This page in:
Global satellite readings have revolutionized our ability to estimate economic activity in situations where government data is unreliable. Currently, the primary signal used to estimate economic activity is the radiance of night lights. In a recent paper, we rely on satellite measurements of nitrogen dioxide (NO2), a direct byproduct of combustion. By doing so, we manage to redress long recognized shortfalls in both national accounts and night light measures.
Our findings show that although night lights remain the preferable indicator of long-term growth in countries with weak statistical systems, NO2 provides additional information on short-term fluctuations in economic activity . We also find that NO2 improves the measurement of economic activity everywhere — including advanced economies. We are also able to re-rank national statistical systems based on their quality — in some cases, significantly — relative to commonly used classifications.
NO2 has clear advantages over night lights as an indicator of economic activity. It is always measured during daytime — when businesses are open and people are at work. It also stays close to the source of emissions —such as engines, vehicles or chimneys — and has a short atmospheric life – less than one day. Importantly, the data are published online shortly after measurement is taken.
Satellite NO2 readings should not be confused with those from ground-level monitoring stations. These are measured infrequently, especially in developing countries. They also have an uneven spatial distribution—it is not the same to measure pollution in an urban intersection or in the countryside. Also, station readings may be altered for political motives, such as the need to meet government targets.
In our paper we use vertical column densities of tropospheric NO2 from an ozone monitoring instrument onboard NASA’s Aura satellite. These columns have a high spatial resolution – equivalent to 13 km × 24 km. Aura is part of the “A-Train”, a line of satellites moving in a pole-to-pole route that crosses the Equator always at around 1:30 PM local time.
An illustration of the information content of NO2 readings can be provided by lockdowns during the Covid-19 pandemic. Figure 1 shows the change in NO2 emissions between the first quarters of 2019 and 2020 in China. Hubei province, where Wuhan is located, had the most significant downturns (blue). Our paper contains similar maps for Europe and the United States. They reveal the different intensity of lockdowns across jurisdictions, presumably for both epidemiological and political reasons, as shown in a previous use of this indicator.
Our paper not only improves on previous economic activity estimates because of its chosen indicator, but also because of its statistical methodology. We build on the error-measurement framework utilized by Chen and Nordhaus and Henderson et al. in their seminal work on night lights. But we introduce a data-driven identification strategy that does not require assumptions that official statistics are always accurate in advanced economies .
Figure 1. Change in NO2 emissions during Covid-19 lockdowns
Our methodological contribution can be described intuitively. Both official GDP and NO2 are correlated with the true, unobservable level of economic activity. In the case of GDP (the first relationship), measurement error declines with the quality of the statistical system. For consistency with previous analyses, we rank this quality based on the expanded World Penn Tables, with category “A” corresponding to the most sophisticated systems, and category “E” to countries that have basically no statistical organizations.
As for NO2 (the second relationship), economic activity can lead to a different intensity of emissions across space and over time. For example, countries engaged in the service sector can be expected to have lower emissions than those specializing in manufacturing or oil extraction . Similarly, we show that countries in categories A and B have trend NO2 declining over time, a possible outcome of their decarbonization efforts.
We combine these two relationships by eliminating the true economic activity, which allows us to express NO2 directly as a function of GDP plus an error term. This is a relationship that can be estimated empirically, as both readings and GDP are observable. But standard regression results would be biased because the error term in this shortcut relationship is correlated with GDP.
To address this identification problem, estimates based on night lights assume that there was no bias in the case of A-rated countries. Instead, we use night lights as an instrument to identify the estimated relationship. The validity of the instrument comes from NO2 and night lights measured at different times, by different satellites, and with different sensors. In addition, auroral activity and the phases of the moon are significant concerns for night lights readings, but not NO2.
Thanks to this unbiased estimation, we have two measures of true economic activity: official GDP and its NO2-based proxy. The error-correction framework implies that the level of true economic activity can be computed as a weighted average of these two measures, with the optimal weight on official GDP is lower the weaker the country’s statistical system.
Our findings show that for A-rated countries like the United States, fully 9 percent of the optimal measure of true economic activity should come from the NO2-based proxy. For B ratings, including Spain and Germany, the optimal weight increases to 28 percent. In C-rated countries, including China, it goes up to 40 percent. In the D and E categories, including Zimbabwe and the Republic of Yemen, it reaches 64 percent and 80 percent, respectively.
The difference between this optimal measure of true economic activity and GDP captures the extent of misstatement in official statistics. In absolute value, the average misstatement ranges from 0.25 percentage points in A-rated countries to 5 percentage points in the E category—substantially more than estimated based on night lights.
Beyond averages, this optimal measure of true economic activity uncovers many significant errors for specific countries and years:
- Pandemic. In 2020, official GDP fell by 1 to 2 percentage points more than true economic activity in several advanced economies with strong statistical systems.
- Multinationals. In another advanced economy, including intangible intellectual property in national accounts led to official GDP overestimating true economic activity by up to 7 percentage points.
- Financial crises. During a major debt crisis, the official GDP of a high-income country declined by 3 percentage points more than its true economic activity.
- Oil shocks. After the opening of a major pipeline, true economic activity in an oil-exporting country increased 10 percentage points less than its official GDP.
- Conflict. In several countries going through war the fall in true economic activity was 10 percentage points less than indicated by official GDP.
- Hyperinflation. Official GDP overstated the fall in true economic activity by 43 percentage points in one middle-income country, but it understated it by 47 percentage points in another.
The variance of the annual misstatement of true economic activity over time also allows us to assess the quality of statistical systems across countries. This data-driven ranking leads to a significant reshuffling compared to commonly used classifications.
Other possible applications of the proposed methodology include the measurement of subnational GDP in large countries, the generation of quarterly estimates of GDP growth in countries where official figures are only annual, and the “nowcasting” of GDP while waiting for official figures to be released.