A guide to fixing the US jobs data problem

The Star

5 days ago

President Donald Trump started a firestorm over the reliability of US economic data when he fired the head of the Bureau of Labor Statistics (BLS) following the agency's July non-farm payrolls report.
The report wiped 258,000 jobs that were previously counted in May and June, the largest downward revision in decades outside of the Covid pandemic.
A miss that big was guaranteed to draw scrutiny.
The BLS's jobs and inflation data are arguably the most important numbers the government compiles because the US Federal Reserve (Fed) relies on them to conduct monetary policy.
The Fed has long been criticised for reacting too slowly to economic developments, an inevitable consequence of relying on lagging government data.
Now the Fed and BLS are likely to face even more pressure to catch up.
The timing and reliability of economic data have real world consequences. The Fed held short-term interest rates steady at its July meeting, but it most likely would have lowered them if the BLS had delivered a more accurate payroll count in May and June.
The bond market certainly thinks so: The two-year Treasury yield, which reliably anticipates the Fed's interest rate moves, shaved 0.25 percentage points immediately following the jobs report in anticipation of a rate cut.
But the problem is much bigger than a single person atop the BLS.
Most of the data the agency compiles is based on voluntary surveys.
They take a lot of time to collect, and response rates have been declining for years.
The response rate of the Job Openings and Labor Turnover Survey (Jolts) was down to just 35% for the March report.
BLS said it recognises its data quality issues and is using new tools to gather information, such as video and web-scraping.
It's not just BLS. Survey-based data has a lot of people scratching their heads lately.
Polls turned out to be less predictive during last year's presidential election than betting platforms such as Polymarket.
Consumer sentiment surveys have also been a less reliable barometer of the economy than market-based gauges such as bond yields and stock valuations, which continue to signal a mostly stable economy despite gloomy sentiment.
That shouldn't be surprising. Market-based data leverages technology to reach many more participants than antiquated surveys, giving it greater depth, speed and accuracy.
I suspect the central bank would agree. The St Louis Fed compared job postings from hiring platform Indeed with job openings in Jolts reports.
The overlapping data tracked closely, except that Indeed's data were less volatile and more frequently available.
If I were a policymaker relying on jobs data, I know which one I'd prefer.
The federal government can do a lot to modernise jobs-related data collection.
It can require public companies to report material changes to their headcount, if any, on a weekly basis.
Businesses keep track of their number of employees in real time, so it shouldn't be burdensome for them to do, and it would improve the speed and reliability of the jobs data tremendously.
Roughly a third of US workers are employed by a public company, about the same percentage as represented in the BLS's jobs survey.
The government can further broaden the data set to include employees of privately held businesses by leveraging tax withholding information.
Most US workers are employees, which usually means their employer withholds taxes from their paycheck at least twice a month and remits it to the Internal Revenue Service, giving the service a current count of payrolls.
It's not a new idea. The Bureau of Economic Analysis contemplated using tax withholding data to estimate wages and salaries nearly two decades ago but ran into obstacles around extrapolating income from withholding.
Simply counting payrolls, however, would be much simpler and certainly more comprehensive, reliable and timely than the survey data the BLS currently collects.
Payroll processor ADP already does it with its National Employment Report, although it only covers the 25 million workers within its reach, or roughly 20% of the workforce.
Monthly changes in ADP's report have been highly correlated with changes in BLS' numbers since 2010.
By the way, a robust payroll database would also allow the government to deliver targeted stimulus directly to workers when needed.
During the pandemic, workers were meant to receive assistance through the Paycheck Protection Programme, but much of the money was pocketed by business owners, creditors and suppliers instead.
Using withholding data to pay workers directly would have been more secure and effective – and can be the next time.
If Trump wants government data to be more 'fair and accurate', as he claimed in the wake of the July jobs report, his administration will have to marshal a broader vision than merely replacing the head of a single federal agency.
A good start would be for BLS and other government agencies to coordinate efforts to transition from surveys to real-time market data. — Bloomberg
Nir Kaissar is a Bloomberg Opinion columnist covering markets. The views expressed here are the writer's own.