Carnero, M. Angeles Koopman, Siem Jan Ooms, Marius
Year of Publication:
Tinbergen Institute Discussion Paper 03-071/4
Although the main interest in the modelling of electricity prices is often on volatility aspects, we argue that stochastic heteroskedastic behaviour in prices can only be modelled correctly when the conditional mean of the time series is properly modelled. In this paper we consider different periodic extensions of regression models with autoregressive fractionally integrated moving average disturbances for the analysis of daily spot prices of electricity. We show that day-of-the-week periodicity and long memory are important determinants for the dynamic modelling of the conditional mean of electricity spotprices. Once an effective description of the conditional mean of spot prices is empirically identified, focus can be directed towards volatility features of the time series.For the older electricity market of Nord Pool in Norway, it is found that a long memory model with periodic coefficients is required to model daily spot prices effectively. Further, strong evidence of conditional heteroskedasticity is found in the mean corrected Nord Pool series. For daily prices at three emerging electricity markets that we consider (APX in The Netherlands, EEX in Germany and Powernext in France) periodicity in the autoregressive coefficients is also established, but evidence of long memory is not found and existence of dynamic behaviour in the variance of the spot prices is less pronounced. The novel findings in this paper can have important consequences for the modelling and forecasting of mean and variance functions of spot prices for electricity and associated contingent assets.
Autoregressive fractionally integrated moving average model Generalised autoregressive conditional heteroskedasticity model Long memory process Periodic autoregressive model Volatility.