Growth from the fourth quarter of 2013 to the first quarter of 2014, originally thought to have been about +0.1% in April, was revised last week to –2.9%. That’s at a seasonally-adjusted, annualized rate (SAAR) – the way the U.S. Bureau of Economic Analysis (BEA) usually reports real GDP growth. News reports varied between shock and concern. Was the anemic recovery over? Or, was it just that this winter was especially harsh?
In reality, these headline growth numbers simply don’t contain all that much information for real-time business cycle analysis. There are many reasons, but two deserve special attention: (1) the statistical noise created by seasonality; and (2) the propensity to revise GDP many years after the period being measured.
Seasonality in GDP is enormous. The chart below shows the annualized growth rates for the seasonally adjusted (red) and unadjusted data (blue) from 1980 to 2004 – when the BEA stopped reporting unadjusted data to save money. The seasonal adjustments swamp the small changes in the adjusted growth rates. If you looked only at unadjusted data, you could say that the U.S. economy goes through a depression in the first quarter of every year, as the level of output plunges on average by 18 percent!
Source: BEA GDP Release Table 8.1.
Even more important, the first-quarter seasonal factor (the percentage gap between the raw data and the adjusted readings) varies considerably. In the 1980-2004 sample, it ranged from a low of 15 percentage points to a high of 25 percentage points, with a standard deviation of 2.5 percentage points. The variability of the seasonal factor in the first quarter reflects in part the climatic variation that occurs across U.S. winters (just think of how much worse 2014 was than 2013). As one of us has discussed in the past, other countries have interesting seasonal patterns as well, with northern Europe basically shutting down in June and southern Europe doing the same in August.
The point is that it is difficult to extract the useful signal – the SAAR economic growth rate that we care about – from the noisy unadjusted GDP data. When the seasonal factor is large and variable, as it is in the first quarter of every year in the United States, it is heroic to draw inferences from a percentage point here or there.
Next there are revisions – lots of them. Not only is GDP data released immediately following the quarter – mid-April for the first quarter of January, February and March – but then government statisticians publish revised data monthly for two months, annual revisions for a couple of years, and more comprehensive revisions roughly every five years (such as the overhaul announced in July 2013). These revisions present economic analysts with the rather irritating specter of studying history that is constantly changing.
And, the changes can be quite big. Looking at what’s happened over the last 50 years or so, the standard deviation of the revisions to quarterly growth (SAAR) from the first to the third release is about 0.7 percentage points. But prior to last week’s release of revised first-quarter data, the biggest revision on record was only 2.5 percentage points (it was down). So, a 3-percentage point revision only three months after the quarter ended is enormous.
Nevertheless, further large revisions may still lie ahead! Statistically, the revisions to economic growth for quarter t between the t+3-month estimate (which we just received last week for the first quarter) to the t+10-year estimate (which will not be available for nearly a decade) have ranged from minus 6 percentage points to plus 7 percentage points over the past 40 years, with a standard deviation of about 2 percentage points. Put differently, there remains a good chance that some of this reported decline will be revised away. We might even wonder whether the economy contracted at all last winter.
To see how important the revisions can be for business cycle analysis, look at what happened to the estimated quarterly SAAR growth of real GDP in the third quarter of 1990 (see chart below). The initial reading showed positive growth of 1.8%. Two years later, the estimate was minus 1.6%. The growth estimate settled down close to zero only after more than 10 years of revisions!
Why should we care? The National Bureau of Economic Research (which officially dates U.S. business cycle turning points) now identifies July 1990 as the start of a recession. It would have been helpful for households, businesses, and policymakers to know this information in real time. That’s why so many observers who care about the cyclical state of the economy turn to data other than GDP – like U.S. labor market reports – that are available quickly and revised quickly, too.
Revisions to Third-Quarter 1990 Real GDP Growth (SAAR)
Source: Real-time database of the Federal Reserve Bank of Philadelphia and authors’ calculations.