Econometric models of temperature effects on country-level GDP are increasingly used to inform global warming damage assessments. But theory does not prescribe estimable forms of this relationship, yielding discretion to researchers and generating considerable model uncertainty. We employ model cross validation to: (1) assess out-of-sample predictive accuracy of 800 model variants; (2) identify the set with superior predictive performance; and (3) characterize magnitudes of model and sampling uncertainty. Model uncertainty is comparable in magnitude to sampling uncertainty, yielding among GDP growth models a 95% confidence interval for GDP impacts in 2100 of -84% to +359%. GDP levels models yield a much narrower 95% confidence interval of -8.5% to +1.8% and centered around losses of 1–3%, consistent with damage functions of major integrated assessment models. Accounting for model uncertainty, we identify statistically significant marginal effects of hot temperatures on the levels of poor country GDP and agricultural GDP. We do not identify statistically significant temperature effects on GDP growth.
Note: This paper was originally published in July 2018 and revised in November 2018 and October 2020.