Abstract
Using a global panel of public firms over more than two decades, we estimate firm-level impacts of abnormal seasonal temperature and precipitation through a flexible machine learning framework with theoretically motivated firm characteristics. While prior firm-level studies typically only find modest average weather effects using linear specifications, our approach uncovers economically large and statistically significant effects on sales, efficiency, profitability, and costs, which are concentrated in weather-exposed industries, labor-intensive and older firms, and firms in less developed regions. Model-implied weather effects also predict earnings announcement returns, suggesting investors underreact to abnormal weather exposure. Finally, using climate projections for 2049, we estimate economically meaningful negative long-run impacts that vary substantially across regions and firms, highlighting the importance of nonlinearities and firm heterogeneity for climate risk assessment.