CMStatistics 2019: Start Registration
View Submission - CFE
A0638
Title: A regime-switching stochastic volatility model for forecasting electricity prices Authors:  Peter Exterkate - University of Sydney (Australia) [presenting]
Oskar Knapik - Kozminski University (Poland)
Abstract: Three crucial challenges outstanding in the area of electricity price forecasting are addressed. Specifically, we show the importance of considering fundamental price drivers in modelling, develop new techniques for probabilistic (i.e. interval or density) forecasting of electricity prices, and introduce a universal Bayesian technique for model comparison. We propose a new regime-switching stochastic volatility model with three regimes, which may be interpreted as negative jump or ``drop'', normal price or ``base'', and positive jump or ``spike'', respectively. The transition matrix between these regimes is allowed to depend on explanatory variables in a novel way, using an underlying ordered probit model. Bayesian inference is employed in order to obtain predictive densities. The main focus is on short-term density forecasting in the Nord Pool intraday market. We show that the proposed model outperforms several benchmark models at this task, as measured by their predictive Bayes factors. In particular, the incorporation of stochastic volatility, regime switching, information from the day-ahead market, and exogenous information from weather reports into the model are all shown to improve its predictive performance, without falling prey to curse of dimensionality problems.