The Atlanta Fed's GDPNow model, which provides a running estimate of quarterly real GDP growth prior to official reports by the US Bureau of Economic Analysis (BEA) 
, is sometimes an outlier relative to professional forecasters. This has been true in spades since the last week of February 2025, as the chart on our GDPNow Home page shows. There are two reasons for this: the first is related to accounting for gold imports (which prior to March 6, 2025, we did not do); the second is due to the accounting of a handful of GDP subcomponents, for which GDPNow has a weaker outlook.
The BEA provides background on how they handle gold imports and exports in the GDP measures of imports and exports (here and here 
). And we have discussed what adjustments we have made to our model to better account for gold imports and exports (here and here ). But it is worth looking more closely at some of these data, plotted in figure 1 below, since I (and perhaps others) had some early difficulties in spotting the surge in gold imports in the official data, a surge that has been reported in the press (here ).
The measure of nonmonetary gold imports, tracked by the orange line in figure 1 and whose monthly value has never exceeded $9 billion in its 35-year history and has never exceeded $4 billion since 2020, is reported in Exhibit 7 of each month's international trade report released jointly by the BEA and the US Census Bureau at 8:30 a.m. (ET) in a regular pattern of release dates . Not included in that report, but released at 10 a.m. by the BEA on these trade release dates , is a second "balance of payments" (BOP) measure of nonmonetary gold imports. That measure spiked to a monthly average of $29.1 billion for January and February of 2025 (the yellow line in figure 1).
Why the disparity in these measures of gold imports? As the BEA notes in an FAQ , the BOP measure of gold includes a portion of "finished metal shapes" imports (the green line in figure 1) that are reclassified by the BEA as gold. Recently, the most important of these commodities by far is identified as harmonized system code 7115900530: "Articles of precious metal, in rectangular shapes, 99.5 percent or more by weight of precious metal, not otherwise marked or decorated, of gold." In other words, gold bars. According to data from USA Trade Online , these imports surged from $2.08 billion in November of 2024 to $28.69 billion in January 2025 before falling back a bit to $22.96 billion in February 2025.
In the 8:30 a.m. (ET) international trade report, the finished metal shapes imports that include these gold bar imports are all classified on a Census basis within "industrial supplies and materials." These also spiked recently, as can be seen in figure 1 (the blue line). Industrial supplies and materials imports are of interest because they, along with five other sub-aggregates of goods exports and imports, are included in the Census Bureau's Advance Economic Indicators (AEI) report often released about a week before the full international trade report.
Unfortunately, the AEI report does not separate gold bars or finished metal shapes from the remainder of the industrial supplies and materials imports aggregate, and this affects our methodology. Is there a way we can utilize the international trade data in the AEI until the full report is released? A number of analysts have noted Swiss gold exports to the United States have surged. According to Swiss-Impex data and Federal Reserve Board exchange rates , Swiss gold exports to the United States spiked from under $400 million in each of the first two months of 2024 to $17.2 billion in January 2025 and $14.8 billion in February 2025. The reason for the spike is likely due to the move of smaller gold bars, stored in London, to Switzerland, where they are refined into the larger gold bars acceptable in the New York market. These Swiss data are often released about a week to ten days before the AEI, and Comex data on gold inventories are available even earlier. Nonetheless, because this data is not available from US government sources and the industrial supplies and materials trade data in the AEI is not further disaggregated, we discard that trade data when using the report to forecast net exports. We do, however, use the remaining trade data in that report as described here .
In terms of the quantitative significance for GDP growth, the January and February 2025 average (BOP) gold imports of $29.1 billion would, if maintained in March, put it $22.2 billion above its fourth quarter average. Annualizing this difference (by multiplying by 12) implies that a "straight" gold adjustment ($265.9 billion ≈ 12*$22.2 billion) would increase the GDP growth forecast relative to the standard GDPNow model by 3.6 percentage points. The difference between the gold-adjusted and standard model nowcasts is a little more than half this large, partly because the model is (implicitly) forecasting less of a gold impact in March and also because the model forecast for goods net exports doesn't put all of its weight on the portion of the model using the monthly foreign trade data (see page 11 of Higgins, 2014 ). With this (smaller) gold adjustment, the model is forecasting smaller (in magnitude), but still slightly negative, first-quarter real GDP growth (reflected in the right most bar and marked by the orange circle in figure 2 below).
In figure 2, the second bar from the right represents the GDP subcomponent contributions to the consensus April Blue Chip Economic Indicators forecast, which is less than one percent but still in positive territory.
Compared to either the latest gold-adjusted GDPNow nowcast, or the one made at the same time as the Blue Chip survey, the largest subcomponent contribution difference between our model and Blue Chip forecasts is for net exports and the change in private inventories. A number of GDPNow users have asked whether the model is underestimating the inventories contribution, given the recent surge in (non-gold) goods imports. While the model does have at least advance Census Bureau inventory estimates for many industries through February 2025, there are other industries (farm, nonmerchant wholesalers, construction, and utilities) not covered in these reports. They account for a bit more than 20 percent of the total inventory stock.
We don't have direct insight into what is happening with these "other industry" inventories, but using figure 3, we can look at whether the historical relationship between the quarterly growth rates of real goods GDP and manufacturing industrial production growth is consistent with their first quarter estimates. Although goods share of GDP is three times as large as manufacturing—value added is 29.9 percent versus 9.9 percent in 2024:Q4—we can see that there is a strong relationship between the two quarterly real growth rates of the two measures of output (with a correlation of 0.76 for 1972–2024). The region between the two dashed lines is the 95 percent confidence region where we would expect that proportion of the points to fall given the strength of the positive relationship of the two growth rates apparent in the plot.
The combination of the first-quarter growth rates of manufacturing IP and goods GDP is well below the 95 percent confidence band for the standard model and nearly on, but just above, the lower boundary for the gold adjusted one. One could interpret this as suggesting that even the gold adjusted model forecast is a little too pessimistic. But we'll have to wait until the official BEA report on April 30 to see if this is indeed the case.
We plan on releasing the gold-adjusted model's nowcast through April 29. Starting on April 30, what is now the gold-adjusted model will replace the standard model, as noted here . If gold imports were to fall and remain at $0 in April and thereafter, the standard model nowcast would likely understate second-quarter growth of imported goods and thereby overstate the GDP growth contribution of net exports. While this isn't likely to happen, it's also plausible that gold imports could meaningfully distort the (current) standard model's nowcast in the second quarter—hence, the switch.
By Pat Higgins, policy adviser and economist in the Atlanta Fed's Research Department