Beyond U-3.

The thesis

The headline U-3 unemployment rate (currently 4.3%, April 2026) excludes three groups that materially change the picture of labor market health:

• Discouraged workers who stopped looking for jobs.
• Marginally attached workers who want work but haven't searched recently.
• Involuntary part-time workers who want full-time employment but can't find it.

The BLS U-6 rate, which includes all three, currently sits at 8.2% — nearly double U-3. This project builds the data infrastructure to track that gap historically, contextualize it against recessions, and monitor the forward-looking indicators that turn before U-3 does.

Key findings

The hidden labor gap is not stable — it widens in stress

Computed via SQL with a CASE-based recession classifier — see analysis_queries.sql →

Eight leading indicators — current state

Not every indicator is flashing red. The thesis is not that recession is certain — it is that U-3 alone gives an incomplete picture, and the indicators underneath show more stress than the headline suggests.

Architecture

Every piece is automated. A single Python entry-point reads 14 FRED series, resamples mixed-frequency data (daily/weekly/monthly → monthly), computes derived metrics, and loads everything into PostgreSQL via SQLAlchemy. Windows Task Scheduler kicks it off on the 5th of each month, two days after the typical BLS Employment Situation release.

FRED API  (14 series)
   |
   v
python/fred_pipeline.py
   |- fredapi pulls all series
   |- resamples to monthly
   |- computes U3/U6 gap, temp-help % off all-time peak
   |- bulk-inserts via SQLAlchemy
   |
   v
PostgreSQL 18  (labor_dashboard)
   |- unemployment_measures
   |- labor_force_participation
   |- epop
   |- leading_indicators
   |
   v
CSV exports  -->  Tableau Public  (2 dashboards)

Schema & SQL

Four normalized tables. Each pipeline run truncates and re-inserts so the snapshot is end-to-end consistent with the latest FRED release.

CREATE TABLE unemployment_measures (
    id      SERIAL PRIMARY KEY,
    date    DATE NOT NULL,
    u3_rate DECIMAL(5,2),
    u4_rate DECIMAL(5,2),
    u5_rate DECIMAL(5,2),
    u6_rate DECIMAL(5,2)
);

CREATE TABLE leading_indicators (
    id                    SERIAL PRIMARY KEY,
    date                  DATE NOT NULL,
    temp_help             DECIMAL(10,2),
    temp_help_pct_decline DECIMAL(5,2),
    quits_rate            DECIMAL(5,2),
    saving_rate           DECIMAL(5,2),
    jobless_claims        DECIMAL(10,0),
    yield_curve           DECIMAL(6,3),
    consumer_credit       DECIMAL(10,2),
    cc_delinquency        DECIMAL(5,2),
    prime_age_lfpr        DECIMAL(5,2)
);

The query behind the headline finding

SELECT
    CASE
        WHEN date BETWEEN '2007-12-01' AND '2009-06-30' THEN 'Great Recession'
        WHEN date BETWEEN '2020-02-01' AND '2020-04-30' THEN 'COVID Shock'
        WHEN date BETWEEN '2001-03-01' AND '2001-11-30' THEN 'Dot-com Recession'
        ELSE 'Normal Period'
    END AS period,
    ROUND(AVG(u6_rate - u3_rate), 2) AS avg_gap_pts,
    COUNT(*) AS months
FROM unemployment_measures
GROUP BY 1
ORDER BY avg_gap_pts DESC;

The 14 FRED series