GDP weakness: Weather problem or measurement problem?

Editor’s note: This post originally appeared on Real Clear Markets on May 20, 2015.

Economic output (gross domestic product or GDP) showed a mere 0.2 percent growth rate in the first quarter of 2015.  Last year, it was even worse, with a 2.1 percent contraction in the first quarter.  Both years had especially harsh winters, but many commentators have now noticed that the pattern of weak first quarters has been around for longer — not just in years with harsh winters. As a result, they are starting to question if the adjustments that the statisticians at the Bureau of Economic Analysis (BEA) make for seasonal effects in their calculations of growth are in fact adequate. 

Unfortunately, researchers cannot adequately address the question of whether these numbers are low due to a seasonal adjustment problem because the BEA does not release the underlying not-seasonally adjusted data, nor does it provide details of how they implement the seasonal adjustment to the data. This secretive practice is bizarre and contrasts with other agencies, who are transparent about this crucial aspect of data construction.

Some researchers have attempted to assess the problem by looking for remaining seasonality in the supposedly seasonally adjusted data. Gilbert et al. (2015) test for whether there is a statistically significant remaining seasonal pattern, and find that the evidence is borderline. Rudebusch et al. (2015) seasonally adjust the official real GDP data, and find that growth in the first quarter moves up from 0.2 to 1.8 percentage points with this double seasonal adjustment. The two exercises are answering different questions, though the latter exercise seems more relevant to an observer attempting to get the best possible estimate of underlying economic momentum. 

Recent commentary on the issue unfortunately conflates the problem of seasonal adjustment with the unusually harsh winters in 2014 and 2015.  It is important to remember that these are two quite different questions. Seasonal adjustment controls for regular variation within the year, including normal weather patterns.  It does not intend to adjust for deviations in weather from seasonal norms. The conjecture is that the seasonal adjustment undertaken by the BEA is not working well.  This might be because the BEA ignores minor seasonal patterns in many components of GDP — while this might be reasonable for these disaggregate components, it leaves a substantial seasonal pattern in the aggregate data because the seasonal patterns are correlated across components. The BEA does not provide the information adequately to assess this conjecture. But in any case, it would be a problem with the seasonal adjustment process that is separate from weather effects from a couple of unusually cold winters.

It is a question of considerable practical importance.  To the extent that the apparent weakness in the first quarter owed to problems in seasonal adjustment, then a sharp bounceback should be expected for the second quarter, leaving the Federal Reserve on course to start raising rates, perhaps in September. But to the extent that is a sign that the economy could be losing momentum, then the Fed might be expected to delay the normalization of monetary policy. The BEA should release the underlying not-seasonally-adjusted data to allow the broader research community to properly analyze the question.