Spatial stochastic frontier models: controlling spatial global and local heterogeneity

In the last decade special attention has been focused on estimating a firm's efficiency and productivity; Stochastic Frontier Analysis (SFA) has been one of the most used techniques that allows the separation of inefficiency from stochastic noise, assuming homogeneous technology is available to all producers and that there is independence between observations. However, this second assumption is violated data are spatial auto-correlated, thus biasing statistical inference. Attention has, therefore, shifted to models that allow the controlling of heterogeneity introducing, in the model or in the error term, contextual variables correlated with inefficiency. In our paper we propose viewing the spatial external factors (natural or artificial) in a new way: instead of identifying ex-ante a multitude of determinants, often statistically and economically difficult to detect, we suggested using an original methodology that, following a classical SFA approach, splits efficiency into three components: the first one is linked to the spatial lag, the second one to the DMU's specificities, and the third to the error term. Finally, we tested our method using simulated data and examined the Italian wine sector, testing the ability to control spatial, global and local heterogeneity.