Component estimation for electricity market data: Deterministic or stochastic?

In the literature, there is a wide agreement that electricity market time series include several components describing the long-term dynamics, annual, weekly and daily periodicities, calendar effects, jumps, and so on. As a result, modeling electricity variables requires the estimation of these components to filter them out and achieve stationarity, and to project them into the future. For some of them, and particularly for the long-term component, two main approaches have been proposed in the literature: the deterministic and the stochastic. The application of statistical tests for discriminating between these two alternatives is not always successful because of the low power of the tests. This work aims at examining in depth the issue of comparison between the deterministic and the stochastic approach in modeling and forecasting electricity demand. For the Italian demand data two different models are applied to the same dataset: the first one treats the components by the deterministic approach representing them through splines and dummy variables, while the second uses stochastic (time-varying) versions of these components stated in state-space form. The final comparison is based on the forecasting performance of the two sets of models and on the features of the component-adjusted time series.