Forecasting the daily Elspot prices : whether weather will work

Abstract

It is always important for participants in a market to receive and process all available information in order to make rational decisions. At Nordpool this is of extra relevance, since participants have to hand in bids on a day-ahead basis. To have an accurate forecast of the prices which regards the relevant information is therefore crucial. This thesis’ objective is to forecast Elspot daily prices and testing if weather variables will increase the accuracy of the forecast. This is done by implementing different ARIMA models and including explanatory variables. The variables used are temperature and precipitation. We find that out of the different ARIMA versions, the SARIMA stands out as the most well defined. This is a seasonal ARIMA model accounting for weekly seasonality. However, after much testing it was not possible to find a model which controls for all serial correlation, there were still some left in the residual. When including the weather variables the model got slightly improved. Most of the improvement originates from the inclusion of temperature. The forecast does a good job of predicting prices one day ahead. It also gives an ok indication of the next four to five days price movements. The inclusion of weather variables does not improve the forecast as much as expected. It increases the certainty of the forecast only slightly in volatile periods. We conclude that the model does a god job forecasting, but it can be developed further. Future research should focus on completely removing the serial correlation, and extracting all possible information from the weather variables.