The Economic Impact of Low- and High-Frequency Temperature Changes

TL;DR

This paper investigates how low- and high-frequency temperature variations differently affect economic growth, finding significant effects mainly from low-frequency changes in Europe and globally, with methodological considerations for accurate inference.

Contribution

It introduces a robust interactive fixed effect approach to distinguish the impacts of temperature frequency components on economic outcomes, addressing previous methodological limitations.

Findings

Low-frequency temperature increases are linked to about 1.3 percentage points reduction in growth in Europe and globally.

High-frequency temperature changes have a marginal effect on U.S. economic growth.

Methodological improvements reveal the importance of controlling for common time effects in temperature-economic impact studies.

Abstract

Variations in the low- and high-frequency components of temperature may have distinct impacts on economic outcomes. Parametric and non-parametric estimates from three panels of data all find significant heterogeneity in the relative importance of the two components, but there is clear evidence in each panel of a common, slowly evolving low-frequency factor that is highly correlated with the low-frequency factor of economic activity. In regressions that quantify the output effects of the components, we find that one-way clustered standard errors often lead to size distortions, and that an additive fixed effect specification does not adequately control for common time effects. Using bootstrap inference to assess estimates from our preferred interactive fixed effect specification, we only find a marginally significant effect of the high-frequency component on growth in the U.S. panel. However, the effect of the low-frequency component is significant in the European and International panels, suggesting that the increase in the low-frequency temperature component over the post-1980 period is associated with a reduction in economic growth of approximately 1.3 percentage points. The findings are corroborated by time series estimation using data at the unit and national levels.