Evaluation of errors in national energy forecasts

Denys Sakva

Physical copy available from RIT's Wallace Library at HD9502.U52 S34 2005


Energy forecasts are widely used by the U.S. government, politicians, think tanks, and utility companies. While short-term forecasts were reasonably accurate, medium and long-range forecasts have almost always been highly erroneous. In the U.S. many energy policy decisions are driven by Annual Energy Outlook (AEO) forecasts prepared by Energy Information Association (EIA). This thesis evaluates accuracy of AEO reports from 1982 to 2003. Parameters evaluated are: total energy consumption, energy consumption by sector, sector specific parameters, and major model assumptions. Error decomposition and regression analysis are used to appraise accuracy of forecasts. I found that often underlying parameters used to calculate more aggregate parameters suffer from errors that are higher by amplitude than forecasted parameter itself. Positive and negative errors cancel each other and conceal higher error in the underlying parameters. Total energy consumption was predicted with higher accuracy than energy consumption by sector. Energy prices were predicted with very low accuracy and errors reach 250%. Almost all parameters suffer from systemic errors and were consistently overestimated or underestimated. I also determined numerical estimates for expected increase in accuracy because of increase in assumptions accuracy.