The US Bureau of Economic Analysis (BEA) estimated 
that real GDP contracted at an annualized rate of 0.3 percent in the first quarter of 2025, the first negative reading since the first quarter of 2022. As figure 1 shows, this reading was still well above the final nowcast of −2.7 percent growth for what was then the standard GDPNow model and meaningfully above, but closer to, the final "gold-adjusted" model forecast of −1.5 percent.
As detailed here, the gold-adjusted model—unlike the former version—subtracts international trade in gold when estimating net exports of goods. Otherwise, the models are essentially identical, and both of their final projections have the change in private inventories (CIPI) adding 0.30 percentage points to first-quarter real GDP growth.
The April 29 GDPNow nowcasts were well below the BEA's estimate 
that CIPI added 2.25 percentage points, accounting for the largest subcomponent contribution forecast error for either of the models. The excerpt below from the GDP release gives a clue as to why the models were so wrong on this CIPI subcomponent (boldface added):
The largest contributor to the increase in investment was private inventory investment, led by an increase in wholesale trade (notably, drugs and sundries). The estimates of private inventory investment were based primarily on Census Bureau inventory book value data and a BEA adjustment in March to account for a notable increase in imports. For more information on the source data and BEA assumptions for inventories, refer to the key source data and assumptions table (available at 10 a.m.).
It's unclear from the release what the size of BEA's March adjustment was. However, if we take the March values for merchant and nonmerchant wholesale inventories (provided in the BEA's key source data and assumptions 
and investment related underlying detail files) and simply plug them in place of the GDPNow forecasts for those values in the Inventories tab of the April 29 GDPNow spreadsheet , the model's CIPI contribution forecast goes up by 0.84 percentage points. Using the model's monthly inventories data-based forecast rather than averaging it with the quarterly-based model forecast, as described on page 17 here , further increases the CIPI contribution to 1.02 percentage points. The estimate cited in the Wall Street Journal assesses the impact of the adjustment to the March inventories data on GDP growth to be 1.2 percentage points.
Real GDP growth was slightly negative despite the large CIPI contribution and a solid 3.0 percent real growth rate for aggregate personal consumption expenditures (PCE) and private fixed investment primarily because net exports subtracted 4.83 percentage points off first-quarter real GDP growth, as shown in the third bar in figure 2. BEA data show that nearly all of the reduction (4.79 percentage points) was accounted for by a 50.9 percent annualized surge in real goods imports that, apart from the pandemic related rebound in the third quarter of 2020, was the largest increase since 1972. In theory, this gain should be offset in the CIPI and/or other spending subcomponents of GDP. But the aforementioned BEA inventories adjustment shows that this cancellation may not be evident in the published monthly GDP source data.
It's unclear whether the reverse phenomenon—spending on goods drawn from inventories that are not accounted for in the published Census Bureau inventories data—can or will occur. But we can anticipate that it is likely that either the BEA's estimate of inventories contribution to first-quarter GDP growth will be revised down or GDPNow's projected contribution of it to second-quarter GDP growth will be revised down on June 27. Until June 27, GDPNow will make its own calculation of first-quarter CIPI in GDP using Census Bureau data on the book-value of inventories and BEA data for the remainder of the CIPI related data. This is because using Census Bureau book-value data usually generate virtually the same CIPI estimate for the prior quarter as what one would get using only the BEA data immediately after the GDP release. This allowed the model to anticipate BEA revisions of CIPI for the prior quarter in second and third release GDP estimates after Census Bureau revisions to monthly inventory book values. However, GDPNow currently calculates a first-quarter annualized CIPI of $94 billion in 2017 dollars, while the BEA calculated it as $140 billion. The published first-quarter CIPI will be revised two more times by the BEA on May 29 and June 26. After the latter revision, the BEA's first-quarter CIPI will be "final" until the annual revision that will take place on September 26 if the BEA follows the same annual revision scheduling pattern as it did in 2022–24. With respect to the July 30 2025:Q2 GDP release, 2025:Q1 CIPI will be "frozen" at the level published in the June 26 GDP (third) release estimate. So GDPNow would switch to the temporarily "frozen" BEA estimate on June 27 (see also the third-to-last FAQ here and/or more formal discussion on pages 13–17 here ). If both GDPNow and the BEA estimates for CIPI remain at their current, but different, estimates through June 26, the GDPNow switch to the higher 2025:Q1 value for CIPI would reduce its topline nowcast by 0.8 percentage points on June 27.
Figure 1 also shows that when we extend the gold-adjusted model forecasts from when it went "live" on March 6 back to February 26—the last forecast date before January international trade data with the surge in gold imports were released—the standard and gold-adjusted model forecasts are nearly identical in this final pre–gold import surge forecast. In the historical forecast accuracy analysis below, we use what was then the GDPNow growth forecasts for the first quarter of 2025 prior to February 26 and the gold-adjusted growth forecasts for the first quarter of 2025 beginning on this date.
Figure 2 shows Blue Chip and GDPNow forecasts for the composition of second quarter real GDP growth. The Blue Chip forecast includes all subcomponents shown in the chart except for "government plus residential investment." We essentially derive this as a residual using the topline GDP and other subcomponent forecasts along with some chain-weighting calculations. The current and now defunct GDPNow forecasts are also shown in the figure, the latter of which doesn't include the gold adjustments to the international trade data described here. The latter being stronger than the former implies that the model sees (net) gold imports falling this quarter relative to the first, which is plausible given the sharp decline in gold imports from January ($32.6 billion) to March ($17.5 billion).
Figure 2 also shows the consensus Blue Chip Economic Indicators (BCEI) forecast for the second quarter of 2025 is weaker than the most recent GDPNow forecast as well as a typical quarter, proxied by the left-most two bars showing average annualized growth between the fourth quarter of 2007 and the fourth quarter of 2019 National Bureau of Economic Research business cycle peaks , and average growth since the latter peak. The subcomponent composition of the Blue Chip consensus forecast is also quite different than any of the others shown. It has both net exports and inventories partially reversing their first-quarter contributions but netting out to a positive, rather than a negative, total contribution. Its remaining subcomponent contributions are negative on balance and all somewhat weaker than their corresponding, "typical," first-quarter, and GDPNow-projected values. GDPNow was about as weak on May 1 as the Blue Chip forecast after it first incorporated April "soft data" in the ISM Manufacturing and consumer attitudes reports. But it was revised up the next day after incorporating better April data on employment and motor vehicle sales .
How might we assess the uncertainty around the Blue Chip and GDPNow forecasts? Prior to the pandemic, GDPNow projections two-and-a-half-months ahead were generally moderately less accurate than professional forecasts, but they were about on par with them within a month of the release. And since 2020, GDPNow has been less accurate on average than it was prior to the pandemic. In particular, using the gold-adjusted model for the first quarter of 2025, from the first quarter of 2021 to the first quarter of 2025, GDPNow's average absolute error for its final forecasts (0.65 percentage points, using a seasonally adjusted annual rate) was greater than it was in its prepandemic history (from the second quarter of 2014 to the fourth quarter of 2019) since it was first published (0.51 percentage points). Moreover, 80-day ahead prepandemic forecasts over this span were about as accurate, on average, as 20-day ahead forecasts starting in the first quarter of 2021.
The pandemic posed a number of forecasting challenges, including fiscal stimulus and supply chain disruptions, that made forecasting more difficult and probably less accurate than it was in the 2010s. The difficulty of forecasting also increases as the probability of a recession rises. The subdued Blue Chip forecast of GDP growth in the second quarter of 2025 is very similar to the consensus in the April Wall Street Journal Economic Forecasting Survey , and the respondents in that WSJ survey put the 12-month recession probability at 45 percent on average. Although lower than the 48 percent to 63 percent recession probability range prevailing in the survey during the second half of 2022 and all of 2023, this latest reading is well above its 17 percent average seen during 2014–19. All of this suggests that, in a probabilistic sense, we should expect forecasts in the second quarter of 2025 to be less accurate than they were in 2014–2019. For some additional (and more formal) statistical evidence, as seen in the chart in GDPNow's home page and here , the difference—or so-called "forecast disagreement"—between the average of the 10 highest Blue Chip forecasts and the average of the 10 lowest Blue Chip forecasts is much larger in the most recent BCEI survey than it was in late March. Research by Constantin Bürgi and Tara Sinclair (you can also see it here ) shows that increased forecaster disagreement in two-and-a-half-month ahead GDP growth expectations are significantly associated with higher recession odds in a probit model. Further, a regression analysis I did with the historical BCEI data (excluding the first three quarters of 2020) implies that the current "Top 10-Bottom 10" BCEI forecast disagreement is associated with an 80-day ahead GDP growth nowcast that is 0.92 percentage points less accurate than it would be if that disagreement was instead at its 2014–19 average. With these differences in mind, figure 3 shows the average absolute forecast errors for the 80-day ahead GDPNow and BCEI real GDP growth forecasts and its associated subcomponent contributions beginning in the first quarter of 2021.
The Blue Chip forecasts for topline growth are a bit less accurate, on average, than GDPNow. On the other hand, they are a bit more accurate for their subcomponent contribution forecasts. Evidently, GDPNow had a more fortuitous cancellation of subcomponent errors than the BCEI did. Nevertheless, these accuracy metrics are of similar magnitude. Since the average absolute forecast error is equal to (√(2/π)) ≈ 0.8 standard deviations in a Gaussian distribution , the post-2020 period accuracy standard suggests that a negative GDP growth rate in the second quarter of 2025 is within the 70 percent confidence interval. But a fairly solid-to-strong growth rate is also within it. Figure 3 also shows that real PCE, CIPI, and net exports account for the largest, and similarly sized, forecast errors.
Figure 4 is comparable to figure 3 but uses shorter-term forecasts. The 20-day ahead Blue Chip forecast errors of real GDP growth are larger than those for GDPNow, but—as we noted before—the subcomponent contribution forecast errors are comparable. I should note that some luck is also evident when we compare GDPNow's 20-day ahead nowcast with its final forecasts, as the latter topline nowcasts are slightly less accurate than the former despite the improved accuracy of the final nowcasts for real PCE, CIPI, and net exports. For these shortest-horizon forecasts, the PCE contribution error metric is smaller than the same metrics for CIPI and net exports. Much of the final PCE forecast improvement relative to its 20-day ahead forecast is the result of incorporating a final retail sales release for the quarter 10 to 12 days before the GDP release.
It's an open question whether another judgmental BEA adjustment to CIPI, or perhaps one of the other subcomponents, could again factor into the forecast errors for GDP growth in the second quarter of 2025. But the above analysis and figures suggest a reasonable likelihood of being able to rule out either negative or relatively strong GDP growth this quarter as we approach the release date. The analysis also shows that net exports and CIPI are among the more difficult subcomponents to forecast—and that difficulty is present even without the added uncertainty surrounding tariffs.
By Pat Higgins, policy adviser and economist in the Atlanta Fed's Research Department