Multi-Day-Ahead Electricity Price Forecasting: A Comparison of Fundamental, Econometric and Hybrid Models

Abstract

Forecasting hourly electricity prices and their characteristic properties is a core challenge for energy generation companies and trading houses. The short-term marketing and purchase of electricity is usually managed with standardized products traded on different markets and with specific temporal resolution and maturity. The size and scope of the electricity price forecasting literature has grown significantly in recent years, with the majority of studies focused on short-term (intraday and day-ahead) or long-term (investment decisions) periods. However, the literature for forecasting the period beyond the day-ahead horizon, which is relevant for trading the aforementioned products or for managing assets over several days, is rather scarce. Our paper fills this gap by developing individual forecasting models covering horizons from the day ahead up to a week ahead. We introduce hybrids of a parsimonious fundamental model and various popular econometric models. In a case study for the German day-ahead market in 2016 we test and compare the different model settings by carefully considering realistic available data and limiting the calculation time to fit typical trading time constraints. We find that the best models across the individual horizons and across all horizons jointly are hybrid model approaches. They combine the strengths of autoregressive models in terms of capturing daily - even non-linear-structures with the immediate reactions of fundamental models to short-term events or fundamental changes in the market.