A GARCH forecasting model to predict day-ahead electricity prices

A GARCH Forecasting Model to Predict Day-Ahead

Electricity Prices

Reinaldo C.Garcia,Javier Contreras,Senior Member,IEEE,Marco van Akkeren,and João Batista C.Garcia

Abstract—Price forecasting is becoming increasingly relevant to producers and consumers in the new competitive electric power markets.Both for spot markets and long-term contracts,price forecasts are necessary to develop bidding strategies or negotia-tion skills in order to maximize profits.This paper provides an approach to predict next-day electricity prices based on the Gen-eralized Autoregressive Conditional Heteroskedastic(GARCH) methodology that is already being used to analyze time series data in general.A detailed explanation of GARCH models is presented and empirical results from the mainland Spain and California deregulated electricity-markets are discussed.

Index Terms—Electricity markets,forecasting,GARCH models, time series analysis,volatility.

I.I NTRODUCTION

P RICE forecasting has become a very valuable tool in the current upheaval of electricity-market deregulation. Companies that trade in electricity markets make extensive use of price prediction techniques either to bid or to hedge against volatility.When bidding in a pool system,the market participants are requested to express their bids in terms of prices and quantities.Since bids are accepted in order of increasing prices until total demand is met,a company that is able to fore-cast pool prices can adjust its own price/production schedule depending on hourly pool prices and its own production costs [1].

Another instrument to facilitate market trading is the bilateral contract system.In this setting,a buyer and seller agree on a cer-tain amount to be transferred through the network at a specific fixed price.This price is agreed upon by both sides beforehand and is also based on price predictions.Most of the deregulated electricity markets use a mixed bag of pool and bilateral panies have to optimize their production schedules

Manuscript received December23,2003;revised September7,2004.This work was supported in part by the Ministry of Science and Technology of Spain and the European Union through Grant FEDER-CICYT1FD97-1598.Paper no. TPWRS-00699-2003.

R.C.Garcia is with the Department of Energy,Transportation,and Environ-ment,German Institute of Economic Research,DIW,14195Berlin,Germany (e-mail:rgarcia@diw.de).

J.Contreras is with the E.T.S.de Ingenieros Industriales,Universidad de Castilla–La Mancha,13071Ciudad Real,Spain(e-mail:Javier.Contr-eras@uclm.es).

M.van Akkeren is with PMI Group,Walnut Creek,CA94957USA(e-mail: marco.vanakkeren@).

J.B.C.Garcia is with the Derivatives Group,Dexia Bank,Brussels,Belgium (e-mail:joaobatista.crispinianogarcia@).

Digital Object Identifier10.1109/TPWRS.2005.846044so that they can hedge against pool price volatility through bi-lateral contracts.Therefore,a good knowledge of future pool prices is very useful in valuating bilateral contracts more accu-rately.

In recent years,several methods have been applied to predict prices in electricity markets.For example,the Transfer Function [2]and autoregressive intergrated moving average(ARIMA) models[3],[4]have been tested in the Spanish and the Norwe-gian markets.In addition,Artificial Neural Networks(ANNs) have been applied to both the England and Wales pool[5]as well as the Victorian wholesale market in Australia[6].Other techniques,such as Fourier Transform[7]and stochastic mod-eling[8]have addressed the same problem.

As mentioned earlier,one key aspect of pool prices is their volatility,at least during certain periods.Note that price volatility is also very important to calculate annual average prices in order to valuate contract prices.

Spot price volatility has been recently studied in several pub-lications.Benini et al.[9]have analyzed several markets,such as Spain,California,England and Wales,and the PJM system. Mount[10]has claimed that a uniform auction worsens this problem as compared to a discriminatory auction in the Eng-land&Wales system.The Californian market has also served as a benchmark to apply Value-at-Risk models[11]or stochastic linear regression models[12].

Generalized Autoregressive Conditional Heteroskedastic (GARCH)models[13],[14]consider the moments of a time series as variant(i.e.,the error term:real value minus forecasted value does not have zero mean and constant variance as with an ARIMA process).The error term is now assumed to be serially correlated and can be modeled by an Auto Regressive(AR) process.Thus,a GARCH process can measure the implied volatility of a time series due to price spikes.For example, California experienced huge price spikes during the summer of 2000that led to the closure of the market[11],[12]until new rules were developed.

This paper focuses on day-ahead forecasts of electricity prices with high volatility periods using a GARCH method-ology approach.Our GARCH models provide24-hour forecasts of the next day market clearing price based on historical data [15],[16].To illustrate our model,price forecasts of the main-land Spain[17]and California[18]electricity markets are presented and discussed.

The paper is organized as follows.In Section II,the general GARCH methodology applied to the Spanish and California day-ahead markets is shown.Section III presents numerical re-sults of the simulations and Section IV states our conclusions.

0885-8950/$20.00©2005IEEE