Application of Public-Domain Market Information to Forecast Ontario's Wholesale Electricity Prices

Abstract

This paper evaluates the usefulness of publicly available electricity market information in forecasting the hourly Ontario energy price (HOEP). In order to do so, relevant data from Ontario and its neighboring electricity markets, namely, New York, New England, and PJM electricity markets, are investigated, and a final set of explanatory variable candidates that are available before real-time are selected. Multivariate transfer function and dynamic regression models are employed to relate HOEP behavior to the selected explanatory variable candidates. Univariate ARIMA models are also developed for the HOEP. The HOEP models are developed on the basis of two forecasting horizons, i.e., 3 h and 24 h, and forecasting performance of the multivariate models is compared with that of the univariate models. The outcomes show that the market information publicly available before real-time can be used to improve HOEP forecast accuracy to some extent; however, unusually high or low prices remain unpredictable, and hence, the available data cannot lead to significantly more accurate forecasts. Nevertheless, the generated forecasts in this paper are significantly more accurate than currently available HOEP forecasts. To analyze the relatively low accuracy of the HOEP forecasts, comparisons are made with respect to ARIMA models developed for locational marginal prices (LMPs) of Ontario's three neighboring markets, and price volatility analyses are presented.