Comparison of alternative functional forms in production (Q1069652)

In this paper, a Bayesian approach to choosing between two non-nested multivariate regression systems is developed. The Bayesian approach involves the calculation of the posterior probabilities of alternative hypotheses and formation of a posterior odds ratio. Specific results are obtained for the comparison of a multivariate regression and seemingly unrelated regression model. These odds ratio results are applied to the choice between translog and Fourier flexible form cost share equation systems. When calculated for aggregate U.S. manufacturing cost data, the odds ratios favor the Fourier flexible form over the translog form. Even in small samples, the Fourier series expansion appears to be an effective approximation technique.