« "Fire-and-forget" fiscal policy | Main | Canada's recovery is 30%-40% complete »

Comments

Feed You can follow this conversation by subscribing to the comment feed for this post.

Nick: "a tighter fiscal policy will influence AD and make the natural rate of interest lower"

really? you sure? why would that be? (I'm not being coy, I don't think this is an obvious point. I am however sleep deprived/ hung over/ in a state of existential bliss after the olympic hockey tournament though so my thinking is feeling very cloudy.)

Adam P.: Man, you must have celebrated a lot last night. Or maybe I did? Anyway, (at least) one of us is misunderstanding something!

Simplest way to explain what I mean: define the natural rate of interest as the (real) interest rate at which the IS curve intersects the "full employment" line (i.e. the point on the IS curve at which the output gap is zero). Assume the IS slopes down, and that a tighter fiscal policy would shift the IS left. That means IS will intersect full employment at a lower interest rate.

Or, start with a consumption-Euler equation, evaluated where C+G=Ypotential. Have a temporary cut in G, which requires a temporary increase in C, which requires a fall in r to satisfy the Euler equation.

Nick: You raise an interesting and timely issue. Though one could argue that expectations of fiscal policy are already built indirectly into the model

John Cochrane of the Univ. of Chicago would likely be sympathetic to the idea of explicitly incorporating some indicator of fiscal policy into a Taylor Rule model. I'm not convinced; I still believe that current US fiscal inaction could lead to all kinds of other distortions and relative price movements other than higher inflation rates. John Cochrane's web-site

Incidentally, I have often have trouble telling whether you are pursuing a normative or positive approach to the problem at hand. In this case, the Taylor Rule model can be used for both: modeling central bank behaviour and providing prescriptions for central bank strategies. Clearly, as I kept reading, it became apparent that you were adopting a normative approach. Thought you might like to know. Note: I like normative approaches and believe economists should use explicit normative approaches more often. In fact, there may be nothing wrong with the way you are presenting and discussing problems; it may simply be me struggling to overcome earlier socialization. ;-)

oh, yes you're right. I wasn't thinking, I had in mind permanent changes in G that would shift around the whole consumption path, changing its level but not its growth rate.

westslope: Thanks!

I don't have a clear enough picture of John Cochrane's views in fiscal policy. If he argues that loose fiscal policy doesn't affect AD if the stock of M is fixed, because it just increases the rate of interest (and velocity is unchanged), then yes, he would want the central bank to adjust its interest rate on reserves to keep M and AD fixed. If he argues that fiscal policy doesn't shift the IS curve, then monetary policy should ignore fiscal.

But I'm really just arguing from the mainstream perspective: that both monetary and fiscal policy can affect AD.

Interesting you had trouble telling whether I was taking a positive or normative approach. I can't have been clear enough. I kept editing and re-editing, writing "ought and will", then "should and will", then just "should". I should have kept my original. I meant both normative and positive. Central banks ought to look at fiscal policy when setting interest rates; and I believe they will.

Adam P: I was trying to think through the same thing too, when writing the post (especially on the exchange rate bit). It's complicated by the fact that G has been temporarily high, and it will now (I assume) be going back to a permanently lower level. The only way I can keep my head clear is to stop thinking about temporary changes vs permanent changes. G can be decomposed into permanent G + transitory G. Transitory G is positive now, and will go back to zero again soon. So that's sort of like a "temporary" fall in G.

"I can only think of three reasons for ignoring fiscal policy: you think it won't change in the near future; you think that fiscal policy doesn't affect AD; you think that the effects of fiscal policy on AD are already captured in some other variable already included in the reaction function."

Well, what do you mean by "ignore"? If you mean, "take no notice of", then I expect that you are right. However, if you mean, "take no action in regard to monetary policy", is there not another reason? Namely that you want more aggregate demand, but monetary policy alone is not increasing it. So if fiscal policy does increase aggregate demand, you still may not do anything, until the Taylor rule (or whatever you use to decide policy) says so.

I may have disagreed with you once or twice in the past, but this time I think you're on to something. And unlike westslope, I thought that this post was one of your best-presented and most lucidly explained. And yes, it is quite possible that my two opinions are related ... ;-)

The CB takes this into account through projections of future inflation and output, by being forward looking, rather than using the original Taylor rule. This makes more sense in abstract, though how good the models are is another question.

Nick,

Three comments:

1. I follow your theoretical arguments. However, in the baseline New Keynesian models that are often used to analyze monetary policy, the natural rate of interest is only affected by technology shocks. Now, we can argue that is a flaw of the model, but that is another topic altogether.

2. You admit that if Ricardian Equivalence holds this is a moot point. I think that there are a lot of sensible reasons why it shouldn't hold. However, the evidence suggests it does hold:

http://www4.ncsu.edu/~jjseater/PDF/PublishedPapers/RicardianEquivalenceEBook.pdf

Does anybody know of a more recent survey?

3. Finally, what is the true extent of fiscal expansion? A recent NBER paper argues that "the aggregate fiscal expenditure stimulus in the United States, properly adjusted for the declining fiscal expenditure of the fifty states, was close to zero in 2009. While the Federal government stimulus prevented a net decline in aggregate fiscal expenditure, it did not stimulate the aggregate expenditure above its predicted mean."

http://www.nber.org/papers/w15784

If monetary policy is going to be concerned with fiscal policy, we had better decide how to measure fiscal expansion first.

I think you are getting closer to my point that interest rates are about currency denominated debt levels, especially in a wealth/income inequality world with a positive price inflation target.

The taylor rule does incorporate fiscal policy; it does so in the sense that fiscal policy leads to changes in output and prices.

This is "obvious" from a control theory perspective. For instance, your car's cruise control adapts to change in terrain. It does so not by directly monitoring the slope, but again by simply observing the speed and acceleration of the car.

The taylor rule is entirely backwards looking. It does not contain a predictive model of the economy. We could build such a taylor rule: it could include an explicit notation of how changes in the policy rate are "expected" to adjust inflation and prices.

Still I think this would be a much different beast.

You might if you care, look at "Smith Predictor" and "Kalman Filter"--two techniques of augmenting linear-control systems (of which the taylor rule is an example).

Min: I meant that the central bank should decide monetary policy without looking at fiscal policy only under those 3 conditions. That doesn't mean fiscal policy shouldn't have been used.

Phil: Thanks!

Lord: yes, if the BoC takes fiscal policy into account in a forward-looking reaction function, that will help. But see my reply to Jon below.

Jon: Yes, but there's a subtle difference between a Taylor Rule and Cruise Control. And that's what I was getting at in my math example. For simplicity, ignore the output gap in the Taylor Rule, so i responds only to the gap between actual and target inflation (inflation=speed, for cruise control). In the TR, the *level* of i depends on the inflation gap. In a car's CC, the *change* in the level of the gas pedal depends on the speed gap. If CC worked like a TR, the car really would go at a slower speed on uphills than downhills.

Now, in practice, I believe that central banks operate more like CC than a TR. If inflation is below target, they don't just set a low i, they *keep on (slowly) lowering* i until inflation returns to target. Since they can't observe N, the natural rate, it makes theoretical sense for them to do this. Plus, if you put a lagged i in an empirically-estimated reaction function, the coefficient seems close to 1.0 (monthly data). In other words, the alleged "interest-rate smoothing" term that econometricians find is really just a cruise control term. Damn, I ought to write a post on this: "Interest rate smoothing vs cruise control". Wonder if I can get Stephen interested?

But, in any case, even if central banks follow cruise control rather than Taylor Rules, if they look ahead (back to Lord's point) and take fiscal changes into account, they will do much better. The feedback loop speed-speedometer-gas pedal-speed in a car operates with much shorter lags than inflation-measured inflation-interest rates-inflation does in monetary policy.

Josh:
1. I think that central bankers have a much richer theory of the natural rate than their models. They know (or should know), that in a fully-specified NK model, the natural rate should depend on preferences, demographics, fiscal policy, weather, endowments, etc., as well as technology.
2. I don't now the current state of play of empirical tests of REH. But if economists actually believed REH (plus that G has no effect, which is separate from REH), then they wouldn't have advocated fiscal policy in the first place.
3. Yes, that's an interesting result about US states. Not sure if it holds for Canadian provinces though. But the important question is not whether (total) fiscal policy changed going *into* the recession, but how it will change coming *out*.

Too much Fed: There's an optimal level of government deficits. (And probably an optimal level of government debt too, though the theory on that is weak).

"But if economists actually believed REH (plus that G has no effect, which is separate from REH), then they wouldn't have advocated fiscal policy in the first place."

I am not sure what anyone really believes regarding fiscal policy. Two years ago the consensus was that fiscal policy was ineffective. In addition, nobody has seen an Old Keynesian IS-LM model published in any recent literature. Nonetheless, these are precisely the models that have been used by stimulus advocates to estimate the multiplier.

It seems to me that there are many who either don't believe the models that the modern literature is using or who rely on ideology rather than economic theory when discussing fiscal policy. I have sympathy for the former point, however, using Old Keynesian models is going in the wrong direction.

Nick, Very good post. I was going to talk about how forward-looking Taylor Rules solve the problem, but Lord beat me to it.

Nick,

While you are at it, maybe you could provide some indication of how total indebtedness in the economy affects the natural rate? Is there a level of total credit to GDP that affects it? 350% (the current level in the U.S.)? 500%? 1000%?

A lower deficit implies either lower private savings or a higher trade surplus (from an accounting identity standpoint). Clearly, the level of indebtedness has some influence on the ability of consumers to reduce savings. Therefore one would imagine the adjustment would have to come from the currency or nominal wage deflation. This is essentially Spain's problem: to expect private savings to fall (and total credit to GDP to rise) is unrealistic given the country's high total credit to GDP ratio.

The real reason the Fed has held and will continue to hold rates at extremely low levels is because they are trying to slow the pace of deleveraging. This is what Bernanke calls "credit easing" (to distinguish it from QE). It seems that as many times as he has refused to call the MBS and Treasury purchases "QE", most economists still refer to it as an attempt to stimulate aggregate demand, as they do low rates.

Josh: You make it sound as if there are no New Keynesian academic arguments in favour of fiscal stimulus. That's not the case:

http://www.nber.org/papers/w15714

The argument is about the fiscal multiplier when monetary policy is stuck at the zero lower bound and the Taylor rule cannot be followed.

Woodford basically comes to the same conclusion as Nick: that fiscal stimulus is effective at the ZLB, and that it needs to be eliminated BEFORE the CB starts to tighten, as any subsequent spending would be ineffective under the standard NK model with the CB following the Taylor Rule.

Nick: What effect would a CB shift from an inflation targeting regime to a price level path targeting regime have on the exit strategies for fiscal & monetary stimulus?

Bob,

I am well aware of Woodford's argument. However, I am not convinced that it is correct. For example, the zero lower bound only really matters if we are in a liquidity trap -- a circumstance I do not find intellectually satisfying.

Also, Cogan, Cwik, Taylor, and Wieland in their assessment of New Keynesian and Old Keynesian multipliers assume that the interest rate is held at zero and yet their results still yield much lower estimates of the multiplier in the former case.

Josh: Good question.
1. What's published in the journals isn't a very good reflection of what people think. First, since governments weren't doing fiscal policy, nobody was writing about it. And they weren't doing fiscal policy largely because we thought that anything fiscal policy can do, monetary policy can do better, *in practice* (lags, political stuff, etc.), not because we thought that fiscal policy can't shift the AD curve (depending on how monetary policy reacts, of course). Second, journals, like newspapers, only report "Man bites Dog" stories. They don't report stuff that everybody believes already. So, when we hit zero, only old farts like me remembered how fiscal policy was supposed to work. To quote myself from one year ago:
"Fiscal policy is like getting grandfather's musket down out of the attic, where we locked it away to stop the kids (politicians, or perhaps ourselves) playing with it. It's an ugly, unreliable, inaccurate, slow, dangerous, weapon to use, and nobody can remember how the damned thing works."

2. As bob says, there's no problem at all building fiscal policy into a NK model. Really, there's no difference at all between NK and ISLM in this regard. Start with ISLM, add a NK Phillips Curve (it needed some sort of PC anyway), replace the LM curve with the Taylor Rule (or some sort of monetary policy reaction function), re-interpret the IS curve as an Euler equation, make sure the expectations all make sense, and voila! (The NK imperfect competition stuff was more important, but that's a separate story). Sure, with perfect capital markets, and rational expectations, etc., you get REH. But that's true in both NK and pre-NK models.

But, I agree, it was still sort of puzzling, how the profession looked at fiscal policy again. I think there's something to the "Dark Ages" story though. A lot of people had forgotten this stuff, or had confused silence with disbelief.

bob: "Nick: What effect would a CB shift from an inflation targeting regime to a price level path targeting regime have on the exit strategies for fiscal & monetary stimulus?"
It depends. Is this a switch to price-level path targeting now? Or if we had always been PLP targeting? What is the base year? I'm going to duck that question. It would really warrant a whole post. Sorry.

Scott: Thanks!

David: In a closed economy, it's not obvious to me how total debt/credit would affect the natural rate. Part of the problem is that if debt is endogenous, it's not easy to pose the question correctly. But, for example, if we asked "what would be the effect of a once-and-for-all never-to-be-repeated cancellation of all debts: well, a big change in the distribution of wealth, but that would have no obvious first-order effects on consumption or savings. I think investment demand would increase, since capital markets can never be perfect, so there would probably be a rise in the natural rate. Another big topic, for another post.

Nick,

I have simultaneously started two debates, so perhaps I should clarify my position.

1. I would agree that a lot of the discussion regarding fiscal policy went dormant because we didn't use fiscal policy. I am also NOT trying to make the case that fiscal policy is ineffective at increasing AD. Nor am I arguing that Ricardian Equivalence should be the starting point of analysis -- on that point I was playing devil's advocate.

Also, you note that one reason why fiscal policy is thought to be worse is because of "lags, political stuff, etc." This is a point of frustration that I have regarding the stimulus debate as well. "Political stuff" is a major problem and potentially reduces the magnitude of the shift in AD.

2. I am not arguing that we cannot incorporate fiscal policy into NK models, but rather that the implications are much different -- even at the zero lower bound. (Cogan, et. al come to different conclusions than Woodford.)

3. The entire debate seems to hinge on the zero lower bound. But, it order for the zero lower bound to matter, we must be in a liquidity trap. I am not convinced that such a thing exists. When you have a model with two assets -- money and bonds -- it is very easy to derive such conditions. However, if we incorporate more assets, it becomes substantially more complicated. In order for a liquidity trap to exist, the marginal rate of substitution between money and ALL other assets must be zero.

In other words, I reject the premise that we need fiscal policy at the lower bound.

If there had been some progress in the last 80 years, perhaps we wouldn't have needed it.

"Also, Cogan, Cwik, Taylor, and Wieland in their assessment of New Keynesian and Old Keynesian multipliers assume that the interest rate is held at zero and yet their results still yield much lower estimates of the multiplier in the former case.

...

2. I am not arguing that we cannot incorporate fiscal policy into NK models, but rather that the implications are much different -- even at the zero lower bound. (Cogan, et. al come to different conclusions than Woodford.)"

OK, so what's wrong with Woodford's critique of the Cogan paper?

Bob,

Let me clarify my position and then respond to your comment.

I myself am a critic of NK models. However, I agree with most of the profession in saying that they are preferable to Old Keynesian models. In addition, NK models are notoriously more pessimistic than Old Keynesian models regarding fiscal stimulus. As a result, I think that our priors should be based on the NK methodology.

On a separate issue, I don't think that the zero lower bound matters all that much. Nonetheless, that is another topic altogether and perhaps best left for another time.

Now to the Woodford paper...

Woodford's critique of Cogan et. al is that they assume that government spending continues past the point in which the interest rate is at the zero bound. In other words, the interest rate is assumed to be zero for two years, but the increase in government spending is expected to be permanent. As a result, this diminishes the effect of fiscal policy once the interest rate goes back to Taylor rule-type determination.

I have no problem with this as a general criticism. However, this ignores the fact that the purpose of the Cogan paper was for a direct comparison with the Bernstein-Romer estimates of the multiplier. These estimates were based on a permanent increase in government spending (as shown in Figure 1 of the Cogan paper).

Now, getting back to the Woodford paper, it is true that he shows the multiplier can be greater than unity. However, the magnitude of the multiplier is dependent on the probability that government spending will remain at the elevated level. Specifically, if the probability is equal to zero, the multiplier is equal to 2.3. In contrast, if the spending is permanent, the multiplier is equal to -5. The conclusion is that as the probability gets larger, the multiplier declines.

The predominant question therefore seems to surround when the multiplier drops below unity and, ultimately, zero. As Woodford notes, the former occurs when the stimulus is expected to continue for 4 quarters or more after the financial shock; the latter when the duration after the shock is 10 quarters or more.

Given that the current stimulus package is based on a sort of 'time release formula' in which spending began in 2009 and continues into 2011, I don't think that the Woodford paper shines an optimistic light on the stimulus. In addition, the Woodford model has vastly different predictions than the Bernstein-Romer model regarding a permanent increase in government spending.

I actually like the Woodford paper (as far as NK models go). However, it is important to consider that the model explicitly ignores distortionary taxes (see the recent paper by Uhlig for precise implications) as well as public choice issues surrounding the nature of the spending. These are certainly issues that matter for practical purposes, but they are not unique criticisms to the Woodford paper. Finally, the reason that I like the Cogan paper is because it uses the Smets-Wouters model, which has been shown to outperform most other models. Thus, the results are not unique to some fringe model. Perhaps Woodford's model performs as well, but there isn't yet evidence to verify such a conclusion.

Josh and bob: I'm enjoying your good argument. Let me try to provide some intuition to the NK results:

Assume fully Ricardian consumers, and that the mpc out of transitory income is zero, and the mpc out of permanent income is one.

Holding the real interest rate constant, you would then get a multiplier of one for a transitory increase in G, and a multiplier of zero for a permanent increase in G (because permanent disposable income falls dollar for dollar with permanent increases in G, via future tax liabilities).

Now relax the assumption that the real interest rate is held constant. Instead assume the nominal interest rate is held at the zero bound temporarily. What happens to the real interest rate depends solely on what happens to inflation. And that depends on how monetary policy responds to fiscal policy *after* the economy escapes the zero bound. By letting me make whatever assumption I wanted about the monetary policy reaction function, what price level it would target after escaping the ZILB, I think I could make the fiscal multiplier anything I wanted it to be.

But at this point an old Keynesian would scream: "But you NKs have assumed Ricardian households! In particular, you have assumed that no household is borrowing-constrained (so the mpc out of transitory income is zero)! So no wonder you NKs have such pathetically small fiscal multipliers! Your C+I+G aggregate expenditure curve is horizontal, not upward-sloping like in our good old-fashioned models! So the only way you NKs can get a fiscal multiplier above one is by assuming that the interaction of fiscal and monetary policy causes expected inflation to increase at the ZILB, so an increase in G causes r to fall!"

Does that make sense?

"Also, Cogan, Cwik, Taylor, and Wieland in their assessment of New Keynesian and Old Keynesian multipliers assume that the interest rate is held at zero and yet their results still yield much lower estimates of the multiplier in the former case."

Do you now see how wrong this statement is?

That's the point I was trying to get you to realize there, but you still don't seem to get it.

Josh: "3. The entire debate seems to hinge on the zero lower bound. But, it order for the zero lower bound to matter, we must be in a liquidity trap. I am not convinced that such a thing exists. When you have a model with two assets -- money and bonds -- it is very easy to derive such conditions. However, if we incorporate more assets, it becomes substantially more complicated. In order for a liquidity trap to exist, the marginal rate of substitution between money and ALL other assets must be zero."

I sort of agree. I too am sort of playing Devil's advocate, and adopting a more mainstream Keynesian approach for the sake of argument. I don't believe monetary policy was necessarily ineffective, though I would still be sympathetic towards fiscal policy as an insurance policy.

Nick said: "Too much Fed: There's an optimal level of government deficits. (And probably an optimal level of government debt too, though the theory on that is weak)."

IMO, the optimal gov't deficit and optimal gov't debt level is zero. I want to make the federal gov't similar to the state and local gov'ts (a user of currency).

I'm assuming the "currency printing entity" is removed from the gov't and the fed.

Bob,

I am assuming that the point that you want me to "get" is that the magnitude of the multiplier is dependent on when the fiscal stimulus ends. I understand this point. Nonetheless, I don't have a problem with Cogan estimating the multiplier by assuming that the stimulus is permanent because the paper is meant to be a direct comparison with the Bernstein-Romer model that operates under the same assumption. Thus, the point that I am making is that NK models are more pessimistic on stimulus than the Old Keynesian variety GIVEN the assumptions that have been made.

There are indeed NK arguments for stimulus. I did not mean to imply that there were not. However, as I detailed in my previous comment, the magnitude of the effects are dependent upon the expected duration of the stimulus. Thus, as a practical matter, I still see NK models as more pessimistic than Old Keynesian models -- which was the point I was initially trying to drive home. Would you agree with this proposition? Or is my view simply biased by the fact that I have little confidence in the government to devise a well-targeted policy?

"I am assuming that the point that you want me to "get" is that the magnitude of the multiplier is dependent on when the fiscal stimulus ends."

No, it's dependent on the duration of zero interest rates relative to the duration of fiscal stimulus spending. You're still not getting it. So long as the rate is "pegged" at zero, fiscal stimulus will continue to have a large multiplier.

If you understood all three of
a) Romer-Bernstein
b) Cogan et al.
c) Woodford

You would realize that Cogan et al. doesn't accurately present Romer-Bernstein (a representation that you seem to be taking at face value) and does not provide any sort of "direct comparison" or argument against it. It relies on an assumption that the Taylor rule is in effect to reach the pessimistic multiplier estimates, while Romer-Bernstein EXPLICITLY assumes zero interest rates for the duration of stimulus spending.

The whole point is that you can't argue against fiscal stimulus at the ZLB by assuming that the Taylor Rule is in effect.

For another iteration of this argument, see Brad Delong:

http://issuu.com/delong/docs/20090629_20090501a_stimulus_stanford.doc/1?mode=a_p

Bob,

I wrote, "I am assuming that the point that you want me to "get" is that the magnitude of the multiplier is dependent on when the fiscal stimulus ends."

You wrote, "No, it's dependent on the duration of zero interest rates relative to the duration of fiscal stimulus spending."

We are saying the same thing. I'm sorry that I didn't write "relative to the duration of zero interest rates", but I thought that was understood. Clearly, I understand this argument as evident by my comment regarding the Woodford paper above.

The point that I am making is that it CAN be a reasonable assumption to assume that the interest rate is at the zero lower bound to estimate the multiplier. Certainly, that is where one can maximize its magnitude. However, Bernstein and Romer assume that the interest rate is at the zero bound through 2012. I don't find that to be a reasonable assumption.

We (or at least I) seem to be having two separate arguments regarding: (1) theoretical estimates of the multiplier, and (2) the relative validity of Cogan, et. al v. Bernstein-Romer. On the first topic, we agree that in these models, the size of the multiplier is dependent upon the duration of the stimulus relative to the zero bound. On the second topic, I do not find it reasonable to assume that we will be at the ZLB for four years.

Well, NOW you agree, but if you do, then that makes many of your earlier posts (Woodford is not convincing, Cogan shows poor multipliers at the ZLB, etc. etc.) inconsistent with your current position. It's difficult for me to keep track of what position I'm actually arguing against. It would help if you acknowledged your errors and clearly differentiated between different positions if you are going to change them throughout the argument.

Anyways, since I guess we now agree on the point I made in my first post about "(1) theoretical estimates of the multiplier", the issue can be put to rest.

On (2) I don't see how you can make a strong argument that it is an unreasonable assumption. Presumably Bernanke was consulted in the process and didn't see it as absurd. Beyond that, as Krugman and many others have noted, the stimulus was too small, the US still has strong disinflationary pressures, and unemployment is projected to stay extremely high for years to come. When and why do you think rates will be raised?

Bob,

You write,


"Well, NOW you agree, but if you do, then that makes many of your earlier posts (Woodford is not convincing, Cogan shows poor multipliers at the ZLB, etc. etc.) inconsistent with your current position. It's difficult for me to keep track of what position I'm actually arguing against. It would help if you acknowledged your errors and clearly differentiated between different positions if you are going to change them throughout the argument."

I have not changed my position. If you go back to the comment in which I said that Woodford is not convincing, I stated:

"I am well aware of Woodford's argument. However, I am not convinced that it is correct. For example, the zero lower bound only really matters if we are in a liquidity trap -- a circumstance I do not find intellectually satisfying."

Thus, I was making an argument about the zero lower bound.

Similarly, regarding the Cogan paper, I merely made the observation that their multiplier is low. You responded by accusing me of saying that there no NK arguments for stimulus -- a claim I never made.

I stated in my previous comment that I agree that the size of the multiplier is dependent on the duration of the stimulus relative to the zero bound in these models. Woodford's paper is a theoretical exercise. Cogan's paper is a practical application. My point in saying that Cogan gets dismal results at the ZLB was to highlight that the mere existence of the ZLB is not sufficient to generate large multipliers. Figure 3 in Woodford's paper highlights this point by plotting multipliers that correspond to the probability that the government purchases remain at their elevated level.

Nonetheless, I thank you for questioning my integrity.

lol... whatever

"I thank you for questioning my integrity."

I didn't before this last post, but I am now.

Cool it guys. I *was* enjoying and learning from your argument.

In fact, rather than arguing with each other, why don't you both argue with me. You have both read more than I have in that particular area. Do I have the intuition right in my 11.46 comment above? Because I'm thinking of doing a post on this sometime.

Well, I think I more or less agree with everything at 11:46, although this catches my eye:

"at this point an old Keynesian would scream"

ooh.. that sounds like fun. Don't tempt me Nick;) I'm trying to be a good mainstream economist here, but sometimes it's difficult to restrain my inner vulgar Keynesian. I can't help it. I was raised on Galbraith.

I was raised on Lipsey and Samuelson! But when I was a lad, the key point of old vulgar keynesianism was the multiplier in the sense of 1/(1-mpc), and the mpc out of current disposable income in these fancy NK models is next to zero, because consumption depends only on *permanent* disposable income (plus interest rates). Plus, people are fully Ricardian, and fully discount the future tax burden of increased government spending.

If half the people in the NK models were assumed to be borrowing-constrained, we would get a baseline multiplier of 2 for transitory increases in G, rather than 1. That's a big difference.

Yep. On top of that I'm not even a fan of the PIH or hardcore Rat Ex or hardcore Ricardian Equivalence to begin with.

While arguments that take MPC into account do provide a reason for the general effectiveness of a dollar in stimulus, I think that where they are really important is in the design of the stimulus; deciding where that dollar goes. A big opportunity was missed in designing the stimulus in terms of MPC because so many people were stuck arguing for or against stimulus in very broad Keynesian vs. Classical terms, or about the overall size of the stimulus.

In general, I think that the old Keynesian models may be better at dealing with "mesoeconomic" issues. This is not based on an exhaustive survey or anything, but my impression of NK is that it is all micro (a lot of which I find dubious, but I often accept for the sake of argument) and macro (most of which is pretty good), with very little in between. It misses a lot of the messy interesting stuff that happens when you start disaggregating households, businesses, modeling institutions etc. I think that's at least one of the reasons why the new DSGE models being used at the CBs were adopted so late, while the old Keynesian models lived on, long after they were supposedly discredited.

Nick,

My apologies to both you and Bob. I was feeling a bit cantankerous this afternoon for reasons that had nothing to do with this blog and I'm sorry that I carried that sentiment into the comments.

In any event, I don't see anything wrong with your characterization of either argument.

I was raised on Friedman and Lucas. I appreciate the Lucas critique, but I do have some quarrels with the Rat Ex literature and some of the underlying assumptions of the models. As Nick has probably realized by now, I am more of a monetarist than anything else (do monetarists even exist anymore?). However, I have always preferred Brunner and Meltzer's brand to anything else.

Nick: you're right that the Taylor function is what's called proportional control. That that means is that the gains (the weights in the taylor function) are correct only for a very specific model how 'i' translates into the observed variables.

Conversely, your car includes the integral and derivative of velocity in its computation, which is why it adjusts for gain errors and appears to anticipate.

Indeed, that derivative term is known as anticipatory control because the rate of change in the observed variables leads to prediction under a locally-linear assumption.

Nonetheless, these techniques are different from building a feed-forward component to the control law based on more detailed modeling the economy.

You are correct, though, that proportional control can be made exact, if the 'plant' is known exactly.

Josh: depends how narrowly you define "monetarism". Narrow monetarists (can we call them "M1"?) have almost died out. But broader monetarist ideas are very much alive. In many ways, we don't think of monetarists ideas as "monetarist" any more, because most people accept them. (Like fish don't notice the water; the hegemony that need not speak its name). Nowadays, "monetarism" means only those monetarist ideas that have failed to gain general acceptance. (It's a bit like "Austrian", really). But I too am perhaps a bit more monetarist than most, in some ways.

Jon: I find your last comment very interesting. I am thinking of trying to do a post on the topic (or a bleg, since I don't know much about it). A couple of questions:

1. How come you know a lot more about control theory than I do? (Is it just an accident? Should I feel bad that I know so little? Do most economists know more?)

2. "Conversely, your car includes the integral and derivative of velocity in its computation, which is why it adjusts for gain errors and appears to anticipate."

So the position of the gas pedal depends on: velocity; acceleration (derivative); and distance? (integral??). Can you expand/explain a little? I sort of understand it a bit, but not very well. Does each of those 3 things perform a separate function? Wouldn't 2 be enough?

Nick,

Perhaps I am an M1 monetarist. I agree that many monetarist propositions have now been widely accepted, but many others are notably missing. The monetary transmission mechanism is assumed to operate solely through 'the' interest rate. Money is completely absent from DSGE models. I see this as a significant problem, but most of the profession does not.

I am also interested in Jon's comment. Do elaborate...

Nick:

I would be more right to say that your car uses the integral and derivative of the speed error. The process begins by first computing the difference between the target and the observed parameter. For instance these are (i - i*) and (y - \bar{y}) in the taylor rule. In the case of your car this is just (s - s*).

So position of the throttle is then set
1) based on the integral of the error "I"
2) based on the derivative of the error "D"
3) based on a proportional scaling of the error "P"

It is possible to implement a controller that is only PD, PI, PID, etc.

These different terms are understood to have certain functions. If all of the dynamics are known perfectly, proportional control is adequate on its own; however, this is not possible in practice. That's the role of the integral term. So long as an error persists, the control function reacts harder until the error reaches zero. The derivative term adds an anticipation.

You need not have all three terms, but they contribute different behaviors to the system.

It should be plain, that the taylor rule is a very primitive "P" controller.

But PID is very long in the tooth. In the past 20-30 years most of the 'new' methods have been MPC (Model Predictive Control) and Linear-Quadratic-Gaussian. Both which I think is what you had in mind--in a sense--when you were discussing using a model of fiscal policy within the Taylor rule.

The comments to this entry are closed.

Search this site

  • Google

    WWW
    worthwhile.typepad.com
Blog powered by Typepad