note · 25 Jul 2026 · inflation

CPI fell to 2.8%. The model’s median path turns back up.

Inflation came in 0.12pp above the archived forecast median and inside its 68% band. The model’s central path turns higher again by Q4.

Data: ONS D7G7 (CPI annual rate), vintage 2026-07-25 · Bank of England IUDBEDR (Bank Rate) · forecast round 2026-07-21, boe-svar, 2026Q1 data edge. Written on 25 July 2026; corrected on 18 Nov 2026 (the q-value, from the current evaluation, and the estimation sample). Later outturns are on the scoreboard.

What happened

UK CPI inflation fell to 2.8% year-on-year in 2026Q2, the third consecutive quarterly fall from the 3.8% peak in 2025Q3:

UK CPI annual rate, quarterly (ONS series D7G7)
2025Q32025Q42026Q12026Q2
3.8%3.4%3.1%2.8%

The forecast round dated 21 July, from a 2026Q1 data edge, had a median of 2.68% with a 68% credible band of 2.24% to 3.16%. The outturn was 0.12pp above the median and inside the band.

What the model’s median path shows

The same forecast round does not extrapolate the fall. Its median CPI path is roughly flat through the third quarter and then turns back up:

boe-svar median CPI path from the 2026-07-21 round, with 68% credible bands
QuarterMedian68% band
2026Q32.71%1.95% – 3.45%
2026Q43.12%2.09% – 4.14%
2027Q13.12%1.88% – 4.49%

By 2027Q1 the 68% band spans 1.9%–4.5%. Re-acceleration is the central case, not a confident call.

From the 0.9% ONS outturn for growth in 2026Q1 (2026-07-25 vintage; since revised to 0.8%), the model’s median growth path rises to 1.1% in 2026Q2 and 1.7% by Q4.

What would change this view

  • 2026Q2 GDP, due from the ONS in August. The model's median is 1.10% year-on-year; a materially weaker print would undercut the recovery half of the story and is the single most informative release for this view.
  • A downside Q3 CPI print. The model's 68% band opens at 1.95%; an outturn near that floor would make the re-acceleration look like model inertia rather than signal.
  • Anything on the fiscal side. This note is a pure forecast read: no policy change is assumed. The transmission from a fiscal announcement to the rate path is not yet modelled here.

Limitations, stated

  • boe-svar is a replication of Brignone and Piffer (2025). This round’s coefficients were estimated on 1992Q1–2023Q2; later observations update the conditioning data, not the coefficients.
  • The model forecasts CPI and GDP. It has no term structure, no exchange rate transmission channel exposed on this site, and no view on market pricing.
  • One scored forecast is not a track record. Treat the model's accuracy as unestablished and read the pseudo-out-of-sample evaluation, where it beats a drifting random walk on 4 of 8 variables at one quarter and 2 of 8 at eight; none of these survives adjustment for the 64 tests run (smallest q = 0.39).

← All notes · The model behind this note · Track record