Nominal GDP targeting has been proposed as an alternative to the Taylor principle. One challenge to implementation is the relatively large revisions in the growth rate of this variable (and don’t get me started on the level).
Figure 1: Q/Q nominal GDP growth, SAAR, from various vintages. NBER defined recession dates shaded gray. Source: ALFRED.
How big are the revisions? The BEA provides a detailed description. This table summarizes the results.
The standard deviation of revisions going from Advance to Latest is one percent (annualized), mean absolute revision is 1.3 percent. Now, the Latest Vintage might not be entirely relevant for policy, so lets look at Advance to Third revision standard deviation of 0.5 percent (0.6 percent mean absolute).
Compare against the personal consumption expenditure deflator, at the monthly — not quarterly — frequency; the mean absolute revision is 0.5 percent going from Advance to Third. The corresponding figure for Core PCE is 0.35 percent.
Admittedly, the estimation of output gap is fraught with much larger (in my opinion) measurement challenges, as it compounds the problems of real GDP measurement and potential GDP estimation; this is a point made by Beckworth and Hendrickson (JMCB, 2019).
On the other hand, the unemployment rate can be substituted for the output gap in the Taylor principle, by way of Okun’s Law. As Aruoba (2008) notes, the unemployment rate is not subject to large and/or biased revisions. (The estimated natural rate of unemployment, on the other hand, does change over time, as estimated by CBO, and presumably by Fed, and others, so there is going to be revision to the implied unemployment gap.)
One interesting aspect of the debate over nominal GDP targeting relates to growth rates vs. levels. If it’s growth rates, there is generally a “fire and forget” approach to setting rates. An actual nominal GDP target of the level implies that past errors are not forgotten (McCallum, 2001) (this is not a distinction specific to GDP; see the inflation vs. price level debate). Targeting the level faces another — perhaps even more problematic — challenge, as suggested by Figure 2.
Figure 2: Nominal GDP in billions of current dollars, SAAR, from various vintages. NBER defined recession dates shaded gray. Dashed red line at annual benchmark revisions. Red arrows denote implied revisions to last overlapping observation between two benchmarked series. Source: ALFRED.
Revisions can be large at benchmark revisions, shown as dashed lines in the above Figure. But even nonbenchmark revision can be large, as in 2009Q3. ( Beckworth (2019) suggests using a Survey of Professional Forecasters forecast relative to target and a level gap as means of addressing this issue — I think — insofar as the target can be moved relative to current vintage.)
None of the foregoing should be construed as a comprehensive case against some form of nominal GDP targeting. But it suggests that the issue of data revisions in the conduct of monetary policy is not inconsequential.
Addendum, 9pm Pacific:
Reader joseph notes the revisions in the natural rate. Here is a graph of various vintages of natural rates (long run) as estimated by CBO.
Figure 3: Unemployment rate, latest release (red), long run natural rate of unemployment rate, February 2011 release (blue), January 2016 (orange), August 2019 (green), all in percentage points. NBER recession dates shaded gray. Source: BLS, CBO via FRED, and NBER.