Month: March 2025

In recent days, there has been a good deal of discussion in the business press and on economics Bluesky about the Atlanta Federal Reserve Bank’s “nowcast” of GDP for the first quarter of 2025. The suggestion is that the US may already be entering a recession.

The first quarter of 2025 is, of course, ongoing; strictly speaking, 2025Q1 GDP has not happened yet. But the Atlanta Fed’s GDPNow series tries to estimate what GDP for the quarter will turn out to be, based on data that is available before the official first release of GDP numbers at the start of the following month.

The Atlanta Fed has been producing these estimates since 2011. The reason this one got so much notice is that it shows “real” (inflation-adjusted) GDP this quarter declining at an annual rate of -2.8 percent, the second lowest value its shown since they started producing it. (The lowest, of course, was for 2020Q2.)

This is obviously significant as an indicator. I think that it does incorporate genuine information about what’s going in the world, and means broadly what it seems to mean. But I want to point out an important caveat, which both suggests we shouldn’t take this number at face value and raises some interesting questions about how we measure the economy.

The Atlanta Fed’s forecast implies a decline in growth of 5.1 points relative to the 2.3 points of inflation-adjusted growth in the fourth quarter of 2024.¹ If you dig into the Atlanta Fed’s numbers a bit, you’ll find that a large part of this is consumption spending, which is projected to be essentially flat this quarter, after contributing 1.5-2 points of growth in each recent quarter. An even larger contribution, however, comes from imports, which added 0.17 points to GDP growth last quarter but are projected to subtract 3.27 points this quarter.

You can see this below, showing the contribution of each component to “real” growth in recent quarters. The last column is the Atlanta Fed’s estimate for the current quarter.

The fall in consumption spending, and the zero real growth in investment spending, represent, I think genuine developments in the world. But I find it impossible to take the imports number at face value.

By the conventions of the national accounts, imports are a subtraction from GDP. So the big negative bar on the right of the figure represents a rapid growth in imports – close to the fastest import growth, in fact, in US history. But in real economies, imports almost always rise when GDP growth is strong, and fall when GDP growth is weak. The prediction that we will have an almost-unprecedented slowdown in growth alongside and almost-unprecedented rise in imports doesn’t fit the historical experience.

Here, for example, is another version of Figure 1, but covering the period of the last major recession in the US, in 2008-2009. Notice how when investment spending (and GDP, though it’s not shown) fall through the floor in the second half of 2008, the contribution of imports turns sharply positive, indicating lower imports. Then when consumption and investment spending begin to rise again, making a positive contribution to growth, the import contribution turns negative. This is the usual historical pattern.

The strong relationship between expenditure growth and imports is, I think, one of the most basic and reliable Keynesian facts about the world. Countries import more when they grow faster, and import less when they grow more slowly or shrink. In the long run, yes, relative prices and competitiveness more broadly may be important. But in the short run of a few years or quarters, income is what matters.

Given this strong Keynesian prior, I have a lot of trouble accepting the Atlanta Fed’s nowcast that we are seeing very weak GDP growth but very rapid import growth. It’s not impossible, but it’s certainly very strange.

Here’s a figure, going back to 1947, showing annualized quarterly “real” GDP growth rates and the contribution of imports. The Atlanta Fed’s estimate for the current quarter is the large red dot in the lower left.  As you can see, it’s not entirely out of line with the historical experience. But it would certainly be an outlier. The great majority of quarters with import growth even close to this saw exceptionally strong GDP growth.

An accounting point: When we teach the national income identity — Y = C + I + G + X – M — we present it as if M was a distinct category of spending. But it really isn’t. Final spending by every unit in the economy falls into one of the other four categories. -M is there to subtract the imported component of the other spending categories. This matters here, because it means it is impossible for anyone to simply import more, without also doing more of one of the other categories. Even if imported materials are just stockpiled in a warehouse, that is inventory investment, at least from the point of view of the national accounts.

Over time, of course, imports might rise independently of other components, if the fraction of imported inputs used to produce consumption or investment or export goods changed. But these are changes that happen only gradually. In the short run, it’s impossible for anyone to spend more on imports without also spending more on something else. And in practice, again, import spending reliably rises when total spending rises, and falls when total spending falls. (The relationship in the figure would look much closer if I used annual data.) A deep recession with a dramatic rise in import spending — what the Atlanta Fed’s numbers imply — is well outside the historic experience.

So what is really going on?

The Atlanta Fed is looking at genuine data. The high import numbers reflect more stuff being declared at US ports; the consumption numbers reflect lower grocery store receipts.

One natural way to make sense of it is that this is a surge of imports as businesses try to get ahead of Trump’s tariffs. Normally, imports are a reasonably stable share of current spending. But in this case, the imported part of future spending has been moved forward to this quarter.

Now, in principle, if this what’s going on, then the higher imports should be balanced by an increase in inventory investment — accumulation of raw materials and goods in process — with no effect on GDP. But the Atlanta Fed is assembling its data from a  variety of different sources; there’s no reason to expect it to conform to the accounting relationships that final GDP has to. If, let’s say, trade data comes in sooner than inventory data — which seems very plausible — then it will look instead like the import share of other categories of spending is increasing. Which would be a subtraction from GDP.

To be clear: I think this is fine. Consistency and transparency are very valuable qualities in public data; they shouldn’t be lightly sacrificed even where some one-off adjustment will clearly yield a better point estimate. I think the Atlanta Fed is right to apply their methods consistently, even if they result in implausible  results in this particular case.

There is, though, another intriguing possibility.

A research report from Goldman Sachs² suggests that the apparent rise in imports is to some significant extent due to a rise in imports of monetary gold. The Goldman Sachs analysts write:

most of the widening in the trade deficit since November has been driven by higher gold imports … as participants in the gold market sought to insure themselves against potential tariffs on gold. Although this may seem like a frontloading effect ahead of potential tariffs, these imports are for the most part … being shipped to the US on the off-chance that physical delivery of the gold is required,…  Importantly, the Bureau of Economic Analysis (BEA) excludes most gold imports when calculating the imports component of GDP.  ….

The same reasoning applies more generally to front-loading by retailers, wholesalers, and producers ahead of tariff increases. Because these developments are unrelated to US production, they should have little net effect on US GDP. In the case of non-gold goods, higher imports should be offset by higher inventories in the national accounts. In practice, it is possible that front-loading exerts a modest drag on reported GDP because imports… tend to be measured more accurately than inventories. We suspect this dynamic is playing out now to some extent… But because front-loading these imports now implies fewer imports later, we think the net effect on 2025 GDP growth should be small.

Again: in the conventions of the national accounts, if businesses buy extra foreign inputs today in order to avoid higher costs later, that should, in principle, be recorded in the national accounts as equal increases in imports and inventory investment, with no net effect on GDP. But if the rise in imports is observed earlier or more accurately that the rise in inventory investment, we will see a spurious decline in GDP.

But what about the point about gold specifically, that the BEA excludes gold imports when calculating the import component of GDP? This is not something I’d ever really thought about or even been aware of. But having now poked around a bit, yes, this is correct. Gold imports show up in the trade data because, of course they do. It’s a good crossing the international border. But monetary gold, gold held as an asset, does not show up in the imports (M) shown in the National Income and Product Accounts, because the NIPAs are organized to track production, and gold held as an asset is not being used in production.

This is a very interesting accounting issue — in fact it’s what motivated me to write this post.

National accounting always faces the fundamental question of the production boundary. Which activities are part of production, and which ones aren’t? GDP (and its subsidiary components, like M) is supposed to be a sum of payments for new production. It’s not supposed to include payments for transfer of ownership of existing assets. But this is not always a clean distinction.

Gold is a weird commodity in this context, because it is both an important input to production (of both industrial equipment and jewelry) and an asset held for its own sake. In principle, when gold is unloaded from a ship and put into a warehouse, there’s no way to know whether it is destined to be an input to the production of some consumption good or piece of industrial equipment, or if it is being held as an asset. Maybe at that moment it hasn’t even been determined.Gold is gold.³

What the BEA does — this is interesting — is to take the difference between US use of gold as a production input and US production of gold, and call that imports of gold for purposes of the NIPAs. The difference between the actual net imports of gold and this number is assumed to be monetary gold. In practice, domestic production and use seem to be pretty close, so NIPA gold imports stay close to zero regardless of what the trade figures show. This procedure seems reasonable enough.

It’s not clear to me if monetary gold imports explain the whole story of the Atlanta Fed’s strange rise in imports, or just a part of it. The way the data is presented the Goldman Sachs report makes it hard to compare magnitudes. But it is true, on the one hand, that a surge in imports ahead of the tariffs not informative about GDP growth this quarter. And on the other hand, the treatment of gold imports in the national accounts raises some profound issues, whether or not it fully explains the apparent import surge.

The fundamental challenge with gold is that is both an important input to production, and an important asset in its own right. This is a challenge for our accounting framework that relies on a sharp line between payments related to production and asset sales.  Gold is hardly unique in that respect.

A very analogous and more generally important case is housing. When a family buys a home, they are buying both a flow of consumption (the use value of living in that house) and an asset (the exchange value they can receive by selling the house, or borrowing against it).

Conceptually, these are two different purchase. But in reality, the homeowner is writing only one check. This is a big problem both substantively and for data. Substantively — well, this goes beyond the scope of this post, but  the fact that people’s providing for their own housing needs also involves taking a position in a speculative asset has some pretty far-reaching effects on our society. From a data standpoint: How are we going to split the one payment of the homeowner into the part that is paying for the use of the hoser right now, and the part that is paying for the chance to profit from the appreciation of the house? It’s not an easy question.

Returning to the Atlanta Fed GDPNow estimates. It’s worth emphasizing that the estimate of zero real growth inc consumption spending, which doesn’t have any practical or conceptual problems as far as I can tell. So even if we set aside the import question, there is reason to say that real-time economic data suggest a sharp slowdown in spending — and therefore output, income and employment — relative to the recent trend. I think we should take that forecast seriously directionally, even if there is reason to be skeptical of the dramatic fall in GDP that they forecast.

If we set aside the import numbers, the estimate is real growth for the quarter of close to zero. Which would still be a sharp slowdown, and lead us to expect a significant rise in unemployment.

At the same time, we should keep in mind — always, and perhaps even more now — that numbers like GDP are not material facts existing out there in he world. They are the result of aggregating an enormous number of private payments in a specific way, which involve a great number of more or less arbitrary choices. If we don’t understand how the numbers are constructed, we will not be able to say much about what they mean.

ETA: Thanks to David Rosnick for helping me think through this.