28th International Symposium on Forecasting - ISF2008, Niza (Francia). 22-25 junio 2008
Resumen:
In this paper, a periodic dynamic regression model is used to carry out a one day-ahead forecast of the Spanish electricity hourly prices. The explanatory variables included in the model are the electricity demand and the wind power generation forecasts supplied by the System Operator. The periodic model switches among different dynamic regression inner models, each one of them specialised in a particular day of the week. This feature makes it possible to cope with the differences that can be observed in the daily seasonal dynamics and in the sensitivity of the price to explanatory variables in working and non-working days. The model has been implemented in a state-space form, using the Kalman filter for the optimal estimation of the state variables of the process. The proposed model is empirically compared to different reference models in order to quantify the improvements due to the inclusion of each of the explanatory variables and the benefits of considering a periodic model. The main outcomes of this work are (a) the wind power generation has been revealed as a very important input variable, that is, its inclusion in the model together with the demand significantly improves the out-of-sample error, and (b) different linear dynamics have been identified in the hourly electricity price time series.
Fecha de publicación: 2008-06-22.
Cita:
A. Muñoz, A. Cruz, J. Zamora, R. Espínola, Forecasting electricity prices with periodic dynamic regression models, 28th International Symposium on Forecasting - ISF2008, Niza (Francia). 22-25 junio 2008.
