John B. Taylor is one of the profession’s most recognized macroeconomists, and for good reason. He has made numerous contributions to theories on wage and price formation and monetary policy. Many concepts are so central that they carry his name. “Taylor contracts” (staggered nominal wage or price contracts that are a central ingredient in many macroeconomics models), “Taylor curves” (curves that simply illustrate the feasible monetary policy trade offs), and, of course, the “Taylor Rule”, which is a specification of a nominal interest rate rule for a central bank.

Originally mentioned in a 1993 paper, Taylor showed that the simple rule—that recommends that the nominal interest rate adjust to inflation and output deviations from trend—tracked actual policy setting by the Fed for the previous six years quite well. This had an enormous influence on subsequent monetary policy theory where a Taylor rule is often used as a simple way of illustrating monetary policy conduct. Also, tons of empirical papers have examined the empirical relevance of Taylor rules in the US and everywhere else.

Now the rule has entered into a new dimension: the legislative sphere. Republicans want to pass a bill where the Fed is required to specify the rule it follows. John Taylor (an open advisor for republicans) has, not surprisingly, supported the idea. Such “Taylor legislation” is, however, slightly milder than one would believe, as it should instruct the Fed to follow a rule of its own liking; not necessarily the Taylor rule per se.

Following this, Krugman and Taylor have had their usual blog shoot-out concerning the cons and pros of such legislation. Most disagreement is on whether the Taylor rule was followed or not before the Great Recession, or whether one is talking about the correct Taylor rule, or whether it was or was not (in part) responsible for the crisis. (Tony Yates points out some more academic and practically based views against legislation that are overlapping with mine in many respects.)

Such disagreement in itself shows the danger of committing to even a particular simple rule. Two leading economists can’t even make head or tails about whether the rule has actually been followed or not. And even though Taylor has emphasized that the Taylor Rule should be seem as a normative prescription (and not just a vehicle for describing policy behavior in practice), his original paper is full of qualifiers that states that one should only see such a rule as a benchmark, and that deviations from the rule would be a good thing under various circumstances. And that makes a LOT of sense, showing that for practical purposes a “rule” is a misnomer and never a good thing (unless it can be made contingent on all foreseeable and non-foreseeable events).

But wait a minute. Hasn’t a long literature following Kydland and Prescott (1977), pointed to the desirability of rules over discretion? Would it not be better to “tie the hands” of policymakers? The answer is yes, but one must also remember (and understand) that the “rules” Kydland and Prescott talked about, were those that solved time-inconsistency problems (credibility problems) of optimal policy. I.e., they are policy prescriptions that are optimal and do not succumb to temptations to deviate to reap short-sighted gains. I.e., they are a description of *optimal commitment behavior*. A Taylor rule *does not* belong to that family in most of the models in which it is used. So yes, it is a rule in a semantic sense, in the sense of describing a given simple behavioral pattern to follow, but it is not a rule in the Kydland and Prescott meaning. So forget about rules vs. discretion arguments.

If one wants to raise the bar for legislation in monetary policymaking, one should acknowledge that policymakers act discretionary, i.e., on a day-by- day basis trying to achieve its *goals. *(Janet L. Yellen’s comments on the FOMC press conference December 17, 2014, has discretion written all over it; see p. 12). Therefore one must provide them with clear goals. There is a vast literature on delegation of monetary policymaking that shows how intermediate goals can bring discretionary policymaking closer to the Kydland/Prescott-style commitment-style behavior. Price level targeting or nominal income growth targeting are some of the alternatives. These are alternatives that can work well, since reaching these goals imply that the policymaker takes the past into account. This is beneficial if one believes that credibility is important: When you respond to past events, you can influence future expectations better to your current advantage (because what you do today will affect future policy and thus expectations).

In view of the above, the Taylor rule raises a conundrum: In environments where expectations are unimportant and credibility problems are absent, rendering the rules vs. discretion distinction irrelevant, a Taylor-type rule can be optimal (cf. Svensson, 1997). On the other hand, if expectations are important, the Taylor rule is rarely optimal. It is not a behavioral pattern that solves credibility problems, just because it has the word “rule” attached.

Actually, one of its celebrated implications, the “Taylor principle” (another term that carries Taylor’s name), should not necessarily be observed in optimal and credible policymaking. The Taylor principle states that the coefficient on inflation in the Taylor rule should exceed one, such that inflationary pressures lead to a more than one-for-one increase in the nominal rate, thereby increasing the real interest rate. But if a central bank enjoys credibility and can manipulate expectations to its advantage, it can obtain stable inflation with rather small fluctuations in the nominal rate. Indeed one could then in practice observe a negative coefficient on inflation. I have shown elsewhere that one can estimate Taylor rules with such properties in model economies where the policymaker is performing as good as it gets. On the other hand ,Taylor rules satisfying the Taylor principle could be an indication of discretionary policymaking. To reiterate: “Commitment versus discretion” is not the same as “Taylor rule versus discretion”.

It is not a good idea to require that a central bank should publish a particular behavioral rule. If it should work, it would be way too complicated to specify, and if it should be simple (as the Taylor rule), policy is just being constrained in an arbitrary way. I don’t think there is a meaningful middle way. It would just lead to endless political debates on whether the “rule” has been violated or not, and if so why. And whether actual policy in the past, good times or bad times, can be characterized by a Taylor rule is rather irrelevant for this type of legislation. The US now, may not be as it was in 1987-92.

As in all branches of economics, good policy design is about shaping incentives appropriately. Requiring by law any decision making agent to directly act in a particular way is usually not a good idea. If it was, then why not solve unemployment problems by requiring all firms to adhere to a hiring rule that secures full employment?

Did Cochrane not show that according to the New-Keynesian theory you can not find the coefficient by a regression http://www.jstor.org/stable/10.1086/660817?seq=1#page_scan_tab_contents

Yeah, he showed that identification is extremely dodgy is such models. In his online appendix to said 2011 JPE paper, he is kind enough to refer to my 2002 wp (my 2011 is based on the last part of that wp). In my 2011 paper, identification is not a problem by construction, but still one can’t make sensible inference about preferences, determinacy etc. from estimated Taylor-type rules. Discretion versus commitment makes a big difference though. The Taylor principle is typically not fulfilled under commitment, whereas it sometimes is under discretion.

Thanks for answering. You achieve determinacy by an optimizing central bank. So your paper doesnt make any suggestion on how to make monetary policy. The Neo Fisher solution that higher rates actually raise GDP and inflation is one of many solutions.

The purpose of the paper was to demonstrate the impossibility of inferring anything from estimated interest-rate rules. It doesn’t, however, rule out a characterization of the optimal policy within the given model, which I indeed perform in the paper. I am not sure how that is not a suggestion on how to make monetary policy.

Okay, correct me if I am wrong. Your optimal monetary policy depends on structural parameters and whether it is rules or discretion. So under some parameters and paradigms the neo-fisher solution where higher rates raise inflation and output is the optimal monetary policy. One of your estimation on commitment data could maybe look like a Ne0-Fisher solution.

Yes, under some circumstances there would be a positive correlation between inflation and nominal interest rates. But typically a negative one under commitment. Anyway, I am a bit in doubt about what you mean by “solution”. Is it a descriptive feature of a certain monetary policy measure in a certain model, or is it a normative thing?

I don’t know if solution is the right word. What I just want to know if the central bank can stimulate the economy by rising the rates under some assumptions in your paper. So is it fair to say, that the central bank can choose by its preferences and if it wants to operate under rules or discretion(assuming it can choose that), whether it wants a positive or negative correlations between nominal rates and inflation. Or does it depend on other variables?

If that is correct then then central bank can easy eliminate the zero rate lower bound by switching to a policy where there is a positive relation between nominal rates and inflation. however, there might be some credibility issues by such an action.

No that would not be possible. The point is that an optimal interest rate value can be expressed by infinitely many equilibrium combinations of the interest rate and current and expected future inflation. Some implying a positive, some a negative correlation between variables. This illustrates that what one may find in the data, e.g. a negative coefficient on inflation, says nothing about indeterminacy, as many believe following e.g. Clarida et al 2000, QJE.

Importantly though, the particular optimal interest rate can in principle be negative (and thus unattainable). So even if one could choose one of the infinite combinations (which IS sort of meaningless in itself), it will not help one to overcome the ZLB.

I don’t understand that. An equilibrium where there is a positive relation between nominal interest rate and inflation can be determinate, that are what people like Cochrane, Williamson and others have shown. So in your paper the sign of the correlation between inflation and the nominal interest rate is due to the central banks preferences and if it wants to operate under rules or discretion. So in principle if the central bank changes these things it could change the sign on the correlation between inflation and nominal interest rates. However, it requires that the central bank can credible do so.

The ZLB is not a problem if you have a positive relations between nominal interest rates and inflation because there are no limit on the nominal interest rates and therefore no limits on how much the central bank can stimulate the economy.

By the way it is a cool blog you have, I just wish I had notice it before. The Danish public economic debate needs intelligent voices, in which it currently haves few of, but then maybe a blog in Danish might be nice.

In the model of the paper, the optimal nominal interest rate can be negative. Both under commitment and discretion. This fact

cannot be escapedby choosing a certain sign on the correlation between the nominal interest rate and macroeconomic variables.To repeat, the purpose of my paper was to show that an observed correlation in data (or, say, an observed Taylor principle or not), does not say anything about whether the data is representing a determinate or indeterminate equilibrium.

I know I take the paper further than you will and what it probably should be. I am trying to understand what your paper says about what the sign of monetary policy is, because I think it actually says something. If you don’t think that your paper can say anything about it, then many of my questions were irrelevant. I hope this make it clear what my questions are about, even though my English is a bit sloppy.

As said, any sign. And observed signs: No info about stability properties. Also, no chance for getting over the ZLB. If you think a higher nominal interest rate is expansive if there is a positive correlation between the rate and inflation, then great: The ZLB will not be a problem in recessions. But then it would be a problem in sufficiently large booms, where you supposedly should lower the interest rate. One cannot change signs of the transmission mechanism at will, and I think that issue is more or less orthogonal to the observed equilibrium correlations in question.

What I was considering was that maybe you could change things like central bank preferences(by hiring a different chairman) and whether to operate under rules or discretionary, and these things might change the sign on monetary policy. I was just curious about if this could be a theoretical solution, I doubt too that this could work in practice.

You are right about ZLB, I was just to used to thinking about our current recession than to consider a boom.