For decades government-issued reports and other traditional forms of data, such as gross domestic product (GDP), have been used to shape forecasts of countries’ export growth. These forecasts depend largely on estimating foreign demand and income.

This common approach has at least two glaring limitations:

  • Timeliness. In many cases, traditionally issued GDP reports reflect conditions that were in place one to two quarters earlier because of data availability. In other words, it’s not real-time information.
  • Reliability. Studies show that the quality, depth and accuracy of reports on growth in foreign GDP—a key element in forecasting demand—can vary, especially in developing countries. As a result, experts evaluating this data might not be getting accurate pictures of what is happening on the ground.

So, what if there were a way to get truly accurate, real-time snapshots? The answer might have been found—quite literally—in outer space.

Jun Nie, a senior economist at the Federal Reserve Bank of Kansas City, and research associate Amy Oksol this year published a study asserting that nighttime satellite pictures of lights on the ground can be reliably used in forecasting U.S. export growth. They studied monthly satellite pictures of several countries’ nighttime lights. They found greater accuracy in the data derived from those images when compared with export growth forecasts based on traditional methods—specifically GDP analysis and a “random walk.”

“Nighttime lights as viewed from satellites make up a unique dataset that provides information on nearly every place on earth,” Nie and Oksol explain in their research paper. “Satellite cameras take pictures of the entire planet at night (so lights can be better seen) and filter the images for various anomalies such as clouds and fires.”

Because of significant advances in satellite technology over the years, these images have a high level of detail. That level of detail makes it possible to assign a luminosity value ranging from zero to 63—with zero being unlit and 63 being maximum light. By using special mapping software, Nie and Oksol were able to use luminosity values to create a numerical lights index for a particular region. The software could then calculate the amount of nighttime light in that region in a given period. These luminosity values also could be aggregated across cities, regions, countries or other geographic areas for a particular point in time.

To test the accuracy of the night-lights approach, Nie and Oksol created a lights index for major U.S. trading partners and compared forecasts using the lights information with traditional quarterly forecasts of GDP growth and the “random walk” approach. To gauge accuracy of each method, Nie and Oksol compared the root-mean-square error (RMSE), a common forecasting metric of the average deviation of the forecast from the actual value.

The conclusions:

  • Monthly lights data generated significantly fewer forecasting errors than the quarterly GDP model.
  • With satellite pictures now available on a daily basis, future forecasting models could be improved by exploiting this higher frequency and truly real-time availability.

Although Nie believes the Kansas City Fed study, which focused on monthly images, is perhaps the first deep analysis of its kind, he points out that the concept of connecting nighttime lights with economic activity is not entirely new. Various studies dating to the 1990s—when only annual satellite images were available—have shown that pictures of night lights can be helpful in measuring a country’s GDP. This is because the amount of light in a particular area correlates to the area’s income levels.

The Earth Observation Group at the National Oceanic and Atmospheric Administration (NOAA) began making monthly satellite images available in 2012, and these pictures are what Nie and Oksol examined in their study. Daily images became available in 2017. Although further analysis will be necessary, this daily availability is “truly real-time” and “opens up some new opportunities” in forecasting, Nie said.

“Overall this dataset is pretty new…and that’s exciting,” Nie said. “There are a lot of things to explore in the future.”

Further resources: Read the research by Jun Nie and Amy Oksol.