This paper develops a novel procedure for proxying economic activity with day-time satellite imagery across time periods and spatial units, for which reliable data on economic activity are otherwise not available. In developing this unique proxy, we apply machine-learning techniques to a...