Base rates & false positives
Months spent in recession make up roughly 13–15% of the post-1950 sample, and the unconditional odds of a recession beginning within the next 12 months run in the same ~15% range — that's the base rate. A signal that fired randomly would still 'predict' recessions at that frequency; a signal that fires too often or too rarely tells you less than it seems. Every recession indicator must be read against this baseline.
The false-positive problem
The yield curve's record — 8 correct calls, 2 false alarms since 1966 — is extraordinary by indicator standards, but it means even the best single signal is wrong a meaningful fraction of the time. Lesser indicators are worse: inverted credit spreads, PMI dips, and claims spikes all produce regular false positives. The question is never 'did it fire?' but 'what does firing imply against the base rate?'
How the composite handles it
Individual signals carry limited information; a panel of 18 weighted signals firing together is much harder to dismiss. Our composite score weights the whole panel — a single isolated warning moves the score modestly, while broad confirmation pushes into the high/severe bands where the historical hit rate is far better.