Estimating the Economic Impacts of Climate Stress on U.S. Dairy Production: Evidence from Debiased Machine Learning and Simulation-Based Estimator Choice

Authors: Duong Ta, Mario Ortez, Bradford Mills, and Michael Cary

Department of Agricultural and Applied Economics, Virginia Tech

Draft v1.5, last updated October 19, 2026. Funded by The Dairy Alliance.

Abstract

We estimate how heat and cold exposure affect milk yield and fat content in a monthly panel of 21 major U.S. dairy states from 2003 through 2025. USDA NASS outcomes are matched to hourly ERA5-Land weather aggregated with county milk-cow inventory weights. State × calendar-month fixed effects absorb each state’s seasonal cycle and year-month fixed effects absorb national monthly shocks, so responses are identified from differences in weather across states within the same month. We compare linear and cubic regressions with LASSO and neural-network learners, with and without automatic debiasing, and select the reported estimator for each stress tail based on its performance in synthetic-data simulations: the cubic regression for heat and the debiased neural network for cold, both with two-way clustered intervals. An additional 24-hour-equivalent day at or above THI 69 lowers milk per cow by 0.29 percent, cumulated over the exposure month and three lags. The cold estimate, −0.22 percent per day below THI 39, has a 95 percent interval that includes zero. Removing heat stress would have raised production by about 1.7 percent, and removing both tails by 2.9 percent. The size of these effects depends on the functional form of the response, which the data cannot pin down. Applied uniformly to 2003–2024 production in the 48 contiguous states, the joint share is worth about $23.9 billion, or $1.09 billion per year. Evaluated at each state’s exposure, the heat share ranges from 0.7 percent of production in Oregon to 4.5 percent in Florida. Heat lowers and cold raises milk fat content. A break search finds no detectable change in the yield responses, but the heat effect on fat content weakens after the mid-2010s. Climate-normal diagnostics give suggestive but specification-sensitive evidence of adaptation.

About the project and its earlier dashboard snapshot

Recommended citation: Ta, Duong, Mario Ortez, Bradford Mills, and Michael Cary. 2026. "Estimating the Economic Impacts of Climate Stress on U.S. Dairy Production: Evidence from Debiased Machine Learning and Simulation-Based Estimator Choice." Working paper, draft v1.5, October 19.
Download Paper