The Atlanta Fed's GDPNow was among the first economic models to draw broad public attention to the large amount of gold imported into the United States in late 2024 and early 2025.
GDPNow's calculations of gold imports contributed to its projection that US gross domestic product (GDP) was on a trajectory to fall into negative territory during the first quarter of 2025. As the potential for an economic downturn caught the attention of financial media, GDPNow took early note of this influx of gold, incorporating it into its calculations.
As it turned out, the economy was not in dire straits, though the first official estimate ultimately showed that GDP contracted slightly. GDPNow had produced that outcome in part because the surge of gold imports made the balance of trade appear to be an unusually strong drag on growth. Net exports—defined as exports minus imports—factor into measuring GDP, and GDPNow had been considering gold as an imported good to be consumed, even though it was largely being moved from foreign vaults into US vaults.
Analysis later would show the abnormally high level of gold imports resulted from owners transferring gold to the United States in anticipation of tariffs, which might increase the value of gold already in the country relative to gold whose later importation could trigger tariffs. A typical monthly import rate of $1 billion to $4 billion in recent times had increased to more than $13 billion in December 2024 and to almost $34 billion in January 2025, according to Pat Higgins, the Atlanta Fed economist who developed GDPNow, citing trade data, in report IDS-0182, from the US Bureau of Economic Analysis (BEA).
Shining a spotlight on the role of gold
The conversation spurred by the dour forecast in the February 28 edition of GDPNow underscored the influence of the popular economic tool. The model showed growth was on track to decline by 1.5 percent in the first quarter, down from the positive 2.3 percent growth rate the model had indicated the previous week. The model's calculations included the January surge in total goods imports in the February 28 advance international trade report but did not separate out the impact of gold imports. The data on gold imports would not be available until the March 6 release of the full trade report.
Between February 28 and March 6, Higgins tweaked GDPNow's mathematical code to calibrate for distortions from surging gold imports, and the Atlanta Fed produced two versions of GDPNow from March 6 through April. (The only other time the code for GDPNow had been tweaked during the quarter being evaluated was during the Covid-19 pandemic in 2020.) In an article explaining the change, Higgins wrote:
"While not on that level [of the pandemic], the unusual widening of the January trade deficit that led to much of GDPNow's sharp decline on February 28, and the circumstances surrounding that decline, was also unprecedented in one respect.
"That is, as we now know from the March 6 full international trade report—but could only strongly suspect based on anecdotal and non-US government data until then—much of the widening of the trade deficit in January was due to an increase in nonmonetary gold imports from $13.2 billion in December to $32.6 billion in January. This accounted for 60 percent of the widening the goods trade deficit."
The alternative model released on March 6 removed gold imports and exports from the calculation of the net exports GDP subcomponent. Through most of April, GDPNow offered twin snapshots of GDP growth: one that took into consideration the amount of gold imports and exports, and one that didn't. The version without gold transfers became the standard on April 30.
Higgins had observed in 2021 that the GDPNow model is subject to big anomalies in the economy. As he wrote in a May 2021 Policy Hub: Macroblog post, "GDPNow will remain susceptible to forecast inaccuracies whenever unusual events (such as the COVID-19 pandemic) hit the economy."
The willingness to update the engine that powers GDPNow on the basis of later data is consistent with a longstanding philosophy that undergirds the tool: that new evidence matters and should be considered.
Behind the data tool's development
Higgins developed GDPNow in 2011 in an attempt to provide a real-time snapshot of the nation's GDP. (The Atlanta Fed began making GDPNow available on its website in 2014.) It is not an official forecast of the Atlanta Fed, the Federal Reserve System, or the Federal Open Market Committee, the latter of which sets short-term interest rates.
Instead, GDPNow is a tool that begins each calendar quarter with a clean slate. As it is provided with newly released economic indicators, GDPNow updates its "nowcast." These economic indicators include data from public and private entities including the US Bureau of Labor Statistics, the US Census Bureau, the Institute for Supply Management, and the US Department of the Treasury. In all, the algorithm incorporates 13 subcomponents that are also used by the BEA in devising its quarterly GDP report.
GDPNow's approach of measuring data accumulated over time sometimes results in the nowcast's updates being highly volatile at the start of each quarter, but they generally become quite accurate over the course of a quarter as the model crunches more data.
Two other Reserve Banks produce nowcasts: the Reserve Banks of New York and St. Louis. The New York Fed relaunched its weekly nowcast in September 2023 after stopping production in September 2021, when the pandemic posed difficulties for its model. The Bank has since revised its formula interim. The St. Louis Fed initiates its model forecast each quarter with a consensus of professional forecasts and revises it based on patterns observed in subsequent data releases. If the data releases come in stronger (or weaker) than economists expected, the model's GDP forecast is revised accordingly.
Higgins observed that present economic uncertainty might affect GDPNow's future nowcasts. "Although gold is not likely to distort GDPNow's remaining nowcasts for the quarter as it did for the last one, this is not to say we should expect it to be as accurate as it was either last year or in the prepandemic era," Higgins said. "GDP nowcast accuracy tends to decline when forecaster disagreement and volatility for various measures of spending increase."
David Pendered
Staff writer for Economy Matters