Another quarter of negative growth?

The Atlanta Fed forecast for the second quarter on July 15 was -1.5% Q/Q SAAR. What this tells us about what will likely be a pre-release and later releases.

Figure 1: GDP (black), Goldman Sachs (pink square), GDPNow of the Federal Reserve Bank of Atlanta (green triangle), IHS-Markit (light blue triangle), in billions Ch.2012$, SAAR. Forecasts for 07/15/2022. Source: BEA and Goldman Sachs, Federal Reserve Bank of Atlanta and IHS-Markit, and author’s calculations.

What does this point estimate mean? From Frequently Asked Questions About the Atlanta Fed:

Since we started tracking GDP growth with versions of this model in 2011, the average absolute error of the final GDPNow forecasts is 0.84 percentage points. The standard error of the forecasts is 1.25 percentage points. These accuracy measures cover initial estimates for 2011: Q3–2022: Q1. Some additional analysis of the GDPNow forecast errors is available in the macroblog posts located here and here. We have made some improvements to the model compared to its earlier versions, and the model’s predictions have become more accurate over time (full track record here). When backtested with revised data, the standard error of the out-of-sample model prediction with the same data coverage that the analyst would have had just before the “preliminary” estimate is 1.15 percentage points for 2000:Q1-2013. A: Q4 period. The figure below shows how forecasts become more accurate as the interval between the date of the forecast and the date of the upcoming release of GDP data narrows.

Overall, these accuracy scores do not provide conclusive evidence that the model is more accurate than professional forecasters. The model appears to perform well compared to other traditional statistical models.

What does RMSE 1.15 percentage points mean. Considering that the RMSE does not decrease significantly from about 25 days to 1 day before the announcement, I will use this number (13 days left until the publication of July 28) to calculate a 90 percent interval for the GDPNow forecast.

Figure 2: GDP (black), Atlanta Fed’s GDPNow (green triangle) and 90% interval (grey+), in billions of dollars Ch.2012, SAAR. Forecasts for 07/15/2022. Source: BEA Atlanta Fed and IHS-Markit, and author’s calculations.

In other words, there is a chance that GDPNow will record positive Q2 growth (or even more negative).

This historical track record should be treated with some caution as the model has been revised several times. In addition, since the April 29 release, additional measures have been taken to account for the complications of seasonal adjustment and the dynamics associated with the sharp decline in 2020 (see chart). here).

Most recently, forecast errors for the last year were -0.9pp, 1.8pp, 0.4pp, -1.8pp.

note that The NBER BCDC does not use the rule of thumb for two consecutive quarters of negative growth to define a recession..