Forecasting Volatility of Electricity Intraday Log Returns with Generalized Autoregressive Score Models

Abstract: We forecast volatility of electricity intraday log returns with Generalized Autoregressive Score (GAS) models. We extend our GAS models with variables representing the difference between the public’s expectation of weather and energy load and the actual outcome using a restricted ARMA(4,4) model. Further, we account for the intraweekly seasonality with dummy variables. Our study is conducted on the four Swedish electricity price zones with hourly data over the period 2013 to 2022. We find that tGAS without exogenous variables is the most accurate model when extreme values are allowed in the log returns. However, GAS models that assume a Gaussian distribution and that are extended with energy load variables are the most accurate when extreme values are winsorized in the data preparation. Further, we discuss the applications of our models for both statistical researchers and risk managers. To the best of our knowledge the GAS model has not been applied the Nord Pool intraday market.