· Automotive Data Group · Market Intelligence · 4 min read
How to read US vehicle export statistics without fooling yourself
US federal trade data is free, detailed and monthly. It is also easy to misread in four specific ways, each of which produces a confident number that is wrong.
Almost everything said publicly about the US used-vehicle export trade is anecdote. A broker says volumes to a particular market have collapsed; three months later everyone repeats it; nobody has looked. This is odd, because the underlying statistics are published monthly, free, by the United States Census Bureau, at a level of detail most industries would envy.
The data is genuinely good. The problem is that it is easy to misread, and a misread produces a number that sounds authoritative. Here are the four errors we see most often.
Error one: summing the rollups
The Census trade API returns country-level rows alongside aggregate rows — world totals, regional groupings, “all countries except” constructions. They share the same dimension as the countries themselves. Sum the response naively and you will roughly double your total, or worse.
Filter aggregates out explicitly before you aggregate anything. If your world total comes out at twice the published world total, this is why.
Error two: assuming the category means what you want it to mean
This is the important one, and it is where most published analysis of this trade quietly falls over.
Harmonised System heading 8703 covers passenger motor vehicles. It does not distinguish new from used. If you pull 8703 exports to a given country and describe the result as “used vehicle exports”, you have published a number that includes every new car shipped in the same period — and for some destinations that is most of the total.
Splitting new from used requires working at a finer Schedule B subheading, and the correct code is worth confirming directly against the current Schedule B rather than inheriting it from a blog post. We include this warning in our own tooling because it is the mistake that is easiest to make and hardest to notice.
Error three: reading value as volume
Census export data is denominated in dollars. Vehicle counts are a different series and are not always available at the same granularity.
A corridor whose dollar value rose fifteen percent may have shipped fewer vehicles at higher prices. In a period of significant used-vehicle price inflation — which the last several years have supplied — value and volume can move in opposite directions for quarters at a time. If your commercial question is “how many cars”, a dollar figure is a proxy, and you should say so.
Error four: mistaking the destination for the destination
Trade statistics record the country of ultimate destination as declared on the export filing. For vehicle exports this is a genuinely soft number. A car declared to one country may be trucked across a border within a week of clearing the port, and the transit hubs that serve this trade are exactly the places where that happens most.
This does not make the data useless. It makes it directional. Treat a hub country’s figures as covering its hinterland, and be careful about drawing conclusions on the difference between two neighbouring markets when one of them is the other’s port of entry.
What the data is genuinely good for
Having listed the traps, the case for using it anyway is strong:
- Direction and magnitude of change. Month-over-month and year-over-year movement in a corridor is reliable even when the absolute level is fuzzy.
- Seasonality. Several years of monthly data reveals a seasonal shape most operators feel intuitively but have never quantified.
- Structural breaks. Sanctions, currency moves, regulatory changes and conflict all leave clear signatures. Dating the break precisely is often more useful commercially than the level itself.
- Relative corridor size. Which destinations matter, in what order. Wrong in absolute terms, broadly right in rank.
The method
Pull the series yourself, filter the rollups, be explicit about which HS or Schedule B code you used and what it does and does not contain, state whether you are reporting value or volume, and publish the query alongside the chart. Anyone who then disagrees with your number can show you why, which is the entire point.
We maintain a small script that does the pull and writes the result to disk with its provenance attached. It ships with our site source, warning comments included.
Automotive Data Group produces corridor-level market intelligence for the US used-vehicle export trade. If you have a question the public data cannot answer on its own, get in touch.