Publications

Abstract

There is a long history of using weather measures as explanatory variables in statistical models. For example, Fisher (1925) examined the effects of rainfall on wheat yields, andWright (1928) used weather as an instrumental variable to identify a demand function for oils. Because weather is exogenous and random in most economic applications, it acts like a “natural experiment” and thus in some settings allows researchers to identify statistically the causal effect of one variable on an economic outcome of interest (Angrist and Krueger 2001). The relatively recent literature on the economic impacts of climate change has turned the spotlight onto quantifying the effect of climate on a number of economic outcomes of interest (e.g., agricultural yields, mortality rates, electricity and water demand). This literature has often found a nonlinear relationship between climate and these outcomes, with extremely warm temperatures being especially important (e.g., Schlenker and Roberts 2009). Climate is a long average of weather at a given location. To identify the causal effect of climate on these outcomes, the literature has generally relied on either climate normals (i.e., long averages of observed weather in a cross-sectional setting) or day-to-day (or year-to-year) fluctuations in observed weather as explanatory variables across time and space. The econometrician’s
choice of a weather versus a climate measure as an explanatory variable critically affects the interpretation of the estimated coefficients in the econometric model: that is, whether the outcome is a true climate response or a short-run weather elasticity.

Download File

Download a PDF (588KB)