Modelling the distribution of day-ahead electricity returns: a comparison

This paper contributes to characterizing the probability density of the price returns in some European day-ahead electricity markets (NordPool, APX, Powernext) by fitting some flexible and general families of distributions, such as the alpha-stable, Normal Inverse Gaussian (NIG), Exponential Power (EP), and Asymmetric Exponential Power (AEP), and comparing their goodness of fit. The alpha-stable and the NIG systematically outperform the EP and AEP models, but the tail behaviours and the skewness are sensitive to the definition of returns and to the deseasonalization methods. In particular, the logarithmic transform and volatility rescaling tend to dampen the extreme returns.