What the New York Fed's model actually produces is a probability, not a verdict. The model uses the slope of the yield curve to calculate the probability of a recession in the United States twelve months ahead. Its FAQ summarizes the underlying evidence this way: the term spread has borne a consistent negative relationship with subsequent real economic activity in the United States, "with a lead time of about four to six quarters," and models predicting recessions or real GDP growth "tend to explain 30 percent or more of the variation" in the measure of activity. A relationship that explains some of the variation, a year or more in advance, is a serious research finding and is not a timing instrument.
The historical record as the New York Fed states it, with its own qualifications attached. The FAQ reports that the yield curve "has predicted essentially every U.S. recession since 1950 with only one 'false' signal," which preceded the credit crunch and slowdown in production in 1967, an episode the NBER did not classify as a recession despite a marked decline in industrial production. On the narrower question of inversions specifically, it states that since 1960 an inversion measured on the ten-year against the three-month "has preceded every recession on record," and that in monthly averages the ten-year rate was at least 12 basis points below the three-month rate before each of those recessions. It also records the near misses that make the threshold matter: in two episodes during the 1990s the spread fell to very low positive levels, 42 and 12 basis points, without inverting, and activity continued in both cases. Two cautions belong with all of this. The FAQ is a summary of a research literature whose citations run to the mid-2000s, so it does not speak to more recent episodes. And the number of US recessions in the period is small, so a record of being right nearly every time rests on a handful of observations.
The FAQ is unusually candid about what is missing, and it is the honest part to quote. In its own words, "no theory establishes a clear connection specifically between yield curve inversions and recessions." The relationship is empirical. Two technical points from the same source sharpen how the signal should be read. First, it is the level of the spread that carries the information, "not the change in the spread, nor even the source of the change in the spread," so an inversion produced by short rates rising is not read differently from one produced by long rates falling. Second, persistence matters: one-day moves of more than 25 basis points in the spread occur about two and a half percent of the time, so "a signal that lasts only one day may be dismissed, but a signal that persists for a month or more should be looked at carefully," which is why the work is generally done on monthly averages.
The part that matters for a household is the arithmetic of the lead time. Suppose the signal is read correctly and a recession does follow. The lead is four to six quarters, the model's output is a probability rather than a date, the depth is not forecast at all, and the NBER dates the recession retrospectively, typically many months after the turning point it names. A decision to change a long-term allocation on this basis requires being right about the signal, right about the timing of the exit, and right again about when to return, which is the general problem covered under market timing. The indicator's usefulness is as a description of financial conditions, not as an instruction.
