2. The inverted curve as a recession predictor: the real track record and its failures

Macro 3: The yield curve and the business cycle

Few market statistics carry the fame of this one: in the United States, every recession since the late 1960s has been preceded by an inversion of the yield curve. The sentence is true and, in the way it is normally used, deeply misleading. This module takes it apart piece by piece, because understanding why a good signal can be useless in practice is worth more than memorizing the signal.

Where the fame comes from

The idea entered the academic literature in the 1980s and 1990s. Campbell Harvey showed in his doctoral thesis that the slope of the curve contained information about future growth that was not in other indicators. Shortly afterwards Arturo Estrella and Gikas Hardouvelis, and later Estrella together with Frederic Mishkin at the New York Federal Reserve, compared the curve with the usual suspects: the stock market, the leading indicators, the monetary aggregates. At horizons of two to six quarters, the spread between the ten-year bond and the three-month bill beat almost all of them, and it was also absurdly simple to calculate. Ever since, the New York Federal Reserve has published a monthly recession probability estimated from that single number.

There is also a plausible causal mechanism, not just a correlation. Banks fund themselves short and lend long. If three-month money costs more than a ten-year loan yields, the margin on new lending is squeezed and banks tighten their terms. An inverted curve does not merely anticipate the slowdown: it can help bring it about.

The track record told honestly

Here begins the part that almost never gets told. The full sample is around eight US recessions in more than half a century. Eight observations. With eight points you cannot do serious statistics on much of anything, still less when the indicator was chosen after looking at those very data. Anyone sifting through fifty economic series from that period will find several with a spotless record, and most of them will be coincidences.

And even with eight cases the record is not clean:

• 1966. The curve inverted, credit dried up, the economy clearly slowed and the recession was never declared. It is the classic false positive, and it is still the example people cite half a century later precisely because there are so few cases to cite.
• 1998. A brief inversion during the turbulence of that summer. The next recession began in 2001, two and a half years later and for different reasons. Calling that a prediction is stretching the word.
• 2022 to 2024. The longest inversion on record, more than two years, and to this day no recession dated by the body that dates them in the United States.

The lag: the detail that makes it almost unusable

Even if the signal were always right, it still could not be traded on, and the reason is the lag. Between the inversion and the start of the recession anything from a few months to more than two years has passed, depending on the episode. In 2006 the curve inverted and the recession began in late 2007, a good eighteen months later. In August 2019 it inverted and the recession arrived in February 2020, barely six months on, but caused by a pandemic that no curve could have anticipated: even the hits have attribution problems.

Translated into decisions: if you sell your portfolio on the day of the inversion, you have no idea whether you are six months early or thirty. And the stock market is not the economy. It tends to keep rising while the curve is inverted, because earnings are still growing. Anyone who liquidated a portfolio of US equities in July 2022, the day the spread crossed into negative territory, missed a rally of close to 50% through the end of 2024, dividends aside, waiting for a recession that was never declared.

Why the signal may have aged

Two underlying reasons, beyond the size of the sample:

• The term premium has moved. As far back as 2005 Alan Greenspan called it a conundrum that long rates refused to rise while the Federal Reserve was tightening. After years of heavy bond buying by central banks, insurers and pension funds, part of the flattening may be mechanical rather than a genuine expectation of cuts. If the inversion comes from there, it does not say what it is supposed to say.
• The signal is public. Everyone watches it: companies, banks, governments. An indicator this closely observed changes the behavior of those observing it, and that can either bring the adjustment forward or bring it about.

What to take away

An inverted curve is not an oracle. It is a reasonable thermometer of one specific thing: that the market considers monetary policy to be restrictive and expects it to have to loosen. That is valuable information, and free into the bargain. What it is not, and never has been, is a sell order with a date on it.

There are also two details that almost nobody respects. The first, that the people who proposed the signal required an inversion sustained through a full quarter, not the one-day crossing that the press celebrates on the spot. The second, that the signal is binary: even if it were always right, it says nothing about how deep the recession will be or how far the stock market will fall, which is exactly what an investor would need to know in order to do anything with it.

There is even a nuance that tends to be ignored: historically the recession has tended to begin once the curve had already steepened again, not during the inversion. It is an interesting regularity and it rests on the same scant half-dozen episodes, so it deserves the same skepticism as everything above. If someone sells you this signal without mentioning the sample size, the variable lag, 1966 and the 2022 episode, they are not teaching you anything: they are selling you something.