Sunday, May 08, 2011

Employment: A comment on the Birth/Death Model

by Bill McBride on 5/08/2011 12:12:00 PM

On Friday, I wrote in frustration:

[A]nyone who adds (or subtracts) the Not Seasonally Adjusted (NSA) birth/death model numbers from the headline SA payroll employment is clueless. Someone sent me this "analysis" today: "... you exclude the 62K from McDonalds hirings, and 175K from the Birth Death Adjustment, and end up with.... +7K jobs". That is complete nonsense. The key issue with the birth/death model is it misses turning points; otherwise it is an important part of the monthly estimate.
As I noted on Friday that "analysis" was complete nonsense. My suggestion is that most people should just ignore the model, but for those who want to understand it, below the fold is a slightly technical discussion with links to complete explanations.

To commentators: Please stop subtracting the BD numbers from the headline BLS numbers! You are misleading your readers.

The monthly payroll report from the BLS is a survey based estimate of the employment situation. A key purpose of the report is to provide a timely snapshot of the number of payroll jobs added or lost each month.

Like all surveys, the monthly estimate is subject to both sampling and nonsampling errors. A primary source of non-sampling error is due to employment growth generated by new business formations. So, as part of the monthly report, the BLS uses a model (the "Birth/Death model") to estimate this employment growth.

Here is a description of the CES Net Birth/Death Model and some frequently asked questions. Historical Birth/Death factors and an overview of the CES birth/death model.

A few years ago several people - myself included - pointed out that the Birth/Death model would miss turning points in employment. I thought the model would overstate the number of jobs added as the economy slid into recession (and understate the number of jobs lost monthly during a recession). Sure enough that is what the annual benchmark revision showed during the employment recession.

Note: Every year, the BLS benchmarks to state unemployment insurance tax reports:
Each year, the CES survey realigns its sample-based estimates to incorporate universe counts of employment—a process known as benchmarking. Comprehensive counts of employment, or benchmarks, are derived primarily from unemployment insurance (UI) tax reports that nearly all employers are required to file with State Workforce Agencies.
However, in general, the benchmark revisions are very small. So most of the time the Birth/Death model is very good.

Unfortunately a few years ago, after several of us pointed out that the Birth/Death model would miss the turning point - several commentators started subtracting the monthly BD number from the monthly payroll number. That was completely wrong. First, the BD number is Not Seasonally Adjusted (NSA), and it can't be subtracted directly from the Seasonally Adjusted (SA) payroll number. Second, even if we could remove the BD number, this would reintroduce the primary non-sampling error - and defeat the purpose of a timely month estimate!

Since we are (hopefully) at another turning point, and the economy is adding jobs, does this mean the employment situation report is now underestimating the number of jobs added? Maybe. We won't know until the next benchmark revision early in 2012), but the errors probably will not be large because the BLS is now "updating the Current Employment Statistics (CES) net birth/death model component of the estimation process more frequently, generating birth/death residuals on a quarterly basis instead of annually."

As I noted above, most people should just ignore this model and use the headline number.