Precision across the rare pre-recession months; higher is better. It punishes false alarms more than AUROC does.
AUROC
How well a forecast ranks months that preceded a recession above months that did not: 0.5 is a coin toss, 1.0 is perfect.
Brier
The average squared error of a probability forecast; lower is better. A forecaster who always says the base rate scores about 0.10 here.
downside beta
How much a holding moved in months when the S&P 500 fell, relative to the market (1.0 = the same).
drawdown
The fall from a previous high to a later low, in percent.
EBP
Excess bond premium: the part of corporate bond spreads not explained by default risk, a gauge of lenders' risk appetite (Gilchrist and Zakrajšek, Federal Reserve Board).
kill rule
A rule written before the model was relied on that says when to stop trusting it and what to publish instead.
NBER
The National Bureau of Economic Research, whose committee officially dates US recessions, usually months after they begin.
probit
A model that turns a weighted sum of inputs (η) into a probability with the normal curve: P = Φ(η).
stagflation
Inflation with a stalling economy, as in 1973–75 and 1980–82, when rising rates pushed bonds down along with stocks.
term spread
The 10-year Treasury yield minus the 3-month bill rate. Below zero (an inverted curve) has preceded most US recessions.
walk-forward
Testing a model by forecasting each past month with only the data and recessions known at the time, as if live.
z-score
How unusual a reading is against its own history: the distance from the median in units of typical variation.