Skip to content

What drives the model

Electricity prices are set by the most expensive plant needed to meet demand in each quarter-hour. Anything that changes how much demand there is, how much cheap renewable output there is, or what the marginal plant costs to run will move the price. EPF’s models learn from information in seven groups.

Where prices have been in the last hours, days and weeks is the strongest single signal. Prices repeat daily and weekly shapes, and a high-price regime tends to persist for a while. For the strategic product, tomorrow’s published prices are the anchor for the rest of the week.

Hour of day, day of week, season, and each country’s own public holidays. A Sunday afternoon in May and a Tuesday evening in January are very different markets, and holidays differ between Spain, Portugal, France and Germany.

How much electricity is expected to be used, and how much of it is left for conventional plants once wind and solar are counted (residual demand). Residual demand is one of the most informative inputs, because it decides which plants have to run.

The day-ahead wind and solar forecasts that system operators publish. On a windy, sunny day the cheapest plants cover most of demand and prices fall, sometimes below zero. Where these forecasts are published they materially improve the day-ahead forecast.

Temperature drives heating and cooling demand; wind speed and solar irradiance drive renewable output. Weather forecasts extend this information beyond tomorrow, which matters most for the strategic product.

Gas is usually the marginal fuel in these markets, so gas and carbon prices set the cost of the plant that sets the price. They explain much of the level of prices over weeks and months.

European markets are coupled through interconnectors, so prices in one country carry information about another. Whether this helps is tested per market: for some markets it improves the forecast, for others it adds noise, and each market’s model uses it only if it earns its place.

Some price moves have no warning in public data: a sudden geopolitical shock to fuel markets, an unplanned outage of a large plant, or a regulatory change. The models adapt once such a change shows up in prices and fundamentals, but they cannot anticipate it. This is one reason EPF is always judged against naive baselines: a forecast earns trust by beating simple rules over long periods, including turbulent ones.