Five Recession Detection Rules Tested Against 75 Years of US Data — One Composite Has Never Been Wrong

Chart: Sahm Rule Trigger Episodes Against NBER Recessions (1948–2026)

Sahm Rule trigger episodes against US recession history

Of 15 Sahm Rule trigger episodes since 1948, 13 were followed by a recession within twelve months. The exceptions are July 2003 and August 2024.

Key Takeaway

Two of the Sahm Rule’s 15 trigger episodes (13%) are not associated with an NBER recession. As reported in the NBER business cycle reference chronology. Both occurred during periods when the unemployment rate rose modestly without a broader labor-market deterioration: the 2003 jobless recovery and the post-pandemic labor-supply normalization of 2024.

Sources: BLS Unemployment Rate (FRED series UNRATE); NBER Business Cycle Dating Committee. Chart: Eco3min Research.

PNG
SVG
PDF
Free to use with attribution · Cite

How to Read This Chart

The black line is the US unemployment rate, plotted monthly from January 1948 to April 2026. The vertical gray bands mark the twelve recessions dated by the National Bureau of Economic Research, the body that determines US business cycle peaks and troughs ex-post. Each green dot marks a month in which the Sahm Rule first crossed its 0.50 percentage point threshold — meaning the 3-month moving average of the unemployment rate had risen at least 0.50pp above its trailing 12-month minimum — and was subsequently followed by an NBER-dated recession within twelve months. The two red X markers, in July 2003 and August 2024, indicate trigger episodes that were not followed by an NBER recession. Related discussion: the assumptions that mislead investors on recessions and the cycle.

The Sahm Rule was published by Claudia Sahm in 2019 in a paper for The Hamilton Project at the Brookings Institution. The 0.50pp threshold and the 3-month moving average specification were chosen by Sahm to balance sensitivity (catching real downturns early) against false-positive rates. Our test extends Sahm’s original analysis through April 2026 and pairs it with four alternative rules drawn from the empirical recession-dating literature. The complete dataset, including all monthly indicator values for each of the five rules and the full record of S&P 500 forward returns from each trigger date, is available below as CSV and XLSX.

Why It Matters That a Single Rule Has Now Been Wrong Once

The Sahm Rule occupies a particular position in macroeconomic forecasting. Unlike the yield curve, which has been studied since the 1980s and which generates roughly one signal per decade, the Sahm Rule was published in 2019 by a former Federal Reserve economist and rapidly adopted by financial media as a near-infallible recession indicator. By the rule’s own historical record at the time of publication, it had correctly identified every US recession from 1948 to 2008 with zero false positives. The May 2024 Sahm trigger therefore received considerable media attention, and the August 2024 confirmation — when the rule officially crossed its 0.50pp threshold at 0.53pp — prompted widespread discussion of an imminent recession.

The empirical fact is that the recession did not arrive. The unemployment rate reached 4.2% in mid-2024 (the post-trigger reading), then drifted between 4.0% and 4.5% through 2025, and stood at 4.3% in April 2026 — roughly twenty months after the trigger. The NBER, which dates recessions ex-post and typically with a 6–18 month delay, has not designated any recession start date in this period. The Sahm Rule, by its own quantitative criterion, has now produced its first false positive in 75 years of testable history.

This page sets out to do something that has not, to our knowledge, been done in a single open-access reference: test the Sahm Rule alongside four alternative published rules (the Triggers Rule, the Two-Month Rule, the Claims Momentum Rule, and a Composite Rule combining Sahm and Claims) against the full NBER record from 1948 to 2026, with the August 2024 anomaly treated as the central empirical object rather than as an inconvenience to be glossed over. The question is not whether the Sahm Rule is “still” valid — it is what an honest scorecard looks like when one accepts that any rule with a 0.50 threshold and 75 years of data is likely to eventually produce an outlier.

Five Rules, One Scorecard

The five rules tested are summarized in the table below. Each is a published, ex-ante rule with documented parameters — none has been retrofitted to the data in this study. The Composite Rule is the only construction specific to this analysis, and it is defined as the simple logical AND of the Sahm Rule and the Claims Momentum Rule, both evaluated in the same calendar month.

The Five Rules — Definitions

The Composite Rule is operationally testable only from January 1969 onward, because R4 requires at least two years of weekly initial jobless claims data to compute the 4-week moving average and the trailing 12-month minimum, and the ICSA series begins in January 1967. For the four NBER recessions before 1969 (1948, 1953, 1957, 1960), only the three unemployment-rate-based rules can be evaluated.

Scorecard — Episodes, True Positives, False Positives

An “episode” is a contiguous block of months during which the rule remained triggered. Consecutive months are counted as a single episode. A trigger episode is classified as a true positive (TP) if it began within twelve months of the start of an NBER-dated recession; otherwise it is a false positive (FP).

Visual scorecard — five rules vs NBER recessions

Each row plots one rule’s trigger episodes (green = followed by recession; red = not followed) against the NBER recession history.

Why the Composite Wins

The Composite Rule’s perfect post-1969 record is not coincidence. It is the mechanical consequence of demanding that two independent labor-market indicators — a stock indicator (the unemployment rate level) and a flow indicator (initial jobless claims) — confirm each other. Episodes where only one of the two fires are exactly the historical false-positive episodes of the standalone Sahm Rule.

Sources: BLS (UNRATE, ICSA via FRED); NBER Business Cycle Dating Committee. Chart: Eco3min Research.

PNG
SVG
PDF
Free to use with attribution · Cite

The Triggers Rule and the Two-Month Rule both detect every recession, but at the cost of dramatically higher noise — the Two-Month Rule fires 56 times with 21 false positives, an extreme signal-to-noise degradation. The Claims Momentum Rule alone is comparable in noise to the Sahm Rule, with 14 trigger episodes and 2 false positives (Feb 1977 and Jun 2023), but it does not improve on Sahm because its false positives occur in different periods. It is the intersection of Sahm and Claims that drops the false-positive count to zero. We turn to that intersection in detail below, after examining the 2024 anomaly that motivated this study.

The August 2024 Trigger: What the Data Actually Showed

The Sahm Rule first crossed its 0.50 percentage point threshold in mid-2024. Real-time observers, examining the BLS’s August 2, 2024 employment situation release (covering July 2024 data initially printed at 4.3%), saw an indicator value at or above 0.50pp — Claudia Sahm herself confirmed the trigger in published commentary that same week. In our revised-vintage dataset (which incorporates subsequent BLS benchmark revisions), the official threshold cross occurs in the August 2024 monthly observation, with a 3-month moving average of 4.17%, a trailing 12-month minimum of 3.63%, and an indicator value of 0.533pp — three basis points above the threshold.

2024 Monthly Indicator Readings

The Sahm trigger lasted exactly one month. By September 2024, the indicator had retreated to 0.43pp and never returned to threshold. The unemployment rate has subsequently oscillated between 4.1% and 4.4%, standing at 4.3% in April 2026. Claims momentum during the entire August 2024 period — when the Sahm Rule fired — was 11.5%, less than half the 25% threshold that defines the Claims Momentum Rule. The Composite Rule never triggered in 2024.

Why the Composite Rule suppressed both Sahm false positives

Initial jobless claims momentum during the 2003 and 2024 Sahm Rule false-positive episodes stayed well below the 25% confirmation threshold.

Mechanism

In both 2003 and 2024, the unemployment rate rose because of structural shifts (jobless recovery; labor supply normalization) that did not produce the wave of layoffs that initial jobless claims would have registered. Claims stayed near pre-trigger levels (5.8% above trailing minimum in 2003; 11.5% in 2024). The composite rule’s second gate held.

Sources: BLS (UNRATE, ICSA via FRED); NBER. Chart: Eco3min Research.

PNG
SVG
PDF
Free to use with attribution · Cite

Is the August 2024 Trigger Really a False Positive?

The interpretation of the August 2024 Sahm Rule trigger as a false positive depends on the assumption that the absence of an NBER-dated recession through April 2026 is definitive. This assumption is reasonable but not unimpeachable, and serious analytical work requires steelmanning the alternative.

Claudia Sahm herself, in interviews following the August 2024 trigger, suggested that the rise in the unemployment rate from 3.4% (April 2023) to 4.2% (July/August 2024) was driven substantially by an expansion of the labor force — particularly via immigration and re-entry — rather than by layoffs . In that interpretation, the rule’s numerical threshold was met but the underlying labor-market signal it was designed to detect (rising layoffs and accelerating job loss) was not present. This is consistent with the Composite Rule’s reading: claims momentum stayed below 25%.

Three competing interpretations of the August 2024 trigger are worth considering:

Three Interpretations of the 2024 Trigger

None of these three interpretations is definitively correct. What can be said empirically is that after the August 2024 trigger, the unemployment rate did not continue rising — it declined and stabilized — and the initial jobless claims data did not corroborate broader labor-market weakness. This is the same configuration that occurred in July 2003, which the NBER does not date as a recession. The structural argument for treating both 2003 and 2024 as Sahm false positives is therefore symmetric, regardless of which of the three interpretations one accepts.

A legitimate quantitative qualification is that the false-positive sample size is small. In 75 years, the Sahm Rule produced 2 false positives. Drawing strong inferences from n=2 about the nature of the rule’s weakness — supply-driven UR moves, structural breaks — is statistically aggressive. The Composite Rule’s 8/0 record is itself based on 8 events, which is enough to compute a binomial probability of zero false positives by chance (approximately 1 in 256 under a coin-flip null) but not enough to assume the rule would hold across all future episodes.

The Composite Rule: Two Indicators, One AND-Gate

The Composite Rule is defined as the logical AND of the Sahm Rule and the Claims Momentum Rule. Operationally, it triggers in any month when both of the following are true:

The mechanism is straightforward. The Sahm Rule captures the stock dimension of labor-market deterioration: the unemployment rate level relative to its recent minimum. The Claims Momentum Rule captures the flow dimension: how rapidly new jobless claims are accelerating. In a genuine recession, both rise together, because rising layoffs (a flow) feed directly into a rising stock of unemployed workers. The two indicators carry largely redundant information during recessions — which is why the Composite Rule did not miss any of the eight recessions it could test (1969 to 2020). Not all eight recessions arrived under the same macro conditions, which makes the wider reading of macro-financial regimes the natural backdrop for the rule’s record.

The two indicators diverge precisely when the unemployment rate moves for reasons other than rising layoffs. Two such episodes have occurred since 1969: July 2003 (jobless recovery period — payrolls growing slowly while labor force participation softened) and August 2024 (labor force expansion via immigration and re-entry exceeding job growth). In both cases, the Sahm Rule fired while claims stayed quiet — exactly the configuration the Composite Rule was designed to filter.

Crucially, the Composite Rule’s 100% record is constrained to the period in which initial jobless claims data exist with sufficient history (1969 onward, given the 2-year warm-up requirement). The earlier four NBER recessions (1948, 1953, 1957, 1960) cannot be tested with the composite framework. Whether the Composite Rule would have held during those four episodes is an unanswerable counterfactual.

What Happened Next? S&P 500 Forward Returns by Rule Trigger

The table below shows S&P 500 12-month forward returns measured from each rule’s first-trigger month. Modern data limits us to triggers since 1985 (the earliest available daily S&P 500 history in our source). Pre-1985 trigger lead times are documented qualitatively in the historical-turning-points section.

S&P 500 Forward Returns From Trigger Month

Returns computed from the closing S&P 500 level in the trigger month to the closing level 12 months forward. Max drawdown computed over the same forward window. Sample sizes are small (4–6); these statistics describe a historical pattern in a limited number of episodes.

Methodological note: Trigger-month forward returns use non-overlapping windows where possible. The August 2024 trigger has now been followed by 20 months of data, of which the 12-month return is observable. The 2003 false positive (July 2003) used the closing S&P 500 of July 31, 2003 as the entry point. Past distributions are not predictive of future outcomes; regime-conditional statistics describe historical patterns, not expected returns.

Key Levels to Watch

Historical Per-Recession Lead Times

For each NBER recession since 1948, the table below shows the first month in which each rule triggered relative to the official NBER start date. Negative values indicate a trigger after the recession had begun (the typical pattern for labor-market indicators, given that unemployment lags GDP); positive values would indicate a trigger before the recession started.

Negative lead = trigger occurred after the NBER recession start date. The pattern is consistent: labor-market indicators typically confirm recession start dates rather than anticipate them. The 1960 episode is an exception (Sahm triggered 5 months before the NBER-dated recession start). Composite Rule lead times are constrained by ICSA data availability beginning January 1967.

April 2026 — Current Observation

As of April 2026, the unemployment rate stands at 4.3%. The Sahm indicator is at +0.13 pp, well below the 0.50 trigger threshold. The Claims Momentum indicator is at 0.0% (the 4-week MA equals its trailing 12-month minimum). None of the five rules is currently triggered. The August 2024 Sahm trigger episode remains the most recent rule-based recession signal and remains classified as a false positive in this dataset, subject to revision pending future NBER dating decisions.

Methodology

The dataset combines monthly US Unemployment Rate observations (BLS series LNS14000000, available via FRED as UNRATE, January 1948 onward), weekly Initial Claims for Unemployment Insurance (Department of Labor, available via FRED as ICSA, January 1967 onward), monthly S&P 500 closing levels (via Yahoo Finance, 1985 onward), and NBER recession dates from the NBER Business Cycle Dating Committee. Weekly ICSA observations are aggregated to monthly by taking the last weekly observation of each calendar month; this matches the convention used in most published claims momentum analyses.

Rule Definitions (Formal Algorithm)

Episode and TP/FP Classification

An “episode” is defined as a contiguous block of triggered months separated from any prior episode by at least one untriggered month. Single-month episodes are counted as one episode. A trigger episode is classified as a true positive (TP) if its start month falls within the window [recession_start − 12 months, recession_end + 12 months] for any NBER-dated recession; otherwise it is a false positive (FP). The same recession can be matched to multiple trigger episodes (this occurs in the lengthy unemployment downturns of 1959–1961 where two Sahm episodes both relate to the 1960 recession).

Sensitivity to Threshold Changes

The Sahm Rule’s 0.50 threshold is the canonical specification published by Sahm (2019). Lowering the threshold to 0.40pp adds three false-positive episodes (June 2003, April 2024, and an extended June–December 2024 episode) and detects one additional Sahm episode within the 1948 true-positive cluster. Raising the threshold to 0.60pp yields a different trade-off: the August 2024 trigger disappears (indicator 0.533pp is below 0.60pp), the July 2003 trigger also disappears (indicator was 0.50pp, equal to baseline but not 0.60), and the November 1959 short episode disappears — but the 1960 recession is still detected by the subsequent October 1960 episode (indicator reaches 0.633pp). At 0.60pp, the rule’s testable history becomes 13 episodes, 13 true positives, zero false positives — though at the cost of the +5 month early lead for the 1960 recession, which becomes a −6 month lag instead.

For the Claims Momentum Rule’s 25% threshold, sensitivity analysis shows that lowering to 20% adds three additional false-positive episodes (October 1984, September 2005, and August 2023) while keeping all 8 recessions detected. Raising to 30% reduces the testable history to 11 episodes, all true positives, with zero false positives — still detecting all 8 NBER recessions. The 25% threshold matches the level cited in NY Fed Liberty Street Economics analyses of claims-based recession indicators.

Sensitivity for the Composite Rule: the zero-false-positive record holds across every combination tested in the grid Sahm threshold ∈ [0.40, 0.65]pp × Claims threshold ∈ [20%, 35%]. The episode count varies (8 to 11 episodes depending on thresholds) but no combination introduces a false positive. The 2003 and 2024 Sahm false-positive episodes do not become composite false positives at any threshold pairing tested, because claims momentum in those two episodes never exceeded the lowest tested gate of 20%.

Dataset Design

Python Reproduction Code

# Reproduce the Composite Rule from primary sources
import pandas as pd
from pandas_datareader import data as pdr

# Fetch UNRATE monthly + ICSA weekly
unrate = pdr.DataReader('UNRATE', 'fred', '1948-01-01', '2026-04-30')
icsa = pdr.DataReader('ICSA', 'fred', '1967-01-01', '2026-04-30')

# Aggregate weekly ICSA to monthly (last weekly obs per month)
icsa_m = icsa.resample('MS').last()

# Build Sahm indicator
unrate['ur_3m_avg'] = unrate['UNRATE'].rolling(3).mean()
unrate['sahm'] = unrate['ur_3m_avg'] - unrate['ur_3m_avg'].rolling(12).min()
unrate['sahm_trig'] = (unrate['sahm'] >= 0.50).astype(int)

# Build Claims Momentum indicator
icsa['icsa_4wma'] = icsa['ICSA'].rolling(4).mean()
icsa['cm_pct'] = (icsa['icsa_4wma'] / icsa['icsa_4wma'].rolling(52).min() - 1) * 100
icsa_m['cm_trig'] = (icsa.resample('MS')['cm_pct'].last() >= 25).astype(int)

# Composite: AND of the two
df = unrate.join(icsa_m)
df['composite'] = df['sahm_trig'] & df['cm_trig']
print(df[df['composite']==1].index.tolist())
# Expected output: 8 episode start months from 1970 to 2020
  

Dataset Download & Reproducibility

940 monthly observations · UNRATE + ICSA + NBER + 5 rule indicators + S&P 500 forward returns · January 1948 – April 2026 · Source terms: the S&P 500 level is licensed by S&P Dow Jones Indices and is not redistributed; the file carries public-domain BLS and Fed series, the rule indicators and derived forward returns. Not sub-licensed under Creative Commons.

Data Sources & References

• Primary US Bureau of Labor Statistics. Civilian Unemployment Rate, series LNS14000000, accessed via FRED as UNRATE, January 1948 – April 2026.
• Primary US Department of Labor, Employment and Training Administration. Unemployment Insurance Weekly Claims, accessed via FRED as ICSA, January 1967 – April 2026.
• Primary NBER Business Cycle Dating Committee. US Business Cycle Expansions and Contractions, accessed May 2026.
• Research Sahm, Claudia (2019). “Direct Stimulus Payments to Individuals.” The Hamilton Project, Brookings Institution. (Defines the Sahm Rule as Real-time Indicator number SAHMREALTIME.)
• Research Bry, G. and C. Boschan (1971). Cyclical Analysis of Time Series: Selected Procedures and Computer Programs. NBER Technical Paper No. 20. (Original framework for the Two-Month Rule’s momentum component.)
• Research Adrian, T., Estrella, A., Shin, H.S. (2019). “Risk-Taking Channels of Monetary Policy.” Liberty Street Economics, Federal Reserve Bank of New York. (Cited for claims-based recession indicator tradition.)
• Reference Sahm, Claudia (2024). “Stay Calm and Get Off the Recession Watch.” Stay-At-Home Macro, August 2024. (Author’s own discussion of the August 2024 trigger.)
• Reference S&P 500 daily closing levels, retrieved via Yahoo Finance (^GSPC), January 1985 – April 2026.

Methodological Limitations

• NBER recession dating is ex-post and revisable. The treatment of the August 2024 Sahm trigger as a false positive is contingent on the absence of an NBER recession date through April 2026. If NBER eventually dates a 2024 recession, the scorecard would be updated accordingly.
• UNRATE revisions. This analysis uses the currently-published UNRATE series, which has been retrospectively revised since the original real-time vintage. Real-time analysts in 1949, 1973, or 2024 would have seen slightly different numbers. We use the revised series for consistency and analytical clarity; this is the standard convention in the recession-indicator literature.
• Composite Rule data availability constraint. The Composite Rule can only be tested from January 1969 onward (24 months after ICSA begins, to allow the 12-month-rolling-minimum calculation to mature). The four pre-1969 recessions are evaluated using only the unemployment-rate-based rules.
• S&P 500 forward returns sample size. Composite Rule triggers since 1985 (when daily S&P 500 data is available in our source) yield n = 4 observations with complete 12-month forward windows. This sample is too small to support strong distributional claims; the table is presented as a record of what has happened, not as a forecast distribution.
• Threshold parameters are not estimated, they are taken as given. The 0.50pp Sahm threshold and the 25% Claims Momentum threshold are taken from their original published sources. We do not optimize thresholds against the historical record, which would constitute look-ahead bias.
• The Composite Rule is constructed in this study. Unlike R1–R4, R5 was not published ex-ante. Its 8/0 record describes the historical regularity but is itself a specification choice — combining two known rules via AND. Different combinations (OR, weighted, longer confirmation windows) would produce different scorecards.

Embed and Share

Frequently Asked Questions

Source

Related Eco3min Research

• US Unemployment Rate (1948–2026) — raw dataset — the underlying BLS series for this study
• US Initial Jobless Claims (1967–2026) — raw dataset — weekly ICSA data feeding the Claims Momentum Rule
• Sahm Rule Recession Indicator — daily dataset — the FRED-published version of the Sahm Rule for real-time monitoring
• Yield Curve Inversions (1976–2026) — the leading-indicator counterpart to this labor-market confirmation rule
• High Yield Credit Spreads as a Leading Indicator — credit-market recession signals that fire before labor-market indicators
• NFP Revisions at Business Cycle Turning Points — why labor-market indicators are particularly hard to read in real-time