Understanding Errors in EIA Projections of Energy Demand

This paper investigates the potential for systematic errors in the Energy Information Administration’s (EIA) widely used Annual Energy Outlook, focusing on the near- to midterm projections of energy demand as measured in physical quantities. Overall, based on an analysis of the EIA’s 22-year projection record, we find a fairly modest but persistent tendency to underestimate total energy demand by an average of 2 percent per year over the one- to five-year projection horizon after controlling for projection errors in gross domestic product, oil prices, and heating/cooling degree days.For the 14 individual fuels/consuming sectors routinely reported by the EIA, we observe a great deal of directional consistency in the error patterns over time, ranging up to 7 percent per year. Electric utility renewables, electric utility natural gas, transportation distillate, and residential electricity all show significant biases, on average, across the full five year projection horizon examined. Projections for certain other fuels/consuming sectors have significant unexplained errors for selected time horizons.Independent evaluation of this type can be useful for validating ongoing analytic efforts and for prioritizing future model revisions.

Each year since 1982, the U.S. Energy Information Administration (EIA) has forecast demand for energy usage – a process made difficult by vagaries in weather, the overall economy, price and supply fluctuations, and other uncertainties. The projections are valuable to state and federal officials as well as oil and gas producers and electricity providers as they plan future operations.
In a paper entitled “Understanding Errors in EIA Projections of Energy Demand,” RFF Senior Fellows Carolyn Fischer and Richard Morgenstern and Research Assistant Evan Herrnstadt examine the accuracy of EIA forecasts over a 22-year period. Their analysis finds “a fairly persistent tendency to underestimate total energy demand by an average of 2 percent per year over the one- to five-year horizon.”

The authors note that since 1994 the projections in the Annual Energy Outlook published by EIA have been based on the National Energy Modeling System, an econometric model. They conjecture that there are undetected biases in the models used to project energy demand, particularly for electric utility renewables, electric utility natural gas, residential electricity, and transportation fuels.

“The EIA is a respected statistical agency with a well-deserved reputation for professional competence, political independence, and transparency,” they authors write. “It is our expectation that findings of the type developed here will themselves be subject to evaluation and, if sustained, will serve as a valuable input to the EIA’s ongoing efforts to revise and improve its modeling capabilities.”