Do jumps and cojumps matter for electricity price forecasting? Evidence from the German-Austrian day-ahead market

Abstract

This paper analyzes the potential for including jumps and cojumps in electricity price forecasting models. The study is carried out on the German-Austrian day-ahead electricity market with a multivariate framework in which each hour of the day is treated as an individual time series. Three models are specified: The ARX model, the ARX-J model (which includes jumps), and the ARX-J-CJ model (which also includes cojumps). Prices are transformed using several variance stabilizing transformations. The forecasting performance of the three models with original and transformed prices is compared using several forecast horizons running from one day-ahead to one week-ahead. Results show that the forecast horizon is crucial in determining whether jumps and cojumps should be included in electricity price forecasting. Jumps and cojumps add important information to forecast prices for horizons longer than 4 days, but there is no gain in forecast accuracy for shorter horizons. The results are of interest to market participants for taking optimal decisions and pricing base week futures contracts.