Methodology and sources

Data updates

Inflation data

All charts use ONS time series from the Consumer price inflation dataset (MM23), downloaded from the ONS website's time series API. We publish the figures unchanged. Headline series:

• D7G7: CPI ANNUAL RATE 00: ALL ITEMS 2015=100
• L55O: CPIH ANNUAL RATE 00: ALL ITEMS 2015=100
• DKO8: CPI 12mth: Excluding Energy, food, alcoholic beverages & tobacco 2015=100
• D7OE: CPI MONTHLY RATE 00: ALL ITEMS 2015=100
• D7BT: CPI INDEX 00: ALL ITEMS 2015=100

Category pages use the CPI annual rate and index (2015=100) for each of the 12 COICOP divisions. Machine-readable copies of every charted series are at /data/series/<ID>.json.

The CPI nowcast

The forecast predicts the next month's CPI annual rate in three steps.

1. Seasonal baseline (v1). For the target calendar month, take the average month-on-month change in the CPI index (D7BT) over past years, excluding 2020–2023 because of pandemic and energy-shock distortions. Apply it to the latest index to get a predicted index and annual rate.
2. Known-price adjustments (v2). Where we already know a price moved, replace the seasonal assumption for that part of the basket:
   • Electricity and gas. In January, April, July and October, the change in Ofgem's price cap, worked out as a typical bill at fixed consumption (2,700 kWh electricity, 11,500 kWh gas) including standing charges. In other months, no change.
   • Motor fuel. The average of GOV.UK weekly pump prices dated between the 8th and the 21st of each month (around when the ONS collects prices), with petrol and diesel weighted by their CPI weights.
   • Food. The recent 12-month trend in food prices compared with its long-run average.
   Each adjustment is (signal − seasonal norm for that item) × the item's effective CPI weight. The weight is the ONS annual weight, price-updated from the January link month.
3. Rounding. The annual rate is rounded to one decimal place, like the ONS headline.

The coefficients are not fitted to the backtest. The only choices are the fixed consumption levels, the collection-window dates and the excluded years, each set in advance for a stated reason.

The current draft (October 2026) uses a seasonal baseline of 0.14% per month over 34 years. See the forecast page.

Scoring

• Error: predicted minus actual annual rate, in percentage points. MAE: mean absolute error. Bias: mean error.
• Direction: whether the forecast called the move up, down or unchanged from the previous print correctly.
• Skill: 1 − MAE ÷ MAE of the naive "no change" forecast. Above zero means the model beats the naive forecast.
• Lock: a live prediction is fixed before release. Drafts are not scored as live predictions.
• Backtest: 24 months (October 2024 to September 2026), run on the current data vintage cut at the month before each target. Labelled as a backtest everywhere.

Licence and attribution

Contains public sector information licensed under the Open Government Licence v3.0: Office for National Statistics (consumer price inflation), Ofgem (energy price cap) and the Department for Energy Security and Net Zero via GOV.UK (weekly road fuel prices).