Unemployment has remained stubbornly high the past few years, which has led investors to become hyper sensitive on any relevant data in this area. Economists have generally been poor predictors of employment data over this period, which makes every employment report that comes out that much more uncertain.
Both weekly initial jobless claims and monthly employment numbers are adjusted by the government to account for seasonality, and in the past couple of months these adjustments have been quite large. Since weekly jobless claims typically lead monthly employment data, we have placed our focus on analyzing this data set.
Strictly looking at the data, it appears that seasonality may be making the jobless claims data look better than it really is, and consequently we expect jobless claims to rise in the coming weeks as seasonality fades away.
We have constructed a model to compare against weekly initial jobless claims in an attempt to validate recent reports. Our model is composed of three data sets:
1) Weekly internet searches for the keyword unemployment. It should not be a surprise that often the first thing someone does when they lose their job is to go on the internet and see how to collect unemployment benefits. The data here shows there is a modest, but consistent lead time relative to weekly initial jobless claims. (50% weighting)
2) The ASA Staffing Index, which tracks the hiring of temporary employment. It is well documented that companies tend to hire temp help before taking on full time workers. The data here shows a longer lead time, from a few to several weeks, consistently leading directional changes to weekly jobless claims. (35% weighting)
3) Daily withholding tax revenue as reported by the department of Treasury. This is real time data showing how much Treasury has taken in employee withholding taxes. This data can be volatile, but is an excellent gauge of not only how many people are working, but how much income they are earning. The lead time here is 2-3 months (15% weighting)
A few additional notes to help view the results. To account for seasonality, we have adjusted all our data to reflect year over year changes. Additionally, since some of these series have unique outlier events such as temp help falling during the holidays and bonuses being paid in December impacting withholding data, we have normalized our data into a 20 day moving average to reduce the noise.
Last, since some of the data series naturally move in opposite directions, we've inverted the ASA Staffing Index and the withholding tax data results so that they move together and areconsistent with a respective positive or negative event. That leaves the chart below showing the weekly jobless claims (black line) and the weighted "Elmwood Employment Index" (blue line) as a 20 day simple moving average.
The results are straight forward. Seasonality has clearly paid a much larger role recently in the government unemployment statistics as compared to more unbiased real time data. At last reading, the gap between these two series has widened to very significant levels. So in order for these two data series to converge again, either employment must improve substantially in the very near term, or as seasonality begins to fade unemployment claims need to rise, to more accurately match the real time alternate data suggested. We believe the latter is a much higher probability event.