Long Run and Short Run Pesticide Inefficiencies in the French Wine Farming Sector[Inefficacité à long terme et à court terme des pesticides dans le secteur viticole français]

While the systematic use of pesticides on farms is associated with several negative impacts on human health and the environment -in particular biodiversity, many policy initiatives strive to provide the right incentives to shift agricultural production towards more environmentally friendly practices. For instance, at the EU level, the Sustainable Use of Pesticides Directive aims to reduce by 50% the use and the risks associated with chemical pesticides by 2030. The case of pesticide use in the French wine sector is quite controversial as vine plants are the most heavily treated crops in France. In France, vineyards account for about 3% of agricultural area but use 20% of the total pesticides. In parallel, New World producers are increasingly greening their wineries, while French wine growers are seen as laggards. At the policy level, in 2018, the French national policy progamme ‘Ecophyto II+' sets more ambitious targets than its EU counterpart by aiming at a 50% reduction of pesticide use by 2025. The effectiveness of pesticide policies depends, on the one hand, on the existence of viable technical solutions that do not compromise yields and profit and that are acceptable to farmers, and on the other hand, on how dependent farms are on pesticides. In this work, we focus on the latter point by relating pesticide dependency to a technological locked-in characterized by self-reinforcing mechanisms, which prevent the adoption of other beneficial alternatives to pesticides even though they may exist. A peculiar feature of technological locked-in is the existence of structural/persistent inefficiencies. Our goal in this work is to disentangle long-run (structural/persistent) and short-run (transient) inefficiencies in pesticide use in the case of French wine growers. To this aim, we estimate a four-error component stochastic frontier that accounts for farm heterogeneity. Two different models are estimated. The first one is a pesticide input requirement technology, and for robustness, a variable input distance function is estimated as the second model. Both models are estimated in two steps. In the first step, a GMM estimation is conducted to deal with endogeneity issues; in the second step, the error components are estimated. The application is to bookkeeping and structural data from a representative sample of French wine growers between 2002 and 2021.