Why Treasuries Are ‘Risky’ Again | Odd Lots

Thank you for having me. It's really great to be here.

So most broadly, I'm interested in how the macro economy, monetary policy and inflation are reflected in bond markets and are in turn shaped by them. I think that's a pretty interesting topic because we did to, you know, to me, the macro economy, inflation, unemployment, those are the big questions. They're not easy to answer about. That makes it all the more intriguing to me. And of course, financial markets are unique in that they're sophisticated. They're forward-looking. So a lot of these things get priced in. And that's what I find very interesting.

Yeah. So that brings me to my research on the policy reaction function.

Most simply, that's what markets or observers more broadly expect the Fed to do in response to economic conditions. So as a numerical example, if forecasters expect 2% inflation, they might forecast a 4% policy rate. If they expect 4% inflation, they might expect a 6% policy rate. So that would be a one-for-one adjustment, a perceived policy reaction coefficient of one. And that will, that matters. This policy reaction function matters for at least two reasons. And one of them comes back to your question. This reflection back, which is that if the policy reaction function is well understood, then that means that as data comes out, unemployment, inflation and so on, the policy rates will move in the direction that the Fed intended and thereby help with monetary policy transmission or maybe even speed it up before the next FOMC meeting.

Yes. But if you see only if the reaction function is understood.

Yes, that's a great question. So the, I mean, I think the main point is that a policy reaction function can and should be measured. And what my coauthors and I have done, I will think of is actually the most obvious way of doing it, which is, I think, always a good starting point. So this is joint work with my, is Michael Bauer at the San Francisco Fed and Addison Jerome at Harvard Business School. So we took two approaches. The first one is, based on forecasts, we didn't run our own surveys because we wanted to know, you know, how have these perceptions changed over time, historically, over decades? And, you know, a policy reaction function is quite similar to a 1993-type Taylor rule. But there are important differences. The first difference is that it's forward-looking. So it's about expectations. The second difference is that it's a little bit shorter-run. It's about the next couple of quarters at most. The next couple of years. And traditional monetary policy rules are used to describe historical data in a backward-looking manner. What we said is, hang on, rather than using this historical data, let's just use forecasts, survey data, interest rate forecasts, and it turns out that every month there's a lot of richness in terms of the forecasts that are being produced for the policy rate. And so we use this data from the Blue Chip of Financial Forecasters, which asks, what is your forecast for the Fed funds rate and what are, what are the underlying assumptions that you used for inflation and output to come up with this policy rate forecast? So it's in a way, it's an ideal setup because it's asking what is the role for the Fed that you're plugging in. And so we just took this, this data that's forward-looking, you know, 30 to 50 forecasters, a couple of forecast horizons up to six quarters out. And we ran the exact same regression that you would run for a monetary policy rule. But, you know, on this different data, forward-looking data. So yeah, if you basically treat this, we can't run experiments on the market economy. But if you treat this data as the best approximation of an experiment on the macro economy, you know, if we run a regression of the forecasted Fed funds rate onto forecasted inflation and forecasted output relative to potential, and so the coefficients there on inflation would tell us, when forecasters are forecasting high inflation, how much higher is the expected Fed funds rate. And vice versa, for example. So that's the first methodology. The second one is closer to what I said about the macroeconomic announcement dates and rates moving in response. That's of what you can use that also to obtain a market-perceived rule, just to say if inflation comes out higher or lower than expected. Let's say inflation comes out higher than expected. And interest rates, say the two-year rate goes up, goes up a lot. Then, you know, that is also a regression that we run. You need a couple of observations. Of course. It's a rolling window, but you can run this regression of how much do yields change onto what's the inflation news on that day. And that gives another way of getting a perceived reaction function. It turns out that both of these methodologies often give very similar answers. And I love that because it makes me a lot more comfortable with either methodology.

Yes, that's, that's interesting. So we have focused on the variation over time and the point estimate of the reaction function. Got it. There clearly is a lot of heterogeneity across forecasters. And we've done, we've looked, we've looked at that a little bit, but we've really focused on how does it vary over time. Got it. But you know, as one example, when there was a lot of agreement, you know, after late 2011 was when the Fed came out and gave very clear date-based forward guidance, interest rates will be at zero at least until mid-2013. At that point, all the Fed funds rate forecasts collapsed to zero. And you know, you see absolutely no dispersion. So that would be an example where, where that collapsed.

Great question. And that's, that's a question that my coauthors and I thought a lot about. I think, you know, just having measurement is extremely useful, whichever approach you choose. Right? So in order, picking a framework or picking a communication strategy, evaluating it, I think is very important. So now what we looked at in terms of what moves the, the needle on the perceived reaction function. Let me take you back to 2020 was not a good time for any of us. But so that was the time of the pandemic. Initially, inflation was extremely low. And in 2021, inflation then picked up and there were some of these really big positive inflation surprises. And that was also the time where there was the discussion of, you know, is inflation permanent, is it transitory, keyword. We need to get a transitory. Yeah, yeah. Foreign. Yeah. The Fed initially just kept interest rates at zero. It was also the time when there was this, now old new framework of average inflation targeting, and in 2021. Yeah. So the interest rates stayed at zero for, for quite a while, even though inflation was running at five, 6%. And the perceived reaction function that we saw at that time was extremely flat. So there was a perception that, that the policy rate would stay at zero pretty much irrespective of economic conditions. And you saw that, for example, in May 2021, that was a particularly positive inflation surprise. That was an inflation surprise of almost half a percent. So annualized, the surprise component was 6%. That's huge. Yeah. And the two-year rate didn't move. Not at all. And you also saw that in the forecasts, which were all very, very close to zero. And that was true for the forecasters who had higher inflation expectations. That was true for the ones who had lower inflation expectations. So that's a little bit the nice thing about our data set, because of course, there was a very big, good shift, transitory, as Powell called it, at some point at the time, a lot of, a lot of people thought inflation was transitory, but not everyone. And by looking at this set of different forecasters, we can also see the ones who thought that inflation would be higher and all of them expected zero interest rates. And that, I think, is quite telling about, you know, the response that, that was anticipated at the time. And so then, you know, towards the end of 2021, the tone started to change. And March 2022 was liftoff, at first just 25 basis points. And then we saw this really, really big interest rate hikes repeatedly. And so after the Fed started to act, that's when our, our data, we see that the perceived inflation response really picks up. It basically goes from 0 to 1. Between say early 2022 towards the end of 2023. So there was a very big increase in perceived inflation response, but it occurred late and only after the Fed had started to act. So we call that the learning from actions channel.