The Pink Slip Signal

Most economic data arrives with a delay. GDP is published quarterly, roughly a month after the quarter ends, and is revised twice more after that. The jobs report comes monthly, with a 5-week lag, and is based on employer surveys that take time to collect and process. Even the unemployment rate — the basis for the Sahm Rule we covered in Episode 5 — is monthly. But initial claims for unemployment insurance are published every Thursday, covering the week that ended the previous Saturday. Five-day lag, weekly frequency, minimal revision. There is no faster way to take the pulse of the American labor market.

The number itself is simple: it counts people who filed a new claim for unemployment benefits during the reporting week. Not people who are still receiving benefits (that’s continuing claims, or CCSA), and not people who exhausted their benefits or never filed at all. Initial claims capture the flow into unemployment — the number of new layoffs, firings, and job eliminations happening right now. Think of it as a speedometer for the labor market: the unemployment rate tells you how many people are out of work, but initial claims tell you how fast that number is changing.

The ICSA series begins in January 1967. In those 59 years, claims have averaged about 360,000 per week. But that average is distorted by the much larger labor force of the 1970s and 1980s relative to the economy. More informative is the range: claims have been as low as 162,000 (reached in the late 1960s when the labor force was much smaller) and as high as 6.14 million (the week of April 4, 2020, when pandemic lockdowns triggered the largest wave of layoffs in recorded history). The current reading of 219,000 is close to the post-pandemic all-time low of 190,000, reached in September 2022.

What makes claims so powerful as a stress indicator is their behavior during economic turning points. In healthy economies, claims are stable and low — people get laid off, sure, but the number is roughly constant from week to week. When a recession begins, claims don’t just rise — they surge. The transition from healthy to stressed isn’t gradual; it’s abrupt. During the Great Financial Crisis, claims went from 321,000 in January 2008 to 587,000 by December — nearly doubling in a year. During COVID, they went from 202,000 to 6.14 million in three weeks. The speed of the signal is the signal.

The pattern is striking. During healthy expansions, claims stay in a narrow band — roughly 200K–320K in the modern era. The number barely moves from week to week, which is why the financial press usually ignores the Thursday release during good times. Then recession hits, and claims explode. The jump from 2007 to 2009 is representative: average claims went from 321K to 574K, an increase of 79%. During the dot-com recession, they went from 299K to 406K. The 2020 pandemic is off the chart entirely — average claims of 1.36 million, with a single-week peak of 6.14 million that shattered all previous records by a factor of ten.

What’s notable about the current era is how long claims have stayed extraordinarily low. The average for 2022 was 214K, for 2023 it was 223K, for 2024 it was 223K, for 2025 it was 226K, and 2026 so far is 212K. Five consecutive years at or below 226K. To find a comparable stretch of sustained low claims, you have to go back to 2018–2019 (219K and 218K), and before that, to the late 1960s when the labor force was half its current size. Adjusting for the growth in the employed population, today’s claims levels are arguably the lowest in American history.

Raw claim numbers need context. A reading of 300,000 meant something different in 1975, when the labor force was 94 million, than in 2025, when it’s 168 million. Ideally, you’d normalize claims by the size of the employed population. But in practice, the absolute thresholds have been remarkably stable over the past 30 years, because the level that signals trouble has evolved roughly in line with workforce growth. Here are the practical thresholds that Wall Street economists and Fed officials have used since the mid-1990s:

The key threshold that practitioners watch is the 4-week moving average crossing 300,000. This has been a reliable recession warning since the 1990s. A single weekly print above 300K can be noise — a hurricane, a seasonal adjustment glitch, an auto plant retooling. But when the 4-week average crosses 300K and stays there, it has historically meant that layoffs are broadening beyond any single sector or region. The 4-week average today is about 215K — nowhere near the danger zone.

Another useful metric is the rate of change. Claims tend to rise gradually at first, then accelerate. During the GFC, the 4-week average crossed 300K in late 2007, hit 400K by mid-2008, and peaked above 650K in early 2009. The entire escalation from “healthy” to “crisis” took about 18 months. During the 2001 recession, the escalation from 300K to the 517K peak took about 9 months. If claims were to start rising from today’s 219K, the historical pattern suggests you’d have months of warning before reaching crisis levels — assuming the shock isn’t a sudden one like COVID.

Initial claims and the Sahm Rule measure related but different things. Claims count the flow of people being laid off. The Sahm Rule measures the accumulation of those layoffs into higher unemployment. In theory, a surge in claims should precede a Sahm Rule trigger, because people have to file claims before they show up in the unemployment rate. In practice, the relationship is close but not mechanical, because not everyone who loses a job files for unemployment, and the unemployment rate is influenced by labor force participation as well.

This distinction explains the 2024 puzzle. The Sahm Rule triggered in July 2024, but initial claims never spiked. Claims averaged 223K for the year — well within the “very healthy” range. The Sahm Rule triggered because the unemployment rate rose, but claims stayed low because the rise was driven by labor supply expansion (more people looking for work), not by layoffs. If you had been watching both indicators, claims would have told you the Sahm trigger was a false alarm. The layoff data was saying: “Companies aren’t firing people.” The unemployment data was saying: “More people are unemployed.” Both were true, but only one was cause for alarm.

This is why the dashboard approach matters. No single indicator tells the whole story. Claims and the Sahm Rule are complementary: claims are faster (weekly vs. monthly), more granular, and directly measure layoffs. The Sahm Rule captures the broader labor market picture, including supply-side effects. When both are green — as they are now, with claims at 219K and the Sahm reading at 0.20 — the signal is unambiguous. When they diverge, as they did briefly in 2024, the divergence itself is informative.

Initial claims at 219,000 place the labor market in the 10th percentile of its 59-year history — meaning 90% of all weekly readings since 1967 have been higher than this. Continuing claims at 1.79 million are at the 14th percentile — also exceptionally low. Together, they paint a picture of a labor market where few people are being laid off and those who are find new work quickly.

The recent trend is flat to slightly declining. Monthly average claims have been in the 208K–230K range for the past 18 months, with no sustained upward drift. This stands in stark contrast to what you’d see if a recession were developing: in every downturn since 1967, claims began rising months before the NBER-dated recession start, typically crossing the 300K level 2–6 months before the peak of the expansion. We are nowhere near that trajectory.

For the stress dashboard, claims are deep green — the strongest labor market signal in the entire dashboard. The danger signs to watch for: a sustained move of the 4-week average above 250K (elevated but not alarming), then above 300K (genuine concern), then above 350K (likely recession). At the current pace, even a doubling would only reach about 440K — a recession-level reading, but one that would take months to develop absent a sudden shock.