Doubtful of China’s economic numbers? Satellite data and AI can help

The Chinese government is notorious for doctoring its official statistics, from using messed-up methodology, to not reporting some key metrics, to out-and-out fabrication. Last year, the governor of one of China’s rust-belt regions admitted for the first time that the province had inflated GDP figures for years. “Officials produce the numbers, and the numbers produce officials,” he said at the time, referring to the idea that massaging data can help one get ahead in Chinese officialdom.

As a result, China’s data releases, while closely watched, are often greeted with skepticism by economists (paywall). But slowly, technology is handing analysts, economic experts and investors new tools that allow them to fact-check official numbers and pronouncements. Generated with the help of satellite imagery and artificial intelligence, these alternatives to the official data are revealing the secrets of China’s real economy.

Ahead of China’s next big data release Tuesday (April 17), of economic growth for the first quarter of this year, here’s a look at what it’s possible to learn about China’s economy from these sources. It’s going to be a long time before these new sources of macroeconomic data can supplant government data, if ever—because governments have the ability to get granular from people and companies at all level of the economy—but they’re still very useful counterpoints for investors.

“Many of our financial customers are looking to gather as much data as possible to make market decisions, so our data provides an additional point for them to consider,” said James Crawford, a former NASA and Google engineer who founded California-based data service firm Orbital Insight in 2013. “Customers can use our data to help inform their decisions knowing that it is grounded in observable truth, is updated frequently, and is methodologically the same across regions.”

Tracking manufacturing growth by satellite

To measure China’s manufacturing expansion, investors usually pay attention to two sets of monthly data called the Purchasing Managers’ Index (PMI). One is released by China’s national statistic bureau, and the other is based on surveys performed by financial service company IHS Markit, and sponsored by news organization Caixin. Both seek information from companies on new orders and stock levels, among other things.

On a scale of 0 to 100, a PMI reading above 50 is a sign of economic expansion, and a reading below 50 represents contraction. Oftentimes the official and Caixin numbers go in different directions due to differences in the firms they sample. The official PMI generally puts more weight on state-owned and large companies, while the Caixin figure tends to focus on private and smaller firms; the former also consists of a far larger sample size, about 3,000 firms.

The Chinese Satellite Manufacturing Index (SMI), created by US company SpaceKnow since 2016, offers a different approach. The index uses 2.2 billion satellite snapshots taken of more than 6,000 industrial areas spanning half a million square kilometers of Chinese territory to get a numeric measure of how well the country’s manufacturing sector is doing. Published every other week, the SMI is read in the same way as the PMI, with the 50-point threshold separating expansion from contradiction.

While the official PMI edged up to 51.5 in March, the Caixin PMI hit a four-month low of 51.0. The SMI for March was at 51.7, up 0.1 from a month earlier.

The SMI is valuable because “it’s independent of the market and the Chinese government, and it provides a different look of the same reality,” said Pavel Machalek, co-founder of SpaceKnow. The SMI is a complimentary but replacement of the PMI readings, he said.

According to Machalke, SpaceKnow uses machine-learning algorithms to analyze specific markers of economic activity from satellite imagery of inventory goods, real estate and surface materials, among other things. For example, if an area of grass-covered land later shows up covered by concrete, Machalek explains, it’s a sign of manufacturing expansion over the time period.

Verifying China’s pollution curbs at Australia’s ports

As part of its anti-pollution drive, China last year ordered over two dozen northern cities surrounding Beijing to cut steel output by up to half during the winter heating season from mid-November to mid-March. The policy is expected to see local steel factories ditch low-grade iron ore for higher grades of ore, to make their steelmaking process cleaner and more efficient.

So is the government directive really working? Quandl, a Toronto-based data service company, tracks shipping data to estimate iron ore sales for Australia’s three biggest mining companies—Rio Tinto (RIO), BHP Billiton (BHP) and Fortescue Metals Group (FMG)—which altogether sell 80% of their iron ore to China. (Australia is the top iron ore exporter to China.)

By counting vessel numbers at Australian ports dedicated to ore exports, Quandl is able to create daily estimates for iron ore sales for each of the three companies, weeks to months before they publish relevant data in their stock filings. Here’s what Lilian Lau, a researcher with Quandl, found out about the trio’s iron ore exports to China from November to February, compared to the same period a year ago:

RIO and BHP, which sell higher-grade iron, saw a surge in their exports to China during the November-February period, while FMG, whose iron ores are mostly low-grade, suffered a decline in sales to China. Lau explains she looks into data from November to February because it takes about three weeks for ships to travel from Australia to China.

“All the above confirms Chinese companies are looking to use higher grade [ore],” says Lau. In other words, she says, the Chinese crackdown is working.

Tracking oil reserves by shadows

The price of oil is volatile in large part thanks to a lack of transparency in the market. We might know how much oil is stored in Rotterdam, one of the world’s biggest oil shipment hubs, but other sources of oil can easily shake up the market without warning, since their storage inventories and capacity are not reliably reported. Beijing is among the many governments worldwide that hold a Strategic Petroleum Reserve (SPR) for national security purposes. However, unlike the US, which regularly reports its SPR data, China rarely does so.

Screen Shot 2018-04-13 at 3.15.26 PM
Orbital Insight‘s computer-vision algorithm automatically detects the shadows from the walls of floating-roof oil tanks. (Orbital Insight)

Orbital Insight uses satellite imagery to detect China’s oil tanks above the ground. Because these tanks typically have floating lids that rise and fall with oil storage to minimize evaporation, the size of the crescent moon-like shadows from the walls of the reservoirs is exactly a reflection of how much oil is in them. When a tank is full, its shadow is the smallest; and when it’s empty, the shadow is the biggest. Using this method, Orbital Insight creates a near real-time estimate of China’s oil supply, and compares it with the monthly figures published by state news agency Xinhua.

It’s worth noting that the Orbital Insight data includes all known floating-roof tanks, government and commercial, while Xinhua—the best official data source—only reports oil reserves in commercial tanks, potentially including fixed roof, floating roof, and underground ones. In that sense, the chart above is not really an apples-to-apples comparison, but gives a more complete look of China’s oil supply. “We have found that floating roof tank storage often serves as a good directional proxy for overall storage in a given region,” said the firm’s founder James Crawford.

Using the same methodology, Orbital Insight covers oil storage from the world’s major economies including the US and OPEC countries. “The markets are never quite sure which governments to trust, so our project has the advantage of being completely objective,” Crawford said.

home our picks popular latest obsessions search