Can a bad economic mood see a recession coming before the official numbers do? Researchers at the Federal Reserve Bank of San Francisco, the Bay Area's regional Fed, set out to answer that, and their conclusion may surprise anyone who has dismissed sour "vibes" as noise.
In a working paper released July 17, economists Nicolas Petrosky-Nadeau, Yeji Sung and Daniel J. Wilson tested whether "soft" data, measures of consumer sentiment, economic-policy uncertainty and the tone of the news, can forecast recessions as well as the "hard" data of jobs, output and prices. The paper is titled, plainly, "Do Vibes Predict Recessions?"
Their answer: yes, roughly as well, and sometimes better. Looking one month ahead, a model built only on sentiment and narratives scored higher on a standard accuracy measure than one built on hard economic statistics, and about the same as a model that combined both.
The soft-data model was quicker to flag rising recession risk, catching a much larger share of the months leading into a downturn, but it also cried wolf more often, producing more false alarms than the hard-data model.
The takeaway is not that feelings beat facts. It is that the two carry different information. Soft data, the authors write, both "proxy for information that later appears in hard-data releases" and "contain distinct information about recession risk" of their own. Combining the two produced the most reliable forecasts, which is why the researchers frame sentiment as a complement to traditional indicators, not a replacement.
AcadeResearch, a research-analysis site that highlighted the paper, wrote in its assessment that the study "puts a defensible number on the marginal recession-forecasting content of soft data at short horizons, and shows that content survives careful controls for publication lags, revisions, and overfitting."
The methodology was built to be honest about what could have been known at the time. The team trained its models on real-time data as it existed month by month, from August 1999 through May 2026, a stretch covering three recessions, so the models were not quietly cheating with numbers that were only revised into place later.
