Development of Regional Power Sector Coal Fuel Costs (Prices) for the Short-Term Energy Outlook (STEO) Model
April 13, 2017
Overview
The U.S. Energy Information Administration's Short-Term Energy Outlook (STEO) produces monthly projections of energy supply, demand, trade, and prices over a 13-24 month period. Every January, the forecast horizon is extended through December of the following year. The STEO model is an integrated system of econometric regression equations and identities that link data on the various components of the U.S. energy industry together in order to develop consistent forecasts. The regression equations are estimated and the STEO model is solved using the EViews 9.5 econometric software package from IHS Global Inc. The model consists of various modules specific to each energy resource. All modules provide projections for the United States, and some modules provide more detailed forecasts for different regions of the country.
The coal module provides forecasts of coal supply (production, stocks, waste coal), trade (imports and exports), consumption, prices, coal coke (production, consumption, trade, and stocks), and raw steel production. The coal module contains 73 equations, of which 23 are estimated regression equations. Some of the input variables to the coal module are exogenous, coming from other modules in the STEO model (e.g., natural gas and petroleum prices) or forecasts produced by other organizations (e.g., weather forecasts from the National Oceanic and Atmospheric Administration). A projection of national coal prices is developed using the coal module, which is passed to several other modules in STEO. The coal module, in conjunction with the STEO electricity fuel consumption module, returns a projection of national coal demand. Figure 1 provides a visual overview of the production, trade, and power sector stocks portions of the coal module. The current STEO coal module documentation can be found at http://www.eia.gov/forecasts/steo/documentation/steo_coal.pdf.
Many equations in the coal module, as well as those proposed in this document, include monthly dummy variables to capture the normal seasonality in the data series. For example, JAN equals 1 for every January in the time series and is equal to 0 in every other month. Dummy variables for specific months may also be included in regression equations where the observed data may be outliers because of infrequent and unpredictable events such as hurricanes, survey error, or other factors. Generally, dummy variables are introduced when the absolute value of the estimated regression error is more than 2 times the standard error of the regression (the standard error of the regression is a summary measure based on the estimated variance of the residuals). No attempt was made to identify the market or survey factors that may have contributed to the identified outliers.
Dummy variables for specific months are generally designated Dyymm, where yy = the last two digits of the year and mm = the number of the month (from “01” for January to “12” for December). Thus, a monthly dummy variable for March 2002 would be D0203 (i.e., D0203 = 1 if March 2002, = 0 otherwise).
Dummy variables for specific years are designated Dyy, where yy = the last two digits of the year. Thus, a dummy variable for all months of 2002 would be D02 (i.e., D02= 1 if January 2002 through December 2002, 0 otherwise). A dummy variable might also be included in an equation to show a structural shift in the relationship between two time periods. Generally, these shifts are modeled using dummy variables designated DxxON, where xx = the last two digits of the year at the beginning of the shift period. For example, D03ON = 1 for January 2003 and all months after that date, and D03ON = 0 for all months before 2003.