Bayesian calibration and number of jump components in electricity spot price models

TL;DR

This paper develops a Bayesian MCMC method to determine the number of jump components in electricity spot price models, revealing structural changes before and after financial crises.

Contribution

It introduces a Bayesian calibration approach for jump processes and demonstrates how the composition of these jumps changes over time in electricity markets.

Findings

Significant reduction in positive price spikes after the financial crisis.

Independent signed sums of jump processes better model price spikes than mean-reverting jumps.

Structural changes in jump components are consistent across markets.

Abstract

We find empirical evidence that mean-reverting jump processes are not statistically adequate to model electricity spot price spikes but independent, signed sums of such processes are statistically adequate. Further we demonstrate a change in the composition of these sums after a major economic event. This is achieved by developing a Markov Chain Monte Carlo (MCMC) procedure for Bayesian model calibration and a Bayesian assessment of model adequacy (posterior predictive checking). In particular we determine the number of signed mean-reverting jump components required in the APXUK and EEX markets, in time periods both before and after the recent global financial crises. Statistically, consistent structural changes occur across both markets, with a reduction of the intensity and size, or the disappearance, of positive price spikes in the later period. All code and data are provided to enable replication of results.