Friday, February 10, 2023

Updating some important coincident indicators

 - by New Deal democrat

We returned to no more significant monthly data today. So here are some important coincident indicators I’ve been particularly following.

Redbook consumer purchases only increased 4.3% YoY last week, the lowest number in almost 2 years. The 4 week average also declined to 4.7%, also a 2 year low:

This strongly suggests that the January retail sales report, which will be published next week, will be negative again in real inflation-adjusted terms, implying a further weakening of YoY employment gains in the months ahead.

The American Staffing Index also had its second negative YoY reading in a row:

This tends to correlate with temporary jobs in the nonfarm payrolls report, and argues the YoY comparison there, which just turned negative there, will worsen in the next monthly jobs report.

On the other hand, withholding tax payments, which were up 5.4% YoY in January, are also higher 5.4% YoY for the first week of February. Here’s an up to date chart from the California Treasurer’s Office (CA being 12% of the US population), showing the stabilization in payments in January:

Finally, here’s the Lewis-Martens-Stock Weekly Economic Index:

This index so far has stubbornly refused to turn negative (which is after all a good thing).

Thursday, February 9, 2023

As Holiday seasonality disappears, initial jobless claims turn higher YoY


 - by New Deal democrat

Initial jobless claims rose 13,000 last week to 196,000. The four week moving average declined -7,500 to 189,250. Continuing claims increased 38,000 to 1,688,000. Below I show all three since initial claims first fell below 300,000 in late October 2021:

Because seasonality played such a role in January’s jobs report, here are initial claims non-seasonally adjusted (red) as well as the seasonally adjusted number:

This week was the first time since Christmas week that the number of layoffs was not heavily affected by the fact that workers who weren’t hired for the Holiday season couldn’t be laid off once it was over. Note that the spike for the five weeks of January for 2022 only partially materialized this year. 

And that, as it turns out, is significant, as shown in the below graph of the YoY% changes in initial and continuing claims:

For the week, initial claims are now higher YoY, for the first time since December. It will take a couple more weeks before we know if the underlying trend truly is higher. Remember that there is no recession signal unless and until claims, particularly on a four week average basis, are higher at least by 10% YoY. Hence the starting point when claims first fell below 300,000 in the graphs.

Wednesday, February 8, 2023

Credit conditions in Q4 were recessionary


 - by New Deal democrat

While we are still in our lull concerning monthly data, on Monday there was a significant update of one long leading indicator that is only reported Quarterly: the Senior Loan Officer Survey.

This survey has an excellent history of over 30 years telling us about credit conditions. Loosening of credit, and an increase in demand for credit means expansion ahead. Credit tightening and a decrease in demand for credit have only occurred shortly before, during, and shortly after recessions.

In the case of Q4 2022, the data speaks for itself.

Here is demand for commercial loans by large, medium, and small sized firms:

With the exception of one quarter in 1994, a falloff in demand of this magnitude has only happened at the time of the 3 recessions since then.

Here is the net percentage of banks tightening credit for loans to large, medium, and small sized firms. Note in this case a higher number means more tightening, so is bad:

The only time conditions have tightened this much has been in advance of or during the last 4 recessions since 1990.

Through Q4, banks were tightening credit, and firms were pulling back in asking for loans. This is recessionary, period.

Which has reminded me that I need to update my comprehensive examination of long leading indicators over at Seeking Alpha. I’ll post a link once that is done (maybe today, maybe not).

Tuesday, February 7, 2023

Scenes from the blockbuster January jobs report 2: revisions do not resolve discrepancies in the reports


 - by New Deal democrat

Yesterday I wrote that the blockbuster January jobs report was essentially the result of two factors: (1) a very low number of potential applicants in the jobs pool with an unemployment rate well under 4% meant that employers were reluctant to let go of workers, which especially impacted the numbers, which particular showed up as (2) the seasonal adjustments expected about 2,000,000 people to be laid off in January, but only 1,600,000 were, as some employers elected to keep Holiday hires on the payroll; also, you can’t fire the people in January that you didn’t hire for seasonal help in October through December.

Today let me deal with the second big reveal in the January report: the 2021 and 2022 revisions, and their effect on the big discordance between the Household and Establishment survey number that began last April.

There were revisions to both the Household and the Employment surveys. 

The Household revisions are easy to understand. Every year there is an adjustment for population growth. But instead of spreading that out over the 12 previous months, the Household survey adds it all at once to the January number. 

As a result, officially in January per the Household survey, 894,000 more people were employed. But 810,000 of that was caused by the addition of 954,000 to the 2022 population estimate. Only the remaining 84,000 of that was the change from December to January.

One way to reasonably estimate what the Household survey would have looked like in 2022 is to spread out those 810,000 job gains over the past 12 months, which amount to 67,500/month. The below graph does exactly that, adding 67,500 to each Household survey number in 2022, and compares it with the revised Establishment numbers:

Note that adding 67,500 to the January Household number would give us 151,500, about 400,000 below the Establishment number. But more importantly, note that there is still a very large disconnect between the Household numbers, which added 1,810,000 jobs since last March, for an average of 181,000 per month, vs. the Establishment survey’s reported gains of 3,649,000, or 365,000 per month.

Secondly, in the last several months there has been debate about whether the Household survey has been picking up weakness that was entirely missing in the Establishment survey. Evidence for this was indicated by an estimate from the Philadelphia Fed, based on the QCEW report for Q2 of last year that only 11,000 private sector jobs had been added during that entire 3 month period. This is important because the QCEW is the gold standard for employment reports, consisting of the actual total from 95% of all businesses via unemployment insurance payments. Subsequently the Census Bureau itself updated its “Business Dynamics Survey” through Q2 of last year, which seasonally adjusts about 70% of the QCEW numbers, and reported an outright *loss* of -287,000 jobs in the private sector during that period. 

To best show this, here is a graph from Prof. Menzie Shinn of Econbrowser, comparing the unrevised Establishment survey reports, with the Household reports, the QCEW gains normalized to the same number in 2021 and applying his estimate of seasonal adjustments, the Philadelphia Fed estimates, and the Business Dynamics Survey result:

Note that Prof. Shinn’s method of seasonally adjusting renewers a very different result from both the Philadelphia Fed and the Census Bureau itself in the BDS.

So at first blush it is very surprising that the 2022 benchmark revisions to the Establishment survey *added* 457,000 jobs to the previous results, and while Q2 2022 numbers were revised downward by -59,000, still showed 1,047,000 jobs gained during that Quarter, as shown in the graph below:

So, did people mistake the message of the Q2 QCEW? As Dean Baker wrote, apparently not:

The QCEW relies on unemployment insurance filings, which give a virtual census of payroll employment. The establishment survey is benchmarked to QCEW annually, but the benchmark takes place with the January data, using the QCEW data from the prior year’s first quarter. The QCEW data from March 2022 were just included in the establishment survey, increasing employment growth in the year from March 2021 to March 2022 by 568,000.”

In other words, while the strong QCEW data from Q1 2022 was included in the benchmark revisions, the weak data from Q2 was not. 

This shows up dramatically if we compare the YoY% changes in payrolls growth as measured by the QCEW census, vs. the same metric from the Establishment survey. This is the best way to measure because the one drawback of the QCEW is that, even though it is comprehensive, it is not seasonally adjusted. But that doesn’t affect YoY comparisons.

Here is the YoY% change monthly in the QCEW survey through its most recent report for Q2 2022:

And here is the same thing for private nonfarm payrolls:

Comparing the YoY QCEW report through Q1 2022 with private payrolls, the former was up 7.2m jobs and 5.1%, and the latter up 6.9m and 5.6%. While the % change is still off, the total number is close, and on par with variances in the past. The median monthly difference between the two measures is 0.3%, and the mean is 0.5%.

For Q2, YoY the QCEW was up 5.7m and 4.0%, while private payrolls after revisions were up 6.3m and 5.3%. This is a very large difference. The median difference is 0.8%, and the mean is 0.9%.

My expectation is that the difference will be resolved in favor of the much more comprehensive final QCEW figures once they are in. Additionally, we’ll get preliminary figures for the Q3 2022 QCEW numbers later this month, and this will tell us if the divergence continued.

In short, even with the revisions there is a large difference between big Establishment gains in the last 9 months of 2022 vs. tepid ones in the Household report. And the divergence between the Establishment sample and the comprehensive QCEW and BDS jobs census figures for Q2 was not taken into account in the January revisions, and remains.

Monday, February 6, 2023

Scenes from the blockbuster jobs report 1: in January, nobody* got laid off!



 - by New Deal democrat

There’s no important new economic data until Thursday this week. Meanwhile, there was lots to digest about Friday’s blockbuster jobs report, which I have now done, so I’m going to spend a couple (maybe 3!) days diving in to the details. Today I’ll deal with how seasonality and a very tight labor market were decisively important in Friday’s report. Tomorrow (and maybe Wednesday as well, depending on the length of what I have to say) will deal with revisions to both the Establishment and Household data for 2022.

Let me start off by running the following thought by you: do you really think that bars and restaurants hired almost 100,000 more people in January, as shown in the jobs report?

Yeah, me neither. And the truth is, they didn’t! Actually, in the aggregate they let go of 183,000 employees. Here’s the seasonally adjusted (blue) vs. non-seasonally adjusted (red) numbers beginning in July 2021:

Here’s the longer term view since the start of the series in 1991:

In a normal year, in January bars and restaurants typically let go of 200,000 or more employees. Only once (in 2006) did they lay off fewer than they did last month. 

The seasonal adjustment (and it’s perfectly valid, I am in no way saying it is inappropriate) translates that relative lack of layoffs in January to stellar job gains.

Similar patterns, while less drastic, appear across the jobs spectrum. Let me give one more example.

Virtually every single metric in the past several months has suggested that the manufacturing sector has entered a downturn. The ISM manufacturing new orders index is at a level that in the past has almost always indicated a recession is occurring. Manufacturing production has turned down slightly in the past several months. The average manufacturing work week has also declined sharply enough from its peak to be consistent with a recession. And in November and December, even after revisions, manufacturing added only 13,000 employees per month. So it was reasonable to expect employment in manufacturing to turn down in January. Instead, on a seasonally adjusted basis 19,000 jobs were added.

So, here are manufacturing jobs seasonally adjusted (blue) and not (red) since July 2021, showing that actually there were 91,000 layoffs, or 0.7% of the December workforce:

Here is the longer term view of the monthly % change in non-seasonally adjusted jobs in manufacturing since the modern era began in 1983, up until the pandemic, adding 0.7% so that the 2023 % would show as 0:

There were fewer layoffs this January than there were most years during that entire 35+ year period. Seasonally adjusted, that means a strong gain (again, a perfectly valid adjustment).

In fact, for private payrolls January’s monthly layoffs of 1.632% of the December workforce was the smallest in the entire 80+ year history of the data (shown as 0 in the below graph):

Government employment also jumped 74,000 in January. This was also heavily influenced by seasonal factors, but also by the return to work of 48,000 employees at the University of California who had gone on strike in November. The resulting non-seasonally adjusted decline for January of -1.5% was the least since the end of the Great Recession (note July, not January, is the month for the biggest NSA declines in government jobs):

Again, let me emphasize that there’s nothing wrong with performing this seasonal adjustment. 

But let’s take a slight detour and look at the entire historical record for the 4 week average of initial claims up until the pandemic:

In January, an average of only 191.750 people filed for initial claims. Except for the 1960s (when the US population was only half what it is now), and March and April of 2019 and last year, this is the lowest monthly average in history. Here’s a close-up view of seasonally adjusted and non-seasonally adjusted initial claims in the last 18 months, showing that the typical early January sharp jump in claims simply did not materialize this year:

In essence, nobody* is getting laid off. There was a seasonal increase in hiring for the Holiday season, but in the aggregate businesses decided not to lay some of those people off, but keep them on the payrolls. That, plus the resolution of the California strike, is why payrolls jumped so much in January. (*hyperbole)

Which raises a question: what will happen in February? As shown on the graph below of non-seasonally adjusted employment during the last expansion, February is typically positive, leading to the peak months of March through June:

Now let’s look at the revisions for 2022 which just took place:

January 2022 was revised down from 504,000 to 364,000, while February was revised up from 704,000 to 904,000. These were the two biggest revisions of the entire year, suggesting that seasonality led those two months astray.

In conclusion, here’s what we have: in a very tight labor market, in the aggregate employers were reluctant to lay off seasonal hires in January, electing to keep them on payroll. This translated into blockbuster job gains on a seasonal basis. But we have to wait for February’s report to see whether this is a sign of renewed strength in the jobs market, or whether employers have less need to hire new workers as a result. In other words, will January’s strength continue in a month where actual hiring, not layoffs, are expected.

Addendum: It’s also been pointed out that you can’t fire in January the people you didn’t hire in October through December. In 2022, there was considerably less Holiday season hiring than in recent years, as shown in this Challenger Gray graphic from early December:

Tomorrow: the story behind the revisions