Satellite data significantly improves oil demand forecasting

A new ECB study demonstrates that satellite observations of nitrogen dioxide (NO₂), a pollutant from fossil fuel combustion, significantly enhance the real-time nowcasting of oil demand.

Integrating NO₂ data into forecasting models improves accuracy by an average of 25 percent compared to simple autoregressive benchmarks and 20 percent against more sophisticated models.

These accuracy gains are particularly pronounced during crisis periods, such as the pandemic, but remain robust in stable times.

The research, covering 10 advanced and emerging economies including the United States, China, and India, which together represent 60 percent of global GDP and NO₂ emissions, highlights the unique advantages of satellite data: daily timeliness, global coverage, high spatial granularity, and free access.

Non-linear models, especially neural networks, yield the largest improvements, underscoring the complex link between energy demand and pollution.

Official oil consumption statistics are typically published with a three-month delay, leaving policymakers without crucial real-time information during periods of rapid economic change.

The study addresses this gap by leveraging satellite-based NO₂ measurements.

Raw satellite data, however, requires extensive cleaning and adjustment due to distortions from cloud cover, snow, or solar angles.

Researchers apply standard filtering procedures, including spatiotemporal averaging, to produce smooth, reliable NO₂ pollution indices.

The predictive power of NO₂ was tested using both linear OLS models and non-linear techniques, such as neural networks, comparing them against traditional predictors like industrial activity, energy prices, and vehicle registrations.

The findings confirm NO₂'s superior performance, even against other high-frequency indicators like Google Mobility indices.