If I were a different sort of person, I would now be accusing some macroeconomists of deliberately misrepresenting the policy implications of their models in order to further their own political agenda.
But I am not that sort of person. I don't generally go for conspiracy theories. And I myself used to make the same mistake that they are now making. Because I didn't understand that model properly. So I figure it's very likely the simple explanation: they don't understand it properly either. But it would be prudent for them to avoid throwing stones.
It's important that we do try to understand our own models properly, so we can try to teach them properly. And at least try to present the policy recommendations of our models without spin. Subject to all our usual human imperfections, of course. And if we don't like those policy recommendations, that's OK too. That should lead us to re-examine our models.
Let G(t) be the level of government spending at time t. Let Gdot(t) be the derivative of G(t) with respect to time. Let r*(t) be the natural rate of interest at time t.
In Old Keynesian models, with an Old Keynesian IS curve, r*(t) is a positive function of G(t) and is independent of Gdot(t).
In New Keynesian models, with a New Keynesian IS curve, r*(t) is a negative function of Gdot(t) and is independent of G(t).
Any good New Keynesian macroeconomist will be able to understand why that is true, but may not have thought about it that way before. In a simple model where C(t)+G(t)=Y(t), where Cdot(t) is a positive function of r(t) by the consumption-Euler equation, and r*(t) is defined as the time path for r(t) such that Y(t) equals potential output Y*(t) for all t, so Cdot(r*(t))+Gdot(t)=Y*dot(t), that's what you get. But they can do the math much better than I can.
Suppose the economy is stuck at the ZLB at time t, and we want fiscal policy to increase r*(t).
The Old Keynesian policy recommendation is to increase G(t).
The New Keynesian policy recommendation is to decrease Gdot(t).
Those are very different policy recommendations. We have slipped one derivative, and changed the sign.
In a discrete time model it is harder to see the difference. Because if we increase G(t) and leave G(t+1) unchanged, we decrease (G(t+1)-G(t)). But if we decrease G(t+1) and leave G(t) unchanged, we also decrease (G(t+1)-G(t)). [Update: two math/typos fixed. Well-spotted primedprimate!] According to the New Keynesian model, both will work equally well in raising r*(t).
I have heard macroeconomists recommend increasing G(t) temporarily, and saying that the New Keynesian model supports this policy recommendation.
I have never heard a single macroeconomist recommend reducing Gdot(t), and saying that the New Keynesian model supports this policy recommendation.
If a policymaker said "We need to start cutting government spending now, and keep cutting until the economy is off the ZLB", would any macroeconomist say that the New Keynesian model supports that policy? Because it does.
If we also recognise that G(t) is not in practice a jump variable, that can be made to jump right now, and that fiscal policy changes take time, and so Gdot(t) is the more realistic control variable, then this inconsistency between what macroeconomists recommend and what the New Keynesian model recommends is even more glaring. Announcing that G(t) will be slowly falling from now on until the economy is off the ZLB works equally well, according to the model, whether or not it is accompanied by an instantaneous jump in G(t). But an instantaneous jump in G(t) is hard to implement unless you have planned it well before time t.
If you think that cutting Gdot(t) at the ZLB won't work, that's OK. But don't use the New Keynesian model to support raising G(t) at the ZLB. Because the New Keynesian model says that raising G(t) only works if and because it means cutting Gdot(t). It says that raising G(t) is just a sugar cube to help the cutting Gdot(t) medicine go down. [Update: not that there's anything wrong with taking the sugar cube, but it is wrong to say the New Keynesian model says the sugar cube is what cures you.]
Worthwhile Canadian Initiative 
Steve: I have modified my views slightly. Here is my new post on that topic. It’s closely related to Paul Krugman’s.
Nick:
I am now a new Keynesian.
So, the model shows that we need to cut taxes now, and then reduce the growth rate of government spending (to gradually bring the budget to balance.)
I love that fiscal policy result!
Surely that is the reason to choose between models of the economy. 🙂
Bill: But remember, the model says you should be increasing Gdot when you are off the ZLB, otherwise G will go to either 0% or 100% of GDP.
And see my Sept 16 03.33pm comment. The model (with half the agents spending all their current disposable income) says we need to increase the growth rate of taxes, and cut the growth rate of government spending!
Why doen’t the math say that under the ZLB condition, when r(t) can not be influenced by normal policy, r(t) is in effect a constant. Then r(t) disappears from the derivatives wrt time. The policy implications are that Gdot should increase under the ZLB condition if an increase in Y to Y* is desired. I end up with the same policy view as ever: fiscal policy is not effective, in the context of flexible exchange rates, unless the ZLB condition is in effect.
Kathleen: “The policy implications are that Gdot should increase under the ZLB condition if an increase in Y to Y* is desired.”
The Old Keynesian ISLM model says that G(t) should increase under the ZLB condition if an increase in Y to Y* is desired.
The New Keynesian model says that Gdot should decrease under the ZLB condition if an increase in Y to Y* is desired.
In both cases Y < Y* because r > r*, and we cannot cut r because of the ZLB so we need to increase r* instead. But they tell us different ways to increase r*.
Are there competing New Keynesian models? I say that having completed 4th year Macro using David Romer’s Advanced Macroeconomics text. In the Romer discussion of New Keynesian models he states: … the most glaring omissions from the model are investment and government purchases (pg 315). There is no discussion of fiscal policy as far as I can tell. When I read Romer, the models make intuitive as well as mathematical sense, but there is only discussion of the Central Bank (monetary policy) and no discussion of the ZLB. I completed this 4th year course only 2 (3?) years ago so I suppose I have a modern version of an economics undergrad education.
Kathleen: I think someone else could give better answers than me. I remember seeing investment put in, but can’t remember where. I think they usually omitted fiscal policy and the ZLB until the recent recession.
Do you, or anybody, seriously believe that any consumers (or any number of consumers large enough to make a macro impact) behave in a way causing a positive economic impact from a negative dG/dt? What in the hell are you smoking?
ML: Do you, or anybody, seriously believe that any rhetorical question like that would make a positive contribution to the discussion? Were you too drunk to read the post?
I think Mayson Lancaster’s comment is merely a restatement of the point Andy Harless was sort of alluding to earlier when he said: “Obviously it’s not realistic in our world of agents who are irrational, short-sighted, mortal, liquidity-constrained, rightly distrustful of government commitments, and so on. But in a world with infinitely-lived, non-liquidity-constrained, perfectly rational agents…”
Models necessarily make simplifying assumption (or, as Box says, “essentially, all models are wrong, but some are useful”). Whether the assumptions are good or not depends on whether they are important in regards to what you are looking at. If a particular implication would not hold if these assumptions are relaxed, then those are bad assumptions, which means this is not a useful model for answering that particular question. That’s how you differentiate the model’s bugs from its features, right?
Does the implication of cutting Gdot to increase AD hold if you relax the assumption of Ricardian equivalence or lack of credit constraints?
pG: “Does the implication of cutting Gdot to increase AD hold if you relax the assumption of Ricardian equivalence or lack of credit constraints?”
See my most recent post. I have relaxed the assumption of Ricardian Equivalence. And my Old Keynesian agents’ consumption function might be motivated by something like credit constraints (very roughly). It seems it does hold. Plus you get the implication of raising Tdot!!
Mayson Lancaster might have been restating Andy’s point, but if so he did it in a very unhelpful way, and out of line with the culture on this blog. (Your restatement, whether or not it was what he was saying, was much more worthwhile.)
But it is still a weird set of results. I wonder if the indeterminacy problem might not be at the root of it.
“By reducing the growth of government spending, and reducing the growth of consumption by the Old Keynesian agents (by growing taxes), the permanent income of the New Keynesian agents is now higher than their current income, and this offsets their reduced rate of time preference, and prevents them wanting to save part of their current income.”
Changing the model so that only some of the agents have Ricardian equivalence and others only have a MPC is not exactly what I would call relaxing the assumption, but merely combining it with another assumption that is outdated because it is less realistic. Imagine a frictionless physics model where they change it so that parts of the universe are frictionless and other parts have 100% friction. Is that really relaxing the assumption?
But I honestly thought you hadn’t realized the result was weird and imagine Lancaster might have thought similarly (agree his comment was stated in an unhelpful way).
pG: “Imagine a frictionless physics model where they change it so that parts of the universe are frictionless and other parts have 100% friction. Is that really relaxing the assumption?”
Good point. I will think about that one.
“But I honestly thought you hadn’t realized the result was weird…”
What’s weirder is that Paul Krugman’s model (and all the other NK models of fiscal policy) have exactly the same weird result. And they just can’t see it! By restating the model in continuous time, it doesn’t really change the model at all. It just forces them to see it. It’s all in the framing!
Well, if it does derive from Ricardian equivalence, that’s the same assumption used in New Classical models which I would suspect have features that override this result. And note, this does argue in favor of my opinion that these models are probably serving an ideological purpose (recall my earlier point about RBC economists not trying to change preferences for leisure even though that is a weird result of their model). As such, I think the profession ought to implement a social science equivalent of medicine’s double-blind studies, employing a little division of labor.
I’ve stated this idea elsewhere but am curious what you would think. Have one person create the model. Have a second person actually run the model and gather the results, where the second person never directly interacts with the first person, is not told what the variables in the model actually represent in the real-world, and is also not told what results are expected.
Note, if the model or data is such that it would be obvious for that second person to identify the variables, a third person could perhaps do a transformation (would that work?). Also, the second person can also serve as a check on whether the first person still publishes results that they didn’t like as well as also noticing “weird results” like this. As such, it would be better if the two people were always from different schools of thought, but that would making the matching problem a bit harder and could also be gamed. As such, I think an automated blind random matching system would probably work better.
I am not an academic economist and have never created any models, so I’m not aware whether something like this is already in place or whether it would be unworkable. But it’s my impression that the norm right now is for that second person to be the same as the first person, or a TA or grad student of the first person.
pG: it certainly is embarrassing when we only like our models when they tell us what we want to hear. But I don’t think it’s just politics. I’m a small G guy myself, but advised the PM against cutting G and raising T when the recession hit (and other economists heard me say that). And I’m pleased to see the government preponed some government investment projects during the recession (which is like increasing G(t) and cutting G(t+1), and fits with the NK model OK, if you can raise G(t) quickly.
But I still don’t like these results.
But I don’t think your suggestion would work. Normally we can see in advance where our models are leading when we build them. This is one of those rare cases where something weird popped out, and nobody saw it coming.
Any ideas how to improve my idea? Knowing in advance where a model is leading would seem even more problematic from a cognitive bias standpoint, where one can build a model with an underlying goal of arguing in favor of a particular policy. Can model-building be broken up into modular pieces like a software project?
When I look at the history of macro thought, I see schools of thought arguing primarily over whether there should be more or less government intervention in markets. That would draw resources away from making models that model the economy and forecast better.
pG: “Can model-building be broken up into modular pieces like a software project?”
That is how it was sometimes done, for the big computer models in the olden days. But then the bits didn’t always fit together into a logically coherent whole. There was no guarantee that there exists a possible world in which all those bits could be true at the same time.
This is how it’s normally done. You have an intuitive idea, then build a model to try to formalise that idea. If you find your model contradicts your idea, you re-examine both your idea and the model. For example, before I built my hybrid model, I figured cutting Gdot and cutting T would work. Did the math, and found I was wrong about the cutting T bit. See my November 16 7.55 comment above, where I got it wrong. Building the model taught me something. In principle it might be possible to check against the data, but when the policy variables are being adjusted deliberately to try to smooth out the economy, you have the exact opposite of an ideal experimental design, where you toss a coin to decide on policy, and look to see whether the coin toss causes the economy to do horrible things.
Sorry for the delay in response (I have a two-year-old).
Hopefully you see where I am getting at, but let me back up. When I took what my school calls graduate macro (this is not a top-ranked school, so it doesn’t teach modeling and is more akin to history of macro thought), I was deeply disappointed. A conservative Classical school arguing with a liberal Keynesian school arguing with a conservative Monetarist and then New Classical school arguing with a liberal New Keynesian school (plus a conservative Austrian School and liberal Institutionalist School etc. on the sidelines). This was nothing like the the steady progression of knowledge I had seen in physics or chemistry, and you also don’t see this degree of political polarization in other social sciences like sociology or political science.
What was most striking was that the New Keynesians and New Classical school resolving their methodological disagreement but then still continuing the same ideological argument over the usefulness and effectiveness of counter-cyclical government policy (the same ideological debate as between Keynesians and Classicalists, really). In most scientific fields, one would have expected new disagreements to occur, creating different divisions of people into new schools.
It wasn’t until I came across Barbara Bergmann’s commentary that the above made any sense:
http://www.jstor.org/stable/25046110
She observed that there was a higher degree of government hiring and political appointments of economists compared to the other social scientists, which, of course, creates incentives upon research. She was referring to the U.S., but I believe it is probably also true in many other nations. While medicine does not seem face the problem of political polarization, it still is concerned with the possibility of bias from medical researchers affecting results due to the placebo effect. Thus, they use double-blind studies.
What is the equivalent of this in economic research? I am not aware that there are any institutions or norms or standards in place to counter the politically polarizing incentives upon the field or the possibility of creating a model to back a particular policy argument. While it is admirable that you yourself do not do this, I think any economist can see that it is unrealistic to assume most individuals will ignore and defy incentives placed upon them (and there are numerous examples of those who have not).
Also, I am not sure how important theoretical consistency is when models include such a high degree of unrealistic assumptions. I would think that more closely approximating this world would far be more useful than creating a possible theoretical world that does not resemble ours very much.
pG: Yes, I see what you are getting at. I don’t think your idea for curing the problem will work, but I don’t have any better idea that would work either. Closely approximating the world would be nice, but without experiments to distinguish them, more than one theory can, more or less, fit the facts.
On Barbara Bergmann’s commentary. Maybe. But then we see exactly the same divides within academic economists, who don’t really have much of a financial incentive to grind any particular axe, except to grind some axe that will get them some publications. If you have a comparative advantage in working on a particular theory, simply because that’s the theory you’ve been taught, you tend to stick with it. Unless something else trendy comes along, and you can see a paper you can publish, regardless of whether or not you believe in that trendy new approach. Freshly-minted PhDs have the most to lose from anything that makes their theory-specific human capital obsolete. Old guys like me don’t have much to lose any more, though we still tend to argue for “our side”.
I don’t know about Canada, but academic economists can be and do get appointed to the Fed. Work by academic economists regularly gets cited in the media by politicians. So you would expect the divide to also exist within academia.
Economics is always going to be a field of numerous models. I think that’s fine. The problem, I think, is that the usefulness of a model to an economist’s career relies upon variables such as whether or not the model is useful to a politician. As I mentioned earlier, this would draw resources away from having models more closely approximate the world.
http://www.voxeu.org/article/failed-forecasts-and-financial-crisis-how-resurrect-economic-modelling
It is always possible to alter incentives. And with that, my son is awake…