Stochastic biases in technical change in U.S. agriculture: A bootstrap approach

Abstract

Having argued that the modeling of technical change as a smooth deterministic function of time is likely to misrepresent the true nature of technical change, this paper reexamines biased technical change in U.S. agriculture using a system of share equations with unobserved components errors, with technology treated as a stochastic unobserved variable. Employing data to represent the aggregate output and input of the U.S. agricultural sector over the period 1947-94, significant factor biases were found that appear to be linearly independent and do not appear to be smooth and deterministic. Technical change in U.S. agriculture appears to have been biased toward saving expenditure on labor at the expense of expenditure on intermediate inputs, with some small saving on the expenditure on capital inputs over the entire period 1947-94. The paper also employs a bootstrapping approach in order to obtain finite sample tests with approximately the correct size under less stringent assumptions about the data generating process than assumed by maximum likelihood (ML) based approaches. Using these finite sample values significantly alters the conclusions reached regarding the nature of technical change.