Hard vs soft data
Hard data measures what happened: payrolls, industrial production, claims, credit spreads. Soft data measures what people report: PMI surveys, consumer sentiment, lending-standard questionnaires. Both fail in characteristic ways — hard data is slow and revised, soft data is fast but can capture mood rather than fact.
When each misleads
Soft data failed famously in 2022–24: sentiment and PMIs implied recession for over a year while employment never broke. Hard data fails the other way: 2020 was real before any lagging statistic could confirm it. The Fed's 2019 'insurance' cuts responded to soft-data deterioration that hard data never fully confirmed — growth held, hence 'insurance' — and 2023 showed the same soft-data alarm resolving into nothing.
How the panel uses both
Our signals are deliberately hard-data dominant — claims, production, spreads, housing — because measurable activity is what recessions actually are. The SLOOS lending-standards survey is the exception: it's soft data that predicts hard outcomes (loan volume), making it a leading rather than lagging soft series.