How weather affects the decomposition of total factor productivity in U.S. agriculture

Abstract

This study illustrates and quantifies how overlooking the impact of weather shocks can affect the measurement and decomposition of agricultural total factor productivity (TFP) change. The underlying technology is represented by a flexible input distance function with quasi-fixed inputs estimated with Bayesian methods. Using agricultural production and weather data for 16 states in the Pacific Region, Central Region, and Southern Plains of the United States, we estimate TFP change as the direct sum of multiple components, including a net weather effect. To assess the role of weather, we conduct a comparative analysis based on two distinct sets of input and output variables. A traditional set of variables that ignore weather variations, and a new set of “weather-filtered” variables that represent input and output levels that would have been chosen under average weather conditions. From this comparative analysis, we derive biases in the decomposition of TFP growth from the omission of weather shocks. We find that weather shocks accelerated productivity growth in 12 out of 16 states by the equivalent of 11.4% of their group-average TFP growth, but slowed down productivity by the equivalent of 6.5% of the group-average TFP growth in the other four states (located in the Northern-most part of the country). We also find substantial biases in the estimated contribution of technical change, scale effects, technical efficiency change, and output allocation effects to TFP growth (varying in magnitude and direction across regions) when weather effects are excluded from the model. This is the first study to present estimates of those biases based on a counterfactual analysis. One major implication from our study is that the official USDA's measures of TFP change would appear to overestimate the rate of productivity growth in U.S. agriculture stemming from technical change, market forces, agricultural policies, and other nonweather drivers.