Production Functions: The Search for Identification

Some aspects of the econometric identification and estimation of production functions are discussed focusing on the issue of simultaneity and reviewing the response to it since Douglas's early studies and Marschak and Andrews critique of them. We look primarily at the work that uses micro data for panels of firms and plants in the framework of fixed effects models and at some more recent extensions of it. We find that researchers, in trying to evade the simultaneity problem, have shifted to the use of thinner and thinner slices of data, exacerbating thereby other problems and misspecifications. We conclude on the need for better data, especially on product prices at the individual observation level and on relevant cost and demand shifters, and for better behavioral theories which would encompass the large amount of heterogeneity observed at the micro level.