Image Source: Pixabay

Adding the job vacancy rate to the McKelvey (Claudia Sahm) recession signal eliminates false negatives and false positives, and provides a much faster signal than Sahm.

This idea was one of the things Adam Taggart and I discussed last week on Thoughtful Money, posted today.

Is the US headed towards recession?

It's likely already in one, says macro analyst Mike "Mish" Shedlock @MishGEA

And "way too much is priced into this market"

WATCH: https://t.co/xQyRScXDAH pic.twitter.com/SMVjBIf10k

— Adam Taggart (@menlobear) August 20, 2024

Credit for the indicator goes to McKelvey for the original idea, and importantly to Pascal Michaillat and Emmanuel Saez for the second signal.

I will refer to the pair and the improvement as PMES, their initials.

Also, and I cannot emphasize this enough, credit goes to Regis Barnichon for the data and idea used by PMES and me in the charts below.

Has the Recession Started?

Please consider Has the Recession Started? by PMES, August 11, 2024.

To answer the question, this note develops a new Sahm (2019)-type recession indicator that combines vacancy and unemployment data. The indicator is the minimum of the Sahm indicator—the difference between the 3-month trailing average of the unemployment rate and its minimum over the past 12 months—and a similar indicator constructed with the vacancy rate—the difference between the 3-month trailing average of the vacancy rate and its maximum over the past 12 months.

Vacancy Rate

The vacancy rate is defined as the ratio of job openings to the labor force. The BLS Job Openings and Labor Turnover (JOLTS) report only dates to December of 2000.

However, Regis Barnichon, in 2010 described Building a composite Help-Wanted Index

This paper builds a measure of vacancy posting over 1951–2009 that captures the behavior of total—print and online— help-wanted advertising, and can be used for time series analysis of the US labor market.

Barnichon says HWI and JOLTS “closely track each other. In particular, the composite HWI does a good job of matching the level of JOLTS job openings over 2000–2009, indicating that the MISM can successfully model the share of online advertising.“

It is that overlap period that validates the second indicator.

Returning to PMES…

In the aftermath of the pandemic, the vacancy indicator started rising in 2022 and peaked in 2023, so it would have delivered a a prediction that was so early as to be misleading. This might be due to the extreme outward shift of the Beveridge curve during the pandemic (Michaillat and Saez 2024).

This shift lead to elevated values of the vacancy rate during that period, and therefore elevated values of the vacancy indicator. But the main advantage of the vacancy indicator is that it does not present the same uninformative blips as the unemployment indicator. For instance, there is no problematic blip in June 2003 (the vacancy indicator is not 0 but it is much lower than the unemployment indicator). Of course, it presents other uninformative blips. For instance it had a peak in July 1967 while there was no recessions then.

Minimum Indicator

To decide whether or not there is a recession, PMES takes the minimum of either McKelvey or Job Openings.

If either one is below 0.3 percent, there is no recession.

Sahm Discrepancy

The Sahm rule is that a recession has started when the unemployment indicator reaches 0.5pp. We saw that the rule works perfectly for 1960–2023. But the rule breaks down just before 1960 because in 1959 the unemployment indicator reached 0.6pp but there was no recession. This issue is easily fixed, however, by raising the threshold used in the rule to 0.6pp. This would make the rule a little slower at detecting recession starts, but it would allow the rule to continue working [from now] until World War 2.

For some reason, the Sahm indicator provided by the St. Louis Fed is sometimes negative. This is strange given that—by definition—a variable cannot be lower than its minimum over the past 12 months. Our indicators are never negative.

The reason for the Sahm negative discrepancy is Sahm does not include the current month in the 12-month lookback period.

A Sahm chart is rife with negative numbers. This never made any sense to me either, but that is why.

I doubt, but cannot prove, that McKelvey had a lookback period that did not include the current month, thus generating a mass of negative numbers.

I use the PMES methodology for the charts in this post and from now on.

Note that Sahm started her indicator in 1960 to avoid a false positive. Also, she takes a one-digit input and then creates a two digit output.

Finally, Sahm claims to have invented the rule. However, credit should go to Edward McKelvey, at Goldman Sachs.

PMES rounds to a single digit. I compute the unemployment rate directly, so I can compute to three digits and round to either one or two.

With that backdrop, here are the charts I created.

McKelvey Recession Indicator

Constructed accurately, there are no negative numbers. Sahm has at least one false positive and a second that would occur if rounded up.

McKelvey proposed using 0.3 as a trigger and that generates at least two more false positives.

I believe the October 2023 trigger is false. It barely skirted 0.3 then collapsed for a few months.

My preferred trigger is 0.4, halfway between McKelvey and Sahm. This trigger has two false positives, straight up.

My next two charts use data generously provided by Regis Barnichon.

Job Openings PMES Recession Indicator

The PMES indicator has two false positives but they do not overlap with the McKelvey false positives as the following chart shows.

McKelvey-PMES Combined Recession Indicator

The above chart plots the minimum of McKelvey or PMES from 1953 until now.

Over the last 11 recessions, there are no false positives or negatives on the McKelvey-PMES Combined Recession Indicator, except perhaps the current readings.

I prefer a trigger of 0.4 rounded to two decimal places or perhaps 0.5 rounded to two decimal places.

The 0.3 that triggered in October of 2023 did not stick and appears bogus no matter how much the NBER revises things.

Recession Lead Time In Months Combined McKelvey-PMES

These are very impressive numbers vs Sahm at 0.50 which has at least one false positive and lag times as great at 7 months.

Combining the triggers eliminates the false positives and negatives and allows the use of a lower trigger (0.30 or 0.40 instead of 0.50 or 0.60) straight up.

Michaillat and Saez note “The minimum [combined] indicator is always faster than the unemployment indicator, except in 2008 when it called the Great Recession 3 months later than the unemployment indicator. The slight delay is because job vacancies took some time to drop at the onset of the Great Recession.”

Depending on upper and lower bounds, Michaillat and Saez compute the current odds of recession at either 40 percent or 67 percent.

I note that since 1953, every time the economy was in the current state, the economy was in recession.

That does not make the odds 100 percent because everything is up to the NBER, the official arbiter of recessions.

A Dramatic Two-Day Change in GDPNow Forecast

On August 17, I noted A Dramatic Two-Day Change in GDPNow Forecast, Here Are the Details

A slew of economic reports on Thursday plus a residential construction disaster on Friday caused a huge decline in the GDPNow estimate.

Recession Underway

July 25, 2024: “All Hell Breaks Loose” In the Next Few Months as Recession Bites

August 2: Unemployment Rate Jumps, Jobs Rise Only 114,000 with More Negative Revisions

August 2: 2024: The McKelvey (Sahm) Unemployment Rate Recession Rule Just Triggered

August 15, 2024: Industrial Production Declines 0.6 Percent on Top of Big Negative Revisions

It seems to me that all hell breaking loose.

More By This Author: