The Eyes in the Sky Don’t See Everything

More than 8,000 satellites have been sent into orbit since Sputnik 1 launched in 1957. Companies, such as Planet Labs, Spire Global, and SpaceKnow track planes, ships, roads, buildings, and containers worldwide. Tellus Labs tracks global crops. Orbital Insight and UrsaSpace Systems provide coverage on oil storage.

Meanwhile, the economic data provided by several countries, notably China and India — but also Hungary, Italy, Ukraine, and Argentina — remains mired in suspicion.

Acting as a headwind against government obfuscation and a tailwind for the satellite images attempting to thwart it are advances in data science, artificial intelligence, and machine learning. This confluence of trends has led to another form of congestion— in trading strategies predicting macro numbers based on alternative data. What’s lacking, however, is any consensus as to whether such techniques actually work.

With satellite imagery becoming such a prevalent part of fundamental analysis, the trio of authors wanted to probe levels of effectiveness —are financial markets now able to anticipate a government’s macro numbers before they are announced?

To shed light on satellite-based forecasts, the authors looked to the clouds.

“No satellite using optical imagery ... can observe activity on the ground, if its view is obscured by clouds,” the authors said, explaining how cloud cover could be used to “switch off” the ability of these satellite-based forecasters to provide reliable estimates.

Once they found their overcast-skies-versus-clear measuring technique, the research team devised a study of the market for US crude oil.

Crude is typically transported through pipelines. In a handful of places along the supply chain —small towns like Cushing, Oklahoma, for example —multiple pipelines intersect, creating centralized storage hubs.

“Such hubs host a substantial proportion of oil storage facilities,” the authors said. “For example, ten storage locations used in our analysis account for up to a third of the entire U.S. inventory storage capacity.”

Because oil is often stored in tanks with floating roofs, satellites are able to observe the differences in the shadows cast inside each tank. A simple geometry of shadows makes it possible to gauge tank content levels. But shadows can’t be observed if clouds cover the storage location. Then the difference between what a satellite sees on a clear day, as opposed to a cloudy day, would be significant.

Exploiting the tell-tale bottleneck often key US crude storage hubs, accounting for satellite travel itineraries, and focusing on the flow of information during oil-trading sessions, the team came up with a novel experiment to gauge the reliability of satellite data connected with the key hubs.

Energy trading in large part revolves around weekly West Texas Intermediary crude inventory surveys taken every Friday by the US Energy Information Administration (EIA). The aggregate inventory numbers are released to the public on the following Wednesday at 10:30 a.m. ET. Not surprisingly, this weekly macro-informational event has turned into a key area for satellite-based forecasting around the world.

According to the authors, satellite-based estimates of oil inventories become less precise when clouds cover key supply hubs. Thus, any question on the efficacy of satellite-based estimates can be answered by examining differences in the oil price impact of the EIA’s announcement in cloudy versus clear weeks.

Days were retroactively deemed completely cloudy if they fell into the top quartile of cloudiest days in the hubs’ regions, per public weather data.

Then the authors pulled Thomson ticker data for liquid front-month futures, in 15-minute increments, to create 261 weekly observations (2014–2018),all of which were divided into various time horizons (e.g., a one-hour window versus a two-, three-, or four-hour windows) to smooth out statistical anomalies.

Unknowns, such as when exactly satellites pass over the hubs, were overcome by averaging. The one major given was that satellite data would be effective only on clear days.

So, what would happen during cloudy stretches?

As suspected, investors get a more reliable signal when it’s clear over key hubs, meaning the EIA’s Wednesday announcement doesn’t create much of a stir; in such weeks, the signals having been gleaned and pre-traded into the price.

But when the ten US sample hubs have predominantly cloudy skies, the satellite-based estimates are less accurate. Completely cloudy skies increased the sampling error by 72.9% relative to completely clear days, “indicating an economically non-trivial relation between our cloudiness measure and the observability of oil storage tanks.”

Prices react more to the Wednesday announcement on cloudy weeks, when the signal from satellite-based estimates is exceedingly noisy. “The respective government announcement has a substantially larger, and statistically significant, price impact,” the authors asserted. The announcement impact in cloudy weeks is five to ten times bigger in magnitude, as much as 55 basis points worth of returns per standard deviation of inventory change. In such weeks, therefore, “the resolution of macro uncertainty is lumpier, resulting in larger price jumps.”

The authors also applied their idea to satellite-based estimates of Chinese manufacturing (PMI data), which is also concentrated in four eastern provinces. As with US crude oil, they found reductions in the price impact of PMI announcements on Chinese stock prices in clear periods.

That there are demonstrable limitations to satellite data might prove sobering for traders rushing into this space over zealously. The authors, however, make an equally strong case for measurable effectiveness of such indicators in clear conditions.

For investors seeking to understand macroeconomic conditions, they now have fresh evidence assuring them that they are no longer beholden solely to government announcements.

“Creative solutions focusing on new data and techniques are providing an informative alter- native in many contexts,” the authors said by email. “And we could expect such applications to expand into many more contexts.”

Given that some of these alternative data sources provide data much more frequently relative to quarterly government announcements, markets can expect to have a better sense of economic conditions on a more regular basis. “This,” the authors added, “can reduce price jumps, associated with surprises arising out of the infrequent nature of government data announcements.”