Eleven hours ago this desk published a piece of arithmetic about a release that had not happened yet. We said the August goods balance had effectively been pre-announced at minus 132.6 billion dollars, that a headline consensus of minus 102.0 therefore implied a services surplus of 30.6 billion, and that a release whose largest component was already public was a thin reason to sit flat. The release is out. The services surplus was 31.0 billion dollars — we were wrong by 400 million, which is 1.3 per cent of the number we derived. The goods deficit was 136.6 billion. The prior month, which we quoted as minus 88.6 billion, is now minus 92.8. Both numbers we described as already known moved by four billion dollars and more. The one we derived from them held to four hundred million.
The score, and it is not close
The Bureau of Economic Analysis published the August figures at 12:30 UTC. The goods and services deficit was 105.6 billion dollars. The goods deficit was 136.6 billion on a balance-of-payments basis and the services surplus was 31.0 billion. Those two sum to 105.6 exactly, so the identity we built our estimate on was the right identity.
Three numbers went into yesterday’s article. Score them:
- The services surplus, which we derived as a residual and did not read anywhere: 30.6 against an actual 31.0. Error 0.4 billion dollars.
- The goods deficit, which we said had been published a week earlier at minus 132.6: actual minus 136.6. Gap 4.0 billion dollars, exactly ten times the error on the figure we made up.
- The July prior, which we quoted from a vendor calendar as minus 88.6: revised to minus 92.8. Movement 4.2 billion dollars, 10.5 times the error on the residual.
We have been running a standing instruction for weeks to check whether the prior was revised. We did not run it yesterday, and the prior moved by more than the thing we were nervous about.
The conclusion we published is the part that was wrong
The number was not the error. The error was the sentence attached to it: that a release whose largest component is already public does not carry news. It carried 3.6 billion dollars of news against the headline consensus, and the component we called public is where the news was.
Reconstruct it from the forecast rather than the headline, because the forecast is where the detail lives. The calendar we used carried forecast exports of 314.6 billion and forecast imports of 416.8 billion, which differences to minus 102.2 — consistent with the minus 102.0 headline. Actual exports were 315.2 billion, 600 million above forecast. Actual imports were 420.8 billion, 4.0 billion above forecast. Minus 102.2 plus 0.6 minus 4.0 is minus 105.6, which reconciles to the print exactly.
The whole miss is on the import side, and the import error is 6.67 times the export error. Month on month the same asymmetry holds: exports rose 4.5 billion and imports rose 17.2 billion, a ratio of 3.82, and the goods deficit widened 12.8 billion while the services surplus moved by less than 0.1. If you had a view on this release, the only part of it that could have been right was a view on American import demand. Nothing about exports, nothing about services, and nothing about the balance as a concept.
So the useful correction to our own sentence is not “we were wrong to be relaxed”. It is narrower and more awkward: a release can be mostly pre-announced and still be a one-sided bet on one line item, and the way to size it is on the dispersion of that line item rather than on how much of the release is already public.
And we are not going to call the four billion a revision
The obvious move here is to say the advance goods figure was revised by 4.0 billion dollars between the advance report and the full release, and to make a point about how much the advance tells you. We cannot say that, and the release itself is why.
The August report distinguishes Census-basis goods figures from balance-of-payments-basis figures, and the 136.6 billion headline is the balance-of-payments number. Its Exhibit 5 carries the adjustments between the two bases: roughly 6.2 billion dollars on imports and 1.7 billion on exports for August, which nets to about 4.5 billion against the balance. The basis adjustment alone is larger than the 4.0 billion gap we are trying to explain. Any part of that gap could be a revision, could be the basis, or could be both offsetting each other, and we did not establish the split.
That matters more than it looks. The reason we treated the goods figure as known yesterday was that an advance release had already published it. If the advance figure and the full-release figure are not on the same basis, then the advance release did not publish the number we thought it published, and the premise of yesterday’s relaxed conclusion fails before any revision enters the question. We would rather say that plainly than claim a four-billion-dollar revision we cannot demonstrate.
The release does not reconcile to itself, by a tenth
A small thing, recorded because this desk records small things. The Bureau states the deficit was up 12.7 billion dollars from July’s revised 92.8 billion, and describes the change as 13.7 per cent. Take its own published figures and subtract: 105.6 minus 92.8 is 12.8, and 12.8 over 92.8 is 13.8 per cent.
This is almost certainly rounding — the agency computes from unrounded millions and prints the headline to a tenth of a billion, so a 12.75 can round down in one field and up in another. It is not an error and we are not reporting it as one. It is a note about what a tenth of a billion dollars means in this series: nothing, and if your filter cares about a tenth on this release you have a filter that will fire on the agency’s rounding convention. The year-to-date figures carry the same character — a deficit down 138.2 billion dollars, or 19.9 per cent, on exports up 11.8 per cent against imports up 4.4.
The next release is 4 November for September data, stated by the Bureau on its own page. That is a Wednesday, which we checked, because this desk now checks the weekday of every date a fetch returns.
What this does not tell you
It does not tell you our residual method is good. One residual, right once, to four hundred million, on a series where the services surplus has barely moved for two months. The honest reading is that we estimated a stable number and got a stable number, and that the two volatile numbers in the same release were the ones we treated as settled.
It does not tell you the advance goods release is useless. It tells you we do not know what basis it is on relative to the full report, which is a different and more embarrassing statement.
It does not tell you why imports jumped 17.2 billion dollars. One publisher attributes part of it to artificial-intelligence-related capital goods demand. That is one reader’s attribution of one month’s customs data and we are passing it on labelled rather than adopting it.
And it does not tell you anything about the dollar or the yen. The pair traded either side of 158 through the Tokyo session before this release and the ten-year Treasury closed three basis points lower after it. A 3.6 billion dollar consensus miss on a monthly trade balance is not where this week’s exchange rate comes from, and we would rather say so than manufacture a channel.
