Instrument rules, target rules, NGDP, complexity and learning

At one extreme you have pure discretion. The central bank does whatever it thinks is best.

At the other extreme you have an instrument rule. The rule specifies exactly how the central bank should set the monetary policy instrument, conditional on the indicators (i.e. conditional on its information). The Taylor Rule is an example of an instrument rule*; it specifies the exact setting of the interest rate as a function of recent data on output and inflation.

Somewhere in the middle you have a target rule. The central bank announces a commitment to a target, but uses its discretion on how to set the instrument, given all the information provided by the indicators, to hit that target. All inflation targeting central banks, AFAIK, fall into this category.

This post is not about pure discretion; it's about instrument rules vs target rules.

This post is also, in part, a response to John Taylor's post on NGDP targeting. I stress the "in part" bit, because I don't plan to respond to everything he said. Mainly, I just want to make some conceptual distinctions.

The choice between NGDP targeting and (say) inflation targeting is a quite separate question from the choice between an instrument rule and a target rule. Central banks do not directly control either NGDP or inflation. All central banks really control is their own balance sheet, plus they control their own communications about how they will adjust that balance sheet, conditional on the indicators they observe, to try to hit some target.

For example, you could believe any one of these:

1. That inflation targeting is best, and that you can best target inflation by following a Taylor Rule.

2. That inflation targeting is best, and that you can best target inflation by letting the Bank use its discretion. (That's what the Bank of Canada believes).

3. That NGDP targeting is best, and that you can best target NGDP by following some instrument rule. Something roughly analogous to the Taylor Rule would be to adjust the nominal interest rate by 1.5 percentage points for every one percentage point NGDP (or its growth rate) rises above target.

4. That NGDP targeting is best, and that you can best target NGDP by letting the Bank use its discretion.

There are two questions: inflation vs NGDP as the target variable; instrument rules vs target rules. That's leaving aside the other questions, like levels vs rates of change (price level vs inflation, and NGDP level vs growth rate), and the precise numerical target for the target variable.

Which is best: an instrument rule or a target rule?

If you knew the model of the economy, the choice would be moot. You can solve the model for the instrument rule that best implements the target rule, so the Bank might as well just commit itself to following that instrument rule. (In fact, it seems to me to be conceptually impossible to build an economic model to demonstrate that a target rule is better than the best instrument rule in that model; it's a self-contradictory undertaking).

Even if you didn't know the model of the economy, the Bank still has to figure out its best estimate of how to set the instrument conditional on the indicators to best implement any target rule. So why not just formalise that best estimate into an instrument rule, since that's what the Bank will be doing anyway?

I can think of two answers:

1. Complexity. There might be hundreds of indicators the Bank might look at and respond to. It might be impossible in practice for the Bank to write down an exhaustive list of all the possible states of the world and how it would set the instrument in that state of the world. I couldn't even do that for describing how I would adjust the steering wheel of my car when I drive to work every morning. Even if the Bank could do that, nobody would understand it.

2. Learning. The Bank of Canada has been targeting inflation with (almost) the same target for 20 years (it lowered the target inflation rate in the first couple of years). But it has learned a lot about how to target inflation in that 20 years (at least, I hope its researchers haven't been wasting their time). If it wrote down its implied instrument rule today, it might be a very different implied instrument rule than the one it would have written down 20 years ago.

Any instrument rule would be constrained to be a simple rule, and would be constrained from learning. The only advantage I can see is if you don't trust the central bank to follow the target rule, and want to monitor the Bank much more closely by requiring it to specify an exact instrument rule, so you can tell immediately if it has broken the rule.

Sure, you can always allow the Bank to update the instrument rule when it learns something. But that would seem to weaken any advantage an instrument rule might have. You can't predict in advance what the Bank will learn, and you would have to decide whether to trust the Bank when it said it needed to change the instrument rule because it had learned something new.

Now, there is one way to avoid this trade-off between a simple static instrument rule vs a target rule where you have to trust the Bank. That's by using the method that Scott Sumner (and others) have advocated. You use market forecasts of the target variable as the Bank's sole indicator.

The idea is that the Bank adjusts the instrument (whatever it is) by some fixed rule in response to the implied prediction of the target variable (whatever it is) made by some futures market. Suppose, for example, we kept the same 2% inflation target, but required the Bank of Canada to adjust the overnight rate by some fixed amount proportional to the deviation between the two year ahead market forecast of inflation and the 2% target.

We still get an instrument rule, but it's now simple and does not incorporate learning. It's the prediction market that handles the complexity. It's the prediction market that does the learning.

*Strictly speaking, the Taylor Rule isn't an instrument rule. It contains at least one unobserved variable, potential GDP (or the gap between actual and potential output). We don't observe potential GDP. The Bank has to guesstimate it. Which means the Taylor Rule still relies on discretion.

Comments

[Becky Hargrove]

I remember a discussion with Scott about this in which I saw the choice as between automatic pilot for the plane, or the ability to manually fly with the automatic pilot as option. He said that we couldn't have both. What have I missed?

[Nick Rowe]

Becky:

Pure discretion: the pilot flies where he wants to.

Target rule. The pilot flies to a fixed destination.

Instrument rule: automatic pilot to a fixed destination.

I think you could have automatic pilot with a manual override for emergencies.

[David Pearson]

Markets can undermine an NGDP target rule when unemployment is high. Actors go long NGDP futures and commodities, short the currency. They force up NGDP even while real growth is still lackluster (the benefit from the tradeables sector growth is offset by declining consumer confidence). This forces the Fed to hit the "manual override" to prevent a tightening in the presence of high unemployment. They rationalize it as allowing a "temporary" NGDP overshoot caused by "transient" headline inflation. Markets attack more once they see the Fed will not defend its target, and the inflation dynamic becomes self-reinforcing. Eventually, the Fed abandons its previous level target altogether, and speculators reap a windfall.

[Luis H Arroyo]

Discretion is innevitable. See the last Minutes of Sweden Bank
http://www.riksbank.com/upload/Dokument_riksbank/Kat_publicerat/PPP/2011/ppp111026e.pdf
and the discussion betwen Sevensson and others about the accuracy of some of the key instrument: potencial GDP, gap, prices...
Sevensson defends a lower interest rate than the current 2%, because He thinks NAIRU is not 6,5%, but 5,5%...
Definitevily, Precision is completly pretentious

[Bill Woolsey]

Pearson:

I don't think this makes any sense.

For what it is worth, a nominal GDP target doesn't prevent tightening when the unemployment rate is rising. That is what happens if "cost push" inflation starts. Real expenditure and real output (and employment) are all pushed below trend in proportional to any increase in the level of money prices and wages above trend. It is like a single market with unit elastic demand.

I am not exactly sure how a long position on futures contracts would be hedged by going short on the currency. Is that why you combine the two transactions?

I am entirely lost regarding consumer confidence.

Confidence in what?

[Kosta]

There are also political considerations. An instrument rule protects the independence of the central bank by allowing them to more explicitly justify their policy choices. Of course, this comes at a cost of having much less flexibility in implementing policy, as the central bank is merely reacting to changes in the instrument it is following. But once the instrument rule becomes accepted, very few people could claim that the central bank is playing politics when it changes policy in response to a change in the instrument.

[JKH]

This is a very good post.

But I think in addition you have to explode the instrument rule dimension to examine the binary cases of non-zero and zero nominal interest rates - corresponding to an interest rate instrument rule and a non-interest rate instrument rule (e.g. QE, base expansion, whatever).

Moreover, as I recall Scott Sumner's original formulation of the futures market mechanism, he proposes cash market OMO as the response to futures market signals whatever the level status of interest rates - i.e. a non interest rate instrument rule in all cases - which I think is a problem (e.g. changing a non zero policy rate requires only miniscule reserve changes at the margin, which are subsequently reversed). I think it’s a problem even for a zero bound instrument rule. But that's just me.

So I think the “degree of difficulty” of these sorts of connections as you propose is much different at positive rates versus the zero bound.

[David Pearson]

Woolsey,
In the scenario, actors bid up inflation hedges: NGDP futures, commodities, FX. The result is higher headline inflation, potentially to the point of an NGDP overshoot. The spike in headline inflation creates a real wage "shock": consumer confidence in their future real incomes falls, and so does real spending on discretionary items. The result: high inflation, slow real growth, and (continued) high unemployment.

In the above scenario, the Fed has two choices: 1) enforce the level target by selling NGDP futures to contract bank reserves; or 2) hit the "manual override" switch and allow a "temporary" overshoot. My argument is that the Fed will not tighten with unemployment at 9%. Instead, they will opt for "2)", which will only embolden speculators to take the overshoot further. As NGDP gets further and further from target, the tightening needed to get back to target increases, and so the likelihood that the Fed will tighten falls furhter. For markets, this sets up a self-reinforcing dynamic that eventually forces the Fed to abandon their target.

You seem to assume the Fed would prioritize its NGDP target and therefore tighten (sell NGDP futures) with 9% unemployment. I'm willing to assign some probability to that outcome: perhaps 30%.

[Scott Sumner]

David, Why wouldn't central banks tighten with unemployment at 9%? European central banks have done that many times in recent decades. If the natural rate is that high, it doesn't take long for the central bank to figure that out.

Nick, Nice post, but I don't see where the Taylor rule is a form of inflation targeting, it seems much more similar to NGDP targeting. You have a coefficient of 1/2 on both inflation and output deviations.

[Reverend Moon]

@Scott Sumner

"If the natural rate is that high"

Can you please explain what you mean by the natural rate of interest? It makes about as much sense to me as saying the "natural" room temperature or "natural" speed limit.

[Bill Woolsey]

Pearson:

It makes less sense now.

If the "rule" is that the central bank must always seek a zero balance on the contract, then ceteris paribus, those believing that nominal GDP will be allowed to go above target will go long on the contract. Breaking the rule would be for the central bank to take the opposite position so that nominal GDP remains on target. Holding to the rule would result in nominal GDP falling below target. Those who speculated on breaking the rule lose money to those who speculated that the rule would be kept. If the central bank broke the zero net position rule to some degree and not enough for nominal GDP to remain on target, then it would share in those profits.

If expected inflation has effects on price setting, which could include a lower exchange rate and so higher demand for import competing and export goods, purchases of "commodities," which raises the quantity and prices of intermediate goods inventories, then yes, this would make going long on the contract sensible, which would result in slowing real expenditure. This is needed to keep nominal GDP on target.

This is what happens when people just set prices higher in aggregate. Real expenditures and real output fall in proportion. That is what prevents a wage/price spiral with nominal GDP targeting.

If, on the other hand, the central bank begins to target the unemployment rate, and particular begins to accomodate inflation so that there will be no downward pressure on real expenditure, then such a policy is inconsistent with nominal GDP targeting. It is also inconsistent with inflation targeting or price level targeting.

[Reverend Moon]

@Scott Sumner

Never mind. It makes sense now that I see that you are using "natural" as a noun.

Rim shot

[David Pearson]

Woolsey,
If the Fed does not defend the target to keep inflation down, this is tantamount to targeting unemployment and accomodating inflation. This is a likely outcome of NGDP targeting, IMO.

The mechanics are simple: the Fed targets NGDP of 100. Markets bid the price to 105. The Fed announces it will allow a "temporary" overshoot rather than selling futures. Markets doubt the Fed's resolve given high unemployment. They bid the price to 110. The Fed refuses to tighten in response to transient factos. They abandon the previous 100 target and adopt a new target of 110. Speculators earn $10 per contract.

[Lorenzo from Oz]

Nice: I get the point about you are not writing about discretion. Still, the Fed has been giving us an object lesson in the problem of discretion all on its lonesome: it's inability to anchor expectations. Indeed, worse, it ability to seriously misdirect expectations, as in its "surreptitious disinflation" as the GFC was hitting.

David Pearson: does not your scenario rely on bidding on futures to be ignoring the underlying economic conditions of output being significantly below trend? After all, the "Fed lacks resolve" story suggests that there is reason to think NGDP needs to be higher than it is. A major reason why the Hawke Government floated the $A in 1983 is that it forced speculators to engage in a "two-way bet" rather than a one-way bet against the set exchange rate, given common judgements about in which direction the exchange rate was problematic. With NGDP futures markets, there is a target but not a set price. So I am not convinced that that sort of "one direction" market pressure would be likely.

Also, Central Banks have shown amazing ability to ignore unemployment, both recently and in past history. Which makes the "market game-playing" story even less plausible.

[JKH]

David Pearson:

"Markets attack more once they see the Fed will not defend its target, and the inflation dynamic becomes self-reinforcing. Eventually, the Fed abandons its previous level target altogether, and speculators reap a windfall... If the Fed does not defend the target to keep inflation down, this is tantamount to targeting unemployment and accomodating inflation. This is a likely outcome of NGDP targeting, IMO."

Quite right, and a good example of what can and will go wrong with NGDPT.

What you've described amounts to explicit or implicit flexible inflation targeting, which is what central banks really want and need. It's why NGDPT will never be adopted. It's too inflexible, and too constraining.

Similar to gold as a standard always failing in the end.

(BTW, the futures market idea requires a rule within a rule - it required an OMO rule as a response to a futures market intervention requirement. The futures market intervention requirement is not itself the operational source of the cash that is injected/withdrawn in connected OMO operations. The latter is determined by a rule, which itself is a function of futures market intervention and not the intervention cash effect itself.)

[JKH]

P.S. David:

Sufficient flexibility in being able to reset the NGDPT means that the idea of the target as a relationship between two variables is itself meaningless. Why would anybody believe any particular setting for the target, in the long or short run? The true "target" in that context can only be flexibility itself.

[Too Much Fed]

"All central banks really control is their own balance sheet, plus they control their own communications about how they will adjust that balance sheet, conditional on the indicators they observe, to try to hit some target."

What about reserve requirements and capital requirements?

[Nick Rowe]

David Pearson: I'm not sure I understand your argument, but if it were correct, it would seem to be an even bigger problem with inflation targeting. Because if I understand you correctly, you are saying that the market would "test the central bank's resolve" if inflation ever rose above target if unemployment rose at the same time. But inflation targeting central banks do seem to be able to keep inflation approximately on target.

Kosta: "There are also political considerations. An instrument rule protects the independence of the central bank by allowing them to more explicitly justify their policy choices."

OK. I would express it slightly differently. An instrument rule doesn't really give the central bank any choices.

JKH: Thanks! Yep. An instrument rule that used an interest rate instrument would need a supplement at the ZLB.

Scott: "Nick, Nice post, but I don't see where the Taylor rule is a form of inflation targeting, it seems much more similar to NGDP targeting. You have a coefficient of 1/2 on both inflation and output deviations."

Thanks! Hmmmm. At the Bank of Canada, they always thought of the Taylor Rule as an alternative way to implement inflation targeting, IIRC. It's not obvious to me whether the Taylor Rule would be closer to implementing inflation targeting or NGDP targeting (levels or growth rate?). But the Taylor Rule has: a coefficient of 1.5 on inflation, compared to only 0.5 on GPD; and it's GDP gap, not GDP; and it's a mix of *level* on GDP and *growth rate* of prices. So I think the Taylor Rule is closer to inflation targeting than NGDP targeting.

Rev Moon: Yep, "Natural rate of interest/unemployment" is more like a name, than adjective + noun. It's misleading terminology, because it doesn't mean what it seems to mean, but we are stuck with it for historical reasons.

Lorenzo: yep, we thought the Fed had an implicit inflation target, but now it doesn't seem to. Now it seems like pure discretion.

JKH on David: I don't get it. If the Bank was likely to deviate from and NGDP target if unemployment rose too high, wouldn't it be even more likely to deviate form an inflation target? Because an NGDP target at least gives half the weight to real GDP, which is correlated with employment.

TMF: fair point. Some central banks (China) also use reserve requirements as an additional instrument of monetary policy. I see capital requirements as more a regulation for financial stability though.

[jmg]

I think the crucial question for implementing an instrument rule is:
What happens when the structure of our models (reality) changes?
"Instrument rule: automatic pilot to a fixed destination."
You have the same autopilot, but the landscape changes.
Then , if you want to fly to New York (fixed destination), you may arrive in Chicago.

[JKH]

Nick,

NGDPT to be good must be good for all time.

We know from history that NGDP can be volatile on the upside as well.

Suppose NGDPT = 5 per cent.

Suppose it spikes to 9 per cent in 2020, with 6 per cent real and 3 per cent inflation.

This can easily happen because there is absolutely no way that the economy behaves in such a way as to “conform” to an NGDP target in all circumstances. The real economy does what it does. And this can happen in 2020 whether the T is level or growth at the time.

What does the CB do in response?

What does it communicate about the timing of its plans to bring NGDP back to 5 per cent?

How much tightening does it do, and how quickly?

Does the market incorporate an NGDP term structure (whether through expectations or an NGDP futures curve) that is inverted as a result?

All of this implies judgement on the part of the CB that becomes very awkward to communicate let alone implement. It can more easily communicate and manage under a flexible inflation target.

I think the joint volatility of two variables is hopelessly awkward for the CB to attempt to manage as a target and it’s better to focus on the pivotal one for monetary policy.

NGDP should be an indicator; not the target.

[David Pearson]

Nick,
Inflation targeting might work better because it does not allow for an inflation dynamic to spark. Max inflation matters, and in the NGDP case, you could have a situation, as now, when the rule would require/allow quite-high inflation to return to target. This could spark a self-reinforcing dynamic that you don't see at 2%.

But yes, the BOE has abandoned its inflation target in the same way I predict for a U.S. NGDP target: allowing a "temporary" overshoot that has lasted far longer than they planned.

[Scott Sumner]

Nick, The coefficient on both inflation and output is 1/2, for changes in the real interest rate (which is what matters in NK models.) So they are treated equally, aren't they?

I'm not saying they are identical---RGDP is different from the RGDP gap. But in practice the Taylor Rule seems much more similar to NGDP than to inflation targeting.

[GeorgeYuri]

Sorry to bud in:

«Les banques en difficulté doivent être liquidées !»

http://trends.levif.be/economie/actualite/banque-et-finance/les-banques-en-difficulte-doivent-etre-liquidees/article-4000005302457.htm

mardi 15 novembre 2011 à 06h38

Le processus décisionnel politique aux Etats-Unis et en Europe a été détourné par les banquiers, dénonce l’économiste Frank Allen : «Pour obtenir de la discipline de la part des banques, il faut leur faire savoir que, si elles font preuve d'imprudence, elles seront liquidées !»

[Alex]

Isn't there a need for discussion of Goodhart's Law here? Goodhartism would be strongly opposed to an instrument rule. (and even more so to hooking it to a futures market - a better example of "to control is to distort" would be hard to find).

[Bill Woolsey]

I think most market monetarists believe that the flow of spending on output is a real thing. For example, policy today determines the flow of spending on output next period.

Given that flow of spending, if firms raise prices, then they will be able to sell less output, and so will produce less.

If firms want to produce and sell more, then they must lower prices.

The critics, on the other hand, treat nominal GDP as the product of output and the price level--two things that are determined independently. For example, policy determines output next period, and output determines prices the following period. There is a "shock" to the price level this period, say, an increase, and it has no impact on what firms can sell this period.

What firms can sell really plays no role in the model. In the model, policy determines output in the next period, not spending and so what firms can sell. Output determines prices, not to close off shortages or surpluses--impacting quanties demanded and what people will buy and so firms can sell. It is just output gaps cause inflation.

[edeast]

bill: that helps my brain.

[Too Much Fed]

Can I have more than one(1) comment?