Guide

Sahm rule explained

Harbor Credit Union's asset-liability committee ran a recession playbook keyed almost entirely to the 10-year minus 3-month Treasury spread. When the curve uninverted in late 2025, the desk added duration — only to watch credit spreads widen and loan delinquencies climb while headline GDP still printed positive. Two months later, the three-month moving average of the unemployment rate had risen 0.52 percentage points above its trailing twelve-month low. That single labor-market momentum reading would have flagged the downturn weeks before NBER's business-cycle dating committee announced a peak. The signal has a name: the Sahm rule.

Economist Claudia Sahm proposed a simple real-time recession indicator: when the three-month average unemployment rate rises by at least half a percentage point relative to its low over the prior year, the economy is already in or entering recession. Unlike NBER's retrospective dating, the rule updates monthly with official labor data and has triggered near every U.S. recession since 1970 with few false alarms. This guide covers the mechanics, how it differs from yield-curve and PMI signals, policy uses in automatic stabilizers, the Harbor Credit Union refactor, a technique decision table versus single-indicator models, pitfalls, and a production checklist — alongside our guides on recessions, unemployment, and business cycles.

What the Sahm rule is

The Sahm rule is a recession indicator derived from unemployment momentum, not the level of joblessness. Formally:

Compute U_3m = three-month moving average of the seasonally adjusted unemployment rate (U-3). Compute U_min = the lowest value of U_3m over the trailing twelve months. The Sahm recession indicator fires when:

U_3m − U_min ≥ 0.50 percentage points.

Sahm introduced the rule in 2019 as both a forecasting tool and a trigger for countercyclical fiscal policy — automatic stimulus payments when the indicator crosses the threshold, insulating households before layoffs cascade. For investors and risk managers, the same arithmetic offers a timely, rules-based recession flag that does not wait for two quarters of negative GDP or an NBER announcement months after the fact.

The rule works because unemployment is a lagging level variable but a coincident-to-leading change variable: the rate is low at cycle peaks, but the rise from that low begins early in contraction, often while equity markets are still debating “soft landing” narratives.

Signal taxonomy

Threshold crossing (recession on)

The canonical 0.50pp trigger is a binary recession signal. Historically it has aligned with NBER recession months in the United States with high fidelity since 1970. Treat a confirmed cross as “recession regime” for risk budgeting, not as a precise month-zero timestamp.

Approach zone (0.30–0.49pp)

Readings below the threshold but rising signal deteriorating labor conditions. Harbor now treats 0.35pp as a “yellow band” that halves incremental duration adds and tightens consumer-credit underwriting overlays.

Reset and false-clear risk

The indicator resets only when a new twelve-month low in U_3m is established — meaning unemployment must fall enough to set a fresh trough. A brief pause in rising unemployment does not clear the signal; only a genuine labor-market improvement does.

Policy trigger variant

Sahm's fiscal proposal ties direct household payments to the same arithmetic, sized as a function of the unemployment gap. Portfolio teams can separate the recession detection use from the stimulus sizing use; detection is what most ALM and macro sleeves need first.

International adaptations

The 0.50pp threshold was calibrated on U.S. U-3 data. Other economies may need different buffers because survey definitions, participation dynamics, and furlough schemes distort monthly prints. Always re-backtest before importing the literal rule abroad.

How to compute and monitor it

Use official Bureau of Labor Statistics (BLS) seasonally adjusted U-3 from the monthly Employment Situation report (typically released the first Friday of each month). Steps:

Download monthly U-3 and compute a three-month centered or trailing moving average (BLS and FRED publish UNRATE; apply a three-month MA in your sheet or code).
Over a rolling 12-month window, record the minimum U_3m.
Subtract that minimum from the current U_3m; compare to 0.50pp.
Archive each print with release timestamp — do not revise history silently when BLS issues annual benchmark updates.

Pair the Sahm indicator with high-frequency labor inputs that move earlier: initial jobless claims (weekly), payroll growth revisions, and Okun's-law implied output gaps. The Sahm rule synthesizes the monthly unemployment path; claims can warn you one to four weeks sooner that a Sahm cross is approaching.

For dashboard design, plot three series together: raw U-3, U_3m, and the Sahm spread (U_3m − U_min) with a horizontal line at 0.50. Annotate NBER recession bars in backtests so stakeholders see lead/lag visually.

How it compares to other recession signals

No single indicator catches every downturn. The Sahm rule excels when labor deterioration is the transmission channel; it can lag or miss supply-shock recessions where unemployment rises late.

Yield curve inversion — Often leads by 12–18 months but produces false positives and, as Harbor learned, can uninvert before recession begins. See our yield curve guide.
Two negative GDP quarters — Popular media rule; not NBER's definition and frequently revised away.
LEI and PMI composites — Broader activity baskets; useful early warnings but noisier month to month. See leading indicators.
Credit spreads — Market-priced stress; can spike in financial crises before unemployment moves.
Sahm rule — Late relative to markets, early relative to NBER; low false-positive history in the U.S.; transparent and replicable.

Production macro sleeves should treat these signals as orthogonal votes in a regime classifier, not competing religion.

Harbor Credit Union refactor

After the 2025 drawdown, Harbor's ALM committee rebuilt its recession playbook around a weighted scorecard:

Sahm spread (40% weight) — primary binary recession flag; yellow band at 0.35pp.
Initial claims 4-week MA (25%) — early warning when Sahm is still below threshold.
3m10y Treasury spread (20%) — retained but demoted; uninversion no longer auto-triggers duration adds.
High-yield OAS change (15%) — financial-stress channel for credit-card loss reserves.

When the Sahm spread crossed 0.50 in March 2026 (hypothetical continuation of the case study), Harbor shifted its investment portfolio to a pre-defined recession sleeve: shorten duration by 1.2 years, add Treasury TIPS, and widen consumer unsecured loss assumptions. The move occurred six weeks before consensus GDP forecasts turned negative — and aligned with NBER's eventual peak dating.

On the liability side, deposit beta models assumed slower outflows because automatic stabilizers and Fed cuts typically follow Sahm triggers, cushioning household balances. Integrating the rule into both sides of the balance sheet prevented the one-sided “curve uninverted, we're safe” mistake.

Technique decision table

Approach	Best when	False signal risk	Timeliness
Sahm rule alone	Simple U.S. recession regime flag for ALM	Low historically; pandemic furloughs were a stress case	Coincident; beats NBER by months
Yield curve inversion alone	Long-horizon cycle positioning	Moderate; long false alarms possible	Early; can reverse before recession
GDP two-quarter rule	Media communication only	High; revisions erase signals	Late and noisy
PMI / LEI composite	Global multi-sector monitoring	Moderate month-to-month	Leading by 3–9 months
Sahm + claims + spreads blend	Institutional risk budgets (Harbor model)	Lower than any single input	Balanced early + confirm
ML black-box recession model	Research prototypes	Opaque; overfits past cycles	Varies; hard to audit

Prefer interpretable rules for governance: boards and regulators can understand a 0.50pp unemployment spread; they struggle to defend a neural net that flipped because feature 47 moved.

Common pitfalls

Using raw U-3 instead of the three-month average — monthly prints are volatile; Sahm explicitly smooths them.
Wrong rolling window for the minimum — must be twelve months of U_3m lows, not twelve months of raw unemployment.
Ignoring benchmark revisions — BLS annual revisions can shift historical triggers; stress-test sensitivity.
Expecting sector timing precision — the rule dates economy-wide recession, not peak earnings per sector.
Pandemic-style distortions — furloughs and rapid reopenings broke normal dynamics in 2020; maintain override protocols for structural labor shocks.
Exporting 0.50pp abroad blindly — euro-area and EM unemployment series need local calibration.
Treating trigger as sell-everything — recessions vary in depth; combine with credit, earnings, and policy response scenarios.
Clearing on one soft print — a single flat unemployment month does not reset the Sahm spread if U_min is unchanged.

Production checklist

Compute U_3m from seasonally adjusted U-3 each month on BLS release day.
Track twelve-month minimum of U_3m and current spread to 0.50pp.
Define yellow-band actions (e.g., 0.35pp) before binary trigger fires.
Pair Sahm with weekly initial claims and payroll revisions for early warning.
Backtest against NBER recession dates; document lead/lag in months.
Stress-test benchmark revision scenarios on historical spreads.
Integrate Sahm vote into multi-signal recession scorecard with documented weights.
Pre-author recession portfolio sleeves (duration, credit, liquidity) tied to trigger.
Separate detection logic from Sahm fiscal-stimulus sizing if not a policymaker.
Log each monthly print with timestamp; avoid lookahead in simulations.
Review pandemic-style override rules annually for labor-definition shocks.
Board-report Sahm spread alongside yield curve and credit spreads every cycle.

Key takeaways

The Sahm rule fires when the three-month average unemployment rate rises 0.50pp above its trailing twelve-month low — a simple, historically reliable U.S. recession flag.
It updates monthly with official data and typically precedes NBER dating by months, though it lags market-based stress indicators.
Harbor Credit Union cut recession playbook errors by weighting Sahm momentum above yield-curve uninversion alone.
Blend Sahm with claims, spreads, and LEI inputs — no single indicator covers supply shocks, financial crises, and labor-led downturns equally.
Use the three-month average and twelve-month minimum exactly; raw monthly unemployment and wrong windows break the rule.