diff --git a/CHANGELOG.md b/CHANGELOG.md index 761d6822..9f86ef59 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -10,6 +10,13 @@ heading here; `tests/test_repo_hygiene.py` asserts that they do. ## [Unreleased] +### Bank of England gap explanation and one multiplier sign convention (#280) + +- **`/models/compare`, 100bp question.** A new unfolded block lists why the two UK rows sit either side of the Bank of England's 1999 range: model, shock, horizon and monetary response, plus what narrows each gap. It is generated from a new optional `gaps` key in `results.json`. No figure in any row changed. +- **Multiplier sign convention.** Every multiplier takes the sign of the OBR's table (EFO July 2015, Box 3.2): GDP gained per unit spent, GDP lost per unit of tax raised, so the usual response is positive. The compare columns, `/obr` and `/obr/validation` now use that one wording ("GDP fall per £", "GDP moved per pound" and "ratio" are gone). No number changed sign: the wording was mixed, the values were already positive. +- **SVAR draws.** The 100bp row's description of the published replication run follows boe-var-model `17f5d82`: 765 accepted draws → 731, effective sample size 368.0 → 333.3 (`results/summary.md`). The row's medians are still the fast fixture's. +- **OBR benchmark source.** The spending and tax benchmarks name EFO July 2015 Box 3.2 as the source of 0.6 and 0.3, and note that the OBR's November 2023 article gives 0.45 for public services and 0.33 for tax (obr-macroeconomic-model `docs/published_model_review.md`, G11). + ### Model re-pin and site review (#278, #279) - **#278: model pins.** OBR `2965e82d` → `0e44ec87` (main after obr-macroeconomic-model #74 and #75), BoE SVAR `c13a9c91` → `c827ab86`, FRB/US `846eeb5c` → `162cfe6a`, US HANK `551a40d3` → `58353e5f`. After a redeploy the hosted tools run these pins. The FRB/US and US HANK compare rows with stored paths re-run identically; the OBR rows were re-run at `0e44ec8`. diff --git a/models/compare/build.py b/models/compare/build.py index 5b56d647..d23225b3 100644 --- a/models/compare/build.py +++ b/models/compare/build.py @@ -54,8 +54,8 @@ "rate-up-100": (GDP, INFL, ("Timing", "", "timing"), ("GDP per pp-quarter", "% of GDP", "ppq")), "rate-down-100": (("Against the rise", "peak GDP, % of GDP: cut vs rise", "result"),), "rate-up-25": (("Against 100bp ÷ 4", "peak GDP per bp", "result"),), - "spending": (("Multiplier", "GDP per unit spent", "multiplier"), GDP, INFL), - "tax": (("Multiplier", "GDP per unit of tax", "multiplier"), GDP, INFL), + "spending": (("Multiplier", "GDP gained per unit spent", "multiplier"), GDP, INFL), + "tax": (("Multiplier", "GDP lost per unit of tax raised", "multiplier"), GDP, INFL), "energy": (GDP, ("Peak price effect", "% on the CPI level", "inflation"), ("Policy rate", "", "result")), } @@ -67,6 +67,19 @@ ("linearity", "Against 100bp ÷ 4"), ("result", "What else moves"), ) +# One sign convention for every multiplier on the page (and on /obr), the +# OBR's own: positive when GDP rises with spending or falls with a tax rise, +# whatever the direction of the policy. The signed effect is the Peak GDP +# column. +SIGN_CONVENTION = ( + "Multipliers take the sign of the OBR’s table " + '(EFO July 2015, Box 3.2): ' + "GDP gained per unit of extra spending, GDP lost per unit of extra tax, " + "so the usual response is a positive number. " + "A tax cut or a spending cut has the same multiplier with the GDP sign reversed. " + "Signed effects are in the Peak GDP column." +) + VERDICT_CLASS = { "plausible": "cmp-v-plausible", "plausible-in-model": "cmp-v-mixed", @@ -405,8 +418,25 @@ def multiples(data: dict, q: dict) -> str: } +def gaps_html(q: dict) -> str: + """Itemised reasons the rows differ from the benchmark, shown unfolded.""" + g = q.get("gaps") + if not g: + return "" + items = [] + for item in g["items"]: + src = item.get("source") + link = f' ({esc(src["label"])})' if src else "" + items.append(f'
{esc(g["intro"])}
{esc(n)}
' for n in q["notes"]) + if q["id"] in ("spending", "tax"): + notes = f'{SIGN_CONVENTION}
' + notes + notes += gaps_html(q) return ( f'obr_shock(var="R", shock=1, periods=4)committed fig3 fixture + PR #17: shock="1 s.d. UK monetary-policy shock (+25bp on impact), scaled ×4"committed fig3 fixture + PR #17: shock="1 s.d. UK monetary-policy shock (+25bp on impact), scaled ×4"frbus_shock(var="rffintay_aerr", shock=1.0, periods=1, horizon=12, start="2026Q3", policy_rule="inertial_taylor")GDP per pp-quarter divides the peak GDP effect by the rate path's size to the peak. HANK's figure uses the realised real-rate path; the other rows use the nominal policy rate, which is close to the real rate in FRB/US because its inflation barely moves. On HANK's rstar path (10 pp-quarters at ρ = 0.9) the figure would be about 0.04%, so HANK's 0.16 and FRB/US's 0.15 are not like-for-like.
HANK reports inflation as a quarterly rate; this page multiplies it by 4 to give an annual rate.
+GDP per pp-quarter divides the peak GDP effect by the rate path's size to the peak. HANK's figure uses the realised real-rate path; the other rows use the nominal policy rate, which is close to the real rate in FRB/US because its inflation barely moves. On HANK's rstar path (10 pp-quarters at ρ = 0.9) the figure would be about 0.04%, so HANK's 0.16 and FRB/US's 0.15 are not like-for-like.
HANK reports inflation as a quarterly rate; this page multiplies it by 4 to give an annual rate.
The OBR emulator is 3–6× below the range and the BoE SVAR is above it. The three are different models answering different experiments.
Source: each row’s stored path (FRB/US and US HANK local runs of 18 Nov 2026 at 162cfe6 and 58353e5) · results.json
GDP, % deviation
| Model | MultiplierGDP per unit spent | Peak GDP% of GDP | Peak inflationpp, annual rate | Verdict | Details |
|---|---|---|---|---|---|
| Model | MultiplierGDP gained per unit spent | Peak GDP% of GDP | Peak inflationpp, annual rate | Verdict | Details |
| Published range US | 0.45–1.82 (CBO); 0.6–1 (Ramey); ≈0.95 (Coenen, no accommodation) | — | — | Reference, sourced | |
| |||||
| OBR emulator UK | 1.01–1.02 | +1.014at q1–q4 | 0no response | OffAn accounting identity; the OBR's own figure is 0.6 | |
| Published range US | 0.45–1.82 (CBO); 0.6–1 (Ramey); ≈0.95 (Coenen, no accommodation) | — | — | Reference, sourced | |
| |||||
| OBR emulator UK | 1.01–1.02 | +1.014at q1–q4 | 0no response | OffAn accounting identity; the OBR's own figure is 0.6 | |
| |||||
| BoE SVAR UK | No instrumentThe BoE SVAR identifies no fiscal shock. | ||||
| FRB/US US | 0.81 on impact; 0.75 year 1 | +0.037per 0.046% of GDP | +0.0008at q2 | Plausible0.81 impact, 0.747 yr 1: inside both ranges | |
| FRB/US US | 0.81 on impact; 0.75 year 1 | +0.037per 0.046% of GDP | +0.0008at q2 | Plausible0.81 impact, 0.747 yr 1: inside both ranges | |
| |||||
| US HANK US | 0.175 (ρ = 0.8) | +0.175at impact | +0.84at impact | QuestionableSign flips while spending is still high | |
| US HANK US | 0.175 (ρ = 0.8) | +0.175at impact | +0.84at impact | QuestionableSign flips while spending is still high | |
| |||||
| DEFINE-UK UK | 2.25 nominal, 2027 to 2040 | not reported | not reported | No benchmarkAnnual, investment-led; not comparable | |
| DEFINE-UK UK | 2.25 nominal, 2027 to 2040 | not reported | not reported | No benchmarkAnnual, investment-led; not comparable | |
| |||||
FRB/US's hosted tool cannot run a sustained +1%-of-GDP experiment: the stored run is a one-quarter impulse of about 0.046% of GDP, and the year-1 figure comes from a committed sustained run (macro #212).
+Multipliers take the sign of the OBR’s table (EFO July 2015, Box 3.2): GDP gained per unit of extra spending, GDP lost per unit of extra tax, so the usual response is a positive number. A tax cut or a spending cut has the same multiplier with the GDP sign reversed. Signed effects are in the Peak GDP column.
FRB/US's hosted tool cannot run a sustained +1%-of-GDP experiment: the stored run is a one-quarter impulse of about 0.046% of GDP, and the year-1 figure comes from a committed sustained run (macro #212).
Source: each row’s stored path (FRB/US and US HANK local runs of 18 Nov 2026 at 162cfe6 and 58353e5) · results.json
GDP, % deviation
| Model | MultiplierGDP per unit of tax | Peak GDP% of GDP | Peak inflationpp, annual rate | Verdict | Details |
|---|---|---|---|---|---|
| Model | MultiplierGDP lost per unit of tax raised | Peak GDP% of GDP | Peak inflationpp, annual rate | Verdict | Details |
| Published range US and UK | 0.3–1.5 (CBO, 2-yr lower/middle); 0.1–0.6 (CBO, 1-yr higher income); OBR 0.3, decaying | — | — | Reference, sourced | |
| |||||
| OBR emulator UK | 0.21 at q4 → 0.37 q12, current prices | not reported | not reported | OffRises where the OBR's 0.3 decays | |
| Published range US and UK | 0.3–1.5 (CBO, 2-yr lower/middle); 0.1–0.6 (CBO, 1-yr higher income); OBR 0.3, decaying | — | — | Reference, sourced | |
| |||||
| OBR emulator UK | 0.21 at q4 → 0.37 q12, current prices | not reported | not reported | OffRises where the OBR's 0.3 decays | |
| |||||
| BoE SVAR UK | No instrumentThe BoE SVAR identifies no tax shock. | ||||
| FRB/US US | 0.23 year 1; 0.32 year 2 | −0.115at impact | −0.003at q4 | UnresolvedYr 1 0.23 unmatched; yr 2 0.32 between rows | |
| FRB/US US | 0.23 year 1; 0.32 year 2 | −0.115at impact | −0.003at q4 | UnresolvedYr 1 0.23 unmatched; yr 2 0.32 between rows | |
| |||||
| US HANK US | No instrumentThe labour tax is endogenous in US HANK, so there is no tax-rate instrument. | ||||
| DEFINE-UK UK | No instrumentNo income-tax instrument in the committed scenarios. | ||||
FRB/US with periods = 4 stacks the tax add-factor to +2.37pp rather than +1pp, and the tool's label does not warn (macro #212). The stored run uses periods = 1.
+Multipliers take the sign of the OBR’s table (EFO July 2015, Box 3.2): GDP gained per unit of extra spending, GDP lost per unit of extra tax, so the usual response is a positive number. A tax cut or a spending cut has the same multiplier with the GDP sign reversed. Signed effects are in the Peak GDP column.
FRB/US with periods = 4 stacks the tax add-factor to +2.37pp rather than +1pp, and the tool's label does not warn (macro #212). The stored run uses periods = 1.
Source: each row’s stored path (FRB/US and US HANK local runs of 18 Nov 2026 at 162cfe6 and 58353e5) · results.json
GDP, % deviation