From day-ahead to mid and long-term horizons with econometric electricity price forecasting models

TL;DR

This paper develops interpretable econometric models for electricity price forecasting across multiple horizons, addressing key challenges with novel approaches to improve robustness and interpretability in Europe's energy market.

Contribution

It introduces new methods for incorporating fundamental data, seasonal expectations, and unit root behavior into long-term electricity price forecasting models.

Findings

Models effectively forecast prices from one day to one year ahead.

Incorporating fundamental and seasonal information improves model robustness.

Guidelines provided for selecting key variables across different horizons.

Abstract

The recent energy crisis starting in 2021 led to record-high gas, coal, carbon and power prices, with electricity reaching up to 40 times the pre-crisis average. This had dramatic consequences for operational and risk management prompting the need for robust econometric models for mid to long-term electricity price forecasting. After a comprehensive literature analysis, we identify key challenges and address them with novel approaches: 1) Fundamental information is incorporated by constraining coefficients with bounds derived from fundamental models offering interpretability; 2) Short-term regressors such as load and renewables can be used in long-term forecasts by incorporating their seasonal expectations to stabilize the model; 3) Unit root behavior of power prices, induced by fuel prices, can be managed by estimating same-day relationships and projecting them forward. We develop interpretable models for a range of forecasting horizons from one day to one year ahead, providing guidelines on robust modeling frameworks and key explanatory variables for each horizon. Our study, focused on Europe's largest energy market, Germany, analyzes hourly electricity prices using regularized regression methods and generalized additive models.