The yield curve plots the interest rate the US government pays to borrow at different lengths of time. Normally, longer-term borrowing costs more than short-term borrowing, because lenders demand compensation for tying up money. When the curve inverts, short-term rates are higher than long-term rates. That is the signal.
The most-watched version is the 10-year Treasury yield minus the 2-year. When that spread goes negative, the curve is inverted. Since 1955, every US recession has been preceded by an inversion. The lead time has varied from six months to two years. There has been one false signal, in the mid-1960s.
The mechanism is not magic. The 2-year yield is essentially a forecast of where the federal funds rate will average over the next two years. The 10-year yield is the same forecast extended out to ten years, plus a term premium. When the 2-year is higher than the 10-year, the bond market is saying: the Fed is currently too tight, it will have to cut, and the cuts will happen before the ten-year horizon is up.
That is a forecast of Fed easing, which historically only happens in response to a slowdown. So an inverted curve is not predicting a recession directly. It is predicting that the Fed will need to ease, and the Fed only eases that aggressively when growth falls.
Two caveats matter. First, the curve un-inverts before recessions actually start. The recession signal is the inversion plus the subsequent steepening as the Fed cuts. Second, an inversion can persist for a long time without a recession arriving, which makes the signal hard to trade on. As a forecast tool it is reliable about direction and unreliable about timing.
The curve is not a crystal ball. It is the bond market’s collective bet on Fed policy. When that bet says the Fed will have to cut hard, it is usually right.