Matthieu Cornec (EDF R&D) & Hugo Harari Kermadec (CREST)

In the early 90's, electricity production has moved from state monopolies to competitive markets, with a new element of risk : the wholesale price uncertainty. Thus, modeling electricity-prices together with exogenous variables is crucially needed for plant scheduling, generation asset management and option pricing. Recent literature on empirical time-series has revealed the difficulty for traditional financial parametric models to catch up with the complex features of electricity prices : mean-reversion, multi-scale seasonality, erratic extreme behavior with fast reverting spikes. We propose to consider an empirical approach, inspired from the Bootstrap literature. We make use of the regenerative structure of the time-series seen as Markov chains to construct data blocks which are almost independent. The empirical distribution on the blocks is then an estimator of the data distribution. Eventually, this approach is applied to the joint modeling of Powernext electricity prices and temperature.