Why does Fiscal Stimulus Work ?
Ah now Kevin Drum wrote about something I know something about. He notes a post by John Cochrane and said he was licking his lips waiting for the Delong/Krugman demolition which, however disappointed him.
I was a little disappointed in their responses. They have plenty of detailed issues with Cochrane, many of which strike me as well taken. But I didn’t feel like they ever addressed Cochrane’s core argument. He isn’t insisting that stimulus doesn’t work.1 Instead, he’s taking aim at the stories economists use to explain why they think stimulus works.
He hands the mike to Cochrane
In his words, here’s the Old Keynesian multiplier story:
More government spending, even if on completely useless projects, “puts money in people’s pockets.” Those people in turn go out and spend, providing more income for others, who go out and spend, and so on. We pull ourselves up by our bootstraps. Saving is the enemy, as it lowers the marginal propensity to consume and reduces this multiplier.
“But Cochrane says that New Keynesian models don’t support this story at all. When you take a look into their guts, NK models posit an entirely different underlying mechanism for why fiscal stimulus works:”
If you want to use new-Keynesian models to defend stimulus, do it forthrightly: “The government should spend money, even if on totally wasted projects, because that will cause inflation, inflation will lower real interest rates, lower real interest rates will induce people to consume today rather than tomorrow, we believe tomorrow’s consumption will revert to trend anyway, so this step will increase demand. We disclaim any income-based “multiplier,” sorry, our new models have no such effect, and we’ll stand up in public and tell any politician who uses this argument that it’s wrong.”
The problem here is that Cochrane’s fantasy has so little connection with reality that it is hard to discuss. Consider Krugman’s New Keynesian model of fiscal stimulus here [update please click this link before criticizing my assertions about New Keynesian models] a. Notice there is no mention of inflation or real interest rates.
It is true that New Keynesians talk about expected inflation and real interest rates. But they do that when they discuss monetary policy at the lower bound. In contrast, if one modified a New Keynesian model with the assumption that the inflation rate is a known constant (this means setting a parameter in the new Keyensian Phillips curve to zero) then the model would imply that there is a government spending multiplier of 1. It just isn’t true that, in New Keynesian models, fiscal stimulus works only through expected inflation.
Expecting a reasoned critique of Cochrane’s claim is like expecting a reasoned critiqque of the claim that 2+2=5.
Now Cochrane is an intelligent person. How does he manage such regular howlers ? Well we have the ambiguity of “no spending multiplier” here this ambiguously means that the government spending multiplier is one (there is no effect on private consumption) and that it is zero. This is typical of fresh water fanatics when confronted by models and or evidence. They argue that the multiplier at the ZLB is zero (as say Krugman believes it is off the ZLB) and then a model or data which suggests that it is 1 is presented as proof that they were right.
I mean really saying this is like 2+2=5 is over charitable. Cochrane’s argument is an exposition of the result that 1=0.
DeLong and Krugman do not have the reputation of being over polite, but they are too polite to point out the arithmetic error which is the heart of Cochrane’s argument.
That’s not all. New Keynesians have no presumption as to the sign of the effect of real interest rates on the level of consumption. This is because they have some respect for data and there is essentially no evidence that real interest rates affect consumption. This can be reconciled with utility maximization (anything can) but by making assumptions about parameters so the effect of real interest rates on consumption is tiny in the model as it is in the data.
Anyone who has the slgightest interest in reality knows that real interest rates are negatively correlated with GDP growth because they are negatively correlated with investment. Someone who thinks that the effect of real interst rates on consumption has an importnt role in any calibrated New Keynesian model demonstrates complete ignorance. They are pure fantasies. This explains why the DeLong and Krugman do not seem to take his claims seriously.
update: I corrected some typos because it seems someone read this recently.
Robert: John Cochrane is right that the way in which fiscal policy works in New Keynesian models is very different from the way it works in Old Keynesian models. (It’s more different than he realises.) And that difference has a policy implication too.
In an Old Keynesian model, if there is an increased desire to save, so the natural rate of interest drops, and if the central bank is unwilling or unable to drop the rate of interest to match it, the fiscal policy response is to increase the level of government spending.
In a New Keynesian model, the fiscal policy response is to *reduce the growth rate* of government spending.
There is a very big difference between “Higher G” and “Lower Gdot”.
My post on this is here: http://worthwhile.typepad.com/worthwhile_canadian_initi/2013/11/new-keynesian-multipliers.html
My post included a link to a New Keynesian model in which the opètimal response to a downturn is to ncrease governm,ent spending. It is a very simple model in a blog post by Krugman, but it proves my claim that Cochrane’s unqualiied statement about New Keynesian models is not true as written. Note in the model there is Ricardian equivalence and the optimal policy is a shift in G.
fHere again. First I should admit that I hven’t clicked the link and I don’t plan to. My post addressed the question of whether old Keyensian fiscal stimulus stimulates in New Keynesian models for the same reason it stimulates in old Keynesian models. The answer is yes. I didn’t discuss the optimal policy in case of recession.
The old Keyensian proposal is to have a temporary increase in government spending. This is beyond dispute as it is what Keynes recommended in The General Theory. Prominent old Keynesians argued for budget balance averaged around the cycle. Only obscure figures (if any) talked about permanent stimulus. This means that the proposedspending increases wpuld cause higher output in a New Keynesian model at the zero lower bound.
It is true that a temporary spending increase as advocated by Keynes and all (or almost all) of his followers implies lower Gdot — that’s what temporary means. It is also true that in a New Keynsian model a credited promise to cut future spending will cause higher demand now while it wouldn’t in old Keyensian models.
I guess my bottom line response is that temporarily high G (as proposed by Keynes) does imply lower Gdot. While it is true that lower Gtod can be achieved without temporarily higher G (by having lower G than previously planned in the future) I never suggested that only old Keyensian fiscal stimulus works in New Keynesian models. I only claimed that fiscal stimulus via (temporarily) higher spending works in the same way — by increasing aggregate demand directly and not just by reducing real interest rates.
I still think the key issue is that Cochrane considers totally wasteful government spending (something not seriously discussed by any Keynesian new or old). So he considers stimulus to be a multiplier greater than one. Basically I think the main issue is that he has decided that 1=0,. I think he also assumes that fiscal stimulus is fiscal stimulus so what is true of temporary tax cuts is true of temporary spending increases. This is an elementary error, but it is common and was in the Wikipedia article on Ricardian equivalence.
The list of people who seemed to think that, in an economy with Ricardian equivlance, temporary government spending increases don’t cause higher aggregate demand is embarrassingly long.
Robert: I have previously read (and have now re-read) that note by Paul Krugman. I have also written posts myself arguing that a temporary increase in G will be expansionary in NK models when the real interest rate is temporarily constrained by the ZLB. Despite Ricardian Equivalence.
I repeat my claim. The reason a temporary increase in G works in NK models is that a temporary increase in G, that is *known* to be temporary, reduces the expected growth rate of G from now on. That is the mechanism that shifts the Euler-IS curve and raises the natural rate of interest. This means that an announced cut in G(t+1), holding G(t) constant, will have exactly the same expansionary effect as an increase in G(t), holding G(t+1) constant. It is the expected falling growth rate of G, not the higher level of G, that is expansionary.
In a continuous time model this difference would be stark. G(t) has no effect. Gdot(t) has a negative effect.
I realise now that I didn’t use to understand the reason why fiscal policy can work in NK models. I understand it better now than I did before. I think many or most macroeconomists, even good New Keynesian macroeconomists, do not properly understand it.
Robert: in response to your second comment:
In an Old Keynesian ISLM model, the natural rate of interest is a positive function of G, and independent of Gdot
In a New Keynesian model, the natural rate of interest is a negative function of Gdot, and independent of G.
We have slipped one derivative, and changed the sign. I think that is a substantive difference. Comparing Old and New Keynesian models, hit by the same shocks, where the objective is to solve for the time-path of G(t) that would stabilise the economy in the absence of monetary policy changing r(t), the two models would give two paths for G(t) that would be out of phase with each other.
I see fiscal stimulus as an “enabler”: allowing the economy to produce what it is physically able to produce by getting the fiscal log jam out of the way of money flow.
First, compare the fiscal (and physical) log jam with a “psychological” log jam — the classic baby sitter sharing example: parents were not using their share-the-baby-sitting chits to full capacity because they were hoarding them for when they “really needed” them. An economist showed them they could issue everyone extra chits and then the baby sitting sharing would really got going.
Compare fiscal stimulus with getting the right mixture of gasoline and air in a carburetor. Republicans always want to divert money to the “job creators” who are supposed to build new factories and open new stores when the old ones are closing for lack of demand in a recession. IOW Republicans always want to flood the engine when it does not have enough air (read: “demand”) to burn the gas that is there already.
Fiscal stimulus just gets the mix of gas and air back in proper proportion.
I know I am doubtless asking you to repeat yourself, but what is the mechanism via which making Gdot more negative in an NK model raises output?
Robert writes increasing G stimulates the economy in an NK model “by increasing aggregate demand directly and not just by reducing real interest rates” but you seem to be only talking about effects via interest rates.
Again, Robert writes “New Keynesians have no presumption as to the sign of the effect of real interest rates on the level of consumption” but if I understand you correctly (and I am not sure I do hence the question) you seem to disagree, and be saying that NK models are all about moving around the level of consumption by manipulating interest rates.
Nick Rowe’s argument works even if nominal interest is at the ZLB and inflation is zero (the extreme case of price stickiness). The point is that NK consumers know that they or teir heirs will pay fror government spending sooner or later. This means that the present value of their planned consumption is reduced by the present value of increased government spending. If G is increased by a costant, the C is reduced by that contant.
The it’s Gdot not G i Ricardia equivalence. It has nothing to do with the issues Cochrane stresses.
You are right and I am wrong. Your comment is completely consistent with Krugman’s post and clicking the link and re-reading the post would have made no difference. Indeed you correctly describe a major difference between NK and old K analysis of the effects of G on Y.
This accurate observation is not related to Cochrane’s post ( to understand it I insist one must understand that he ha redefined no stimulus effect to mean the multiplier = 1 having previously defined that phrase as the multiplier = 0). It doesn’t have much to do with the debate which is about temporary spending increases so more G and lower Gdot. But it is valid math which is perfectly consistent with Krugman’s post.
Sorry I was very cranky yesterday.
It’s my understanding that the standard stimulus multiplier in the
theory literature is to measure the response of correctly measured GDP value to useless _or_ useful but separable government consumption.
The optimal policy is not necessarily the one that creates stimulus,
so the stimulus multiplier in these models is simply a way to connect
the model to some Old Keynesian beliefs and stories. The optimal
policy is the one that maximizes the social welfare. The theory
models are really interested in the optimal policy. If I am correct, Waldmann’s post reveals serious confusion about these two topics that are different.
The reason why this stimulus multiplier definition is the standard is
that it doesn’t require an additional dimension to the marginal
utility of private consumption. By my understanding, in the canonical NK model, if one assumes that government consumption is a direct substitute for the private consumption, government consumption doesn’t ever stimulate, even in a recession or ZLB — it’s a trivial case. I think this is another way to show how the Old Keynesian logic is missing from the NK models; in an Old-Keynesian model even direct substitute government consumption will have an impact.
The two remaining cases, useless government consumption, and useful but separable government consumption are both analytically tractable and non-trivial.
The useful but separable government consumption is both tractable and non-trivial because the government consumption changing doesn’t directly change the marginal utility of private consumption.
Therefore, not all effects of government spending are automatically
fully offset by the consumers. The offset is only thru the permanet income channel, meaning that the saving will increase smoothly
(consumption decline smoothly) enough to cover the future taxes
required to pay for the government spending. It also allows for the
optimal long-run average government spending to be non-zero, which
useless government consumption usually doesn’t.
Also, what is worth noting here is that the government doesn’t produce
anything in this model, it just consumes. This is important. Whether
the government spending is either useless or useful and separable
doesn’t usually matter for the fiscal stimulus multiplier. The experiment in the model is not that government goes and hires a bunch of unemployed people and has them do something, the experiment in the model is whether the government takes some of the output now, either without taking compensation in form of taxes, or by borrowing now and later taking without compensation later. There’s usually no distinction to the fiscal stimulus multiplier whether the government consumption was a complete waste or useful but separable. The social welfare (or utility) is obviously different, but production is the same regardless.
This in my opinion underscores the fact that the whole concept of fiscal stimulus multiplier is asinine. The optimal policy is not asinine, and we should be talking about the optimal policy and the utility losses from suboptimal government policies. This is one of the main reasons why I would recommend everyone read the recent Mankiw & Weinzierl paper.
So my conclusion is that all of both Cochrane’s and Nick Rowe’s arguments about what the model says about fiscal stimulus multiplier are completely correct under both the assumptions that government consumption is useless or that government spending is useful but separable.
(As a side note, if the we make government consumption a complement to private consumption, for example buying cars and giving them to people, the fiscal multiplier effect is going to be much smaller because current private consumption will go down.)