No More Neutral Rate?

Last Friday, Reuters published the article "No more neutral rate? The shine comes off the Fed's r-star," by Howard Schneider. He refers to a paper published at a recent conference that low rates lead to a lower estimated "neutral interest rate" (r*). I am unsure about the exact argumentation behind the research that the article is based on, but this is a theme I have discussed in the past.

(Note: I mis-clicked, and published a day early.) 

The interesting part of the Reuters article is as follows:

“In some sense lower rates beget lower rates,” said Piti Disyatat, one of the authors of the paper and a research economist with the Bank of Thailand. If Fed policy decisions also affect the neutral rate, “its ability to act as a benchmark is undermined,” he said.
The paper was among those presented at an economic conference in Boston a decade after the 2008 financial crisis ushered in an unprecedented era of zero and negative interest rates across the globe. They chip at some of the assumptions that have underpinned central banking since the inflation shock of the 1970s.

After quick searching, I could not find the paper presented at the conference. However, I found an earlier paper for which Piti Disyatat is a co-author, and I would make a wild guess that methodologies would be similar.

The earlier paper was a Bank of International Settlements working paper, "Monetary policy, the financial cycle and ultra-low interest rates," by  Mikael Juselius, Claudio Borio, Piti Disyatat and
Mathias Drehmann.

I would briefly summarise that paper as that it is based on Vector Autoregression analysis. They estimate the effects of various shocks on the economy, and then use them to examine the counter-factual case with different monetary policy settings. The path of realised rates determines the estimate of r*.

Breaking Mainstream Assumptions Completely

I would argue that it is a lot easier to do this by scrapping the assumption that interest rates matter at all. The generation of a counter-factual is straightforward:

1. Take observed real GDP growth, inflation, real interest rates.
2. Generate a new scenario where only the real interest rate changes.*

If we base our analysis on the U.S. experience from 2003-2018, we can summarise the trajectory of variables as follows (I am approximating dates):

• steady state** from 2003-2007,
• all heck breaks loose in 2008-2009,
• bounce in 2010-2011,
• boring steady state from 2011-2018.

Actual real rates were higher in the first steady state (2003-2007), and lower in the second (2011-2018). The estimated r* (using any standard methodology) is also higher in 2003-2007 than in 2011-2018.

If we ignore the exciting bit in between the two steady states, this has to happen by definition. In both steady states, growth rates and inflation were relatively stable. The economy was not continuously accelerating in one direction or another -- which is assumed to happen in every mainstream empirical work that I am aware of. Therefore, the estimate of r* has to converge (in some sense) to the stable realised real rate. (The non-zero output gap allows for a a bit of a gap.)

What happens if we fix every single variable except the realised real rate? Exactly the same thing: the estimate of r* has to converge to the realised real rate.

In plain English, if interest rates have no effect on the economy (putting aside interest income changes) and if the Fed held the real rate at 3%, r* would be estimated at being at (around) 3%. That is, we would have had "secular stagnation" with a wildly positive r*.

How Bonkers is that Counter-Factual?

I apologise to any conventional economists that sprayed coffee over their monitor when I suggested that changing the interest rate has no effect on any other economic variable. That is a strong assumption: but how do we know whether it is correct or not?

One can point to research that demonstrates a non-zero effect of interest rates on the economy; in fact, such an effect is estimated in the aforementioned BIS working paper. However, we can explain the results by pointing out that fiscal policy (once we take into account automatic stabilisers) was highly correlated with monetary policy over the period in question, and so the effect they ascribe to interest rates was actually due to fiscal policy (which I believe is only partly represented in the analysis). Until we can ensure that no other variables that are correlated with the monetary policy stance are explaining changes to the economic trajectory, a belief in r* is non-falsifiable.

I am agnostic on the question whether interest rates have literally no effect on the economy, but I would just note that I am unaware of any conventional research that convincingly deals with the question of falsifiability.

Concluding Remarks

This is obviously an important question for business cycle analysis. This article is a re-wording of my existing thinking on the topic; I hope to dig deeper into the topic over the coming months.

Footnotes:

* If we were modelling everything in the economy, this would create an inconsistency once we take into account the effect on interest income changes. If interest rates change, interest income flows have to change (otherwise, accounting breaks). That said, since debt outstanding has a relatively long duration, the interest income flows are not that large. Meanwhile, we are typically not accounting for every economic time series in our econometric analysis, so we do not have a complete set of stock-flow consistent accounts in the first place.

** Admittedly, a steady state housing bubble.

(c) Brian Romanchuk 2018