FT Alphaville has written a fair amount about seasonal distortions in economic data but thought this latest piece of research from Nomura was worth highlighting (mostly because it involves time-travel).
It pokes a small but important hole in the surprisingly low 348k new US claims for unemployment insurance which were filed in the week ending 11 February, the fewest since March 2008.
Nomura suggest that, although the decline is indicative of a positive trend, the scale of the improvement is almost certainly being warped by those darn seasonal factors — adjustments which have been distorted by the extra-large economic shock in 2008-09.
As the chart below shows, over the past three years claims have fallen from yearly highs marked in August-September to lows in February, only to stagnate in the following months:
So far, so normal.
As mentioned, we have written heaps about seasonal distortions over the last while, particularly where manufacturing surveys are concerned.
However, this approach differs by trying to bring to light the distorting effects of the 2008-09 financial crisis by showing how it affected data released prior to September 2008.
Says Nomura:
The way the deep contraction altered historical jobless claims (via revisions) is an indication of how distorted seasonal factors will affect future jobless claims reports
Seasonal adjustment responds to large shocks in predictable ways. Moreover, the effects are projected both forward and backward in the adjusted data. That means it is possible to identify the impact of large shocks on seasonal adjustment by looking at revisions to seasonal adjustment factors for prior periods
The “echo” of the recession appears to have changed actual seasonally adjusted jobless claims data in the way suggested by the hypothetical analysis. Each year the Department of Labor incorporates new jobless claims data and adjusts historical seasonal factors for the prior five years of the series (most recently in March 2011). Cyclical variation misidentified as seasonality distorts both prior and future seasonal adjustments.
Here, meanwhile, is a pretty chart from Nomura demonstrating just that (click to enlarge):
The dashed line shows seasonally adjusted claims from September 2008-August 2009 (note the right-hand scale is inverted). The most striking changes to historical data occur in January (increases) and February-March (decreases). Correspondingly, in January 2009 the number of initial claims slowed from a period of rapid increase in the SeptemberDecember 2008 period, and then sharply increased again in February 2009.
And Nomura thinks as much as one half of the decline since early January this year may reflect distortions in seasonal adjustment. Looking ahead, the best guess is that distortions are likely to be neutral in the next couple of months, and then turn modestly negative in the spring:
We expect the US economy to continue to grow at a modest pace in the coming months, but a lesson from the past few years of jobless claims data [as illustrated in chart one above] is that the February decline should not be expected to persist. And, while our analysis indicates that February-March seasonally adjusted claims are understated, it also suggests overstating in early January. As such, the undistorted trend likely lies between the extremes, closer to 370-380k.
However, says Nomura, should jobless claims stay near 350k into April and May we would have a real signal of a positive trend… finally.
Relating links:
US unemployment “progress” – FT Alphaville
The missing GDP – FT Alphaville
Seasonality bias in the ISM… fixed? – FT Alphaville
‘Tis (still) the seasonality, ISM edition – FT Alphaville
‘Tis the seasonality, hold the jolly – FT Alphaville