What’s Happening With Unemployment? Unfortunately, It’s Pretty Hard To Tell.

We all know the unemployment rate is very high. Overall nearly 40 million Americans have filed for unemployment this year (roughly 1/4 of the workforce), and while some have gone back to work, most stores and offices are still not fully open. But it’s important to know the “rate,” because policymakers and business leaders use the data as a signal: is the economy getting better or worse? Should I rush back to work or wait until the Pandemic slows down? Should policymakers throw more money into the economy or not?

Well, last week I was digging around, and unfortunately, it’s hard to tell what’s really going on.

Last week the BLS announced that the US economy had a miraculous recovery. We created more than 2.5 million jobs, put over a million people back to work in retail, and the unemployment rate plummeted. Donald Trump and Jim Cramer immediately told us “The Economy is Back!” and the stock market reached new highs.

*Update: today the market absorbed new data and made a major correction of -7%.*

Yet these numbers didn’t feel right to me. My neighborhood and all my clients were telling me otherwise. So I decided to dig in.

First I called my friends at ADP, who have actual data on one out of every six employees in the United States (that’s how many people they pay). They told me that their data, which they carefully adjust to match the BLS industry mix, showed that the US lost 2.7 Million jobs in May. Almost the polar opposite of the BLS. I was kind of floored. Their data made much more “logical sense” to me so I got them on the phone.

Yes, their data is based on real payroll data, representing tens of millions of employers. The BLS data, by contrast, is based on a “household survey” (to households) and a “payroll survey” (to employers). While surveys are useful, people don’t fill them outright (and often ignore them) while ADP has real-world data on employers. And ADP pays one out of six people in the United States, so the data is very representative.

Being an analyst, I felt very unsettled. The BLS data seemed wrong. So on Sunday I talked with Heather Long at the Washington Post and she told me the BLS had a “misclassification error” but there was no political influence taking place. She was not aware of the ADP data and told me the BLS does payroll surveys also. I continued to be confused.

That evening my wife heard Ky Ryzdall interview Erica Groshen, the former head of the BLS. She is now a Cornell professor so I called her. Ms. Groshen was very supportive of the BLS and assured me that the agency was not politically motivated. Apparently these misclassification errors are not uncommon: the BLS regularly “revises” its numbers in the months following publication.

Well, yesterday Ms. Groshen published a pretty good explanation of what happened. The BLS data from May includes a question about whether people are working, and if you have been “furloughed” or “sent home” but may still have an employer, you are classified as employed. So the “actual rate of employment” was heavily overstated.

Her analysis points out a “consequential underestimate” of the official unemployment rate. Take a look. It was off by almost 40% in March, 30% in April and 25% in May.

The actual unemployment rate, according to Ms. Groshen, is more like 16.4% – and the unemployment rate in April, according to her, was 19.5%. (Which feels right to me, given the data on unemployment claims.)

But Wait. Did Jobs Go Up Or Down? Could ADP Be Right?

Which gets me back to ADP. ADP is a serious, hard-working, highly credible company. They have no reason to mislead anyone about anything. The ADP data shows that the US economy lost jobs in May.  Here is their data:

I talked with ADP about this and their numbers, which come directly from payrolls (ADP has anonymized near real-time data about when an individual drops off or joins a company), shows a major drop in services (-1.97 million jobs), a significant drop in manufacturing (-719,000 job), and a major drop in trade, transportation, leisure, and hospitality. And ADP also showed a loss of 22,000 construction jobs.

As you can see, the numbers are almost opposite to BLS surveys in many cases. In particular, the BLS huge growth in construction, manufacturing, services, and healthcare, while ADP sees losses in every one of those categories. Could it be a timing issue? Perhaps, but in either case, these samples are within a week of each other.

What Are We To Do?

I’m flummoxed. It makes perfect sense that the economy is starting to improve. People are opening up stores and many states are relaxing restrictions. But are we in the middle of a massive recovery that warrants a stock market rally to near all-time highs? It’s pretty hard to tell.

Yesterday the Federal Reserve (who apparently has their own jobs data which they keep confidential) announced that we would have high unemployment for years and that our zero-interest-rate policy may go on for two years or more.

(In response to this companies are starting to borrow like crazy – Amazon just issued debt at a .4% interest rate- and even philanthropies are borrowing money to fill their endowments.)

As far as the BLS data goes, they reserve the right to change their data (they “adjust” the numbers regularly) and both ADP and Ms. Groshens recommended we wait for their adjustments. Which is all we can really do.

My main finding this week is that we shouldn’t over-promote the BLS data when it first comes out, and the prospect of political interference does seem possible. Can you imagine the reaction from the Administration if the BLS announces that the unemployment rate is actually over 16%?

I asked the ADP folks about this and their reaction was “this is a messy topic and most analysts don’t really understand where all these numbers come from.” I have to agree.

As far as the BLS methodology goes, I’d suggest they ask the top 20 payroll companies to send them data once a month. That would give them near-real-time data on almost 80% of the American workforce, and we could see detailed data on jobs, hours worked, wages, and breakdown by age, race, and job role in detail. Right now I still see people quoting the 13.3% rate.

Remember that the “future of work” has already arrived. Full-time, part-time, gig work, furloughs, layoffs, job swaps – they’re all important. We need to upgrade our data capture tools to reflect what’s really going on.

I will continue to look at this data and try to make sense of it. Meanwhile, I suggest you read everything you can get your hands on and analyze your own business data in detail.