This morning's Employment Situation Report was stronger than expected. By historical standards, the "beat" isn't especially large (175k vs 149k forecast), but most market participants (at least those expressing opinions) are surprised there was a 'beat' at all. The great debate on the impact of uncommonly cold/snowy winter is at the heart of this expectation. In my conversations, even those who really don't think much of the weather impact are still willing to admit it exists in some small form.
The Bureau of Labor Statistics agrees. Buried in the labyrinthine caverns of their data collection lay nuggets we can mine and ultimately refine into a usable conclusion, but this is not for the faint of heart. There are a lot of words and numbers in the rest of this analysis, but the end result will be a number that's likely somewhere close to the number of jobs missing from Nonfarm Payrolls due to the weather. (Hint: it ends up being about 50k, and this was reasonably-well circulated among the investment community, but a number is just a number without a detailed breakdown. This is that breakdown):
Reprinted with permission from an update that went out to MBS Live Subscribers earlier today:
UPDATE: What does all that "Weather Stuff" mean in the Payrolls Data?
Update Issued: 3/7/2014 9:41 AM
You're not alone if connecting these dots is a bit confusing. Hopefully this will be somewhat of a roadmap for the significance of the weather-related data in this morning's payrolls report.
Background facts:
- There are two surveys in the Employment Situation: Establishment and Household. NFP comes from Establishment, Unemployment rate from Household.
- In order to NOT count in NFP, a worker would have to be absent for the entire pay period.
- Just over 20% of workers have 1-week pay periods
- The Household survey asks workers if they had a job, but missed the whole week due to weather! (These people would be counted as employed in the Household survey, but not in the establishment survey).
From there, let's look at what happened last report vs current...
- In January's data (reported early Feb), 262,000 workers said they missed the whole week due to weather. That's historically low, and no big deal (although on an unrelated note, that does lend some credence to how some economic data was surprisingly strong in January vs December).
- The jump up in February's data (today) was big. There are now 601,000 workers saying they didn't work for the whole week due to weather.
- In both cases, 20% of these workers can be assumed to be 'missing' from the NFP figure (because 20% of workers have 1-week pay periods and NFP only counts you if you worked during that pay period).
- Such absences will ALWAYS detract from NFP (i.e. the fact that we can assume they're 'missing' isn't some scandalous revelation... it's always like that), but what we're interested in is the DIFFERENCE from the last report, and more importantly from the average/median February data.
So, from the last report where 262k workers didn't work that week to 601k claiming the same this week, we have a difference of
601k - 262k = 339k
20% of those have one week pay periods and therefore are not likely counted in NFP.
339k*20% = 67.8k
More important is the difference in this February vs the average past February (helps us assess "how many more jobs would we have seen without the weather?" The $64k question...)
The average number of workers reporting no work during the survey period for the past 10 Februaries is 356.8k (tally the Feb column on this page and divide by 11).
601k (this report) - 356.8k (average report) = 244.2k
So the difference between this report and the average is 244.2k. In other words, 244.2k fewer people than normal worked during the survey week. And 20% of those folks only have 1-week pay periods, meaning they wouldn't be counted in NFP. So.......
244.2k*20% = 48.84k
CONCLUSION: It's strongly possible that payrolls would be at least 48k higher if the weather effect had been in the middle of it's historical range (one of those months spiked to over 1000k, so using the historical median as opposed to the average makes this figure jump to over 70k).
