Kenza Elass : kenza.elass[at]univ-amu.fr
Camille Hainnaux : camille.hainnaux[at]univ-amu.fr
Daniela Horta Saenz : daniela.horta-saenz[at]univ-amu.fr
Jade Ponsard : jade.ponsard[at]univ-amu.fr
The Covid-19 crisis has highlighted innovative high-frequency dataset allowing to measure in real-time the economic impact. In this vein, we explore how satellite data measuring the concentration of nitrogen dioxide (NO2, a pollutant emitted mainly by industrial activity) in the troposphere can help predict industrial production. We first show how such data must be adjusted for meteorological patterns which can alter data quality and pollutant emissions. We use machine learning techniques to better account for non-linearities and interactions between variables. We then find evidence that nowcasting performances for monthly industrial production are significantly improved when relying on daily NO2 data compared to benchmark models based on PMIs and AR terms. We also find evidence of heterogeneities suggesting that the contribution of daily pollution data is particularly important during “crisis” episodes and that the elasticity of NO2 pollution to industrial production for a country depends on the share of manufacturing in the value added. Available daily, free-to-use, granular and covering all countries including those with limited statistics, this paper illustrates the potential of satellite-based data for air pollution in enhancing the real-time monitoring of economic activity.