U.S. Employment

U.S. Employment is measured by the Bureau of Labor Statistics (BLS) in two ways:

• The Current Employment Statistics (CES) Survey
  • Monthly employment survey asking a sample of employers “How many people do you have on your payroll?”
• The Quarterly Census of Employment and Wages (QCEW)
  • Large administrative program that collects employment and wages via each state’s Unemployment Insurance (UI) program

Why two measurements?

• QCEW is a census
  • census - complete count of a population
  • Problem is that it is on a 4 month lag.
  • More realistically it is an 8 month lag.
• CES is produced monthly by a survey
  • survey - using random sampling to gather information about a population.
  • Only one month lag! - gets users information faster
  • Subject to sampling error

Who uses this data?

• The Federal Reserve relies on this information to set interest rate and make other important economic decisions
• States use it to make decisions about their own policies

QCEW Revisions

• QCEW data for a given quarter released ~5 months after that quarter
• Every release previous quarters data are updated
• We will call each months subsequent releases cut 1, cut 2, and so on..
• We call the difference between cuts, the QCEW revision
• QCEW Revisions can be found on the BLS website here
• We will care about cut 1 and cut 2 in this presentation

Benchmarking

• CES Benchmarks to QCEW for the month of March of the previous year.
  • Benchmarking is the process of setting a value to a perceived more credible value.
• CES releases two benchmark revisions, i.e. “How far away from the ‘truth’ am I?”:
  • Preliminary Benchmark in August using cut 1 of the QCEW
  • A Final Benchmark in February using cut 2 of the QCEW
• My question: How do QCEW revisions affect the differences between the Preliminary and Final Benchmark CES revision?

CES Revisions

• Final Revisions on BLS website here
• Preliminary Revisions
  • Current year’s can be found here
  • Also needed to gather old revisions using the Wayback Machine to find archived Preliminary Benchmark announcements since they overwrite them each year.

CES Revisions

Final - Prelim

Why am I interested in this difference?

• If CES uses cut 1 of QCEW for the Preliminary Benchmark Revision and cut 2 for the Final Benchmark Revision, we should expect this difference to correlate with QCEW revisions.

QCEW Levels Look Like This

If we look at the difference between cut 2 and cut 1, we get…

Revisions are strictly positive starting in 2021

• If I overlay the difference between Prelim and Final CES revision
• red points are CES differences. Black just to highlight March.

• We can make a good guess at the change in revision from Prelim to Final for this missing point

This point is also stands out (June 2022)

Philadelphia Fed

• The Philadelphia Fed does quarterly benchmarking to cut 1 of the QCEW

• In June 2022 they claimed “In the aggregate, 10,500 net new jobs were added during the period rather than the 1,121,500 jobs estimated by the sum of the states; the U.S. CES estimated net growth of 1,047,000 jobs for the period.”

• Thats a claim that BLS was off by 1 million jobs!

• June 2022 Report

This outlier contributed greatly to the error in the fed report

June 2022 Outlier

• Further differences between the Fed’s number and CES explained by:
  • subsequent revisions in the QCEW
  • seasonal adjustment
• The outlier itself was caused by employers submitting erroneous “third month zeros” when submitting Unemployment Insurance tax data

June 2022 Outlier

• No smoking gun found but possibilities include:
  • Calendar effects
    • When do employees get their paychecks and are they getting counted for that month?
  • Payroll provider issues

Conclusions

• Administrative data can have its own risks
• Revisions to the QCEW impact the reliability of the products created by its users

Presentation Materials

Git repo: https://github.com/kerr14333/employment_revisions Slides: https://rpubs.com/kerr14333/employment-revisions