Yield Curve Recession Indicator Has Correctly Picked 8 For 8

During their latest episode of the VALUE: After Hours Podcast, Harvey, Taylor, and Carlisle discuss Yield Curve Recession Indicator Has Correctly Picked 8 For 8. Here’s an excerpt from the episode:

Tobias: For those guys who’ve just come in late, it’s Cam Harvey, the creator, the inventor, the discoverer of the 10-3 inversion. Cam, a couple of things that you mentioned while you were going through that, that I just want to dig into a little bit. I think that I saw you appear on a Research Affiliates seminar, this is a few years ago now, and you were discussing the 10-3. Before I saw you present it, I didn’t actually realize that– Actually, it was the Meb Faber podcast from a few years ago. I didn’t actually realize that it was the 10-3, because it’s often quoted in the press as being they look at the 10-2. Do you know how that happened or why the 10-2 became substituted in collective mind over the 10-3?

Campbell: Yeah, it’s a very strange situation. So, my dissertation in 1986 uses– actually, I look at the 10 to 3 month and also the 5 to 3 month, and the 5 and the 10, highly correlated. But it’s really important to anchor with a short-term instrument like the three month. I chose the three month because it’s liquid and we measure GDP quarterly. So, let’s do something over a quarter. So, that indicator since 1986, we’ve had– So, this is the out of sample period. In the out of sample period, it is four out of four with no false signals. So, the way I’m looking at it– [crosstalk]

Jake: Which makes it eight for eight then now? Is that accurate?

Campbell: Yeah. So, think of it from the late 1960s. So, that was in my dissertation. Indeed, this is like another story. Most of the key analysis in my dissertation is over this period from the mid-1960s to 1985. In that period, there’s four recessions and there’s four inversions. It looks like four out of four perfect indicator. But my committee is saying, “Well, that could be lucky. You can’t get four out of four.”

It turned out that even though it could be lucky, they liked the idea, were willing to go forward with it for a few reasons. One of the overwhelming reasons was theory was sound. So, theory predicts this. Then if we see it in the data, then well, maybe it’s still lucky but still you’ve got some basis for it that some good strong intuition. That was number one reason. Number two reason, they were fascinated that my indicator got the double dip recession in the early 1980s. So, we had a recession, then a strong recovery, then another recession. These macroeconomic forecasting services, nobody got that. And then the third thing and people of Chicago especially like this that the alternative to paying tens of thousands of dollars a year to these econometric services is a very simple model that is as good or better and it costs the price of a Wall Street Journal, which at the time was $0.25. Yeah, this is good.

Jake: I thought you were going to say because it shows the efficiency of the bond market, the information that’s contained within there.

Campbell: Yeah, the $0.25 will drive the price down to what the efficient price should be.

Jake: Yeah. [laughs]

Campbell: Yeah, that’s where we’re going.

Jake: So, four for four in research, four for four cents in Post, and now recently another inversion has happened.

Campbell: Yeah. So, let’s go through that. That’s going to require some unpacking. I don’t want to leave this hanging, your previous question, well, why is it 10-3 versus 10-2?

Jake: Oh, yeah. Sorry.

Campbell: So, given that the indicator is four for four out of sample and eight for eight to the 1960s, I don’t see a particularly good reason to switch the model to something else that doesn’t have the same theoretical underpinning. So, if it was the case that out of sample, let’s say, I got two out of four, whereas this other indicator got four out of four, then okay, it seems like the model is broken, so let’s fix it. But that’s not what happened. So, the Fed in particular started talking about the 10-2. And by some metrics, it might fit the data a little better. However, it’s given a false signal in 1998.

Tobias: Right.

Jake: Mm.

Campbell: So, again, it’s just not really a good reason to abandon something that’s working.

Jake: Somebody else wanted their name on the 10-2? [laughs]

Campbell: Yeah, maybe. That should be suspicious of that just on its own. So, what I often say is, well, if we want to get the best possible fit in sample,” so calibrated to the data, it’s not the 10-2. It could be the eight-year, 2 month minus the 16 month.

Tobias: [laughs]

Campbell: So, there’s tens of thousands of combinations you could try to get the best possible fit, but we all know that when you do that, that model will most likely fail out of sample because it’s been overfit.

Jake: Right.

Campbell: So, my model is not overfit and it’s done well. So, eight out of eight, but the big question is, “Okay, great. That’s a fabulous historical record. What about today?” So, at the end of December, we had a full quarter inversion. So, my model is about quarters, right? Not months or weeks or days. So, that a 10-year minus 3 month, the spread was negative. So, short rates higher than long rates on average over that quarter. And at that point, I usually put a post up saying code red, and the track record is pretty impressive.

So, this time around, I did put the post up and I said, “The model is giving code red, but let me explain to you why I believe that my model is giving a false signal.” I guess that got the attention of a number of people.

Jake: Yeah.

Campbell: So, Harvey going against his model. Many people could go against my model. Many pundits are against my model already, but it’s a little different when you’re the whatever–

Jake: [crosstalk].

Tobias: Discoverer.

You can find out more about the VALUE: After Hours Podcast here – VALUE: After Hours Podcast. You can also listen to the podcast on your favorite podcast platforms here:

Apple Podcasts