At the heart of these revisions is a horribly flawed birth-death model used by the BLS. My calculation closely matches an estimate by Bloomberg’s chief Economist.
This is a complicated post that goes into some of the more arcane procedures the BLS uses to produce the monthly jobs report.
But please read on. I will tie everything together after a discussion of the data used in this post.
As a bonus, I have confirmation from Bloomberg’s chief economist who arrives at similar numbers.
BED Notes
- BED stands for Business Employment Dynamics. It is based on reporting of about 9.1 million private-sector businesses making it very accurate.
- The BED report provides an accurate count of job creation by new businesses, and job losses by businesses going out of business.
- BED is a large subset of the Quarterly Census of Employment and Wages QCEW which covers 11.6 million businesses
- The problem with BED is timeliness. On July 24, the BLS just released BED data for the fourth quarter of 2023 (Oct, Nov, Dec).
- October to July is a 9 month wait. QCEW data has an even bigger lag.
Birth-Death Notes
- In order to get around the huge time lag of BEDs, the BLS uses a Birth-Death Model to estimate net job creation by new businesses and businesses going out of business.
- The BLS explains “The net birth-death forecasts are not seasonally adjusted, and are applied to the not seasonally adjusted monthly employment estimates to derive the final CES employment estimates.”
- CES stands for (Current Employment Statistics). It’s part of the monthly jobs report.
- The BLS does not calculate and does not know how many jobs the Birth-Death model adds or subtracts (to or from) the monthly jobs report.
CES Notes
- CES, the monthly jobs report is based on about 670,000 businesses vs 9.1 million for BED and 11.6 million for QCEW.
- CES does not have timely data about business openings or closings so the BLS uses a birth-death model to estimate.
- Every month in my synopsis of the jobs reports I state “Do not subtract the reported Birth-Death number from the reported headline number. That approach is statistically invalid and wildly inaccurate.” Yet every month I see people doing that calculation.
Synopsis!
In normal times, not entering into recession or out of recession, the birth-death model is subject to mostly random fluctuations.
The monthly jobs reports are not very accurate, but errors generally balance out over time.
It’s another matter near recessions and the last one was one for the record books.
It’s important to note that BED and the Birth-Death model numbers are both by the BLS and are supposed to measure the same thing (net new job creation or losses by businesses).
BED Net Jobs vs Birth-Death Model
In the Covid-Recession, massive numbers of businesses closed then six months later massive numbers of businesses opened.
Let’s throw away all that and just look at what’s happened between 2022 and now.
I repeat the lead chart for convenience.
BED Net Jobs vs Birth-Death Model Net Jobs
Chart Details (Both Sets of Numbers from the BLS)
- 2023 Bed Q2-Q4 Sum: 484,000 jobs
- 2023 Birth-Death Q2-Q4 Sum: 1,263,000 jobs
- The difference between the two reports is 779,000 jobs.
Here’s a second opinion.
Statements from Anna Wong, Bloomberg Chief Economist (via email)
Labor-market watchers broadly have been split into two camps over the past year: those who saw resilience — justifying the Federal Reserve’s higher-for-longer rate stance — and those who saw momentum fading — which could warrant bringing rate cuts forward.
Bloomberg Economics has been in the latter camp, which is gaining traction of late. Cooling in the labor market has long been evident, in our view, particularly in real-time firm birth-and-death data.
The Bureau of Labor Statistics’ fourth-quarter Business Employment Dynamics report confirmed nonfarm payrolls data have overstated labor-market strength for almost a year.
As such, we think the labor market hasn’t just approached an inflection point — rather, it’s likely passed it. If the Fed waits until September to cut — still our baseline — it’s probably too late.
The nonfarm payrolls report, one of the timeliest labor-market indicators, is among the most-followed economics releases. Data are collected through the BLS’s Current Employment Statistics (CES) program. However, the trade-off for timeliness is reliance on forecasts for the net employment change from firm births and deaths — the true value of which is only known with a very long lag.
That creates a potential for CES revisions at key points of an economic cycle. For example, CES-based payrolls data for 2023 won’t fully reflect underlying firm net closures until early 2025.
For a more timely read on firm births and deaths, we rely on two other BLS reports: the Quarterly Census of Employment and Wages (QCEW) — which provides the most comprehensive employment data among official statistics — and the Business Employment Dynamics series (BED) — a longitudinal sub-sample of the QCEW. Both cover a substantially broader set of firms than the CES survey, which only covers 670k establishments.
Based on that data, we estimate that nonfarm payrolls as reported by the establishment survey will likely be revised down by 730k for the last three quarters of 2023.
QWEC Establishment Counts Have Plunged Since Mid-2023. The 4Q23 BED data confirm that assessment.
The reason CES birth-death factors are exaggerated is the lagged availability of firm deaths data, and the forecasting equation of those factors is backward-looking, and heavily influenced by the elevated rate of business formation during the pandemic.
Looking beyond 2023, we use Chapter 11 bankruptcy and business formation filings to estimate what net employment from firm opening and closures looked like in 1Q24. The results suggest the downtrend continued. Bankruptcy data for 2Q24 isn’t available, but business formation data for 2Q24 has cooled further, and is down 5.3% since 4Q23.
Bottom line: The labor market has been cooling for awhile — the deterioration isn’t sudden. Given its dual mandate, the Fed is likely behind the curve on cutting rates. As such, we expect the unemployment rate to reach 4.5% by the end of 2024.
Pending Negative Revisions
Let’s use Anna Wong’s estimate of 730,000 overstatement vs my calculation of 779,000.
730,000 represents a BLS jobs overstatement of 81,111 jobs every month for 9 months.
On top of that, does anyone believe the birth/death total for 2024 Q4 of 653,000?
That’s the highest quarterly total dating to at least 2017 Q1. And we will not know how flawed that number is until the end of 2025.
Jobs Much Weaker than Expected, the Unemployment Rate Ticks Up
Counting negative revisions, there was unexpected weakness across the board in June, especially private and manufacturing payrolls.
Jobs are from the CES monthly survey, discussed above. Employment and unemployment are from a separate BLS Household Survey.
The discrepancy is massive. But BED and QCEW confirm which one is accurate.
Yet, every month, economists go gaga over garbage headline jobs numbers.
On July 5, I noted Jobs Much Weaker than Expected, the Unemployment Rate Ticks Up
Job Stats vs One Year Ago
- Nonfarm Payrolls (Blue): +2,611,000
- Employment (Red): +195,000
- Full Time Employment (Yellow): -1,551,000
BLS Nonfarm Payroll Prior Revision Magnitude
On July 7, I asked How Much Faith Do You Have in BLS Job Reports?
Let’s check out BLS job report revisions from January 2023 through June of 2024 in a series of four charts.
There were only three upward second revisions in 18 months.
In the above post I go over QCEW data, a bit more detailed than the BED subset and estimate a pending revision of ~900,000.
These drop-by-drop revisions may eat into that, or not. But the birth-death second quarter whopper of 653,000 is not in QCEW data and we need to wait until late 2025 to know what it really happened.
Recession Has Already Started
I repeat my July 8, 2024 recession call: Weak Data Says a Recession Has Already Started, Let’s Now Discuss When
I’ve seen enough. A recession has started. Let’s discuss starting with a very good indicator that has few false positives and no false negatives.
“All Hell Breaks Loose” In the Next Few Months as Recession Bites
For more details a second opinion on recession, and more charts, please see “All Hell Breaks Loose” In the Next Few Months as Recession Bites
Payrolls are on track to be revised downward, but we believe that’s not because payrolls are overcounting but because QCEW is undercounting,” said Sam Coffin, an economist at Morgan Stanley. “Because QCEW is based on UI records, it likely misses counting those who are not authorized to work. If one’s not authorized to work, one also is not eligible for unemployment insurance benefits. In contrast, the payroll survey asks that employees be counted regardless of legal status.”
I agree that big revisions are going to unfold and probably impacts rate cuts and politics in a very strange way:
quarter 2024 data from the QCEW on August 21, 2024, at 10:00 a.m. (ET). Preliminary benchmark revisions for March 2024 for all 50 states, the District of Columbia, and selected metropolitan areas at the total nonfarm level will be made available shortly thereafter..
This was just published from CBO and there’s a lot of overlapping stats coming from many agencies related to how immigration dynamics alter the recession narrative:
“ The overall unemployment rate is mostly unaffected by the immigration surge because two factors offset one another. In 2024, during the peak of the surge in CBO’s projections, the unemployment rate of new arrivals is relatively high (because it takes them time to find work), which increases the overall unemployment rate. That effect is offset by an increase in the demand for goods and services following the surge in immigration, which reduces the unemployment rate. On average, the unemployment rate is about one basis point lower from 2024 to 2034 because of the surge. (A basis point is one one-hundredth of a percentage point.) Although the unemployment rate is largely unaffected, the number of people who are unemployed in the economy increases as the labor force grows.”
It’s important to see the entire picture but we don’t have the data yet
When you constantly take real, live, and accurate data, and manipulate it to match the Agenda/Narrative you are pushing, then you also constantly have to Revise said data. That’s the easy part…
Like all lies, which is essentially what it is, when putting the information back together, it can get really messy, ugly, and so distorted (like now), that even the real, live, accurate data is no longer accurate or believable.
Distortion is an ugly way to carry yourself, and your thoughts. While Jealousy Is the Worst imo, to distort is right up there…
Even with this rain I can hear the husky buzzing of motorcycles on the highway going too fast. Too fast. And why? Because some days nothing matters. Nothing at all. And we see that more and more. Nothing and nobody matters. The social niceties tattered by trauma now gone. Just fight the way through the day. Find the money to pay the bills.
Find a place to stop worrying if only for a few moments.
Don’t pay any attention to the people around you. They don’t matter. Now. Ride in the rain as fast as you can.
There are still some who believe in kindness. They are overwhelmed with others’ needs. They are drooping in the rain now like the maple leaves against a grey sky. They are tired. They keep on going until they don’t. How much heart is enough? Where is the end to this time? This pressing want and anxiety?
The hard rains that are coming will sting bitterly. It is said this will be worse than the Great Depression by those who study such things.
But we don’t believe that. We can’t. How can we? A human cannot live without hope and that sounds about as hopeless as can be.
https://off-guardian.org/2024/07/26/moths-in-the-rain/
The Little Red Hen has moved on. Mary Mapes Dodges’ 1874 version of the hen finding and planting wheat and baking bread, and then not giving any to the other animals that didn’t help is no longer relevant. In some later versions, the hen breaks down and gives them some, but only if they promise to help the next time. In today’s version, the little red hen trades grain futures, and does a leveraged buyout of a bakery, fires the employees, and installs robots. When the farm economy collapses, the little red hen takes her Bitcoin and settles on a 10,000-acre farm in New Zealand.
Survival of the fittest. It’s the American way. And it’s promoted by many here.
I find it best to focus on helping family and friends. Nothing wrong with helping strangers, but you can’t help everyone. There are limits to what any individual can do.
If you look after yourself and your friends and family, then you don’t become part of the problem when times get tough. You can be part of the solution.
I have been following this path all my life. And I’m still waiting for that depression that so many keep predicting. It will probably never happen in my lifetime. But I’m well prepared if it does.
Too much kindness leads to a lot of cynicism.
So… what are the option when the US/world… crash into recession?
Dropping rates and infusing stimulus… will surely only drive inflation even higher… pushing huge numbers of people into the abyss of insolvency and much deeper despair…
Eventually the raging inflation will give way to a deflationary death spiral as discretionary spending craters…
We are experiencing a perfect storm https://fasteddynz.substack.com/p/the-perfect-storm
You summed it up pretty well I’d say
Concur.
I have been listening to the same dialogue my entire life: people complaining about economic stats being either incompetently wrong; or brilliantly manipulated.
Sorry to tell you all. They are neither.
Eeonomic stats will never be accurate; they are simply best guesses and they are continually updated as more data comes in.
Which is why I don’t waste too much time on them; better to follow the longer term trends that they hint at. And let someone else delve deep into the nitty-gritty short term details.
That’s why I pop in here from time to time. Mish puts a lot of time and effort into analyzing these best guesses. And after doing so for years now, he can barely improve on them. Thank you Mish for trying!
Meanwhile, is anyone else enjoying the volatility in the stock markets lately? A traders delight.
One new recommendation. Baytex energy dropped today after posting results. They spent the majority of their capex in the first half, which hurt their results. But they should reap the rewards in the second half.
Cheers!
For those positioning for a Bear market with risk off based upon the idea that unemployment rising would be the event that triggers consumer retrenchment,
all to think is wow.
Overstating job growth by 730,000 is a big deal. Instead of trading in and out for swings. it would appear sitting with a position the better option. When it breaks the move would be fairly dramatic.
Consumer spending in far greater perilous condition.
add: J Powell and the Fed are setting monetary policy based upon 730,000 non existent jobs. Crackup imminent for consumer if this analysis proves out.
This tracking of revision magnitudes would be FAR more useful if they were shown going back for a much longer time range.
Danielle DiMartino Booth has too much botox on her face. This also reflects the politics and the state of our economy
A one- hour political youtube video is beyond my tolerance level. If someone is too lazy to post a transcript, which is easily generated with speech-to-text software and cleanup editing, it cannot be important. Her face is irrelevant.
rip it as an mp3 and listen at your leisure. no need to actually watch any videos.
She also has humongous fake knockers.
It is easy to jump to concluding that a) BLS is incompetent, or b) BLS fraudulently reports data for political reasons, etc.
IMHO, it is possible, even likely, that the variables in the Births/Deaths model (and other models) simply fail to capture what happens in the real world. As a general rule, government does not respond well to crises, except by making laws and throwing money around.
In a previous post, on the ‘inverted yield curve’, I deliberately referred to real retail sales:
https://www.macrotrends.net/1371/retail-sales-historical-chart
The Covid impact on real retail sales is very visible. A sharp decline (economic shock), a sharp recovery, a slight lull, and then a second sharp increase, before beginning a gradual, yet erratic decline to the present. That’s REAL SALES, not inflated sales. Does it match to your Covid consuming behavior?
Assuming the data is trustworthy, this pattern transfers from retailers to suppliers, to producers, to their suppliers….but with time lags, anticipation of trends that might not be justified, and the usual impact of the economic constructs of accelerator and multiplier, etc. There has never been a phenomenon like it in 30 years, with the possible exception of 9/11. Predicting sales in this environment with accuracy–good luck. Employment decisions? Staring a business? Etc.
Remember, your prior experience is a steady growth in real retail sales (from mid-2009 to Dec 2019)…
BTW, where were we on the business cycle when Covid struck?
I’m with Mish, expect more revisions.
That Macrotrends chart is inaccurate. It’s a mathematical error to adjust retail and food service sales using the headline CPI. Headline CPI doesn’t accurately measure the cost changes specific to retail sales.
Headline CPI is currently driven by the services sector; goods inflation (which weighs heavier in retail sales) is quite mild now. So right now, applying the full headline CPI to retail sales is an over-correction. I think that’s partly why the graph has been drooping for the last 1-2 years.
Similarly (but in the opposite direction), the 2021 “COVID spike” on the graph was partly because there was a surge in retail-goods prices that was NOT captured in headline CPI. At that time the inflation hadn’t spread to services yet. So the CPI under-corrected for the growth back then. I think everyone will agree there was a surge in retail sales then (due to all the stimulus spending, while the services sector was still partly shut down)… but I doubt it was as large as the graph shows.
For retail sales, what I’d like to see are aggregated unit sales data – are people buying as much actual stuff as they used to – but those are hard to come by.
Even better would be unit sales per capita, ideally broken out by age, income and geography demographics. That would show which groups of people are genuinely better off now than before.
Note my proviso above: assuming the data is trustworthy. And yes, there are better measures we would like to see. Whether they significantly affect the outcome, who knows? While stimulus boosted sales, so did pent-up demand. For my purpose, ‘eliminating’ the inflationary effect post covid is important–but I don’t have the interest to crunch numbers. Orders get placed and fulfilled for real things, which affects real employment. The inadequacy of BLS models to fill in for inadequate data in erratic situations (covid and after) may be cause of revisions.
The big issue, as you state, is whether ‘people (are) buying as much actual stuff as they used to.’ BTW, I think that might be the case post 2022 in the chart–an early slowdown phase at the consume level. If true, one ‘predictor’ of recession’ with a lead of several months, would be a significant decline in real retail purchases. An increase in credit card debt might be another predictor (living beyond one’s means).
When will the BLS announce the revisions? The day after the elections?
Falsifying economic data is petty theft for a government that commits lawfare, election fraud coups, and attempts to assassinate their political opposition.
And just like that, poof. The jobs up and vanish like a fart in the wind. Dark comedy economics.
Given the importance of employment data, I find it very hard to believe that the United States cannot dramatically improve this system to generate data that is both timely and accurate.
The current system relies on models which are known to be systematically wrong at economic turning points, precisely when this data is most important to people.
Continuing a system which harms people precisely when they need it the most is an evil behavior pattern. It would be considered malpractice in medicine. It would generate lawsuits in industry. It would lead to customer boycotts in retail or banking. Why is it tolerated in econometrics?
It is tempting to consider malfeasance; however, a more likely culprit is simply how people think. Beginning in pre-school, kids are conditioned to memorize and repeat back. Conditioned thinking dominates, even when it might be less than successful. The Covid response was a prime example–this is how we treat a medical problem…If there were alternative (perhaps better) ways, they were censored.
A buzzword in education is critical thinking; however, what results in schools is more conditioned thinking purporting to be critical. IMHO, critical thinking requires two elements to begin: 1) crap detection, and 2) open minds. If you already believe in global warming, you are not critically thinking; and if you believe there is no such thing, you are also not thinking critically. Creative thinking is similarly in short supply, especially in the government–the reasons should be obvious.
All of this is relevant to why econometric models fail. Crap for data is part of it? Then, how economists are trained–heavy on statistics/multi-variable regression models, and some use of times series analysis. Artificial intelligence? SImulation models? Not so much. Critical minds willing to break rules and explore new avenues? Not in the government.
Why are the models wrong at turning points? Because the models have ‘memory’ and want to continue straight is part of it. Also, the variables that induce non-linear behavior are not in the models. Do the model include ‘shocks’ like December 2019? LMAO
I find the monthly unemployment household survey most useful for political interpretation. As fr revisions, oI simply cannot comment either way until and unless Mish decides to dive deeper and go back as far as he can so we can understand how that has worked over time. Going back only to 2022 tells us very little.
By contrast, when I have mentioned the U3/presidential election connection, I have gone all the way back to 1948. I put an asterisk on 2012, because of a methodological change in 2011. I don’t know anything about the birth death model including how long it has existed and how it’s been used over time, which means I cannot know how or why its accuracy has or has not fluctuated in the past.
Therefore, I pretty much ignore his discussion of that, other than to ask for more data, a request he has ignored, which is certainly his privilege. Oh well. Take what you need and leave the rest, as the song went.
And yet there are still help wanted signs up all over town. Shrug.
Which partially explains the increase in jobs while full time employment has been terrible. Help wanted signs are for zero skill labor so all this tells you is that if you want a part time job washing dishes at 2 different restaurants your opportunities are endless.
or as long as people are eating out and making dirty dishes and employing waitresses. The funny thing that no one mentions in economics anymore, is demographics and aging, which inserts many factors and many explanations into providing a rationale picture.
Economics has become mostly examining the entrails of a slaughtered animal for signs from the gods. A mind numbing amount of dodgy statistics, when all you really needs is several hundred trained observers around the nation, telling what they are seeing and experiencing. Trends can be correlated and events observed. We have cheap real time communication, something not available in the last great depression.
Can those jobs support a family or even pay your rent though