Our long-term forecasts extend well beyond the short-term horizon (up to 46 days), covering months and years ahead. We use historical weather years as the basis for every forecast, drawing on historical EC OP forecasts to build our historical weather view.

Data availability varies by model: weather history goes back to 1999 for solar and load, 2006 for hydro run of river, and 2011 for wind - and, as a consequence, price.

Long-term scenarios run on weather years from 2011 to 2025, the period for which wind data is available, generating a rich distribution of plausible futures that reflects the natural variability of weather and is fully accessible to you for scenario analysis.

As with our short-term approach, our long term forecasting combines machine learning and time series analysis, with a strong foundation in fundamental market drivers such as production, consumption, and fuel prices. To ensure transparency and avoid the "black box effect," our models operate in two sequential steps:

We begin by generating fundamental forecasts - load, wind generation, and solar generation.

For long-term forecasts, weather inputs come from historical reanalysis data rather than live weather model runs. Each weather year from 2011 to 2025 is treated as an independent scenario and converted into forecasts of demand, wind, and solar production for the next five years. For each year, we apply:

This produces a distinct hourly fundamental forecast for every weather year - 15 scenarios in total - each carrying its own unique intra-day and seasonal shape.

Notes : Forecast quality depends critically on the data used to train the underlying models, so we continuously refine our historical actuals: training relies on recent actuals with an expanding historical window, prioritizing official grid operator data over aggregated sources where available - for example, Ned-NL over ENTSO-E for Dutch wind and solar generation.

The fundamental outputs (in MW) are then passed into our machine learning-based price model, following the same two-step architecture as the short-term system. In addition to renewable generation and load, the price model incorporates:

Because each weather year produces a distinct set of fundamental inputs, the price model generates a correspondingly distinct price trajectory for each year - capturing realistic intra-day price dynamics (spikes, negative prices, etc.) that a smoothed average profile would mask.

Long-term price forecasts are generated at hourly granularity, and the model is retrained weekly.

Once all 15 weather-year forecasts are generated, we derive a full statistical distribution across them (mean, median, Q25, Q75, min, max), giving you a realistic sense of the range of outcomes that actual weather conditions could produce - anchored in 15 years of observed meteorological data rather than theoretical assumptions.

Note: Long-term hydro run of river and poundage forecasts are not yet fed into the LT price models - only load, solar, and wind currently are.

The real power of the long-term framework lies in scenarization. Because every weather year is forecasted independently, you are not limited to the statistical summaries: you can interrogate any individual year directly, or apply custom scenarios on top of it.

Via the Kpler terminal or API, you can:

This makes the long-term forecasts a versatile tool not just for "what is the expected outcome" but for answering "what is the range, and what drives it".

Long-term forecasts are available directly in the Kpler terminal. You can use the Scenario Builder to create your custom scenarios : choose between individual weather years or statistical views, and apply custom scenarios to model specific assumptions about load growth, capacity, or fuel prices.

Long-term forecasts (load, generation, day ahead prices) are queryable across all combinations of:

Note : Fundamentals are run once per month, typically on the 1st. The run_date parameter defaults to the closest available run on or before the requested date. Note that ad hoc reruns can occasionally occur - for example, if an issue is identified after a run, we may rerun and produce a run_date that does not fall on the 1st of the month.

For detailed endpoint documentation, refer to the our developer portal

Long-term forecasts are also available on FTP under :

File naming conventions:

Note : Files are published per run date under each zone/year/month folder (e.g. .../AT/2026/06) and cover the full forecast horizon in a single file. Ignore any legacy files found at the year-level root (e.g. .../AT/2026, dated until 2030) - these predate our 2025-08-18 methodology change, when forecasts were split per horizon instead of by run date.

For questions or feedback, please reach out to your Kpler account manager or contact our support team.