Bayesian Predictive Distributions for Imbalance Prices With Time-Varying Factor Impacts

Abstract

A dynamic Bayesian model is developed to estimate the time-varying nature of the drivers of the system imbalance prices in the British electricity market. We find that the key exogenous factors that significantly influence prices have impacts that evolve substantially over time. Thus, by modeling their evolution with time varying parameter estimation and making conditional forecasts on the latest estimates, more accurate forecasts are produced. Furthermore, using a Bayesian approach allows predictive distributions to be developed, as would be required for value-at-risk compliance purposes. These densities are also found to be more accurate at the extreme quantiles than a conventional GARCH model with static parameters. We validated the superior performance of this Bayesian time varying predictive density method with the same data as in a previously published benchmark model.