US forecast track record
No model here forecasts the US economy yet, so there is nothing to score.
No US record yet
What the two US models do instead.
- FRB/US answers what a shock does; its validation makes no predictive claim.
- US HANK replicates Auclert et al. (2021); first-order, stylized; one published magnitude is matched, the rest are not validated.
- The UK record exists because BoE SVAR forecasts and accepts being scored.
What starts a record
The same rules as the UK record, the day a US model makes a predictive claim.
- Archived first: rounds committed to Git before the outturn.
- Append-only: under the same CI gate as the UK record.
- Scored: against realised data, whatever that shows.
The Fed’s baseline
The near-term path FRB/US shocks deviate from.
FRB/US October 2026 LONGBASE. Not a PolicyEngine forecast; no bands; never scored.
| Quarter | GDP y/y | CPI y/y | Unemp. |
|---|---|---|---|
| 2026Q2 | 2.2% | 3.8% | 4.3% |
| 2026Q3 | 1.8% | 3.9% | 4.2% |
| 2026Q4 | 2.3% | 3.8% | 4.1% |
| 2027Q1 | 2.3% | 3.6% | 4.1% |
| 2027Q2 | 2.3% | 2.7% | 4.1% |
Source: Federal Reserve FRB/US LONGBASE
Source details
Source: longbase_baseline_yoy.csv, from the Fed’s package data. Shown beside each outturn on US growth, inflation and jobs.
What the US side runs
Four routes on the microsimulation; none takes a reform into a US macro model.
- No reform with macro feedback. FRB/US and US HANK do not take PolicyEngine reforms; no mapping exists from a US statutory reform to a US macro model.
The routes, with the command for each
| What you get | Models, in the order they run | What the chain does | Command |
|---|---|---|---|
| One household, exactly | PolicyEngine microsimulation | arithmetic over the statutory rules; no sampling, no weights | pe-macro household-impact --country us --people '[{"age":35,"employment_income":50000}]' --household '{"state_code":"CA","county_fips":"06037"}' --reform '{"gov.irs.credits.ctc.amount.adult_dependent":1000}' |
| The whole population, one policy year | PolicyEngine microsimulation | revenue and distribution; a static costing | pe-macro score --country us --reform '{"gov.irs.credits.ctc.amount.adult_dependent":1000}' --model microsim |
| A macro shock carried into household earnings | FRB/US → PolicyEngine microsimulation | the wage-bill change applied through the automatic stabilisers | pe-macro frbus-shock-incidence --var rffintay_aerr --shock 1.0 --year 2027 |
| The same question in the HANK member | US HANK → PolicyEngine microsimulation | impulse responses around a calibrated steady state | pe-macro hank-shock-incidence --kind monetary --size -0.0025 --year 2026 |
About the engine
- PolicyEngine microsimulation runs both countries; the US household rows are the UK commands with --country us.
- It reports taxes, benefits, net income, revenue, distribution over a single policy year.
- The incidence rows take a macro shock, not a reform, and push its earnings effects through households: they answer who bears a shock, not what a reform does to output.
- Over MCP: household_reform_impact, population_reform_impact, frbus_shock_incidence and hank_shock_incidence; recommend_model declines a US reform that needs macro feedback.