Estimating the CAP greening effect by machine learning techniques: A big data ex post analysis

Abstract

Abstract Greening payment represents one of the main and controversial novelties of the current Common Agricultural Policy (CAP) 2015–2020 programming period. Such payments bind a portion of farm subsidies to compliance with specified practices, such as crop diversification. Unlike previous ex ante simulations, the present contribution attempts to estimate the ex post impact of greening payments in terms of land use change using a parcel-level constant sample (2011–2017) dataset of approximately 4.5 million observations. First, Markov chains and a weighted χ2 test detect a discontinuity in farmland transition probabilities only in farms that are initially non-compliant with the greening rules. Such a discontinuity is not observed in farms that are not eligible for or already compliant with the greening rules. This evidence, even if indirect, suggests that the greening payment has induced farmland conversion in farms with a lower degree of crop diversification. The greening impact on farmland allocation in this farm group was subsequently simulated using machine learning techniques. This policy has reduced maize monoculture and increased nitrogen-fixing crops, fallow land and other cereals in the targeted farms. Environmental gains (reduction in greenhouse gas emissions –GHG- and input use) and farm economic losses due to land use change have been derived, providing the first tentative cost-benefit analysis of such policy tool. Due to data limitations, indirect costs and benefits of greening (improvement in pest management, land quality and biodiversity) have not been assessed. More research and detailed environmental monitoring data are required to assess such indirect effects and to provide a more comprehensive cost-benefit ex-post analysis of greening policy