QUANTIFYING THE CREDIBILITY OF ENERGY PROJECTIONS FROM TRENDS IN PAST DATA - THE UNITED-STATES ENERGY SECTOR

For energy forecasts to be useful in modelling or in policy efforts, the associated uncertainties must be known reliably. We analyse the actual errors in past forecasts of over 170 energy producing and consuming sectors of the US economy. We find that the often assumed normal distribution fails to model frequency of extreme outcomes (those lying far from the mean) accurately. Triangular distributions perform even worse as they assign zero probability to the outliers. We develop a simple one-parameter model that can be used to estimate a probability distribution for future projections. In addition to energy forecasts, our method can be applied to any field where a history of forecasting is available.