CARDON RESEARCH PAPERS IN AGRICULTURAL AND RESOURCE ECONOMICS

The University of Arizona is an equal opportunity, affirmative action institution. The University does not discriminate on the basis of race, color, religion, sex, national origin, age, disability , veteran status, or sexual orientation in its programs and activities. Economics (AREC) collaborated during the fall semester 2005 on a project to improve forecasts of next-day electricity load. The project was conducted as part of an AREC M.S. class in applied econometrics. Students developed econometric models for forecasting next-day hourly load profiles, and delivered results to AEPCO in a formal business presentation in December 2005. The particular econometric models developed are known as ARIMA (autoregressive, integrated, moving average) models which use only past load data to forecast next-day load profiles. The models were calibrated for five distinct seasons in 2004: winter, spring, pre-monsoon, monsoon, and fall periods. The ARIMA models were estimated using rolling samples of 28 days of (672) hourly load observations for one week in the five seasons. ARIMA forecasts yielded reasonable results: forecast errors at coincidental peaks were generally at or below 5 percent. The time of day of coincidental peaks was usually forecast correctly. ARIMA forecast also captured the shape of 24-hour profiles adequately. The UofA ARIMA forecasts were compared to AEPCO pre-planning forecasts for two weeks in the summer of 2005—a pre-monsoon week in June and a monsoon week in August. ARIMA models usually provided modest improvements relative to AEPCO forecasts. Mean absolute percentage errors (MAPE) for the ARIMA forecasts tended to be slightly smaller than the MAPE of AEPCO's forecasts. ARIMA forecasts can be updated daily by AEPCO econometricians with less than ½ hour of work. In the short term, AEPCO staff can use ARIMA forecasts to complement their pre-planning forecasts. ARIMA and pre-planning forecasts can be compared systematically to yield further improvements in forecasting next-day load profiles.