Measuring plant level energy efficiency and technical change in the U.S. metal-based durable manufacturing sector using stochastic frontier analysis
© 2019 This study analyzes the electric and fuel energy efficiency for five different metal-based durable manufacturing industries in the United States over the time period 1987–2012, at the 3 digit North American Industry Classification System (NAICS) level. Using confidential plant-level data on energy use and production from the quinquennial U.S. Economic Census, a stochastic frontier regression analysis (SFA) is applied in six repeated cross sections for each five year census. The SFA controls for energy prices and climate-driven energy demand (heating degree days HDD and cooling degree days CDD) due to differences in plant level locations, as well as 6-digit NAICS industry effects. Own energy price elasticities range from −0.7 to −1.0, with electricity tending to have slightly higher elasticity than fuel. Mean efficiency estimates (100% = best practice level) range from a low of 33% (fuel, NAICS 334 - Computer and Electronic Products) to 86% (electricity, NAICS 332 - Fabricated Metal Products). Electric efficiency is consistently better than fuel efficiency for all NAICS. Assuming that all plants in the least efficient quartile of the efficiency distribution achieve a median level of performance, we compute the decline in total energy use to be 21%. A Malmquist index is used to decompose the aggregate change in energy performance into indices of efficiency and frontier (best practice) change. Modest improvements in aggregate energy performance are mostly change in best practice, but failure to keep up with the frontier retards aggregate improvement. Given that the best practice frontier has shifted, we also find that firms entering the industry are statistically more efficient, i.e. closer to the frontier; about 0.6% for electricity and 1.7% for fuels on average.
Volume / Issue
Start / End Page
International Standard Serial Number (ISSN)
Digital Object Identifier (DOI)