Econometric Modeling of Regional Electricity Spot Prices in the Australian Market

TL;DR

This paper develops a Bayesian econometric model for Australian electricity spot prices, capturing regional interdependencies, price tails, and shock transmission, validated through forecasting performance.

Contribution

It introduces a novel econometric approach combining spatial equilibrium, copula models, and nonparametric distributions for regional electricity prices.

Findings

Model accurately captures price distribution characteristics.

Forecasts outperform benchmark models.

Demonstrates shock transmission across regions.

Abstract

Wholesale electricity markets are increasingly integrated via high voltage interconnectors, and inter-regional trade in electricity is growing. To model this, we consider a spatial equilibrium model of price formation, where constraints on inter-regional flows result in three distinct equilibria in prices. We use this to motivate an econometric model for the distribution of observed electricity spot prices that captures many of their unique empirical characteristics. The econometric model features supply and inter-regional trade cost functions, which are estimated using Bayesian monotonic regression smoothing methodology. A copula multivariate time series model is employed to capture additional dependence -- both cross-sectional and serial-- in regional prices. The marginal distributions are nonparametric, with means given by the regression means. The model has the advantage of preserving the heavy right-hand tail in the predictive densities of price. We fit the model to half-hourly spot price data in the five interconnected regions of the Australian national electricity market. The fitted model is then used to measure how both supply and price shocks in one region are transmitted to the distribution of prices in all regions in subsequent periods. Finally, to validate our econometric model, we show that prices forecast using the proposed model compare favorably with those from some benchmark alternatives.