Stochastic non-parametric frontier analysis in measuring technical efficiency : a case study of the North American dairy industry
MetadataShow full item record
Regulatory institutions governing many industries in Canada are similar to those of the United States. Some differences in regulations and institutions can be found in those industries for which the two countries compete in export markets, and many agricultural products fall into the latter category. With respect to the production and export of dairy products, Canada has recently implemented policies that are substantially different from those found in the U.S. Differences in dairy policy have been the source of several recent trade disputes between the two countries. Despite efforts to the contrary by participants in the major policy agreements governing agricultural trade (i.e., CUSTA; NAFTA, and WTOA), the regulated structure of the Canadian dairy industry has been maintained. The U.S. and New Zealand have challenged the marketing practices of the supply managed by Canadian dairy sector. These policies have a direct impact on the productive efficiency of dairy farms. In this regard, the dairy industry in Canada and the U.S. provides a natural context for an experiment allowing us to compare the relative performance of otherwise almost identical producers under different agricultural policies. The objective of this thesis is to estimate and compare the technical efficiency of a large set of dairy producers in Canada (Ontario and Quebec), with their counterparts in the U.S. (New York and Wisconsin) by using a stochastic nonparametric frontier regression analysis. Our motivation for using stochastic nonparametric frontier estimates comes from the fact that there are problems inherent in the structure of stochastic parametric frontier models. Specifically in the latter models, the literature has shown that the efficiency scores are sensitive to the choice of both functional forms and the distribution assumptions made about the one-sided random component of the composed error term. To solve this econometric model, an iterative procedure called a smoothing process is used to estimate the mean response function and its parameters constructed in a generalized additive model. Using the method of locally scoring smoothing, the parameters of the regression function are estimated by employing two separate nonparametric techniques: locally weighted scatterplot smoothing (LOWESS), and spline smoothing. After estimating the response function and its parameters, the technical efficiency scores are computed. These efficiency indices are also compared with the one obtained from conducting a stochastic parametric (translog) frontier function. The results show that the overall mean technical efficiency obtained from translog function for all regions is higher than that of the corresponding values obtained from the nonparametric approaches. Both parametric and nonparametric methodologies indicated evidence of differences between the mean technical efficiency of dairy farms in all regions. This means various policies implemented in the two countries significantly impacted the performance of dairy producers. The direction of these differences was in the favor of U.S. dairy farmers, who produced milk more efficiently than their Canadian counterparts. This implies that the regulated dairy industry in Canada has led to lower technical efficiency of Canadian dairy farmers. Canadian farmers surely benefited financially from the implementation of supply management over the duration of this study, but from an efficiency perspective, policymakers might to realize that the current support policy is only sustainable at a cost. Furthermore, Canada's commitments to international agreements such as the WTO may no longer readily allow the federal government and the provinces to pursue some elements of the current supply management policy.