The OBR's CPI inflation modelling

Consumer price inflation (CPI) plays a central role in the Office for Budget Responsibility’s (OBR) forecasts for the public finances. Inflation affects the cash value of government spending and revenue; directly through its impact on welfare uprating, tax thresholds, and debt interest costs, and indirectly through the nominal size of the economy, which drives tax revenues. Therefore, relatively small changes in the CPI forecast can have material implications for borrowing and debt over the medium term.

To inform our CPI forecast, we use a range of models and tools designed for different horizons and components, which together inform the Budget Responsibility Committee’s forecast. We combine granular near-term projections with medium-term component models linked to broader drivers of inflation and use cross-checks to ensure internal consistency. This article publishes details of the modelling toolkit in line with the 2025 external review recommendation to share more transparently our analytical infrastructure.¹ More detail on the model specifications can be found in the annexes to the PDF version of this article.

Our CPI forecast is organised around two connected horizons. In the short term (generally, the next six months), inflation is strongly affected by mechanical influences, many of which are already known: base effects from earlier price movements, seasonal patterns, the timing of annual price resets, and announced policy or administrative changes. Past shocks can also continue to influence inflation over this horizon.

Further out (beyond six months), these mechanical effects are less predictable, so inflation is assumed to be increasingly shaped by underlying drivers – both the effects of past shocks, which are expected to be somewhat persistent, and our assumptions about the future evolution of the economy. These drivers include domestic cost pressures (notably labour costs), import prices and exchange rate movements, energy and other commodity prices, and the evolution of the balance of supply and demand. In some parts of the framework, inflation expectations also enter directly or are used as a cross-check on the forecast. Medium‑term inflation also needs to be consistent with the UK’s monetary policy framework. A high-level summary of our approach can be seen in Figure 1.

Historical inflation dynamics differ markedly across CPI components (Chart 1). Energy inflation is highly volatile and closely linked to global commodity markets and regulatory settings, while services inflation tends to be more persistent and domestically driven. Core goods inflation tends to be more volatile than services due to its sensitivity to import prices, exchange rate movements and supply chain pressures. Other components, such as alcohol and tobacco, are shaped heavily by policy on excise duties and indexation. Reflecting these differences, we forecast CPI largely at component level, using tailored approaches where price-setting behaviour or policy exposure differs materially.²

We then combine component forecasts using CPI weights to produce the headline path, using top-down cross-checks – including the Phillips curve from our small macroeconomic model and a mark-up-based model,³ where headline CPI inflation is explained by import prices, earnings growth, inflation expectations, and the output gap. In this framework, inflation expectations play a direct role, while the output gap serves as an indirect proxy for firms’ pricing power and cyclical movements in profit margins, rather than profit margins being modelled separately.⁴ In the final rounds of the forecast, we incorporate the effects of fiscal policy – both for the impact of the fiscal stance on the balance of supply and demand in the economy, and for individual measures that are expected to have material direct effects on CPI and RPI inflation. These judgements are informed by analysis provided by HM Treasury and other government departments. Throughout, we apply judgement to maintain consistency across CPI components, horizons, and the wider macroeconomic forecast.

In the short term, our approach reflects the importance of seasonality, known future price changes, the pass-through of previous shocks, administrative drivers and other announced policy changes. The core of the short‑term inflation forecast (STIF) is a bottom‑up, item‑level methodology that projects month‑on‑month inflation rates for individual CPI items over the next six months. The six-month horizon reflects the point at which we judge monthly projection errors compound sufficiently to erode the accuracy advantage of the bottom-up item-level approach relative to driver-based models.

Baseline projection – for each item and calendar month, the baseline projection is informed by historical averages of that item’s monthly inflation rate in the same calendar month (capturing recurring seasonal and timing effects). This framework is well suited to short horizons where inflation often reflects timing, base effects and recurring price changes.⁵

The baseline seasonal approach is complemented with explicit modelling of components where price changes are driven by observable factors or administrative processes. Fuel prices, for example, are forecast using linear regressions linking petrol and diesel pump prices to oil prices, combined with assumptions about fuel duties and VAT. Electricity and gas prices are projected using forecasts for the Ofgem energy price cap, which maps wholesale energy prices (both historical and based on the futures curve), policy parameters and non‑wholesale costs into regulated retail tariffs. Other components heavily subject to excise duties or administrative decisions, such as alcohol, tobacco, transport, social rents and education, are forecast using announced or assumed policy rather than pure statistical extrapolation. These item‑month projections are then combined using CPI weights to generate a monthly path for major sub‑aggregates and headline CPI.

STIF cross-checks – alongside the bottom‑up projection, we maintain a suite of monthly statistical models for headline CPI and key components, used primarily as cross‑checks. These include univariate autoregressive models with seasonal dummies and multivariate autoregressive distributed lag specifications that add a small set of economically meaningful indicators – such as activity or slack measures, earnings growth, import costs, energy prices, and financial variables – chosen to match the component’s pricing behaviour.⁶ Outputs from these models are used to inform judgements about risks around the baseline STIF path, rather than to generate the central forecast mechanically.

Beyond the near term, CPI inflation can be modelled either top-down (using a single equation for headline inflation) or bottom-up, by forecasting the main CPI components separately and aggregating them. We focus on a bottom-up approach for four main reasons:

1. CPI components are driven by different forces, making stable top-down relationships harder to fit;
2. it helps keep key assumptions consistent (for example, oil, electricity and gas price conditioning assumptions feeding through coherently into fuels and utility bills);
3. it provides a transparent framework for applying policy and regulatory settings; and
4. the component detail is necessary for the fiscal forecast, for example in mapping nominal consumer spending into tax bases that face different VAT treatments.

The medium-term inflation forecast (MTIF) translates the wider macroeconomic forecast into component paths for CPI inflation once base effects have faded and fewer price changes have been pre-announced. By the forecast horizon, we assume that monetary policy has acted to close the output gap and return CPI inflation to the Bank of England’s 2 per cent target. We group MTIF models by dominant drivers; while specifications differ, most combine inflation persistence with measures of external cost pressures and demand conditions.⁷ The role of inflation expectations only appears directly in our equation for other non-tradables inflation, while profit margins are not generally modelled as a separate driver, with cyclical variation in pricing power instead proxied by measures of slack such as the output gap.

Energy components – these include gas, electricity, petrol, diesel and other fuels, and are largely driven by wholesale prices and regulatory or policy frameworks. Projections are grounded in market conditioning assumptions (such as oil and gas futures curves) combined with institutional features such as the timing and structure of Ofgem price cap updates and announced policy settings (for example, fuel duty indexation or freezes).

Globally-exposed components – the prices of food and non-alcoholic beverages, and other tradable goods, are more influenced by foreign competition, global cost pressures and exchange‑rate movements – though domestic cost pressures can also contribute. Medium‑term models for these components therefore emphasise import prices and external inflation, typically allowing for gradual pass‑through over several quarters, reflecting delays between movements in world prices, wholesale costs, and final retail prices – allowing for the effects of both known past changes and assumptions about future paths. Persistence terms capture the tendency for inflation rates to adjust only gradually to new cost pressures, reflecting price‑setting frictions and adjustment costs along supply chains, while lag structures reflect the time between movements in world prices, wholesale costs and final retail prices.

Predominantly domestically driven components – the prices of other non‑tradables and private rents are more closely linked to domestic economic conditions. The model for other non-tradables places greater weight on labour-cost growth, inflation persistence, inflation expectations and indicators of spare capacity, though import prices still play a role. In the ‘other non-tradables’ equation, the output gap also acts as an indirect proxy for firms’ pricing power, capturing cyclical variation in mark-ups rather than modelling profit margins explicitly. Models for private rents depend primarily on past inflation and earnings growth but include additional housing market variables such as the housing stock per capita and average mortgage rates.

Policy- and indexation-driven components – these components are driven primarily by policy settings or formal indexation conventions rather than relying on estimated behavioural equations. For these items – including social rents, education, alcohol and tobacco – we translate relevant uprating rules and administrative timetables into projected paths for the corresponding CPI indices, applying announced parameters at the points in the year when changes typically take effect.

Inflation forecasting is inherently uncertain, particularly when the economy is affected by large external shocks, changing policy settings and evolving economic relationships. No single model can capture all the forces that shape inflation across horizons or reliably characterise how those forces interact when new disturbances occur.

For this reason, our approach is pluralistic. The models provide guidance rather than a mechanical forecast, and we complement them with ad hoc analysis where warranted – for example, when energy shocks, supply chain disruptions or other unusual developments are likely to affect inflation in ways not entirely captured by historical relationships. Across time horizons, statistical and economic models inform, rather than replace, forecast judgement.

Acknowledgements

The author would like to thank Scott Bowman, Callum Bridge-Adams, Rosanna Colthorpe, Charlotte Bunney, and various other OBR staff past and present for their valuable contributions to this article.

Downloads

The OBR's CPI inflation modelling (article PDF)