The BEA published its annual revision of GDP data, including an update of GDP for the last 13 quarters.
Changes in the numbers are to be expected. Changes of the sort that actually occurred are odd, and require deeper examination.
Fortunately, this macroeconomist has a copy of the unrevised numbers for comparison.
The sort of revisions that can reasonably be expected are those in which the level moves in one direction for most observations (indicating a re-evaluation of the counting of some data), or random adjustments in growth rates (indicating some shifting of GDP from one quarter to its neighbors). Other changes should raise red flags.
Instead, what we see is a decline in growth rates for 10 of the 13 revised quarters. It isn't impossible for this to happen, but it isn't very likely.
A regression of the old growth rates* on the new growth rates has an intercept of .0005 and a slope of 0.865. The t-statistic for the test that the intercept is zero is 2.55, while the one for a test of whether the slope is 1 is 2.13. The former suggests a significant downgrade in growth over the last 13 quarters, while the latter suggests an adjustment in which the (few) high growth quarters in this time frame had their reported growth cut. Of course, this also suggests that any quarters of negative growth in the old numbers would be bumped up towards zero - except for the fact that there weren't any.
I'd be much happier with a revision whose intercept and slope were insignificantly different from 0 and 1. Finding that they weren't bugs me.
Am I being cynical to wonder why this is happening in an election year? Especially one in which the incumbent isn't running, so that a new regime can measure its accomplishments from this newly lowered baseline?
Right now, only the raw numbers are available. An explanation for all this is coming in the August Survey of Current Business. Stay tuned.
* As is standard in macroeconomic time series analysis, I used the differences of natural logs, rather than growth rates.