pricecast

Synthetic hourly electricity price time-series for prospective energy system studies

import pricecast
from pricecast.entsoe import fetch_year

prices, drivers = fetch_year("BE", 2025)          # day-ahead price, load, solar, wind
model = pricecast.load("BE")                      # or pricecast.fit(prices, drivers)

future = pricecast.scale_drivers(drivers, solar=1.39, wind=1.38, load=1.30)
year = model.predict(future)                      # the shape of a 2030 year
year = pricecast.rescale_to_level(year, level=88) # its level, from a fuel-and-carbon scenario

Returns an hourly series of day-ahead prices in EUR/MWh, ready to feed an LP or MILP design model: a heat pump with storage, a battery, a thermal store, an electrolyser.

The average day of 2025, each series over its own annual mean. The model (fitted on 2023-24, driven by 2025's load, solar and wind) keeps the midday trough; reusing a past year cannot know about the panels installed since.

Why

A study that sizes equipment for 2030 needs a 2030 price year. The usual stand-in is a past year reused as it was, which carries the past's solar fleet, the past's price level, and the past's weather. pricecast instead generates the year from what a scenario does provide - installed solar and wind, demand and fuel prices - so the features a design reacts to follow the scenario:

• how deep the price falls at midday, and how far it climbs in the evening;
• what a solar kilowatt-hour is worth against the average one;
• what the electricity for a kilowatt-hour of heat costs.

The model is one least-squares regression of hourly price on the clock and on three drivers - load, solar and wind generation - fitted on recent years. A full year is generated in one shot, with no autoregression on the price.

It is not a forecaster. Nobody has next decade's hourly solar output, so it does not compete with forecasting models. It produces a plausible shape and leaves the level to an outside scenario, because the level is set by fuel and carbon prices and cannot be extrapolated from price history.

What it has been tested on

Five European markets (BE, DE, FR, ES, PL), fitted on 2023-24 and judged on 2025 by what a design optimiser decides, not by hourly error:

Full results, and where it falls short, are in the accompanying paper.

Building a future year

The recipe, in short:

1. Fit on the two most recent complete years, or start from pricecast.load.
2. Choose one weather year and take every weather-dependent input from it: the model's drivers, and your asset's PV output, heat demand and ambient temperature. Prices respond to the same weather; mixing years makes heat look 2-10% cheaper than it is.
3. Scale the drivers by target-over-weather-year capacity and demand (scale_drivers), from ERAA, TYNDP or national plans.
4. Run the model (PriceModel.predict).
5. Set the level as the market's recent ratio times a gas plant's running cost in the target year (level_ratio, gas_plant_cost, rescale_to_level). Err low: an overestimated level costs a design far more than an underestimated one.
6. Check the reach (share_beyond_training): where several percent of hours lie beyond the net load the model was fitted on, floor the year at the training years' 1% price quantile (training_floor).

2030 from ERAA 2025 capacities and demand on 2025 weather. Germany and Poland (13% and 7% of hours beyond the training net load) need the floor.

What it does not do

• Negative hours. It smooths away part of the midday collapse in solar-heavy markets: 1.8% negative hours against 5.7% in reality in 2025. The residual noise (noise=True) restores the share, but in hours uncorrelated with the real ones - use it for revenue estimates, never for sizing.
• Extrapolation. It is linear in its drivers; far beyond them it must be floored, and it has no storage driver, so future spreads are upper bounds.
• Drifting coefficients. The price response to solar steepens as the fleet grows. Correcting the coefficients for that was tested and does not pay; refit on recent years instead.
• Other markets. Day-ahead only - not intraday, balancing or forward.

Install

pip install git+https://github.com/Matt-haug/pricecast
pip install "pricecast[entsoe] @ git+https://github.com/Matt-haug/pricecast"   # with ENTSO-E download

Requires Python 3.9+, numpy and pandas. Downloading data needs entsoe-py and a free ENTSO-E Transparency Platform token, passed as token= or set as the ENTSOE_TOKEN environment variable.

Data

The shipped models are fitted on ENTSO-E Transparency Platform data for 2024-2025: day-ahead prices, actual total load and actual generation per production type. What ships is the fitted coefficients only; the raw series are not redistributed and are downloaded with your own token.

Contributing

Issues and pull requests are welcome - see CONTRIBUTING.md. Reports that a generated year disagrees with a market you know well are especially welcome. This project follows the Contributor Covenant. Generative AI was used in building it; see AI_USAGE.md.

Citing

Use the "Cite this repository" button, or CITATION.cff. Each release is archived on Zenodo with a DOI.

Licence

MIT - see LICENSE. Changes are recorded in CHANGELOG.md.