Empirical analysis of farm level spatial interactions and their role for aggregated policy impact analysis
Final Report Abstract
The project improved our understanding of the importance of spatial interactions for farm growth and specialization decision. A central theme in the project is the focus on the endogeneity problems of spatial regression models and identification strategies to address these problems. In this respect, a novel spatial econometric approach is developed that aims to reduce omitted variable bias resulting from spatially correlated and unobserved confounders. This approach is applied to study farm growth, survival and diversification decisions in applications using large farm level census dataset from Norway, France and the Netherlands. The results highlight the importance of spatial interaction in those areas and underline the importance to account for different spatial interaction processes at different scale as well for heterogeneity in the spatial interaction effects. Additionally, the previously unused panel structure of the Norwegian data is exploited in order to identify spatial interaction effects in context of farmers conversion decision to organic production, showing that negative spillovers from deconversion are equally important as positive spillovers. In a second strand of work, the project explores novel deep learning approaches to improve our capabilities to study the effects of farm subsidies on farmers growth, specialization and survival decisions. The developed approach allows to run policy counterfactual simulation considering detailed changes in the subsidy scheme, is more flexible in consider dynamics and heterogeneity and allows to treat farm growth/survival as a multidimensional process jointly considering a large number of farm activities. Finally, building on this work a systematic literature review of machine learning approaches is developed aiming to provide an econometric perspective on machine learning. This review introduces the most important ML concepts and methods highlighting differences and similarities to our econometric approaches. Additionally, it systematically explores various areas in agricultural and applied economics where ML approaches may help to overcome limitations of our econometric toolbox and where they open up new research possibilities. Further, current limitations of ML approaches are discussed and areas where econometricians could make contributions are identified.
Publications
- (2018): Reducing omitted variable bias in spatial interaction models by considering multiple neighbourhoods. Spatial Economic Analysis 13(4):457-472
Storm H., Heckelei, T.
(See online at https://doi.org/10.1080/17421772.2018.1468571) - (2019): Heterogeneous impacts of neighbouring farm size on the decision to exit: evidence from Brittany. European Review of Agricultural Economics 46(2):237-266
Saint-Cyr, L., Storm, H., Heckelei, T. and Piet, L.
(See online at https://doi.org/10.1093/erae/jby029) - (2019): Identifying effects of farm subsidies on structural change using neural networks. Discussion Paper 2019:1, Institute for Food and Resource Economics, University of Bonn
Storm, H., T. Heckelei, K. Mittenzwei and, K. Baylis
(See online at https://dx.doi.org/10.22004/ag.econ.287343) - (2020). Machine Learning in Agricultural and Applied Economics. European Review of Agricultural Economics 47(3):849–892
Storm, H., K. Baylis, T. Heckelei
(See online at https://doi.org/10.1093/erae/jbz033) - (2020): Beyond the single farm - A spatial econometric analysis of neighborhood effects in farm diversification in the Netherlands. Land Use Policy 99 (2020) 105019
Vroege, W., M. Meraner, N. Polman, H. Storm, W. Heijman, R. Finger
(See online at https://doi.org/10.1016/j.landusepol.2020.105019)
