Modeling the formation of expectations: The history of energy demand forecasts

Abstract

Since 1973 forecasts of future energy consumption in the United States have fallen dramatically. Forecasts of consumption in 1985 made as recently as 1974 were too high by nearly a factor of two. Forecasts of consumption in 2000 have fallen by a similar ratio. The large errors raise questions about the forecasting methods used. This paper tests the ability of adaptive expectations and univariate trend extrapolation to explain the history of energy demand forecasts. A behavioral model of the trend estimation and forecasting process is developed. Energy demand is forecast by extrapolation of the expected growth rate of consumption. The expected growth rate is determined by the past rate of growth of actual energy consumption. The model explicitly considers delays in the measurement of energy consumption and in individual and organizational response to changes in the apparent trend. The model is shown to fit the forecast data well for three different forecast horizons. Adaptive trend extrapolation thus accounts well for significant evolution of energy demand forecasts during a major period of change in energy consumption. The results are reconciled with the fact that many of the forecasts in the sample were based on complex models and do not appear to be simple extrapolations, and implications for behavioral modeling of expectation formation are discussed.