A1054
Title: The coverage probability of forecast intervals in the presence of unpredictable and predictable spikes
Authors: Jonas Andersson - Norwegian School of Economics (Norway)
Samaneh Sheybanivaziri - Norwegian School of Economics (Norway) [presenting]
Abstract: Methods to compute forecast intervals for electricity price forecasts are systematically compared. In particular, we investigate the complication caused by price spikes. The electricity prices are modeled with a mixture model with two regimes, one for regular prices and one for spikes; a specification that we argue captures the most essential features of distributional and temporal properties of electricity prices. Our first findings are that, not surprisingly, fitting a model accounting for the possibility of spikes helps in getting a correct coverage probability. Also, using simple models, e.g., ARMA models, in combination with a non-parametric bootstrap approach, often give coverage probabilities close to the nominal levels. Ignoring the spikes is often not that consequential for the coverage probabilities for nominal levels 95\% and 99\%. 80\% prediction intervals are conservative, i.e., have a coverage probability well above 80\%. While giving an approximately correct coverage probability, simple ARMA models yield substantially wider prediction intervals than the correctly specified mixture model.
