Forecasting of daily electricity prices with factor models: utilizing intra-day and inter-zone relationships

Abstract

This paper investigates whether using hourly and/or zonal prices can improve the accuracy of short- and medium-term forecasts of average daily electricity prices. We consider a 6 years period (2008–2013) of hourly day-ahead prices from 19 zones of the Pennsylvania–New Jersey–Maryland (PJM) interconnection and the PJM Dominion Hub in Virginia, U.S. The predictive performance of four multivariate models calibrated to hourly and/or zonal day-ahead prices is evaluated and compared with that of a univariate model, which uses only average daily data for the Dominion Hub. The multivariate competitors include a restricted vector autoregressive model and three factor models with the common and idiosyncratic components estimated using principal components in a semiparametric setup. The results indicate that there are statistically significant forecast improvements from incorporating the additional information, essentially for all considered forecast horizons ranging from 1 day to 2 months, but only when the correlation structure of prices across locations and/or hours is modeled using factor models.