The Sahm Alarm

Claudia Sahm spent years at the Federal Reserve Board studying how to get money into people’s hands faster during recessions. Her frustration was that policymakers always acted too late. By the time the National Bureau of Economic Research officially declared a recession — sometimes a year after it had already started — millions of Americans had lost jobs, missed rent payments, and fallen behind on bills. The stimulus checks arrived after the worst of the pain. She wanted a real-time tripwire: an indicator that would flash red at the very start of a recession, not months after the fact.

The rule she published in 2019 is disarmingly simple. Take the national unemployment rate (UNRATE). Compute a three-month moving average. Compare that average to the lowest three-month average from the prior twelve months. If the difference is 0.50 percentage points or more, a recession has either just started or is about to. That’s it. No regressions, no leading indicators, no confidence intervals. One subtraction, one threshold.

The elegance isn’t in the math — it’s in the insight behind it. Unemployment doesn’t rise gradually during healthy expansions. It stays flat or drifts lower, sometimes for years. The three-month average of unemployment bottomed at 3.4% in early 2023 and barely budged for over a year. Then in mid-2024, it started rising — to 3.9%, then 4.1%, then 4.2%. A half-point rise from the recent low doesn’t sound like much, but Sahm demonstrated that such rises are never gradual. When unemployment starts climbing, it tends to keep climbing. The 0.50 threshold isn’t arbitrary — it’s the level that historically has separated transient noise from the beginning of a self-reinforcing deterioration.

The reason is feedback loops. When companies start laying off workers, those workers spend less. When spending falls, other companies lose revenue and lay off their own workers. When more people are unemployed, consumer confidence drops, businesses defer investment, and the credit cycle tightens. By the time unemployment has risen half a point, these feedback loops are typically already in motion. The rule doesn’t predict recessions in the way the yield curve does — it detects them. It’s not saying “a recession is coming.” It’s saying “a recession is here.”

Since December 1959 — the first month in the SAHMREALTIME dataset — the United States has experienced ten recessions as defined by the NBER. The Sahm Rule triggered in every one of them. In most cases, it triggered within a few months of the NBER-dated recession start. In some cases, it triggered before the official start date. And in one case — July 2024 — it triggered with no recession at all. The table below is the complete scorecard.

Several patterns jump out. First, the rule almost always triggers after the NBER recession start date, typically by 2–4 months. This might seem like a weakness — after all, it’s supposed to be an early warning. But consider that the NBER itself doesn’t announce recession start dates until 6–12 months after the fact. The December 2007 recession wasn’t officially declared until December 2008. The Sahm Rule triggered in April 2008 — eight months before the official announcement. It’s not early compared to reality, but it’s extremely early compared to when we learn about reality.

Second, the peak Sahm reading correlates loosely with recession severity. The three worst post-war recessions — 1973–75, 2007–09, and 2020 — produced peak readings of 3.83, 3.90, and a staggering 9.50 respectively. The milder recessions of 1990–91 and 2001 peaked at 1.53 and 1.63. The 2024 false alarm peaked at just 0.57 — barely above the threshold, barely sustained, and quickly reversed. In hindsight, the shallowness and brevity of the 2024 trigger was itself a signal that something was different.

Third, notice the “months above 0.50” column. In genuine recessions, the Sahm Rule stays elevated for 12 to 26 months. The feedback loops that Sahm identified take time to unwind: once unemployment rises, it takes a recovery, not just a stabilization, to bring the three-month average back below the 12-month low by enough margin. The 2024 episode lasted just 3 months above the threshold (July, August, September). By October it had dropped to 0.43, and by March 2026 it was at 0.20. No genuine recession in the dataset has ever resolved that quickly.

The July 2024 trigger was the most-discussed economic data point of the year. When the Bureau of Labor Statistics reported that the unemployment rate had risen to 4.3% in July (later revised to 4.2%), financial markets sold off sharply. The Sahm Rule value jumped to 0.53. Headlines declared that the recession indicator with a perfect track record had fired. Claudia Sahm herself went on television to explain that she believed her own rule was wrong this time — an extraordinary statement from the creator of the most famous recession indicator in modern economics.

She was right that the rule was wrong, and the reasons illuminate both the strengths and limitations of any indicator built on historical patterns. The unemployment rate had risen from its 2023 low of 3.4% to 4.2%, but the nature of the rise was different from every previous recession trigger. In past recessions, unemployment rose because companies were laying off workers in response to falling demand. In 2024, unemployment rose primarily because labor supply expanded — immigrants entering the workforce, people re-entering the labor force after pandemic-era absences — while hiring was merely slowing, not contracting. The unemployment rate went up not because more people were losing jobs, but because more people were looking for them.

The evidence for this interpretation was in the details. Initial jobless claims — the subject of the next episode — never spiked. They stayed in the low 200,000s, well within the range of a healthy labor market. The layoff rate, measured by the JOLTS survey, remained near record lows. Job openings fell from their 2022 peak but stayed well above pre-pandemic levels. The Sahm Rule captured a real rise in unemployment, but the rise came from the supply side of the labor market, not the demand side. The feedback loops that typically turn a half-point rise into a snowball — layoffs leading to less spending leading to more layoffs — never engaged.

Immigration played a particularly important role. The Congressional Budget Office estimated that net immigration to the United States surged in 2023 and 2024, adding roughly 3 million workers to the labor force that wouldn’t have been there under pre-pandemic trends. These new labor market entrants initially showed up as unemployed (looking for work but not yet employed), pushing the unemployment rate higher without any corresponding deterioration in the existing workforce. The Sahm Rule, which simply measures the level of unemployment relative to its recent low, couldn’t distinguish between “more people are losing jobs” and “more people are looking for jobs.”

The chart shows the 2024 episode in context. The Sahm Rule crossed the 0.50 threshold in July, peaked at 0.57 in August, and was back below the line by October. Meanwhile, the unemployment rate rose from 3.7% to 4.2% — a real increase, but one that stabilized rather than accelerating. By April 2026, unemployment sits at 4.3% and the Sahm Rule has fallen to 0.20, because the three-month average is no longer far from its 12-month low. The “low” has effectively reset as unemployment stabilized at the higher level.

This is worth understanding mechanically. The Sahm Rule measures the speed of deterioration, not the level of unemployment. An economy with 4.3% unemployment that has been at 4.3% for a year will have a Sahm reading near zero — because the current average isn’t rising above its recent low. An economy with 3.5% unemployment where the rate just jumped from 3.0% would have a Sahm reading of 0.50 — a trigger — even though 3.5% is a lower, better level. The rule captures momentum, not position. This is by design: Sahm showed that it’s the change in unemployment, not the level, that best identifies recession onset.

Of the 795 monthly observations in the SAHMREALTIME dataset, 177 (22%) have been at or above the 0.50 threshold. But that overstates the frequency of danger: those 177 months are concentrated in just 10 episodes (9 recessions plus the 2024 false alarm). The other 618 months — 78% of the dataset — have been below 0.50. In a normally functioning economy, the Sahm Rule spends most of its time near zero or slightly negative, registering nothing worth worrying about. It only becomes interesting when it starts moving toward 0.30, and only alarming when it crosses 0.50.

The speed of the rise matters too. During the GFC, the Sahm Rule went from 0.00 in April 2007 to 0.50 in April 2008 — a full year of slowly building pressure. During the 1973–75 recession, it took from November 1973 to March 1974 (four months) to cross the threshold. During COVID, it went from 0.00 in February 2020 to 4.00 in April — two months, the fastest trigger in the dataset. The 2024 episode was relatively gradual: the rule climbed from 0.20 in January to 0.53 in July (six months), suggesting a slow grind rather than a sudden shock.

The core strength of the Sahm Rule is that it uses the most fundamental economic data available. The unemployment rate is published monthly by the Bureau of Labor Statistics, based on a survey of 60,000 households. It isn’t revised much. It isn’t affected by seasonal adjustments in the way that GDP or retail sales data can be. And it measures something that directly matters to people: whether they have a job. Unlike the yield curve (which reflects bond market expectations), the VIX (which reflects options pricing), or credit spreads (which reflect corporate borrowing costs), the unemployment rate captures actual economic pain in real time.

The core weakness was exposed in 2024: the rule can’t distinguish between demand-driven and supply-driven changes in unemployment. A rise in the unemployment rate means the same thing to the formula whether it comes from mass layoffs at Ford and General Motors or from three million immigrants entering the workforce and spending a few months searching for their first job. The Sahm Rule was designed in an era when major shifts in labor supply between survey months were rare. The post-pandemic immigration surge was unusual enough to break the pattern.

Another limitation is the rule’s binary nature. It’s either triggered or it isn’t. There’s no distinction between a reading of 0.50 (just barely triggered, as in 2024) and a reading of 3.90 (the GFC peak). Real-world recession severity varies enormously, but the threshold gives you only one bit of information: yes or no. The severity scale in the table above is our attempt to add nuance, but the original rule was explicitly designed as a simple on/off switch for automatic fiscal stabilizers — specifically, to trigger direct payments to households the moment a recession was detected.

The rule also has a timing limitation for financial stress monitoring. Because it triggers during recessions rather than before them, it’s a lagging indicator compared to the yield curve or credit spreads. By the time the Sahm Rule fires, the stock market has typically already fallen, credit conditions have already tightened, and the initial shock is underway. For the purposes of our stress dashboard, the Sahm Rule serves as confirmation rather than warning. If the yield curve inverts, credit spreads widen, and then the Sahm Rule triggers, the recession is no longer a possibility — it’s a certainty. If the first two fire but the Sahm Rule stays below 0.30, there’s still a good chance the economy absorbs the shock without tipping into contraction.

The Sahm Rule reads 0.20 as of March 2026, placing it at the 67th percentile of its historical distribution. That percentile sounds high, but remember: the distribution is dominated by readings near zero during long expansions, plus a long tail of extreme readings during recessions. A reading of 0.20 is normal. It means the three-month average unemployment rate is two-tenths of a percentage point above its recent low — within the range of ordinary fluctuation.

The unemployment rate stands at 4.3%, up from the 3.4% low in early 2023 but essentially flat since mid-2024. This stabilization is key. The Sahm Rule’s danger comes from acceleration — unemployment rising faster and faster as feedback loops engage. What we’ve seen instead is a step-up and a plateau: unemployment rose from the post-pandemic lows to a new, slightly higher equilibrium, and has stayed there. This is consistent with a labor market that has normalized after the extraordinary tightness of 2021–2023, not one that is deteriorating.

For the stress dashboard, the Sahm Rule signal is green. The reading of 0.20 is well below the 0.50 trigger, unemployment is stable, and initial claims (as we’ll see in Episode 6) remain near historic lows. If the Sahm Rule were to start climbing again — back toward 0.30, then 0.40, then 0.50 — it would be cause for concern. But a gradual drift upward from 0.20 would give us months of warning. The rule moves slowly in non-crisis environments, which is both its limitation and its utility: when it does move, it means something real is happening.