| Duality theory states that all the economically relevant aspects of a production technology are summarized by both the production function and the cost function. When data permit, the notion of using empirical results from both production and cost frameworks as a mutual check for specification errors seems perfectly sensible. However, the fact is that inferences about the underlying technology based on production and cost models are often widely divergent. In assessing the causes of these divergences, this dissertation researches the suitability of various series expansions (flexible forms) to economic modeling, contributes to an understanding of the errors in variables problem, and develops cost and production systems with consistent behavioral assumptions.;In Monte Carlo experimentation, it is demonstrated that modest price or input measurement errors (because they are likely to be correlated with errors in the share equations) can cause estimates of Allen partial elasticities of substitution to be substantially biased. This errors in variable problem alone can account for divergent results of the sort reported by Burgess, Humphrey and Moroney, and later pondered by Theil. Price and input measurement errors and producer optimizing error are shown to have different effects on estimates produced by cost and production models. Maximum likelihood approaches to the problem are investigated, but the results are disappointing.;The importance of imposing consistent temporal and behavioral assumptions in order to achieve consistent inferences from the dual models regarding the structure of production is demonstrated. This 'discovery' necessitated the refinement of the standard cost and production specifications and the development of exact analogs. These new models include a static production system imposing a weaker maintained hypothesis of cost minimization, a short run equilibrium (SRE) production system, and a fully dynamic production model with maintained hypothesis of cost minimization.;Using the 'region of convergence' criterion (where the larger the region, the better), Taylor series approximations are found to be preferred over Laurent series expansions, and among Taylor series expansions, the Translog will have the largest region of convergence. Monte Carlo experiments suggest that the Translog cost share system will generally outperform the highly touted Fourier form of equivalent order. |
