentives: labour supply, saving, the capital stock over decades	UK climate-policy scenario deltas — nothing quotable today
Method	Static tax-benefit microsimulation — direct rule evaluation, no behavioural response	Structural macroeconometric emulator — Gauss–Seidel over 372 equations per quarter	Bayesian structural VAR — posterior sampling with zero and sign restrictions	Large-scale macroeconometric model — 284 equations solved under VAR expectations	Two-asset HANK (Auclert et al. 2021) — first-order sequence-space impulse responses	Dynamic general equilibrium (OLG) — steady-state root-find and transition-path iteration	Ecological stock-flow consistent (E-SFC) — upstream R code run unmodified at a pinned commit
Country	UK & US	UK	UK	US	US	UK (OG-UK)	UK
Data vintage	UK & US statute; enhanced FRS 2023–24 microdata for UK population runs	OBR Economic and Fiscal Outlook, March 2026	ONS, BoE and FRED — estimated through 2025Q1, conditioned to the 2026Q1 edge	The Fed's `model.xml` and LONGBASE database, April 2026	The published Econometrica 2021 calibration; no data vintage	ONS, OBR and BoE aggregates as calibration targets; OG-UK 0.3.2	DEFINE-UK 1.1, April 2026, at pinned commit `846081a`
Microsim connection	Is the microsimulation — every bridge starts or ends here	Static costing in, macro feedback out — `score_reform` via `HHDI_ADDFACTOR`	CPI path → benefit-uprating incidence (`svar_inflation_incidence`, experimental)	Wage/hours path → employment-income incidence (`frbus_shock_incidence`, experimental)	Wage/labor IRFs → employment-income incidence (`hank_shock_incidence`, experimental)	Two-way: statute in via PE tax functions; long-run changes back into a second microsim run (`og+microsim`)	Scenario income deltas → employment-income incidence (`define_scenario_incidence`, experimental; hosted via `define_payload`)
Reform scoring	Yes — a PolicyEngine parameter + a new value, applied directly to households	Yes — shocks to exogenous model variables; static-costing bridge from PolicyEngine reforms	No — the baseline/conditioning member: it reads the economy reforms are scored against	No — shock experiments only; no reform bridge, so `score_reform` refuses `model="frbus"`	No — stylized shocks only; the labor tax is endogenous, so `score_reform` refuses `model="hank"`	Yes — a PolicyEngine policy translated into estimated tax functions	No — no reform bridge; `score_reform` will not accept it
Out of scope	GDP, inflation, interest rates, or macro feedback	Arbitrary reform incidence; borrowing; capital spending (`CGIPS` is dead) and corporation-tax magnitudes (`TCPRO` converges slowly: 12 quarters is ~40% of its plateau)	The causal effect of a specific statutory reform	PolicyEngine reform effects, anticipated-policy paths, or MCE scenarios	Forecasts, detailed tax reforms, or nonlinear/state-dependent dynamics	A short-run forecast or independently validated reform effect	Levels, forecasts, reform scoring; microsimulation incidence (planned, post-validation)
Run surface	Hosted	Hosted (CLI, MCP, Python API)	Hosted (CLI, MCP, Python API)	Hosted (CLI, MCP, Python API) — raw shocks only	Hosted (CLI, MCP, Python API) — stylized shocks only	Local only	Local CLI (`pe-macro define-scenario`); hosted MCP tools return run instructions
Typical runtime	Sub-second per household; minutes for population runs	Seconds to minutes per scenario	Minutes per full estimation + identification run	Seconds to minutes per simulation	~18s cold per variant (cached); instant warm	~17+ min per steady-state solve (two per score); hours for transition paths	Instant from the cached pinned run; a fresh upstream run takes tens of minutes in R 4.3

Compare two models

First model Second model

validation

How each model is validated.

The seven models support different kinds of check, grouped below into four classes with one headline each.

Model	Verification class	What it is checked against	Headline result
pe-microsim	Checked against implemented legislation	Implemented UK and US legislation, rule by rule	Household maths exact; population adds survey error
obr-macro	Replication with a published anchor	March 2026 EFO; HMRC ready reckoner; ONS outturns	Anchored GDP 0.15% MAPE — by construction; free-running, 4.48%
boe-svar	Replication with a published anchor	Brignone & Piffer (2025); ONS outturns	FEVD 37.4% / 42.3% vs the paper's ~40% / ~50% — CPI ~8pp short
frb-us	Replication with a published anchor	The Fed's `pyfrbus` on an identical shock; published multiplier ranges	Shocked paths match pyfrbus to 1.4×10⁻⁸, the Fed's own release-to-release gap; baseline tracking is an identity
us-hank	Replication with a published anchor	Auclert et al. (2021), via the authors' toolkit	β solves to 0.976274 vs the paper's 0.976; Y, K and r are imposed inputs, IRF magnitudes unvalidated
psl-og	Calibrated counterfactual — no ground truth	ONS/OBR aggregates as targets, not a validation set	Targets met by construction; no ground truth
define-uk	Partial replication — macro block and scenario design-gate pass	Manual Table 4; ONS outturns, OBR EFO, DESNZ emissions — no scenario figures are published	Baseline 2025 growth 4.66% vs a 1.31% ONS outturn — deltas only, never levels

The same questions, answered by other shops.

Imposed matches are anchors, not validation. The stronger evidence is out-of-sample forecasts, independent costings, and agreement with another implementation.

Question	Ours	Theirs	How comparable
Long-run effect of a 1pp basic-rate rise (psl-og)	Not yet produced — the committed run is a transition path to 2030, not the long-run steady state WP 22 reports	OBR Working Paper 22, Table 5.1: GDP/person −0.1%, labour supply −0.2%, productivity +0.1%	Future benchmark — same scenario; the target a later run will be judged against, not a match claimed today
Yield of a 1p rise in the basic rate, 2026–27 (obr-macro + pe)	£6.46bn static	HMRC ready reckoner (June 2025): £6.9bn, rising to £8.2bn by 2028–29	Adjusted — −6.4% in year one, below HMRC at every matched year; HMRC is post-behavioural on administrative data, ours static and survey-based. A benchmark, not a validation
Government-purchases multiplier (frb-us)	0.72 in year one (inertial Taylor rule); 0.90 in year two with the funds rate fixed	Ramey (JEP 2019): government-purchases multipliers sit "in a surprisingly narrow range of 0.6 to 1"	Adjusted — both sit inside Ramey's range, but it pools estimates across horizons and monetary regimes, so neither of ours is matched to it. Against the pegged-rate cross-model range (1.1–1.2), our 0.90 sits below
Income-tax multiplier (frb-us, US)	0.22 → 0.32 over years 1–2	OBR's published UK assumption: 0.3 for income tax and NICs	Loose — different country and model class
Forecast error, 2024Q3 onward (boe-svar)	RMSE 0.32pp (GDP growth, CPI) from one frozen 2024Q2 origin; the 2025 CPI hump missed by up to 0.6pp. Over 49 origins, no skill against a drifting random walk	OBR FER (July 2025): one-year GDP growth under by 0.4pp on average; BoE MPR (Aug 2024): late-2025 CPI ~2.4% modal vs a 3.8% outturn (~1.4pp miss)	Adjusted — different statistics, matched vintages; full diagnostics: boe-svar validation

evidence guides

Read the source literature and replication notes.

Six working papers (full PDFs) and one feasibility note — one document per model.

pe-microsimworking paper

Statute as code for the UK and US: why household arithmetic is exact, population estimates are not, and how the static costing benchmarks against HMRC.

read →

obr-macroworking paper

How the OBR's model was transpiled and anchored to the EFO — and why the anchored fit is not forecast skill.

read →

boe-svarworking paper

Replicating Brignone & Piffer (2025): the identification, the FEVD shares, and the forecast diagnostics against naive benchmarks.

read →

frb-usworking paper

Parsing and solving the Fed's FRB/US model, and matching pyfrbus on shocked paths at its own noise floor.

read →

psl-ogworking paper

The OG-UK calibration, tax-function estimation, and the limits of a model with no ground truth.

read →

us-hankfeasibility note

The open US HANK landscape, and why the Auclert et al. (2021) toolkit was the foundation to build on. Not a replication write-up — that gate is on the us-hank validation page.

read →

define-ukworking paper

The authors' unmodified code at a pinned commit, the Table 4 gate, thirteen pinned manual defects, and why only deltas are served.

read →