There used to be a debate over interest rate control vs. base control of monetary policy. We always knew (at least some of us always knew) that interest rate control couldn't work in theory, but it seemed to work in practice, so eventually even the die-hard defenders of base control quietened down, or were ignored, and we all became "horizontalists" in practice. Modern central banks conducted monetary policy by setting a short term interest rate, so the LM curve, at least in the short-run, was horizontal.

It is time to revisit that debate. We need to understand why interest rate control cannot work in theory, why it seemed to work in practice, and why it now no longer works in practice.

I am following the same theme as my earlier post on whether quantitative easing is trying to raise or lower interest rates, and as many of Scott Sumner's posts, like this one, and especially this comment.

Why can't interest rate control work in theory?

It has long been known that for every equilibrium of a monetary economy there exists another equilibrium (or a whole set of equilbria) in which all real variables are the same as in the first equilibrium and all nominal variables are different in the same proportion (multiplied by some scalar k). Don Patinkin's "Money, Interest and Prices" formalised this insight, but it has been known at least since David Hume's essay "On money". This insight need not rest on any particular theory of the economy, and could even be re-formulated to apply to disequilibria. It rests only on the idea that real variables, not nominal variables, are what matter. All behavioural functions are homogenous of degree zero in real variables (including the real stock of money). Monetary units should matter no more than whether we measure in metres or centimetres.

If you take this homogeneity insight, and add the assumption that the supply of money is exogenous, you get the Quantity Theory of Money (a change in the supply of money will cause an equi-proportionate change in all nominal variables), and the Neutrality of Money (a change in the supply of money will affect no real variable).

Post-Keynesian horizontalists (and we are all horizontalists now, unfortunately, because that's the underlying problem) reject the Quantity Theory because they reject the assumption that the supply of money is exogenous. But that misses the point. A revised Quantity Theory can be re-formulated taking any nominal variable as exogenous -- the price of gold, or nominal GDP futures, for example. The homogeneity insight does not depend on any definition of money supply being exogenous.

One implication of the homogeneity insight is that interest rates stay the same when all nominal variables change. This is true for both real and nominal interest rates, provided the whole time-path of present and future nominal variables changes by some scalar k. (We think of the nominal interest rate as being a nominal variable, but it isn't really, because it has the units 1/time, rather than $; instead it represents the rate of change of a $ nominal variable, just like the rate of inflation.)

If we think of interest rate control as the central bank setting the time-path of the rate of interest, the price level is indeterminate. There is nothing to stop the price level (and the stock of money, and all nominal variables) jumping onto any of the other equilibrium time-paths.

Monetary policy via interest rate control can't work in theory.

Why did it seem to work in practice?

The standard argument of why interest rate control works in practice is that prices are sticky. The economy can't jump from one nominal time-path to another. It takes time for monetary policy, seen as a gap between the interest rate set by the central bank and the neutral or natural rate of interest, to affect inflation. Provided the central bank could adjust the rate of interest more quickly than prices can adjust, the central bank can keep the price level determinate. If the actual rate is below the natural rate, and inflation rises relative to target, the bank must quickly raise the interest rate in response, and raise it more than the increase in inflation (the Taylor Principle), so that the real rate of interest rises relative to the real natural rate, to bring downward pressure on inflation. Provided the bank could respond faster than inflation, and faster than expected inflation, interest rate control could work in practice.

Interest rate control is like riding a bicycle. You can't keep the steering fixed and expect to stay upright. You need to keep moving the steering, and move it faster than your tendency to fall over, if you want to stay upright. And also like a bicycle, you need to steer left if you want to turn right. If you want higher nominal interest rates you first need to lower interest rates, so that inflation starts to rise, and expected inflation starts to rise, at which point you can raise interest rates, and raise them higher than originally, so that inflation and nominal interest rates eventually settle down at some new higher level.

That's how it was supposed to work in practice. Why did it stop working in practice?

I think the short answer is that *something* moved more quickly than central banks' response to that something. But what exactly was that *something*?

The first and obvious candidate for the *something* that moved too quickly is expected inflation. If the actual rate is above the natural rate, actual inflation will fall, and expected inflation will fall. If expected inflation falls more quickly than the central bank can lower the rate of interest, it will continue to fall, because the gap between the actual rate and the *nominal* natural rate gets wider and wider. The system is unstable.

With a credible commitment to an inflation target, that's not supposed to happen in practice. Expected inflation is supposed to remain "well-anchored" to the inflation target. But what that means is that the public trusts the central bank to move quickly enough to validate that trust. If the public believes that the central bank is too far "behind the curve" to validate those expectations, it would be irrational to continue to believe that future inflation will stay at target. And as soon as expected inflation comes adrift from the target, the bank must move even more quickly to get ahead of the curve.

There are always lags in monetary policy. It takes time for the bank to realise that the actual rate is above the natural rate, to lower the actual rate, and for the lower actual rate to stop the disinflationary pressures. If people trust that the bank will eventually get ahead of the curve, the curve stays fairly flat, and it is easier for the bank to get ahead of it. But if people lose that trust that the bank will eventually get ahead of the curve, that loss of trust makes the curve steeper, and makes it harder for the bank to get ahead of the curve. The loss of trust can become a self-fulfilling equilibrium.

Is it just a coincidence that the Federal reserve does not have an explicit inflation target, and yet is the most important central bank in the world, and at the "epicentre" of the financial crisis? It is easier for expected inflation in the US to become un-anchored, because there is no explicit anchor in the US.

A second candidate for that *something* that moved too quickly is the *real* natural rate itself. If the actual rate is above the natural rate, it is not just inflation that will be affected. Real economic activity will also fall, and will be expected to fall. When people expect future real demand for goods to be lower, investment demand falls (the accelerator), and real consumption demand falls (the multiplier), so the real natural rate also falls.

If the actual nominal rate of interest is above the natural nominal rate, the gap widens for two reasons: expected inflation falls; and the real natural rate falls.

Again, if people trust that the central bank will eventually get ahead of the curve, expectations of future real activity will remain well-anchored, just like expected inflation. But if people lose that trust, the curve gets steeper, and it becomes harder for the bank to get ahead of the curve.

But there have been many times in the past when central banks needed to lower interest rates, because they saw disinflationary pressures, indicating that the actual rate must have been above the natural rate. Why did it not always end in tears?

The financial crisis must have been the final straw that exposed the inherent instability of interest rate control. When the financial system is fragile, and bits of it break, the natural rate moves more quickly than normal, and the financial system also becomes more fragile when the actual rate is above the natural rate. Falling expected inflation, and falling expected real activity, weaken an already fragile financial system, and this lowers the natural rate by more than it would with a robust financial system.

With a strong financial system, maybe the inherent instability of interest rate control is a problem in theory, but not in practice. The curve is flat enough that people's trust that the central bank can eventually get ahead of the curve is a self-fulfilling trust. But with a weakened financial system, maybe the curve is just too steep. People lose trust that the bank can eventually get ahead of the curve, and this loss of trust makes the curve steeper still.

We knew that implementing monetary policy via an interest rate instrument was an unstable control-mechanism in theory, but it seemed to work well enough in practice, so we shrugged our shoulders and kept doing it. But it only worked because we trusted it to work. There were always two equilibria: one with validated trust in which prices were determinate and fairly stable; a second with validated mistrust in which prices were indeterminate. We were lucky to be living in the first equilibrium. The financial crisis eliminated that first equilibrium. It will be hard to go back to it, even when the financial crisis is over.

We need to switch to some other instrument for monetary policy, one which does not cause that inherent instability. The homogeneity insight tells us that any nominal variable (anything with $ in the units) will work in principle, and make the price level determinate. But some will work better than others. I am not advocating a return to base control, or the gold standard. Though each of these would make the price level determinate, the equilibrium real money base, and real price of gold, are too variable. I've written a long enough post, so I am going to leave that second question unanswered.

But whatever nominal variable we choose for central banks to buy and sell and control the price of, we could use immediately for quantitative easing, as well as permanently for the new monetary regime. Zero nominal interest rates do not mean that monetary policy is powerless; they just mean we were holding onto the wrong monetary lever. That damned lever has a very peculiar connection to the price level, and in the long run has no connection at all.

Adam P: I agree with all your points about calvo pricing. But I want both a forward and a backward-looking term in the PC. And you can get that backward-looking term in (introduce inflation inertia) if you drop the Calvo assumption that the firms changing prices today are a random and hence representative sample of all firms. Rather, the sample of firms changing prices today will be biased towards the firms with the oldest prices. But that biased sample makes the maths hard.

Posted by: Nick Rowe | April 21, 2009 at 03:17 PM

Agreed that we need a backward-looking term as well. I was thinking that it's the firms whose current prices are the farthest from optimal, many of these would be firms with the oldest prices but some would be firms who mis-judged future inflation or demand for their goods in the past.

I still have a lot of sympathy for the old Lucas story about noisy signals...

Posted by: Adam P | April 21, 2009 at 03:50 PM

A little late to the party I guess ...

Seems to me that inflation expectation being positive and well anchored is a big part of what makes prices sticky. I suppose that's probably obvious.

But if an asset bubble inflates and pops suddenly then inflation expectation would become unhinged or even negative (at least for the asset class in which there was a bubble) and prices would become fluid. That is, everyone suddenly realizes that there was a bubble, and prices are too high and must adjust down. If the bubble was really big, and the pop violent, there might even be a discontinuous gap down in prices.

If housing and the effect of interest rates on housing is an important part of how monetary policy is transmitted to the real economy, then wouldn't it be enough for inflation expectation as it relates to housing to become unhinged/negative/indeterminate to pretty well sabotage interest rates as a control mechanism for the economy? And wouldn't the popping of the housing bubble create first indeterminate expectation (at the very top), then deflationary expectation (on the way down), then indeterminate expectation (at the bottom), and finally positive expectation (on the way back up)?

Posted by: Patrick | April 24, 2009 at 07:03 PM

Much better late than never! Busy? It's good to see you at the party.

I don't think it's obvious that well anchored expectations make prices more sticky. Depends on your model of price adjustment, and if by more "sticky" you mean move less in some absolute sense, or mean move less relative to the equilibrium price.

I hadn't been thinking about real asset prices, like houses, or stocks. I should have been. I think they are an important part of the transmission mechanism. Tobin's Q, if you like.

Your conjectures sound about right, I think.

Posted by: Nick Rowe | April 24, 2009 at 07:48 PM

"Much better late than never! Busy? It's good to see you at the party."

I have to read these "wonkish" posts several times very carefully and go look up the stuff I don't understand before I say anything. Sometimes that takes a while.

I admit that I have to think about the price stickiness and expectation thing some more. I have an intuitive sense that expectation and stickiness should be related, but maybe I'm not understanding something ...

Posted by: Patrick | April 24, 2009 at 11:01 PM

"I have to read these "wonkish" posts several times very carefully and go look up the stuff I don't understand before I say anything. Sometimes that takes a while."

I should probably do the same, before posting them ;-)

Define p* as the frictionless equilibrium price, i.e. the price firms would set if they adjusted price continuously and had no menu costs etc.

If expectations fluctuate more (for some exogenous reason, demand and/or supply will fluctuate more, and so p* will fluctuate more.

So when firms did change their prices, they would tend to change them by a bigger amount, since they would tend to be further away from p*.

And, they would also tend to change their prices more frequently.

That's what my intuition tells me.

Posted by: Nick Rowe | April 25, 2009 at 05:19 AM