On a day when the Fed will likely overshadow everything with unearned attentiveness, the BLS released its annual benchmark for the Establishment Survey. Actually, it was just the preliminary estimate for the coming benchmark revision, which was curiously downward. It bears repeated that the BLS does not estimate the total count of either employed persons or payrolls but rather each monthly release is but a stochastic simulation of the ‘likely’ variation. This is not inappropriate, as the ability to collect and distill surveys and data and match them against broader inventories is quite limited. What is not appropriate is the certainty each monthly number is given by economists and then the media.
Almost all employers in the US are required to file an employee count for March each year, which provides the opportunity for the BLS to match its numbers against what it can glean from the state UI filings. This is the first of the annual benchmarking process, in arrears, but it bears emphasizing that it is not the last as even here sampling processes and incomplete data ‘requires’ statistical modeling to aid in the ‘final’ number.
The history of these annual revisions offers a limited interpretation on just the sort of trend-cycle issues that have plagued so many other data points. If you review that chronology, the trend-cycle variation stands out as large – especially in recessions the past quarter century (all three, so far). You find, as you might expect given trend-cycle influence (almost always a positive bias, with the exception of 2005, until outright recession proves otherwise), there are negative revisions clustered around contraction; before, during and after. Likewise, positive revisions suggest closer to cycle peaks or strong points.
This post was published at David Stockmans Contra Corner on September 17, 2015.
Recent Comments