Uncertainty and learning in a technologically dynamic industry: Seed density in U.S. maize

Abstract

The large and sustained yield gains achieved since the introduction of maize hybrids in the 1930s (about 1.8 bushels per acre per year) have been accompanied by a remarkably parallel and steady increase in seeding density. This increase occurred in an environment characterized by rapid technological innovation, including genetic engineering, and commercial hybrid varieties with short life cycles. An important question, then, is whether and how breeders and farmers have learned about the optimal planting density. In this paper, we use unique and detailed U.S. farm-level data, consisting of more than 400,000 planting choices from 1995-2016, to assess the nature of learning about seeding density. Importantly, we control for unobserved confounders through both hybrid and farm-level fixed effects. We find that the variance in planting rates for a given hybrid has decreased over time, and that farmers tend to plant a given variety at higher rates over time. This is consistent with Bayesian learning in which risk-neutral farmers possess priors consistently below the true optimal rate. We cast doubt on risk aversion as a credible explanation for this finding by analyzing the contrasting evolution of soybean planting rates (a crop with exogenously different agronomic determinants of seed density). We interpret our results as evidence of inertia: the initial bias in maize farmers' priors is tilted towards the optimal planting rates of varieties planted in the past. One implication of the finding that farmers historically underinvested in seeding rates is that eliminating this tendency could result in productivity gains.