Created 3/5/1998
Go to DeLong's Home Page

Comment on Rich Clarida, Jordi Gali, and Mark Gertler, "Monetary Policy Rules in Practice: Some International Evidence"

J. Bradford DeLong

I like this paper. I like this paper enough to make Berkeley's macroeconomics graduate students read it and its cousins when I teach them this fall. I like this paper a lot for three reasons.

The conclusion (B) could lead to an interesting discussion of European Monetary Union, and of the likelihood that it will succeed. But I am not qualified to undertake such a discussion.

And, besides, discussions that start from the premise that a paper is excellent and flawless are boring, or are boring to everyone except the paper's authors.

So let me, instead, talk about an aspect of the paper that makes me uneasy. And the best way to approach this is to look at the first two lines of Clarida, Gali, and Gertler's table 3: Clarida, Gali, and Gertler's GMM-estimated U.S. monetary policy reaction functions for 1979:10-1994:12, and for 1982:11-1994:12.

Look at the second column, at gamma, at the response of the federal funds rate to a shortfall of production below potential. In the second line this coefficient is 0.56 with a standard error of 0.16: since the end of 1982 the Federal Reserve has responded to a shortfall of production below potential by pushing the interest rate pedal toward the floor--and not because output is a cause of lower future inflation (that effect is captured in column 1, in beta), but because the Federal Reserve does not like low output.

But look at gamma estimated over the entire sample: 0.07 with a standard error of 0.06. The central bank's response to a fall in output (independent of output's role as a forecaster of future inflation) is effectively zero, and is very tightly estimated.

Now here's the question: how can tacking an extra one-quarter of additional data points onto the front of a sample:

(a) move an estimated coefficient by three times its standard error; and

(b) tighten its standard error by 60 percent?

A small amount of very different new data can move a coefficient far, but when it does so it usually raises the coefficient's standard error significantly.

So what is going on?

I think that what is going on is that this GMM estimator is producing an equation that is implicitly part of a two-equation system: there is not just a Federal Reserve reaction function, but an inflation process in there somewere. I think that adding 1979-1982 to the sample period changes the estimated inflation process a lot--and in so doing future inflation becomes much more heavily affected by the output gap.

Thus the correlation between low output and low interest rates that in the post-1982 sample the computer sees as showing the "soft heart" of Alan Greenspan is, in the full sample, seen by the computer as the Federal Reserve's hard-hearted forecast that higher unemployment now means lower inflation later.

So: I want to see the estimated inflation process that is implicit in the GMM estimate of the reaction function. I want to see this inflation process explicitly laid out. I can't judge the reaction function without knowing whether the inflation process associated with it is reasonable.

Now which do we believe, the full sample results or the post-1982 sample? I believe the post-1982 sample. The whole point of the Lucas critique is that a big policy change will cause big changes in the inflation process. And table 3 is telling us that there is a big policy change at the end of 1982.

Thus I think that the full-sample estimated inflation process is a mongrel mixture of the inflation process from two different policy regimes, and hence is not adequate for either subperiod.

So what does believing the post-1982 results and throwing away the full sample results tell us?

It tells us that on a scale of "soft-heartedness", of gamma, the reaction of a central bank to high unemployment over and above whatever reaction follow from the fact that high unemployment is a sign of lower future inflation. On a scale of soft-heartedness where the Bank of Japan is one and the Bundesbank is 3, the post-1982 FOMC is between 7 and 10.

The authors characterize the Bank of Japan and the Bundesbank as following soft-hearted inflation targeting. In this metric--relative to the Bundesbank and the Bank of Japan--the FOMC is following bleeding-heart liberal, people-for-the-ethical-treatment-of-animals-style inflation targeting.

Now I think this conclusion is by and large correct--that the FOMC has placed a much higher weight on nudging output back to potential than the Bundesbank or the Bank of Japan. It fits narrative and anecdotal evidence as well. I remember an early 1994 conversation with my immediate boss, Assistant Secretary of the Treasury Alicia Munnell:

"Brad," she said, "the unemployment rate is now lower than our estimate of the NAIRU."

"Yep," I said.

"Brad," she said, "short-term real interest rates are negative--and have been negative for two years."

"Yep," I said.

"Brad," she said, "what is the FOMC doing?"

Well, it is very hard to argue with success. And Federal Reserve policy in the 1980s and 1990s has been an unqualified success. But Federal Reserve policy has also been substantially different from Bundesbank and Bank of Japan policy.

This paper gives us a very nice way or characterizing and quantifying this difference.

Professor of Economics J. Bradford DeLong, 601 Evans Hall, #3880
University of California at Berkeley
Berkeley, CA 94720-3880
(510) 643-4027 phone (510) 642-6615 fax


This document:

Search This Website