Everything that needs to be said has already been said.
But since no one was listening, everything must be said again.

Andre Gide


Saturday, October 31, 2015

NeoFisherism without rational expectations

Is the Fed's "zero-interest-rate policy" (ZIRP) inflationary or deflationary? You'd think that macroeconomists would have a straight answer for such a simple question. But we don't. As usual, the answer seems to depend on things.

Someone once joked that an economist is someone who sees something work in practice and then asks whether it might work in theory. Well, appealing to the evidence is not much help here either. We have examples like that of Volcker temporarily raising rates (by lowering the money supply growth rate) to lower inflation in the early 1980s. But then we have the present counterexample of ZIRP, which seems to be having little effect in raising inflation. Indeed, the Fed has consistently missed its 2% inflation target from below for years now (see Understanding Lowflation).

Some economists suggest that there are theoretical reasons to support the notion that ZIRP is deflationary. The proposition that targeting a nominal interest rate at a low (high) level results in low (high) inflation is known as "NeoFisherism." The idea goes back at least to Benhabib, Schmitt-Grohe and Uribe (2001) in their The Perils of Taylor Rules. The idea has been taken seriously in policy circles. My boss, St. Louis Fed president Jim Bullard wrote about it here in 2010. You can read all about the recent controversy here: Understanding the NeoFisherite Rebellion.

The basic idea behind NeoFisherism is the Fisher equation:

FE1: Nominal interest rate = real interest rate + expected inflation.

One interpretation of the Fisher equation is that it is a no-arbitrage-condition that equates the real rate of return on a nominal bond with the real rate of return on a real (e.g., TIPS) bond. FE1 implicitly assumes that the risk and liquidity characteristics of the nominal and real bond are identical. Steve Williamson and I consider a model where the nominal bond (potentially) carries a liquidity premium, in which case FE1 becomes:

FE2: Nominal interest rate + liquidity premium = real interest rate + expected inflation.

I'm not aware of any economist that disputes the logic underlying (some version of) the Fisher equation. The controversy lies elsewhere. But before going there, let me describe the way things are supposed to work in neoclassical theory.

Start in a steady-state equilibrium where FE1 holds. Now consider a surprise permanent increase in the nominal interest rate. What happens? Well, a higher nominal interest rate increases the opportunity cost of holding money, so people want to economize on their money balances. However, because someone must hold the outstanding stock of money, a "hot potato effect" implies that the equilibrium inflation rate must rise (the real rate of return on money must fall). In the new steady-state equilibrium, real money balances are lower (the price-level and the inflation rate are higher than they would have been prior to the policy shock). If people have rational expectations, then absent any friction, inflation expectations jump up along with actual inflation. If there are nominal rigidities, then inflation may (or may not) decline for a while following the shock, but in the long-run, higher interest rate policy leads to higher inflation. [Aside: my own view is that a supporting fiscal policy is needed for this result to transpire; see here: A Dirty Little Secret.]

The conventional wisdom, however, is that pegging the nominal interest rate is unstable. Suppose we begin, by some fluke, in a steady-state where FE1 holds. Now consider the same experiment but assume that people form their expectations of inflation through some adaptive process; see Howitt (1992). For example, suppose that today's inflation expectation is simply yesterday's inflation rate. Then an increase in the nominal rate must, by FE1, lead to an increase in the real interest rate. An increase in the real interest rate depresses aggregate demand today (consumer and investment goods). The surprise drop in demand leads to a surprise decline in the price-level--the inflation rate turns out to be lower than expected. Going forward, people adapt their inflation forecasts downward. But given FE1, this implies yet another increase in the real interest rate. And so on. A deflationary spiral ensues.

For those interested, refer to this more detailed discussion by John Cochrane: The Neo-Fisherian Question.

Now, the thought occurred to me: what if we replace the assumption of rational expectations in my model with Williamson (cited above) with a form of adaptive expectations? What would happen if we performed the same experiment but, beginning in a steady-state where there is an "asset shortage" so that the FE2 version of the Fisher equation holds. My back-of-the-envelope calculations suggest the following.

First, because expected inflation is fixed in the period the nominal interest rate is raised, FE2 suggests that either the real interest rate rises, or the liquidity premium falls, or both. In our model, there is substitution out of the cash good into the credit good. But because there is a cash-in-advance constraint on the cash good, i.e., p(t)c(t) = M(t), it follows that a decline in the demand for c(t) corresponds to a decline in the demand for real money balances--that is, the price-level p(t) must jump up unexpectedly (for a given money supply M(t)).

Now, given the surprise jump in the price-level, an adaptive expectations rule will adjust the expected rate of inflation upward. What happens next depends critically on the properties of the assumed learning rule, the policy rule, etc. For my purpose here, I make the following assumptions. Suppose that the economy remains in the "asset shortage" scenario and assume that the government fixes the money growth rate at exactly the new, higher, adaptively-formed, inflation expectation. In this case, the economy reaches a new steady-state with a higher interest rate, an arbitrarily higher inflation rate, and an arbitrarily lower liquidity premium (conditional on the liquidity premium remaining strictly positive). [Note: by arbitrary what I mean is that the new inflation rate is determined, under my maintained assumptions, by the initial surprise increase in inflation, which may lie anywhere within a range such that the liquidity premium remains positive. In the absence of a liquidity premium, the inflation rate would rise one-for-one with the nominal interest rate.]

I hope I made my point clear enough. The claim that increasing the nominal interest rate leads to higher inflation does not depend on rational expectations as is commonly claimed. A simple adaptive rule could lead to higher inflation expectations. The key is whether the price-level impact of increasing the nominal interest rate is positive or negative. If it's positive, then people will revise their adaptive expectations upward and, depending on the learning rule and policy reaction, the ensuing inflation dynamic could play itself out in the form of permanently higher inflation. The NeoFisherite proposition is possible even if people do not have rational expectations.

Postscript Nov. 01, 2015
Tony Yates suggests the cost channel in Ravenna and Walsh (2006)  together with least-squares learning might deliver the same result. It's a good idea. Somebody should try to work it out!