EM estimation of multivariate dynamic models for predicting electricity prices

Abstract

In order to make short-term predictions of electricity prices, linear dynamic models in their state-space formulation have been studied. A computer implementation of the EM (Expectation - Maximization) algorithm has been made for maximum likelihood estimation for a Multivariate EWMA model, (exponentially smoothing). In this approach the problem includes a large number of parameters to be estimated as we have implemented the possibility of eliminating superfluous parameters. Finally, we present the results of the hourly spot price forecasts in Powernext, Nord Pool and OMEL markets.