GDPNow Still Looks Backward

Is the US economy set to contract? Talk of a downturn spiked after the Atlanta Fed’s GDPNow model flipped from 2.3% annualized Q1 GDP growth to -1.5% contraction last week ... and slipped further to -2.8% Monday. (Note: It will update and include Wednesday’s growthy Services ISM with Thursday’s Wholesale Trade report.) But the worry is premature and risks being flat wrong. Look under the hood at how GDPNow works—an attempt to blend incoming data to calculate the unreported quarter’s growth—and it becomes clear this isn’t a magical tea leaf for investors to read.

As we have explored before, GDPNow is one of a class of “nowcasts” estimating current-quarter GDP based on available data. This isn’t an official forecast of the Fed or the Atlanta Fed. It is a tool. Friday, a pair of economic releases that feed GDPNow caused the gauge to drop into contraction. The culprits? The advance estimate of January net exports (exports minus imports) and personal consumption expenditures (PCE, aka consumer spending).

Exhibit 1: Evolution of Q1’s GDPNow Real GDP Estimate

Source: Federal Reserve Bank of Atlanta and FactSet, as of 3/4/2025. *Range of top 10 and bottom 10 average forecasts.

On Monday, the Census Bureau’s construction spending report for January and the Institute for Supply Management’s (ISM’s) February manufacturing purchasing managers’ index (PMI) depressed GDPNow further. The model tries to replicate how these data will factor into actual GDP but in “real time” rather than waiting for February and March reports. Two months into Q1—with only January data and one February release available—GDPNow is relying on very incomplete datasets to compose a full quarter’s picture. It is, therefore, volatile historically.

Furthermore, let us dig more carefully into what drove the drop. Imports swamped exports in January’s advance estimate. This caused net exports’ -0.41 percentage point (ppt) detraction to Q1 GDP based on earlier estimates to balloon to a huge -3.70. Part of this was likely importers frontrunning tariffs. An East Coast dockworkers strike (averted with a January deal) may also have had some effect.

But many observers suggested a weird quirk is most responsible: A January surge in gold imports (taking advantage of a rare, one-off arbitrage opportunity between London and New York).^[i] While we can’t yet confirm this is the case—the advance, or preliminary estimate, doesn’t break down industrial imports to the product line—this is key. Such trading and monetary gold imports aren’t registered in the US Bureau of Economic Analysis’s tally of GDP. Hence, should this gold theory prove correct, that huge January import detraction could wind up having next to no effect on actual Q1 GDP when reported. Regardless, a sharp rise in imports doesn’t really show economic weakness. Rather, it may show healthy demand and/or businesses’ anticipation to meet expected demand.

The other factor in the big flip? January’s PCE dip, which slashed its contribution from 1.53 ppt to 0.87. But again, this was just for January. Look at Exhibit 2’s trend and the quarterly average (dashed yellow line), which are used to calculate quarterly growth rates. Besides possible lingering seasonal adjustment issues—January 2024’s PCE also dipped—that doesn’t mean February and March will. Will one month’s decline extend for the rest of the quarter? GDPNow makes no effort to say.

Exhibit 2: Consumer Spending’s January Dip in Perspective

Source: FactSet, as of 2/28/2024.

Monday’s January construction report saw spending fall -0.2% m/m, led lower by residential building’s -0.5%.^[ii] That drove the residential investment component of GDPNow from a slight 0.06 ppt contribution to a -0.20 ppt detraction. February’s manufacturing PMI, which ticked down to 50.3 from January’s 50.9, appeared to have the bigger effect on the model Monday because it is used to “indirectly forecast many of the GDP components” not yet available for February.^[iii] While the headline move was miniscule, its new orders subindex dropped -6.5 ppts to 48.6—below 50, indicating contraction—and the employment subindex also sagged -2.7 ppts to 47.6. This helped drive the model’s estimates for PCE and business investment growth from 1.3% and 3.5% annualized Friday, respectively, to 0.0% and to 0.1%.

We see a big problem with this, though, in the same way that The Conference Board relies on manufacturing PMI subcomponents for its Leading Economic Index: PCE and business investment are mostly services-based, which makes them less than reliable factors to forecast GDP. Moreover, efforts to translate a breadth-based indicator (PMIs measure the percentage of firms reporting growth, not the magnitude) are imperfect at best.

As more concrete data come in revising initial manufacturing PMI-based estimates, GDPNow tends to become more accurate, but at this early stage it can be all over the place. Even on the days right before quarterly GDPs’ first official releases—about a month after quarter end—GDPNow’s average error is plus or minus 0.77 ppt. (Exhibit 3) The average error at quarters’ start? 2.06 ppts. This far out from April 30’s Q1 GDP “advance estimate,” the error bars around GDPNow’s current estimate for it remain substantial.

Exhibit 3: GDPNow Tracking Error on Eve of Quarter Being Forecast

Source: Federal Reserve Bank of Atlanta, as of 3/4/2025.

Before nowcasts, legions of economists did variations of this exercise (and still do) with their own proprietary models—and underlying assumptions. Meanwhile, other nowcast models like the New York and St. Louis Feds’ suggest Q1 GDP growth: 2.9% and 1.5%, respectively.^[iv] Why the difference? One of the reasons is because the New York Fed’s nowcast incorporates disposable personal income (not included in GDP), which offsets PCE’s effect. Vitally, it doesn’t include the advance international trade report. It waits for the final data, which won’t arrive until March 6. The St. Louis Fed’s version relies on professional forecasters and uses incoming data to adjust its nowcast.

All in all, we think this best demonstrates not where the economy will necessarily finish Q1. There are still two months of data to come! The reactionary alarm says more about how much sentiment has reset itself from a heady finish to 2024. People seem back to hunting for rather niche signs a recession looms—and that helps extend this bull market’s wall of worry.