There has been a lot of talk of yield curve inversion lately.

Sounds interesting, how do I do that?

Which immediately leads to some confusion: which term gap? and for how long? what does that portend, and when?

I found this little interview snippet quoted below from the fellow who did the original research, Cam Harvey.

Executive summary: stick with the 10 year minus three month. You can use 5 year minus 3 month if you like but it’s pretty much the same result.

If that gap stays negative for a month or so (not just a brief blip), expect a recession starting 12-18 months later.

*Q: And as I read about, or hear pundits talk about, the yield curve and*

*inversion, there are always arguments about whether it’s the 5-year/2-year*

*slope, 10-year/3-month, 10-year/2-year, you know, pick your slope.*

*What have you found?*

*A: I think we need to go with the original research. In 1986, I looked at two parts*

*of the yield curve, the 5-year note minus the 3-month bill, and the 10-year bond*

*minus the 3-month bill. The crucial thing is to use a very short-term interest rate.*

*Those two yield spreads are highly correlated. I get nervous when I see people*

*talking about the 10-year minus the 2-year. I got all these emails last week about*

*the 5-year minus the 3-year inverting, saying “Happy Yield-Curve Inversion Day!”*

*An investor can data mine to find a piece of the yield curve—maybe it’s the*

*11½-year minus the 7¾-year—that has an inversion. That doesn’t mean anything.*

*I think we need to go with the original un-data-mined theory, which is based on*

*a longer-term rate and a 3-month rate, and the inversion importantly needs to*

*last for a quarter. If it’s a day, so what? GDP is measured quarterly, so we need to*

*measure this quarterly also.*

The update: perhaps surprisingly, it hasn’t gone negative yet this cycle.

Last time it turned negative in any lasting way was around late May or early June 2019.

It very nicely predicted the pandemic recession, which is pretty impressive.

It’s still batting 1000 out of sample since 1988.

Predicted every recession, and hasn’t raised any false alarms.

In a way it’s a shame that knowing the date of a future ecession is not the same at all as knowing the date of the next bear market.

Jim