Stein on Financial Stability in Monetary Policy

Fed governor and Harvard Professor Jeremy Stein gave an important speech on March 21, Incorporating Financial Stability Considerations into a Monetary Policy Framework. I have a few minor criticims, specifically on standard errors, causal mechanism, and Lucas critique. But it's great for Jeremy to think out loud this way, and give me occasion to do the same. You should read the whole thing.

Stein's bottom line:

...all else being equal, monetary policy should be less accommodative--by which I mean that it should be willing to tolerate a larger forecast shortfall of the path of the unemployment rate from its full-employment level--when estimates of risk premiums in the bond market are abnormally low.

This view has put Stein a bit in the camps of the hawks, meaning simply those who for one reason or another think the time to raise rates is sooner rather than later.

This is an interesting framing. Why did Stein say "forecast shortfall of the path of the unemployment rate from its full-employment level" and not just "more unemployment?" Stein is pitching the argument, I think, at the other FOMC member's sensitivity to unemployment.  If the Fed ultimately cares about unemployment a year from now, the probability of a shock that would unexpectedly raise unemployment matters as much as the Fed's expected value. His idea: a bit of tightening might raise the level a bit, but lower the variance.

"How Do You Measure Financial Market Vulnerability?"  Stein thinks about leverage measures, and concludes they are not useful in real time, that to the extent they can be measured, they are better addressed with regulation rather than interest rates. Most of all

How, if at all, does monetary policy influence the evolution of the ratio? Without an answer to this question, it is hard to say how much one would want to alter the stance of policy when, say, the ratio is abnormally high relative to trend.

He concludes that the Fed should watch risk premiums -- the expected excess return on long term treasuries and corporates -- and be ready to tighten if risk premiums seem too low. Essentially, the Fed should add a new term to the Taylor rule,

interest rate = phi_pi*inflation + phi_u*unemployment + phi_r*risk premium. 

(my interpretation, not the speech.)

As an illustration, consider the period in the spring of 2013 when the 10-year Treasury yield was in the neighborhood of 1.60 percent and estimates of the term premium were around negative 80 basis points (3). Applied to this period, my approach would suggest a lesser willingness to use large-scale asset purchases to push yields down even further, as compared with a scenario in which term premiums were not so low.

But measuring the term premium is tricky stuff. It's not just the spread between long bond and short bond yields. If long bonds are 1.60% and short bonds are 0%, it might just be that everyone expects interest rates to rise in the future, and expected returns are the same for holding any type of bond. The "risk premium" is how much of that spread exists over and above (or in this case, under and below) people's expectations of rising interest rates.

So how do you separate the yield spread into expectation and risk premium  components? Footnote 3:

The 10-year nominal rate hit 1.63 percent on May 2, 2013. An estimate of the term premium based on the oft-cited methodology of Kim and Wright (2005) was negative 0.78 percent on this day.

OK, how do Kim and Wright come to this conclusion? Basically, by running regressions. They (we) examine, in the past, what configuration of bond prices and other variables have been followed by interest rate rises ("expectations hypothesis"), and what configuration has been followed by good returns to bond investors ("risk premium")?

This is an imprecise business. Regressions have standard errors big ones. Regressions vary even more by specification -- which variables do you put on the right hand side. Having written two papers on bond risk premiums, I can attest those standard errors and specification uncertainties are large.

At a minimum, I think Stein would do all of us a favor if he would include standard errors and specification errors.  My guess though is that they would be at least one if not two percentage points. The risk premium was somewhere between negative 2 and positive 2 percent, not -0.78%. That might undermine his case (!), but perhaps the Fed can write an internal memo that everyone has to quote numbers with standard errors. So, the natural rate of unemployment is not 6.500%, but has at least a percent or two uncertainty as well.

The same point holds for the much more important credit spreads. If the BAA bond spread is 1%, does this mean a 1% chance of default (including recovery)? Or does it mean that the price is temporarily low and people holding BAA bonds will earn on average 1% more on other assets? Solid research breaking out this spread, also by examining historical correlations, is just beginning.

More deeply,  the historical correlations come from a sample in which the Fed was not affecting long rates.  I don't think QE did much to long rates, but the Fed does, with some sort of "friction" or "segmented market" in mind. That would make those regressions pretty useless now. If you force the weather forecaster to say it will be sunny, the usual correlation between forecast and reality will fail.

More deeply still, there is a classic Lucas Critique problem. Historical correlations can be counted on to move as soon as the Fed exploits them for policy. If low short rates were correlated with low credit spreads which were correlated with subsequent financial turmoil in the past, will raising short rates raise credit spreads and lower financial turmoil now?

The cure is to understand the causal structure, but here we're all really at a loss. Everyone writes about how low interest rates lead to a "search for yield" and low risk premiums, but how? Economic theory pretty much divorces the level of interest rates from the risk premium between different securities. If anything, simple correlations go the other way: low interest rates have happened in the depths of recessions, when risk premiums are highest.

Stein knows all of this of course.

Of course, there are many caveats. Foremost among them is the fact that the ability of increases in the EBP ["excess bond premium"] to predict future economic activity may not reflect a causal link from the former to the latter. Perhaps there are economic slowdowns that are caused entirely by nonfinancial factors, and, when investors see one on the horizon, they get skittish, causing the EBP to rise. If so, it would be wrong to conclude that easy monetary policy--even if it does, in fact, cause lower risk premiums--has any causal effect on the probability of a future slowdown. So at this point, the evidence that I have reviewed can only be thought of as suggestive. 

Making progress on these difficult issues of causality will likely require a clearer articulation of the underlying mechanism that leads to such pronounced asymmetries in the relationship between credit spreads and economic activity. If a causal link is, indeed, present, what is there about it that leads increases in spreads to have a much stronger effect on the economy than decreases? I suspect that the answer has to do with something that mimics the effect of leveraged losses to financial intermediaries--and the attendant effect on credit supply. For example, GZ document that their EBP measure is closely correlated with the credit default swap spreads of broker-dealer firms. The reason could be that losses on their inventories of risky bonds erode the capital positions of these firms, which might in turn compromise their ability to provide valuable intermediation services. Alternatively, a similar mechanism may play out with open-end bond funds, whereby losses cause large outflows of assets under management, again compromising the intermediation function and aggregate credit supply.

So, if there is a correlation between the level of the short rate, the term premium and the risk premium, and a correlation between those and financial stability, it's not about fundamental business cycle risk, it's something about frictions in the intermediation system. We are awfully far from understanding that process, and especially understanding it well enough to manipulate it!

This statement also somewhat contradicts Stein's earlier view that we shouldn't watch and respond to leverage: "How, if at all, does monetary policy influence the evolution of the [leverage] ratio?" asked Stein above.  But this is awfully speculative on how monetary policy affects risk premiums, and through them financial stability.  Finally, frictions by definition don't last forever. We are talking, not about a month or two of higher rates and higher risk premiums, but about rates and premiums that last for years. Do these frictions really last for years?

Stein is duly cautions

...let me emphasize the conjectural nature of these remarks. Even if this broad way of thinking about the problem turns out to be useful, there is a ways to go--in terms of modeling and calibration--before it can be used to make quantitative statements. Thus, at this early stage, I would not want to claim that one is likely to get policy prescriptions that differ significantly from those of our standard models. We will have to do the work and see what emerges. 

But understanding all this will take years.  Do we really get to wait?  Is Stein really making speeches to spur a decade long research agenda? Given the equally tenuous theorizing on the "dove" side about the relation between low interest rates and long-term unemployment or the employment-population ratio, should it wait? How should the Fed act with so much uncertainty about basic cause and effect? I'm glad I'm not on the hot seat.

I  applaud the closing comment.  Recessions are really about risk premiums.

...one of the central and most widely shared ideas in the academic finance literature is the importance of time variation in the risk premiums (or expected returns) on a wide range of assets. At the same time, canonical macro models in the New Keynesian genre of the sort that are often used to inform monetary policy tend to exhibit little or no meaningful risk premium variation.  Even if most of the specifics of what I have had to say in this talk turn out to be off base, I have to believe that our macro models will ultimately be more useful as a guide to policy if they build on a more empirically realistic foundation with respect to the behavior of interest rates and credit spreads.