Yesterday’s unemployment figures were treated with a sigh of relief in most financial markets. Yes, unemployment did rise, but so – supposedly – did job creation, to +163,000. It didn’t take long, however, for Zero Hedge to take that apart:
In other words, of the 163,000 jobs “added”, 429,000 was based on purely statistical fudging. Doesn’t matter – the flashing red headline is good enough for the algos.
While ZH is known for his cynicism, the statistical adjustment is not foreign to statistics. The process is better known as normalization. Essentially, it is a way of making sure data is not so different that it can’t be used for analysis.
The best example of data normalization involves economic or cost figures over several years. Unless you take out the effects of inflation from one year to the next, your data will be skewed and you’ll be unable to gauge anything. Hence the difference between nominal GDP (which includes inflation) and real GDP (which doesn’t). The same concept applies when comparing cost from year to year.
Of course, even a simple matter like normalizing for inflation has its concerns. Does one use the Consumer Price Index? The Producer Price Index? The GDP deflator? The indicators determined by the government (for its own cost projections)? In the case of the last one, the military has at least 15 different inflation indicators depending upon the branch and the budget category.
If that’s not complicated enough, let’s try job number normalization.
In this case, the Bureau of Labor Statistics is trying to account for changes in weather or industrial schedules that could skew a monthly job number. Or, to put it another way, BLS is trying to create “average months” in terms of weather and schedule so it can determine how much job growth (or loss) is due solely to economic conditions.
Of course, just like inflation, the method of normalization is far from certain. After all, one has to consider the jet stream, El Nino/La Nina, solar effects, “climate change”…
…and that’s just the variables for the weather.
Now, in a humming economy (or a recession), the trends can be seen through any normalization issues. Yet in an economy like this – somewhere between slowth and stall – the data can get badly skunked by the very attempt to set it right, unless that attempt itself is perfect.
There is reason to be concerned about the July attempt. As ZH noted, July’s adjustments have risen every year for the last half decade. Population increase could justify that, but the labor force’s decrease over that time suggests something else again.
It would be better for all if BLS opened up about how they normalize their data, instead of just the top-level adjustment number. Then we can have a more informed conversation about these statistics.
Cross-posted to Bearing Drift
