Machine learning in agricultural economics

With the substantial growth in novel data sources and computational power, machine learning holds great potential for economic analysis. However, like any new approach, the strengths and weaknesses of these tools need to be considered when deciding where and how they can be successfully applied. In this chapter, we introduce key ML methods, from penalized regressions, to tree-based methods to neural networks, relating these approaches to common econometric practice. We then explore the potential afforded by ML to fill gaps in our current methodological toolbox. We discuss use cases like the need for flexible functional forms, the use of unstructured data, and large numbers of explanatory variables in both prediction and causal analysis. We also highlight the challenges of complex simulation models including calibration, validation and computational demands and identify places where machine learning can help. We highlight these issues drawing from existing examples in agricultural and applied economics. To unpack the “black box” of ML, we present numerous approaches used in computer science and statistics for model interpretability. Finally, we highlight some ethical issues around the use of ML. We argue that economists can play a vital role in adapting ML methods for the use in economics by combining them with our domain knowledge of economic mechanisms, and our approach to causal identification.