Flexible versus designated technologies and interfuel substitution

Abstract

Zero demands for some types of energy are frequently observed in …rm level data sets. To deal with this problem, two alternative solutions are retained in the literature: …rstly, it can be assumed that the consumption of some kinds of energy is not allowed by the existing technology; secondly one can suppose that …rms decide to use only some of the di¤erent kinds of energy allowed by the technology. In this paper we estimate the inter-fuel substitutions resulting from these two di¤erent situations. We take into account that increasing the price of an energy form possibly modi…es the number of energy used by the …rm. We show that even in this case an estimation of the elasticity of substitution can be obtained. The empirical results are based on a French manufacturing energy consumption survey.

Interfuel Substitution: A Meta‐analysis

“… The minimum value for all the elasticities is a theoretically inconsistent negative value and there is a wide range of estimates in the studies. The average sample size is 358 with samples as large as 25,490 (Bousquet and Ladoux, 2006) and as small as 20 (Agostini et al ., 1992). As noted above, the data are dominated by time‐series studies – 71%, with just 4% of observations from cross‐section studies.…”

“…The average sample size is 358 with samples as large as 25,490 (Bousquet and Ladoux, 2006) and as small as 20 (Agostini et al ., 1992).…”

“…Assuming that coal is the lowest quality fuel and electricity the highest quality (Stern, 2010), the further apart in quality two fuels are, the lower the elasticity of substitution between them is, which intuitively makes sense. The three non‐coal elasticities are estimated with greater precision because the Bousquet and Ladoux (2006) study provides estimates for these but not for the elasticities involving coal. The cross‐section estimates of the three coal elasticities are based on Halvorsen's (1977) study, which uses small cross‐sections of firms.…”

“…the difference between FE and OLS), and time‐series data in the meta‐regressions. There are only two cross‐sectional studies in our sample –Halvorsen (1977) and Bousquet and Ladoux (2006). So the effect of cross‐section versus panel data may not be accurately estimated.…”

“…The means and standard deviations are unweighted. The results weighted by sample size would look very different due to two papers (Fisher-Vanden et al, 2004;Bousquet and Ladoux, 2006) with much larger sample sizes than the other papers. Because not all studies use data for all four standard fuels, each meta-regression uses a subsample that drops observations that have missing values for the dependent variable.…”