GREAT BRITAIN / DAY-AHEAD RESEARCH

Tomorrow’s price, with its evidence

This is a live forecast and we are learning in public. Every hour of tomorrow’s Great Britain day-ahead reference price, published before the outcome is known, with its uncertainty and its sources attached. Below it is the record: every day this model has an archive for, what it said, and what the price turned out to be. Most of those days are reconstructions rather than forecasts published at the time, each one marked as such, and the record is three weeks old. It includes the days we get wrong, because those are the days worth showing.

For information only. These forecasts are experimental and unproven at extreme prices. Do not rely on them for trading, dispatch or investment decisions. Read the limitations

HISTORICAL REPLAY / NOT A LIVE FORECAST

Wednesday 15 July 2026

Low Carbon Contracts Company reference price, pounds per MWh, London delivery time. See how it has done so far, day by day ↓

Research reference

Median forecastShaded rangePublished actualBenchmark

Inspect delivery hour18:00 BST

Each step is one delivery hour.

SYSTEM CONTEXT

Forecast inputs
for this delivery day

Metered wind forecast—

National demand forecast—

Wind and demand are separate inputs. These charts do not establish causation or total GB renewable generation.

Read the hourly values Accessible data table

Historical research, 15 July 2026. Prices in £/MWh.
London hour	UTC start	Median	Lower	Upper	Actual	Benchmark

Reconstructed input cutoffNot availableTargetLCCC IMRP / hourlyMethodHistorical research / windLive scorecardNot eligible

THE RECORD, WHILE IT IS STILL SHORT

What we said, and what the price turned out to be

The model follows the level and flattens the swings. Daily mean forecast against the daily mean settled price. The dashed red line stops where settlement stops, so the last few days carry a forecast and no outcome yet.

The average delivery day across every scored day, London time. The shading is the gap between what we said and what happened, and it changes colour where the miss changes direction.

Read the hour by hour scorecard Every hour, permanently scored

Mean absolute error, signed bias and band coverage by London hour. Bias is forecast minus settled price, so a negative figure means the forecast was too low.
London hour	Days	Forecast	Settled	Error	Bias	Median error	Band width	Coverage	Days too low

HOW FAR IT UNDER REACTS

WHERE THE ERROR COMES FROM

WHY IT CANNOT SEE THE TURN

WHAT WOULD FIX IT

Every day, on one vertical scale, including the bad ones. The dashed red line is the settled price, the solid blue line is what this model said, and the shading is the band exactly as published that morning. The figure under each date is that day’s mean absolute error. Days marked tomorrow or still ahead have not been delivered yet and have no outcome to show. Awaiting price means the day is over and the settlement run has not published it, which takes about three days.

HOW IT CHANGES

Every change to the model is dated,
and scored against the days before it

A change to how the forecast is made gets a new identifier, a line saying what changed and why, and, once enough days have settled to tell, a line saying what it was worth. Changes that turned out to be worth almost nothing stay on the list with that written next to them.

Reconstructed days are rebuilt under whichever policy is current, so the record before a change shows what today’s model would have said rather than what was said at the time.A published forecast is never rewritten.

READ THE SMALL PRINT, IN NORMAL TYPE

What this does.
What it doesn’t.

A public research space for the next delivery day. It is not a two-year price outlook or a ten-year financing case.

What price are we forecasting?

The hourly LCCC Intermittent Market Reference Price (IMRP), a GB day-ahead benchmark using EPEX and N2EX indices. It is not an executable quote on one exchange. Prices are in GBP/MWh.

What does the middle line mean?

A median price estimate, not expected trading revenue. The average of hourly medians is neither the expected daily mean nor the median of daily baseload.

How should I read the shaded range?

As an experimental marginal range for each hour, with a nominal 80% target. It is empirically adjusted from prior errors. Coverage is not uniform across conditions and price spikes remain difficult.

What makes a forecast eligible for the live record?

Sources must arrive by 07:00 London on the preceding day, and the forecast must be completed, archived and published by 09:15, which is ahead of the day-ahead auction that sets the price. The cutoff used to be 09:10. It was moved because Elexon publishes its wind forecast at fixed times and never between 05:30 and 08:30 UTC, so on all 21 days measured the two cutoffs admitted exactly the same wind forecast, and the earlier one turns a five minute publication window into two and a quarter hours. The earlier cutoff is not free. Elexon’s national demand forecast covers only the first five hours of the delivery day until 07:45 UTC and the whole day after it, on 11 of 11 days measured, so a 07:00 London cutoff in summer misses it. This model does not read demand, so it loses nothing today, and a model that did would need the cutoff moved back. London summer time is UTC+1. Late previews, historical reconstructions and missing days are kept separate and are never counted in the on-time record.

Why does it miss most on the days that matter?

Because the price it learns from is three days old. The settled reference price is published about 58 hours after delivery, so the freshest price this model can read when it forecasts tomorrow is the day before yesterday’s, and its price features are lags and averages of prices that old. Early September ran daily mean prices of £153.6, £73.7, £96.4, £145.5, £124.3 and £159.2/MWh inside a single week. The fix is a price input that is not three days old, and the day-ahead auction result is public the afternoon before delivery. That is the next change, and it will be recorded above with what it was worth.

Which limitations remain?

The historical price-availability rule is an explicit compatibility assumption. The research seed is time-limited. Controlled learning, durable storage, scheduling and publishing still require integration and acceptance testing. No guarantee of improvement, trading suitability or bankability is made.

Recorded limitations

Every figure above is regenerated by the scripts in research/audit_scripts/ and written to research/evidence/.

Source: Low Carbon Contracts Company, IMRP actuals. Contains public sector information licensed under the Open Government Licence v3.0. Forecast inputs: Elexon Insights WINDFOR and NDF. The research archive supplies the reconstructed model results shown here.