Employment is the most important measure of economic strength. It is also crucial to overall satisfaction, consumer confidence, and spending. Because it is so important, any data release gets close attention. Sadly, the typical tradeoff is in full force:
The best data is not timely, and the timeliest data paints a fuzzy picture.
This post analyzes the most important sources.
Initial Jobless Claims
This is an excellent source. The claims are reported to the Department of Labor by states with a delay of only one week. Because of strong seasonality, the best view is seasonally adjusted. The four-week moving average is widely followed and helps to smooth the series.
The biggest drawback is that it tells only half of the story – people who have lost jobs. It is mistaken (but popular) to rely on this indicator alone. It can be disrupted by weather or other major events. Occasionally states do not report on time.
The Bespoke chart provides perspective for the recent results.
Business Dynamics and Quarterly Census of Employment and Wages
This report is the most accurate overall look at the employment story. It is the only place where we learn about job gains and the sources of these gains. The data are aggregated from state employment offices. Since businesses have no incentive to inflate the number of jobs, paying extra taxes, this is a solid, non-survey source. Other data series, like the payroll jobs reports, are annually benchmarked against these results, making sure that the monthly estimates do not stray far from reality. This is one major source of revisions.
The drawback is that it takes about eight months to compile and report the data. By the time it is released, few are paying any attention. Here is one table from the most recent report. It illustrates some of the highlights.
The table shows several key points:
- Monthly job gains are on the order of 2.5 million and the losses 2.2 million. The widely reported payroll report is the NET job change. This is important, but just the tip of the massive churn of jobs created and lost.
- We can look at the actual net job change to see if our estimates in that quarter were very accurate.
- Most jobs gained and lost are in existing establishments.
- Job gains from new establishments represent an important contributor.
The Job Openings and Labor Turnover Survey is the best source to evaluate the tightness of the labor market. It is not a good method for evaluating overall job growth, although many try to use it that way. The most important aspects of the survey are the comparison of job openings to job seekers and the evaluation of the job structure – the fit of workers to openings.
ADP Private Employment
ADP reports on private employment, usually two days before the “official” BLS report. The ADP approach formerly used their own data about payrolls covered in their service business. Recently they improved the method to provide a better fit to the BLS series. It now includes the following:
- ADP matched-pair growth rates by industry
- Lagged values of BLS estimates of growth of employment by industry with industry specific restrictions
- Unemployment Insurance Claims (UNI_US)
- Oil Prices
- The Michigan Consumer Sentiment Index (CSENT_US)
- The Composite Index of Leading Indicators (LEAD_US)
This report is a strong addition to our toolkit because the method differs significantly from the BLS approach. It helps to have estimate from competing sources. Here is their latest result.
BLS Establishment Survey (Payroll jobs)
The establishment survey gets the most attention from Wall Street. Strong points include that it focuses on jobs (not people) and the survey collection is done from businesses. The businesses report the number of payroll positions for the week containing the 12th of the month, so the report is relatively current.
The results are so complex that it is easy for pundits to spin the interpretation. Historically, the report has been criticized on many grounds – mostly unwarranted. The BLS statisticians are not politically motivated nor under instructions to change any results. The birth/death adjustment effectively corrects what would otherwise be an undercount. (See Business Dynamics above). Revisions reflect additional data and improve accuracy. The seasonal adjustments are done using the most advanced statistical methods. Etc.
My own objection is that it is a difficult way to measure change. The overall number of jobs is about 150 million. The BLS approach is to estimate the number one month, via survey, and then repeat the process the next month. The difference is the estimate of net job change. This is like weighing a 7500-pound elephant in two consecutive months and attempting to identify a seven-pound weight change!
The sampling error for the headline number, a 90% confidence interval, is about 120K. This is much larger than the changes that regularly move markets.
Revisions are also misunderstood. Mostly they reflect additional responses. There is some tweaking of seasonal adjustments. Revisions do not eliminate the sampling error, which is based on a full response. And the direction of revisions cannot be predicted, unless you have an accurate theory about the non-responding businesses.
BLS Household Survey (Unemployment and participation data)
The household survey counts people, not jobs. One person might have two or more different jobs, for example. This is an important separate look, providing key information about the unemployed. There are different measures aimed at determining whether people have left the labor force and for what reason. The data provides an opportunity to dig in and examine whatever measure fits your needs.
A drawback is the sampling error, which is +/- 450,000 for the overall number of employed. The BLS has a methodology for comparing the payroll and household surveys, but they are really directed at different purposes.
We would always prefer simple and decisive answers from the data, but we won’t get it from the employment report. The punditry will discuss changes in subgroups without ever mentioning the confidence interval of those changes. Producing many pages of data does not make it more accurate or valuable.
My best advice? Do not take any single report too seriously. Look for trends in the overall series and consistency with other indicators. If there is a wild reaction from this single report, it is probably wrong. Long-term investors can use it as an opportunity.
My base case right now is a real GDP growth rate of 2.5%. Tomorrow’s numbers will not reliably deviate from that.