978-1-4244-1744-5/08/$25.00 ©2008 IEEE Keywords—Electricity price, forecasting, SARIMA and GARCH models, GLS models with autocorrelated errors. Abstract—In the last years, electricity markets have been changing from monopolies to liberalized markets, and consequently, with the entrance of new providers, there has been an increase of competitiveness. With the creation of open markets utilities faced the need to understand the way electricity prices change, in order to better develop bidding and hedging strategies. This paper focuses on the Spanish electricity market. In order to develop a model capable of evaluating and predicting the Spanish electricity prices, we use data from January 1998 until August 2005. In addition, as one of the aims is to get short and medium term forecasts, we analyze the data in a daily and monthly basis, using different techniques. In the daily prices, we use SARIMA models complemented with GARCH models. For the monthly prices, two different approaches are considered: time series models, and generalized least squares models with autocorrelated residuals. I. INTRODUCTION HE changing from monopolies to liberalized electricity markets has widespread in the last years, and the creation of the open Iberian electricity market, the MIBEL, has raised many questions, namely the need to better understand and predict the behavior of electricity prices. The aim of this paper is to develop a model capable of evaluating and predicting the Spanish electricity prices, in short (daily) and medium (monthly) term ranges. For that purpose we use average daily prices available from January 1998 until August 2005. We remark that in order to get short term forecast we analyze the data on a daily basis (i.e., average daily prices of the electricity), whereas for the medium term forecast we look at the average monthly price of the electricity. We note that electricity is an untypical commodity, due to its non-storability. This non-storability leads to a very volatile market, as any fluctuation on demand must be covered by power generation. In fact, production and consumption must be perfectly synchronized, resulting in an energy balance between the injection of power in generating points and its use at demand points (plus some additional power to cover transmission losses); see [2]. There are many works dedicated to the modeling of particular electricity markets. One of the common points is that the electricity price depends on both economic activity and weather characteristics, through the demand of electricity. It is well known that the demand is subjected to three types of seasonality: intra-day seasonality (related with different consumption levels during the day and night), weekly seasonality (results from the business activity, with major differences between workdays and weekends), and annual seasonality (that reflects mainly the effect of seasonal conditions). Due to the relation between price and demand, it is expected that electricity prices also reveal these types of seasonality. For example, [7], [10] propose models (extensions of the ones proposed in [13]) that incorporate the seasonality of the data. In addition to seasonality, the behavior of electricity prices has also some interesting characteristics, usually difficult to model. One of such characteristics is the presence of price jumps followed by notorious decreasing in prices, as pointed in [1], who considers a two factor mean reverting model involving jumps. In this work we model the Spanish electricity prices using mainly time series techniques, as in [4]-[6]. We stress that there are many ways to model the data, as, for example, the combination of both stochastic calculus and time series models, as in [3], and models based on neural networks, in [8], and spatial error models in [12]. Statistical Models to Predict Electricity Prices Cláudia Nunes 1 , António Pacheco 1 , Tânia Silva 2 1 Technical University of Lisbon and CEMAT Avenida Rovisco Pais 1049-001 Lisbon, Portugal E-mail:cnunes@math.ist.utl.pt apacheco@math.ist.utl.pt 2 EDP – Energias de Portugal Praça Marquês de Pombal, 13 3º 1250-162 Lisbon, Portugal E-mail: tania.silva@edp.pt T