CPI forecast

How this number was built

1. Seasonal baseline. Over 34 past years (excluding the 2020–2023 shock years), the CPI index rose by an average of 0.14% between September and October.
2. Known price changes. We adjust for things already known before the release, each scaled by its weight in the CPI basket:

   Adjustment	Price signal	Effect on monthly CPI
   Electricity (Ofgem price cap change)	-0.3%	-0.02pp
   Gas (Ofgem price cap change)	+8.0%	+0.08pp
   Petrol and diesel (GOV.UK weekly pump prices)	+3.1%	+0.11pp
   Food (recent trend vs long-run average)	trend	-0.01pp
   Total adjustment		+0.17pp

   Scroll sideways to see all columns →

   "Effect" is the weighted difference between the signal and what the seasonal baseline already assumes for that item.

3. Roll forward. Predicted monthly change 0.32% takes the index from 144.1 to 144.56, against 139.8 a year earlier, which gives an annual rate of 3.4%.

Full details are on the methodology page.

Draft log entries for October 2026: #1 seasonal-avg-v1 3.2% (draft, not locked); #2 seasonal-avg-v2 3.4% (draft, not locked). The page shows the newest entry.

Backtest: how the method would have done

This is a backtest, not a live track record. We reran the method over the 24 months from October 2024 to September 2026, each time using only data published before that month's release. These are not predictions we made in advance.

Caveats

• 24 months is a small sample. The gap between v1 and v2 is a few hundredths of a point, which could be noise.
• The backtest uses today's data vintage, cut off at the month before each target. CPI is rarely revised, but this is not a true real-time test.
• The "direction" score for the naive model is low by construction, because it always predicts no change.
• The Ofgem and fuel signals are mapped mechanically, so a big price-cap move with an unusual pass-through could produce a large miss.
• Forecasts are rounded to one decimal place, like the ONS headline.

See every backtest month on the scoreboard