Implications for the Dirichlet Processes Mixture Linear Model on U.S. Crop Yield Predictions

Agricultural production is often exposed to several sources of risks, uncertainties and random shocks. These risks and random shocks make the evaluation the influence of policy interventions in the agricultural sector difficult due to the existence of limited data about random events. Policymakers often try to evaluate policy interventions with the objective of appropriately accounting for random shocks in the data-generation process. This article examines the implications for the risks, uncertainties and random events on the prediction of crop yields. We propose the Dirichlet process mixture model (DPMM) as a measure to account for randomness in a hierarchical linear model (HLM), and we examine its implication for the prediction accuracy of corn and soybean yields among the northern great-plain states. The proposed model employs a Gibbs sampler as the clustering algorithm for inference on the random density functions. This approach allows for an unsupervised Bayesian-cluster assignment of data with similar characteristics. A panel data set of county‐average yield data is analyzed using 104 counties for corn and 66 counties for soybeans from 1972 to 2018. The results show that the proposed model significantly improves the prediction accuracy for crop yields when an HLM is used