Comparison of extended mean-reversion and time series models for electricity spot price simulation considering negative prices

• Published in: Energy economics 2012-07, Vol.34 (4), p.1012-1032
• Main Authors: Keles, Dogan, Genoese, Massimo, Möst, Dominik, Fichtner, Wolf
• Format: Article
• Language: English
• Subjects: Applied sciences; Commodity markets; Comparative analysis; Computer simulation; Econometric models; Economic data; Economic models; Economics; Electric energy; Electric power; Electricity; Electricity prices; Electricity spot prices; Energy; Energy economics; Energy policy; Energy prices; Errors; Exact sciences and technology; GARCH models; General, economic and professional studies; Market prices; Markets; Mean-reversion process; Methodology. Modelling; Negative prices; Power marketers; Price models; Prices; Reversion; Roots; Simulation; Stochastic models; Studies; Time series; Time-series models
• ISSN: 0140-9883; 1873-6181
• DOI: 10.1016/j.eneco.2011.08.012

This paper evaluates different financial price and time series models, such as mean reversion, autoregressive moving average (ARMA), integrated ARMA (ARIMA) and general autoregressive conditional heteroscedasticity (GARCH) process, usually applied for electricity price simulations. However, as these models are developed to describe the stochastic behaviour of electricity prices, they are extended by a separate data treatment for the deterministic components (trend, daily, weekly and annual cycles) of electricity spot prices. Furthermore price jumps are considered and implemented within a regime-switching model. Since 2008 market design allows for negative prices at the European Energy Exchange, which also occurred for several hours in the last years. Up to now, only a few financial and time series approaches exist, which are able to capture negative prices. This paper presents a new approach incorporating negative prices. The evaluation of the different approaches presented points out that the mean reversion and the ARMA models deliver the lowest mean root square error between simulated and historical electricity spot prices gained from the European Energy Exchange. These models posses also lower mean average errors than GARCH models. Hence, they are more suitable to simulate well-fitting price paths. Furthermore it is shown that the daily structure of historical price curves is better captured applying ARMA or ARIMA processes instead of mean-reversion or GARCH models. Another important outcome of the paper is that the regime-switching approach and the consideration of negative prices via the new proposed approach lead to a significant improvement of the electricity price simulation. ► Considering negative prices improves the results of time-series and financial models for electricity prices. ► Regime-switching approach captures the jumps and base prices quite well. ► Removing and separate modelling of deterministic annual, weekly and daily cycles are necessary to simulate them adequately. ► ARMA and mean-reversion models seem to be more appropriate for the simulation of electricity prices.