The Yield Curve, Read Like a Weather Map

You now have two numbers from this course. The overnight rate, which a committee sets, and the ten-year rate, which the market sets. Lesson 3.1 said the market's number is the one that prices the board. Lesson 2.5 noticed the two had come apart.

Here's the thing to do with two rates: subtract them. Draw a line from the short one to the long one, and the shape of that line is the most watched picture in finance. It has called every American recession since the 1970s before the scorekeeper dated it, and it's been doing something for the last four years that it has never done before.

Weather maps don't tell you it's raining. You can see that. They tell you what the pressure is doing, which is where the rain comes from. The question this lesson answers: how does one line drawn between two interest rates warn of a recession a year before the scorekeeper dates it?

Chicago, 1986. A doctoral student named Campbell Harvey is finishing a dissertation under Eugene Fama, who will later win a Nobel for showing how hard markets are to beat. Harvey's question is narrow: is there anything in the bond market that tells you, in advance, when the economy is going to shrink?

He finds it in the gap between a long rate and a short one. When the short one climbs above the long one, when lenders will take less to lend for years than for months, a recession follows. He tests it on the four recessions in his data, from the 1960s to the early 1980s. Four for four.

Then the test that matters, the one that happens after the paper is written. The curve inverts in 1989, in 2000, and in 2006. A recession follows each one within a year and a half.

Asked about it decades later, Harvey put it this way: "I can predict with 100 percent accuracy there will be a recession, at some point. That's basic economics. The issue is when." His line answered when, within about a year, every time.

One student, one subtraction, and a signal that has been right about the next year for forty years, published in advance, where anyone could check it.

Now the part of the story that's still being written. In July 2022 the curve inverted again: the two-year above the ten-year. It stayed inverted for five hundred and thirty seven trading days, into August 2024, the longest stretch in the series, and at its deepest more than a full point.

By every prior reading, a recession was due by the end of the following year. The scorekeeper has dated none. This week the line sits at plus a third of a point: positive, barely.

Ask the tycoon's question. Who made the rule? Nobody; the curve is the sum of a million lenders' guesses. Who gains from reading it? Whoever moves before the stall instead of during it. Who paid for ignoring it? In every case before 2022, whoever borrowed short at the top and got repriced on the way down. And the open question, this time, is whether the signal broke or the stall is late.

Start with what the two numbers are. The two-year yield is roughly the market's average guess at the committee's overnight rate over the next two years. The ten-year is the same guess over ten, plus something extra for the risk of waiting that long.

That extra has a name, the term premium, which means the pay a lender demands for locking money away for a decade of unknowns: inflation, supply, a government that might issue more.

Lesson 2.5 flagged a ten-year sitting well above the overnight rate while the committee cut. Part of that's the market's guess at future policy. Part is the premium, and the premium is where the long cycle's pile shows up in the curve. The yield curve is the line through those points and the ones between; the spread is the simple subtraction, long minus short.

Now read the shapes as seasons, because they're the short cycle from lesson 2.2 seen from the bond desk.

Steep is the refill. The committee has cut the overnight rate, so the short end is low. Lenders won't lend for ten years at a low rate, so the long end stays up. A big positive spread. Early in the cycle, credit cheap, the heat not built.

Flattening is the heat. The committee starts raising. The short end climbs toward the long end. The spread shrinks. Mid-cycle: the bowl is on its way out.

Inverted is the warning, and here's the mechanism inside it. The short rate is high because the committee put it there. The long rate is lower because lenders expect the committee to cut, and soon, and by a lot.

Why would they expect that? Because they expect the stall. An inverted curve is the market saying, in the only language it has, that today's rate is too high to last, and the reason it won't last is the recession it's about to cause.

Re-steepening is the turn, and it's the part people read backwards. The curve doesn't un-invert because things got better. It un-inverts because the committee starts cutting the short end fast, into the stall. The steepening happens on the way into the recession, not out of it. In each of the last three cases before this one, the curve was already back above zero when the scorekeeper's dated peak arrived.

One more piece, because it's the reason the signal works at all. Banks borrow short and lend long. That's the whole business. When the short rate is above the long rate, lending long loses money, so banks stop. Credit stops. Lesson 2.1 told you what happens next, in order. The curve isn't predicting the stall from outside. It's the price signal that causes it.

Which is also why the signal can fail without being wrong. If banks find another way to earn while the curve is inverted, or if the borrowers who matter don't depend on bank credit, the stopping doesn't stop as much. That's one of the three possibilities on the board for the 2022 case, and it's the one the desk finds least comforting, because it means the machine changed shape while the gauge stayed the same.

The whole series, the ten-year minus the two-year, daily since 1976. Every dip below the line is an inversion. Six of them lasted more than a month. Read them against the scorekeeper's dates.

Five inversions, five recessions, with a lag of ten to eighteen months from the first day below zero. No false alarms in that table: every time the two-year closed above the ten-year for a month or more, a recession was dated within a year and a half. That's the record Harvey's dissertation set up in 1986 and the market kept adding to for thirty five years.

Then the sixth row. July 2022 to August 2024. The longest inversion in the series, the deepest since 1980, and no recession dated. The lag from the first inversion is now four years and counting.

Either the signal has failed for the first time, or the stall is arriving late, or the ground moved under the signal: a committee cutting into warm prices, a bond market charging for a pile of debt rather than pricing the next two years of policy, both of which lesson 2.5 put on the board.

The desk doesn't decide which. It writes the three possibilities down and dates them.

The incentive map. Who set the short rate that inverted the curve? The committee, on purpose, to cool prices. Who set the long rate? Lenders, voting on the committee's future. Who gained from the inversion? Holders of two-year paper, paid more than ten-year holders to wait a fifth as long. Who paid? Banks, whose business of borrowing short and lending long ran backwards for two years, and everyone whose credit dried up as a result.

Here's the reveal. The curve isn't a forecast that comes true. It's a price that makes itself true: when short money costs more than long money, the lending that runs the machine stops paying, so it stops, and the stall follows from the stopping. The one time it hasn't followed is the time the desk is living in.

A digital tycoon reads the curve the way a sailor reads pressure: not to know the weather, to know what's likely and how soon. Three reads.

First, know the shape today, in one number. Ten-year minus two-year, from the same public series the chart uses. This week: plus a third of a point. Positive, flat, coming off the longest inversion in the record. Write that next to the date. Next quarter, write it again.

Second, borrow to the shape. In a steep curve, short borrowing is cheap and long is dear; that's the time to lock long if you'll need it. In an inverted curve, short borrowing is the expensive kind and the kind that gets pulled first. The office's runway rule, cash to cover the months a client's client can't roll, is a curve rule wearing a plain name.

Third, don't argue with a price. When the curve is inverted, someone is willing to take less for ten years than for two. That person isn't a forecaster; they're a lender putting money behind a view. The office can disagree with the view. It doesn't pretend the view isn't there.

Our paper fund reads it by its rules: a written reason on every move, no leverage, so an inverted curve can never force a sale, and a score kept against an asset with no yield curve at all.

• T10Y2Y, the 10-year minus 2-year Treasury constant maturity spread, daily from June 1, 1976. Spells of more than 20 trading days below zero, computed from the series: August 18, 1978 to May 1, 1980 (423 trading days, minimum −2.41 on March 20, 1980); September 12, 1980 to October 23, 1981 (278 days, −1.70 on December 17, 1980); January 4 to June 29, 1989 (123 days, −0.45 on March 28, 1989); February 11 to December 26, 2000 (220 days, −0.52 on April 7, 2000); August 17, 2006 to March 20, 2007 (147 days, −0.19 on November 15, 2006); July 6, 2022 to August 26, 2024 (537 days, −1.08 on July 3, 2023). Value on September 11, 2026: +0.33. FRED.
• NBER business cycle peaks: January 1980, July 1981, July 1990, March 2001, December 2007, February 2020. No peak dated after February 2020 as of this writing. NBER.
• Lags from first inversion day to the dated peak: 1978 to Jan 1980, about 17 months; 1980 to Jul 1981, about 10 months; 1989 to Jul 1990, about 18 months; 2000 to Mar 2001, about 13 months; 2006 to Dec 2007, about 16 months. Arithmetic on the two sources.
• Campbell Harvey's 1986 University of Chicago dissertation linked yield curve inversions to subsequent recessions, using a long Treasury yield minus a 90-day bill; the 2018 Duke piece reports it "has gone on to predict all three recessions since: in 1991, 2000-2001, and the global financial crisis" with "no false signals to date," and quotes Harvey: "I can predict with 100 percent accuracy there will be a recession, at some point. That's basic economics. The issue is when." Duke Fuqua, July 10, 2018.
• 10-year 4.95 percent and 2-year 4.56 percent on September 10, 2026. FRED.
• The February 2020 recession was preceded by a brief 10-year minus 2-year inversion in August 2019 of fewer than 20 trading days, below the threshold used in the table; Harvey's own bill-based measure inverted for longer in 2019. FRED, T10Y2Y and T10Y3M.