Modeling electricity spot prices using mean-reverting multifractal processes

TL;DR

This paper introduces two mean-reverting multifractal models for electricity spot prices, estimating parameters from Nordic market data, and finds the damped MRW better captures price features with a correlation time of about six months.

Contribution

The paper develops and applies two novel mean-reverting multifractal models to electricity prices, providing new methods for parameter estimation and model comparison.

Findings

Electricity spot prices have unique scaling exponents different from stock markets.

The damped MRW model better captures certain features of spot prices.

Estimated correlation time is approximately six months.

Abstract

We discuss stochastic modeling of volatility persistence and anti-correlations in electricity spot prices, and for this purpose we present two mean-reverting versions of the multifractal random walk (MRW). In the first model the anti-correlations are modeled in the same way as in an Ornstein-Uhlenbeck process, i.e. via a drift (damping) term, and in the second model the anti-correlations are included by letting the innovations in the MRW model be fractional Gaussian noise with H < 1/2. For both models we present approximate maximum likelihood methods, and we apply these methods to estimate the parameters for the spot prices in the Nordic electricity market. The maximum likelihood estimates show that electricity spot prices are characterized by scaling exponents that are significantly different from the corresponding exponents in stock markets, confirming the exceptional nature of the electricity market. In order to compare the damped MRW model with the fractional MRW model we use ensemble simulations and wavelet-based variograms, and we observe that certain features of the spot prices are better described by the damped MRW model. The characteristic correlation time is estimated to approximately half a year.