Emulator of the OBR macroeconometric model
OBR emulator
How far to trust each configuration, the check against outturns, and the limits.
How far to trust it, and where it stops
Anchored to the EFO, GDP drifts to +0.42% and consumption to +0.69% above it by 2027Q4; unemployment follows the EFO by construction. Weak when run free. Errors are MAPE against the EFO path.
| Check | Against | Result | Verdict |
|---|---|---|---|
| Solver convergence | Tolerance in every quarter, 15 demo shocks | 12 of 12 quarters | pass |
| Anchored fit, real GDP | October 2026 EFO, 2025Q1–2027Q4; consumption 0.40% | 0.24% MAPE | by construction |
| Held add-factors, real GDP | October 2026 EFO, 2026Q1–2027Q4; consumption 0.85%; 7 of 8 in band, two of them identities on EFO inputs | 0.63% MAPE | weak |
| Free-running, real GDP | October 2026 EFO, 2025Q1–2027Q4; the emulator’s closure, not OBR forecast error | 4.24% MAPE | weak |
| ONS outturns | q/q growth, 2025Q2–2026Q1, ONS release of 12 Nov 2026; the EFO missed 2026Q1 by 0.25pp | misses up to 0.27pp | weak |
| Income tax multiplier | OBR 0.3 on impact, fading; q1, q4 and q12 in current prices; size close, profile reversed | 0.10 → 0.21 → 0.37 | weak |
| Spending multiplier | OBR 0.6, fading; the shock lands in the GDP identity | 1.01 → 1.04 | fail |
| Bank Rate +1pp | Bank of England −0.2% to −0.35% (1999); sign right | −0.06% peak | fail |
Evidence by dimension
| Dimension | Level | What would change it |
|---|---|---|
| Implementation fidelity | moderate | Every behavioural equation run; inactive channels scoped |
| Predictive validation | weak | Rolling-origin tests vs benchmarks and outturns |
| Identification | not applicable | Stays not applicable |
| Policy counterfactuals | weak | Fiscal shocks vs official costings and multipliers |
| Uncertainty | weak | Sensitivity envelopes for judgement and closure |
| Vintage reproducibility | moderate | Three historical EFO vintages reproduced end to end |
About this emulatorIndependent; not an official OBR product
An independent implementation of the OBR's published model code and forecast data. Not produced, maintained or endorsed by the Office for Budget Responsibility; not official OBR estimates.
Three configurations, three tests
Every reform score uses the anchored configuration. The other two are checks.
How each one scoresFull figures, windows and the free-running caveat
- Anchored
- GDP MAPE 0.45% (0.447%) to 2031Q1, every quarter converged at
0e44ec8; CI gates tolerance, the expenditure identity and signs (step 1, unemployment). - Held add-factors
- Re-run on 18 Nov 2026 (
docs/forecasting_framework.mdat0e44ec8): GDP 0.63%, consumption 0.85%, scored on 2026Q1–2027Q4 from the EFO’s own 2026Q1 values (scorecard). - Free-running
- Raw dynamics with GDP endogenous, not OBR forecast error. Every computed line is in step 3.
Bands and gatesWhat counts as in band, what CI enforces
in band rates within 1.0pp · net balances within 1.5% of GDP · levels within 10% MAPE.
gated CI enforces the anchored tolerance and the multiplier band; the HMRC comparison and the outturn backtest are computed for the working paper and refreshed by hand.
The costing it is fed1p on the basic rate, against HMRC
This compares the microsimulation’s static costing with HMRC’s ready reckoner. The emulator takes that costing as given, so it tests the input, not the emulator: a benchmark, not a validation.
The figures, the interpolated 2028–29 bar and why the 2030 row has no counterpart are on the microsimulation’s validation tab.
Static revenue, £bn a year
- ours
- HMRC
Source: Hosted runs
Source details
Source: basic_rate_costing.json; the working paper’s panel B used the earlier 4.x figures (6.46 and an interpolated 6.92).
What it cannot be used for
Each limit opens to its evidence. Only income tax and NICs carry behaviour.
The spending multiplier is close to 1 by construction, against the OBR’s own published 0.6.
Why the multiplier is close to 1Spending lands straight in the GDP identity
The OBR’s 0.6 is for day-to-day public services and welfare spending. A spending shock lands straight in the GDP identity, so £1.25bn a quarter in returns about £1.27bn to £1.30bn of GDP a quarter (figures).
- From model pin
0e44ec8the shock is nominal-consistent by default (obr-macroeconomic-model #75): nominal spending moves by the same cash amount, so the government deflator is unchanged and borrowing rises by the cash cost, +£1,336m at q1 and +£1,419m at q12. Receipts have no equation in the listing and do not respond. - The real-only shock, the earlier default, spilled into the investment deflator and moved business investment by +£295m on impact; that artefact is gone (about −£2m).