Has the Fed Been Fooled by Phony Jobs Numbers?
James Bullard, the president of the Federal Reserve Bank of St. Louis, is on record as saying the Fed was prepared to start cutting its monthly bond purchases up until "weaker data came in" earlier this month. In all likelihood, he was looking at the monthly jobs report, which showed disappointing growth during the summer. Yet a new paper presented by Johns Hopkins economist Jonathan Wright at the Brookings Institution's Panel of Economic Activity indicates that the Fed may have been misled by meaningless data. The evidence suggests that employment growth was just as anemic in July, when the Bureau of Labor Statistics reported 104,000 new jobs, as it was in February, when the BLS reported that 332,000 jobs were added.
The problem is due to some peculiarities in the formula for seasonal adjustments. Weather, the school calendar and holidays all affect how many people are working in any given month, creating a lot of volatility in the raw jobs numbers. For example, the BLS reported that 1.2 million jobs were lost in July and 378,000 were added in August. Thanks to seasonal adjustments, however, most people think that 104,000 jobs were added in July and 169,000 were added in August. The truth is somewhere in between.
These seasonal differences don't affect estimates of the total number of Americans working -- about 136 million -- nor do they change the fact that employment has been growing slowly and steadily for several years. Yet many journalists, traders and Federal Reserve policy makers prefer to obsess over the monthly changes in the number of people working rather than focus on those salient facts. In fact, many people focus on the truly tiny differences between the monthly changes in employment forecasted by economists and the actual changes reported by the BLS.
This is silly, but it wouldn't matter except for two recent developments. First, the Fed has become hypersensitive to monthly jobs data. Second, the process by which the BLS smooths out its raw data seems to have been corrupted by a statistical artifact. As Wright explains, the job losses associated with the Great Recession were concentrated at the end of 2008 and the beginning of 2009 -- the coldest months of the year. That distorted the BLS's seasonal adjustment algorithm, which uses the past three years of data to determine the "normal" pattern of employment growth in different months.
This was first suspected by economists at Nomura and Goldman Sachs, as Cardiff Garcia reported at FT Alphaville. However, Wright's paper presents the first conclusive evidence. He starts by imagining that the job losses associated with the downturn were distributed evenly over the course of the year, rather than concentrated in the winter.
Wright then plugs his alternative scenario into the BLS's own model. Finally, he compares the result of his "improved" seasonal adjustment with the BLS's official numbers. As you can see in the chart below, the distortions have been large and consistent, although the effect has faded somewhat with the passage of time.
The big takeaway is job growth since the trough has been steady and slow. This has some surprising implications for Fed policy. It isn't in Wright's paper, but during the conference he presented another slide showing his chart of seasonal distortions with the start and stop dates of the Fed's asset purchase programs added on top. It turns out that the Fed stops buying assets after a few months of good job data only to restart buying once job growth slows, even though a proper measure of employment would show no acceleration or deceleration in job growth. In other words, the Fed has been reacting to meaningless data. (To be fair, staff economists at the Fed were aware of the seasonal issues caused by the Great Recession and updated their own measure of industrial production accordingly.)
As the years grind on, the distortion caused by the Great Recession will fade. Unless changes are made, however, the flaws with the BLS's methodology -- and the bizarre obsession we have with tiny monthly changes to very large numbers -- will remain.
Corrects numbers of jobs added in July and February in first paragraph.
This column does not necessarily reflect the opinion of Bloomberg View's editorial board or Bloomberg LP, its owners and investors.