Open data

The data behind the prices

Tilth publishes a daily price estimate for nine UK fertiliser grades. It is free to read, free to quote, and free to build on. This page names every input that moves the number, how the model is built, and what you may do with it.

9
Grades tracked: Daily
Estimate cadence: £/tonne
Delivered to farm: CC BY 4.0
Open licence

The catalogue

The nine grades

One modelled estimate per grade, per day, in pounds per tonne. Each card names the reference index, the daily signal, and the currency conversion behind that grade's number, and links to its live price page.

Every figure is an indicative, forecasted estimate. None of them is a transactable quote, and nothing on this site can be bought at the published price.

How the estimate is made

Each grade's daily estimate combines three public inputs onto a UK delivered-to-farm reference level.

Reference index

A nutrient or commodity index sets the primary monthly move. Nitrogen, phosphate and potash grades use the European Commission agri-food indices; muriate of potash uses the World Bank commodity series. The model reads their month-over-month returns, so the indices' native euro pricing cancels out.

Daily driver

The daily move comes from a major listed producer, or a commodity input, for each grade: Yara for most nitrogen grades, Mosaic for phosphate, Nutrien for potash, ICL for polysulphate, and Dutch TTF natural gas futures for nitrate sulphur. That return is the signal that moves the estimate between monthly reference updates.

Currency conversion

Producer equities quoted in Norwegian krone or US dollars are converted to pounds before they enter the model; TTF gas is quoted in euros. The European Commission indices need no conversion, because only their returns are used.

A grade earns the full modelled label only where its drivers beat a flat-carry baseline under walk-forward validation. Where the evidence is weaker we do not hold the line flat: the grade still tracks its drivers day to day, on a smaller response that is flagged as lower confidence, so you can weigh it for yourself. The model is never frozen either: candidate drivers and fitting methods are trialled against the live model under the same rules, and a change ships only when it survives statistical significance testing. The section below sets out exactly how that bar works.

The band around each estimate is computed at a fixed one-week horizon from each grade's own recent volatility. It reflects how much that grade genuinely moves, and it is the model's own statement of how far it trusts the figure.

How we check the model before publishing

Every figure on this site is a modelled forecast, published for consultation and to be taken as it is. It is not advice and it claims no authority beyond its own working, which is documented here so you can judge it for yourself.

01 · Model form
The model works in month-over-month returns, not price levels. A regression with no intercept maps driver returns onto movements of the UK delivered-to-farm reference level, so when the drivers are flat the model predicts no change: the reference level owns the level, the drivers own the movement. Each grade's response is fitted with the estimator that best survives validation for that grade: where a grade's prices move in occasional sharp jumps, the fit minimises the average absolute miss, the same measure the accuracy gate scores, so a single spike month cannot drag the coefficient.
02 · Walk-forward testing
Every grade is tested strictly forward in time: the model is fitted only on months that precede the month it must predict, one month at a time, with at least three years of history before the first test. No future data ever leaks into a fit. A grade earns the full modelled label only if its out-of-sample error beats holding the reference level flat by at least ten percent. If that edge decays the grade does not drop to a flat line: it steps down to a smaller, lower-confidence daily response on the same drivers, with a small hysteresis so the switch does not flap month to month.
03 · Measuring our own error
One-step-ahead forecast errors are serially correlated, so simple error statistics overstate how much evidence sits behind them. We correct for that rather than quote the flattering number: error ranges are estimated by resampling runs of consecutive months instead of single months, model-versus-baseline comparisons account for correlated errors, and we count only genuinely independent observations when judging how much history actually backs a grade. Each grade is also re-tested on rolling training windows, to check its behaviour does not depend on keeping old history around. For readers who want to reproduce the working: a stationary block bootstrap, a Diebold-Mariano test with a Newey-West variance, and an effective sample size on the loss series.
04 · Publish guards
A daily move larger than three standard deviations is held rather than published. Derived figures for a grade, such as price per kilogram of nutrient, publish only while the latest UK market print falls inside the model's own uncertainty band. The API labels every figure with its method and stage, so you always know how the number you are reading was produced.

Sources and attribution

The reference indices behind the model are public, free to use, and named here so you can check our working.

European Commission, agri-food data portal

EU nutrient price indices for nitrogen, phosphorus and potash. Reused under Commission Decision 2011/833/EU, which applies a Creative Commons Attribution 4.0 licence.

View the agri-food data portal

The World Bank, commodity price data

Global commodity price series. Reused under a Creative Commons Attribution 4.0 licence.

View the World Bank commodity price data

What we changed

We do not republish either source. We calibrate month-over-month returns from them and apply those returns to a UK reference level to produce a daily modelled estimate. The figures on this site are ours, derived, and are not the source data.

No endorsement

Neither the European Commission nor the World Bank endorses Tilth, is associated with Tilth, or bears any responsibility for the estimates published here. Any error in our figures is ours.

Take the data

Use the index in your own software or let an AI agent work with it through MCP. Both routes are public and documented in one place.

Public data · API and MCP · no key

For applications

Use the public API

Retrieve current estimates and price history as JSON or CSV, with uncertainty bands, methodology and attribution included.

Open API documentation →

For AI agents

Connect through MCP

Give an agent typed tools to discover products, retrieve prices and history, compare a supplier quote and create a price alert with explicit consent.

Open MCP documentation →

Embed the live price

Put today's index figure for any grade on your own site with one iframe. It stays current by itself and credits the index with a built-in link, which satisfies the attribution requirement automatically.

Paste this HTML

<iframe src="/embed/price-widget?product=AN_UK" width="320" height="150" loading="lazy" title="Tilth fertiliser price widget"></iframe>

Licence and attribution

You may

Read, quote and republish the figures, in print or online.
Chart them, embed them, and build tools on top of them.
Use them in research, commercially or otherwise.

You may not

Present the figures as transactable quotes, or as anything other than modelled estimates.
Represent Tilth as endorsing your product, or imply a partnership that does not exist.

Attribution

Credit Tilth and link back to this page. That is the whole price.

Source: Tilth (tilth.uk)

Ready-to-paste HTML credit

Publishing our figures on a website? Paste this once and the attribution requirement is met:

<a href="https://www.tilth.uk/data/">Source: Tilth, the independent UK fertiliser price index</a>

Who owns this

Tilth is owned by nobody who sells fertiliser. No merchant, no distributor and no manufacturer holds a stake in it, funds the index, or sees who reported a price. That is the entire point of it.

Questions, corrections, or a story

If you think a number is wrong, tell us. If you are writing about fertiliser prices and want the methodology explained properly, we will explain it.

Get in touch