Pre-registration
US 10-year minus 2-year Treasury yield-curve inversions are followed by meaningful labour-market weakening in most completed post-1976 episodes. The pre-registered test treats a completed inversion episode as a monthly average 10y-2y spread below zero for at least two consecutive months, with new episodes allowed after at least three non-inverted months. The claim is supported if at least 70% of completed episodes are followed by a rise in the unemployment rate of at least 0.75 percentage points within 24 months of episode start, and the median 24-month unemployment increase is at least 1.0 percentage point.
Falsification criterion — what would disprove this
This hypothesis is considered falsified if:
SUPPORTED if at least 70% of completed inversion episodes have a 24-month unemployment-rate increase >= 0.75 percentage points and the median increase is >= 1.0 percentage point. REFUTED if fewer than 50% pass or the median increase is < 0.5 percentage points. Otherwise PARTIAL.
formal test & threshold
test: completed_inversion_episode_unemployment_followthrough threshold: pass_rate >= 0.70 AND median_unemployment_increase_pp >= 1.0
Method
- Template
event_study- Clustering
none- Sample
- 1 countries · 1976 – 2026
- Evidence type
- associational
Custom replication script detects completed inversion episodes from daily FRED yields aggregated to monthly means, then evaluates unemployment-rate increases within the next 24 months.
Data
| Variable | Source | Transform |
|---|---|---|
unemployment_rate outcome | fred:UNRATEtier 1 | max increase within 24 months of inversion start |
ten_year_two_year_spread treatment | fred:DGS10tier 1 fred:DGS2tier 1 | monthly mean DGS10 minus monthly mean DGS2 |
● ready · ● pending · ● reconstruct-needed
Detailed result card
Result card - yield_curve_inversion_unemployment_us_1976_2026
Verdict: supported - 7 of 9 completed episodes pass; median increase 1.60pp
Episode Test
Support threshold: at least 70% of completed inversion episodes pass the +0.75pp unemployment follow-through test and median increase >= 1.0pp.
| Start | End | Baseline unemployment | Max unemployment within 24m | Increase pp | Pass | |---|---:|---:|---:|---:|:---:| | 1978-09-01 | 1980-04-01 | 6.0 | 7.8 | 1.8 | yes | | 1980-09-01 | 1981-10-01 | 7.5 | 10.1 | 2.6 | yes | | 1982-02-01 | 1982-06-01 | 8.9 | 10.8 | 1.9 | yes | | 1989-01-01 | 1989-06-01 | 5.4 | 6.4 | 1.0 | yes | | 1989-08-01 | 1989-09-01 | 5.2 | 6.9 | 1.7 | yes | | 2000-02-01 | 2000-12-01 | 4.1 | 5.7 | 1.6 | yes | | 2006-02-01 | 2006-03-01 | 4.8 | 5.0 | 0.2 | no | | 2006-06-01 | 2007-03-01 | 4.6 | 5.6 | 1.0 | yes | | 2022-07-01 | 2024-08-01 | 3.5 | 4.2 | 0.7 | no |
Interpretation
The test is associational: yield-curve inversion is treated as a timing signal, not a causal mechanism. Episodes that start fewer than 24 months before the latest unemployment observation are excluded as censored.
Strongest opposing argument
Every hypothesis ships with its charitable opposing argument. The framework earns credibility by handling objections at their strongest, not weakest.