Title: Electricity prices forecast analysis using the extreme value theory

Authors: Mario Domingues de Paula Simões; Marcelo Cabus Klotzle; Antonio Carlos Figueiredo Pinto; Leonardo Lima Gomes

Addresses: c/o Faculdades IBMEC-RJ – Graduação, Av. Presidente Wilson, 118 – Centro, 20030-020 – Rio de Janeiro – RJ, Brazil ' C/O IAG/PUC-Rio, Rua Marques de São Vicente, 225 – Gávea 22451-900 – Rio de Janeiro – RJ, Brazil ' C/O IAG/PUC-Rio, Rua Marques de São Vicente, 225 – Gávea 22451-900 – Rio de Janeiro – RJ, Brazil ' C/O IAG/PUC-Rio, Rua Marques de São Vicente, 225 – Gávea 22451-900 – Rio de Janeiro – RJ, Brazil

Abstract: The present work attempts to evaluate the risk attached to electricity price forecasts. Initially, an analysis of prices series from different observation frequencies and, as expected, the volatility attenuation as a function of decreased observation frequency, for the same data, is observed. Next, a price forecast is made using a widely established and well used ARMA model. The distribution of residues of this forecast is modelled by a Gaussian curve and a generalised Pareto distribution, as well as its empirical distribution, following which the risk metrics VaR and CVaR are calculated. The Gaussian approximation shows to be appropriate for the estimation of forecast errors at low quantiles, up to 95%, for both daily and hourly data, but underestimates CVaR. The GPD distribution proves to be accurate and safe for the use of CVaR at any observation frequency, while it is introduced a novel GPD combination technique for the use of CVaR at extreme quantiles.

Keywords: electricity prices; risk assessment; VaR; CVaR; EVT; GPD; generalised Pareto distribution; out-of-sample forecasting; price forecasting; extreme value theory; ARMA model; modelling.

DOI: 10.1504/IJFMD.2016.076973

International Journal of Financial Markets and Derivatives, 2016 Vol.5 No.1, pp.1 - 22

Received: 16 Sep 2014
Accepted: 17 Apr 2015

Published online: 16 Jun 2016 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article