Forecasting marginal costs of a multiple-output production technology

This paper presents the results of fitting a scaled translog restricted profit function to ‘pseudo’-data formed by repeated runs of a large linear programming model of domestic and international refining. The translog approximation is designed to estimate the marginal cost of producing eight petroleum products given the amounts of each product demanded and the price of crude oil. We test the model against out-of-sample data from the refinery model and historical data. The model is used in the US Department of Energy's Annual Energy Outlook forecasting system.