Difference between revisions of "Probabilistic Energy Projections"
(Created page with "Lumina is making available a set of probabilistic projections of key energy quantities. The US Energy Information Administration (EIA) provides an Annual Energy Outlook (AEO)...")
Revision as of 19:01, 4 December 2019
Lumina is making available a set of probabilistic projections of key energy quantities. The US Energy Information Administration (EIA) provides an Annual Energy Outlook (AEO) with projections of key quantities often 25 years into the future. These quantities include US prices, consumption, production, and net imports (where relevant) for oil, natural gas, electricity, and electricity. Most researchers and organizations in the USA involved in energy forecasting and investment use these AEO projections at least as a baseline for comparison.
Each AEO Report include a Reference Scenario and often many other scenarios. EIA terms these "projections" not "forecasts". They are based on current US policy and legislation, with no attempt to forecast effects of any proposed changes in policy. It makes no claims about the probability that the future values of these quantities will be less than or greater than the Reference scenario or any others. To its great credit EIA also issues an annual AEO Retrospective that compares AEO projections from past years since 1992 against the actual values of quantities, so you can see their accuracy and errors.
Max Henrion (from Lumina), in collaboration with Evan Sherwin and Ines Azavedo (both previously at Carnegie Mellon University and now at Stanford), has developed a probabilistic forecasting model in Analytica for selected quantities from the AEO Retrospectives. It uses a time-series method called Exponential Trend Smoothing (ETS) fitted to the actual historical values of selected quantities. This method estimates the uncertainty in the forecast based on the error distribution in how well the ETS forecast using earlier years fits actual values in later years -- i.e. the empirical error distributions for the ETS model. It then applies this error distribution to estimate the uncertainty in forecasts into the future.
Lumina makes these probabilistic forecasts available free for download. They currently for these ten quantities: US Petrol Consumption, US Crude Oil Production, Petrol net imports, Natural Gas Price, Gas consumption, Gas production, Net US Gas imports, Electricity price, Electricity sales, World oil price. These results are for years 2015 to 2040 (actual historical values for 2015 to 2017) with 200 Monte Carlo samples.
You can download these probabilistic projections in two formats:
These results represent serial correlation (dependency) from one year to the next for each quantity, but not correlation across quantities. If you are interested in obtaining forecasts for additional quantities, a larger sample size, or samples that model correlations across quantities, or learning more details on how these forecasts were generated, please contact support@Lumina.com.