Climate Change and Agriculture Reconsidered

Despite the existence of a large and growing literature on the potential impact of climate change on agriculture, there still exists some disagreement about the magnitude and even the sign. Our own research suggests that the impact on U.S. agriculture is likely to be strongly negative, based on a series of studies in which we link farmland values to climate variables, and crop yields to both climate and yearly weather variables. Results are significant, robust, and consistent across data sets and methods. A recent but influential study by Deschênes and Greenstone (2007b) reports dramatically different results: based on regressions of agricultural profits and yields on weather variables, they conclude that the impact of climate change will be either insignificant or positive. In this paper we reconcile these conflicting results. We present evidence showing that the differences stem mainly from three sources: (i) data and coding errors in DG’s weather data, agricultural data, and the construction of climate-change scenarios; (ii) the particular climate change scenario which is used for impact predictions; and (iii) standard errors that are biased due to spatial correlation. Correcting DG’s data and coding errors makes predictions for climate-change impacts unambiguously negative in all but one specification. The exception is a profit regression with state-byyear fixed effects where the standard errors are very large because state-by-year fixed effects absorb almost all variation in weather. Department of Agricultural and Resource Economics, University of California at Berkeley, 207 Giannini Hall #3310, Berkeley, CA 94720. Email: [email protected] and [email protected]. ♠ Department of Agricultural and Resource Economics, North Carolina State University, Box 8109, Raleigh, NC. Email: michael [email protected]. ♥ Department of Economics and School of International and Public Affairs, Columbia University, 420 West 118th Street, Room. 1308, MC 3323, New York, NY 10027. Email: [email protected]. Agriculture is the sector of the economy most directly linked to climate and thus likely to be affected by climate change. To date, however, there exists considerable disagreement about not only the magnitude of potential impacts but also the sign. A recent paper by Deschênes and Greenstone (2007b), henceforth DG, criticizes the hedonic model, a cross-sectional approach that regresses farmland values on climate variables for the US first introduced by Mendelsohn et al. (1994), and proposes to use random year-to-year weather fluctuations in a panel of US agricultural profits and yields. DG find no statistically significant relationship between U.S. agricultural profits and weather variables in the same years. DG also find no statistically significant relationship between corn and soybean yields (output per acre) and weather. They argue that if short-run weather fluctuations have no influence on agricultural profits or output, then in the long-run, when adaptations are possible, climate change is likely to have no impact or even prove beneficial. They conclude that “the preferred estimates indicate that climate change will lead to a $1.3billion (2002$), or 4.0 percent, increase in annual agricultural sector profits. [...] The basic finding of an economically and statistically small effect is robust to a wide variety of specification checks [...]. Additionally, the analysis indicates that the predicted increases in temperature and precipitation will have virtually no effect on yields among the most important crops (i.e., corn for grain and soybeans) [...].” In this comment we revisit their paper in an attempt to reconcile their findings with others in the literature, which suggest a less optimistic outcome. We present evidence showing that the differences stem mainly from three sources: (i) data and coding errors in DG’s weather data, agricultural data, and the construction of climate-change scenarios; (ii) the particular climate change scenario which is used for impact predictions; and (iii) standard errors that are biased due to spatial correlation. Correcting DG’s data and coding errors makes predictions for climate-change impacts unambiguously negative in all but one specification. The exception is a profit regression with state-by-year fixed effects where the standard errors are very large because state-by-year fixed effects absorb almost all variation in weather. The first part of DG’s paper argues that the hedonic approach does not produce robust results. We replicate the same checks using a well-specified hedonic model in the online appendix and show them to be robust. For example, in Schlenker and Roberts (2009) we find a strong relationship between corn, soybean, and cotton yields and weather. The relationship is robust and very similar if derived from time-series variations in weather or cross-sectional variations in climate. Holding fixed the locations where crops are grown, we predict yields losses of 30-46% by the end of the century under the slowest warming scenario and 63-82% under the fastest warming scenario. These predictions also accord with our research that uses the hedonic approach, e.g., (Schlenker et al. 2006).