The IS-LM model is the simplest version of the “old Keynesian” approach to macroeconomics. You don’t hear much about IS-LM in the research literature these days, but it’s been coming up a lot on the blogs. Tyler Cowen tells us what he doesn’t like about the model; Brad DeLong and Paul Krugman rise to its defense; Stephen Williamson takes up the gauntlet, and Scott Sumner weighs in.
None of them, in my opinion, has touched the main issue, which is that IS-LM provides absolutely no framework for policy analysis because it makes no assumptions, and draws no conclusions, about what people are trying to accomplish. If you don’t know what people are trying to do, you can’t possibly know how best to help them.
Suppose, for example, that, as Paul Krugman believes, the current state of the economy is being driven by a “liquidity trap”, which means that people are hoarding money instead of spending it, and therefore consuming less, which is why employment is so low. Two weeks ago on this blog, I posed the following question to the IS-LMers:
Why aren’t you thrilled with the current state of the economy? … Why, as the stock of money continues to grow, shouldn’t the joy of hoarding eventually compensate for the annoyance of not having food on the table?
The IS-LM model provides no answer to that question. A model that can’t decide whether the current US economy is in a state of Nirvana is not a useful model for policy evaluation.
Back in July, 2010, I posted a toy model that is designed to address those questions — not because this model is sophisticated enough to be terribly useful, but to illustrate for blogreaders, in general terms, what a useful model might look like.
Greg Mankiw and Matt Weinzierl have recently provided just such a model. Like IS-LM, their model is fundamentally Keynesian. Unlike IS-LM, it is capable of evaluating the relative desirability of various policies. Would we be better off in a world where everyone floats in a vat of money until we all starve to death? The Mankiw/Weinzierl model, reassuringly, says “no”. The IS-LM model, unhelpfully, says “that’s not my department”.
For example: IS-LM says that in certain circumstances, government purchases can increase current output. It gives no indication of whether that output is worth its cost. That’s fine, I suppose, if you fetishize output over individual welfare. But Mankiw and Weinzierl’s model, being designed to address actual policy questions, is capable of saying that we might be happier with some kinds of output than with others. This allows it to conclude, for example, that well designed tax policies are more likely than spending programs to increase the right kinds of output. IS-LM can’t say that, because IS-LM takes no position on what counts as a better outcome.
This is not an endorsement of Mankiw/Weinzierl as the be-all and end-all of macroeconomic policy analysis. It is simply to say that whether you’re a new Keynsian, an old Keynsian, a neo-classical, an Austrian or a Marxist, you can’t do policy evaluation without a model that allows you to ask whether your policies are making people happier. IS-LM is not that model.
Edited to Add: A few hours after I posted this, it was announced that Tom Sargent and Chris Sims had been awarded the Nobel prize for developing macro models that, among other things, are designed to be useful for policy evaluation in precisely the way that IS-LM is not. In that sense, the post could not have been timelier. Congratulations to Sargent and Sims, and to the Nobel committee for a brilliant choice.
Would it be fair to say that a “model useful for policy” must be a “model with a utility function”?
Yes, actually, it does.
For example, the I=S equilibrium means that what firms want to invest in must be equal to what people and the government want to save. If it is not in equilibrium, the curve must lay either to the far left or right. Depending on where or how you want to shift the curve, to reach equilibrium levels, you could either a) increase/decrease taxes or b) increase/decrease government spending.
The LM relation is for monetary policy, and thus would be useful from a central banking perspective. The IS relation would be useful for a matter of fiscal policy. The model is crude, but I don’t think your criticism of it is valid.
This is probably hopelessly niaive, but can we not combine models, so the IS-LM can still be useful. We actually know from other sources that starving to death whilst hoarding money is not a good outcome. Why can’t we look at the outcomes of IS-LM and asses them against external criteria?
Mike H: A model useful for policy must be, at the very least, a model with a social welfare function. That social welfare function, in turn, takes something as inputs, and it’s hard for me to imagine how those inputs could plausibly be anything other than the values of utility functions. So I’d say that to be useful for policy, a model a) surely needs a SWF and b) almost surely needs utility functions. ISLM has neither.
Supply and demand is a model useful for policy analysis. Yes, it can be backed up with utility and profit maximizing analysis, but does not need to be. I can simply assert that, empirically, the demand curve slopes down and the supply curve slopes up and do all kinds of policy analysis, for example the effect of rent controls, without mentioning a social welfare function.
Steve, this may be a silly question, but doesn’t the existence of demand imply the existence of utility? That is to say, by looking at a demand curve, aren’t you implicitly looking at how much people want the product or service demanded? So can’t you construct some measure of social welfare just by looking at aggregate demand?
How is a model with a utility function built in useful for anything but confirming your initial bias? If I input the premise that drugs are wonderful producers of bliss, I’m going to get a result that says drugs should be legal, and maybe subsidized. If I input the premise that drugs ruin lives…then I’m going to get the result that drugs should at least be highly regulated/taxed, and more likely banned. Is anyone going to have their mind changed by the fact that my result came from a model that assumed my premise, rather than just arguing the premise directly?
” A model useful for policy must be, at the very least, a model with a social welfare function.” I don’t see how this can be right. “Useful” is just the wrong claim here. Operations researchers came up with a model to reduce the exposure of ships to German submarines. That was useful for the admiralty. But the model only talked about the likelihood of sinking, not its desirability (or lack thereof). Engineers model the level of flooding levees can withstand. That can be useful in deciding if it makes sense to build the levee.
A theory may need a utilty function to be dispositive, but not to be useful.
David W: Putting a utility function in the model does not entail making assumptions about the parameters of that function; those parameters are to be estimated. So you can have a model that says utility depends partly on food consumption and partly on drug consumption, and then estimate from data (together with the model’s implications) whether drug consumption enters positively or negatively.
Figuring out how to do this in a useful and insightful way is a substantial part of what this morning’s Nobel prize was awarded for.
Several posters have made the same point: According to Steve mathematics is now useless in policy discussions.
Ken: IS-LM can be useful in policy discussions in the same sense that mathematics can be — but IS-LM, like mathematics, fails to provide the one crucial ingredient that you’d expect to get from *economics*, which is an illumination of the tradeoffs.
??? It doesn’t illuminate tradeoffs.
Suppose you increase government spending and hold everything else constant. This shifts the IS curve to the right, and thus affects output and the interest rate. You can then conclude that an increase in output comes at the expense of an increase in the interest rate. How is that not a tradeoff?
@Steve: I confess I am having a litle fun with you but I will point out I am doing to you what you do to Krugman: teasing out an implication of your assumptions. Mostly I am picking at the vagueness of your “useful”.
Let me try to be a bit more explicit though. I certainly agree that it would be nice if the theory gave us a hypersurface of points that are a maximum of a SWF. Presumably a SWF derived from the utility functions that gave all the various curves and surfaces used in the theory. Then the theory could provide clear guidance because it could tell us when we are off the surface and which policies would move it towards it. Is it your claim that the ISLM is not such a theory and the new one is? Is that what you mean by “useful”?
Well, and I know I’m piling on here, to be useful to policy analysis, a model does not *have* to “illuminate” trade offs. Trade offs are important, but surely not all-important–after all, there are tradeoffs.
And yes, I will now stop calling you “Shirley”. :>)
Steve, I know what you are getting at but I think its kinda silly.
I mean you could just drop in “SUM { U(Ct) } for all t” and you would be done. That is to say that people like consumption and dislike the lack of it. Since, a IS-LM suggests that a recession will have simultaneously lower consumption and lower investment you would conclude that people are happier at full employment.
You don’t even need leisure in the utility function because most IS-LM type analysis assumes you cant do any better than full employment output anyway.
Steve –
Just caught one of your other comments, re tradeoffs.
What IS-LM is trying to tell you is that there aren’t any trade offs. This pure free lunch territory.
Now, you might think that’s wrong, but I think its important to see why its not crazy.
People seem to really get upset about recessions. Like really upset. And, it seems like just about everyone is upset.
Its not like a situation with say a minimum wage were Sally is sad because she doesn’t have a job but Bill is happy because he is earning economic rents.
Instead, just about everyone is pissed and they are really pissed.
One interpretation of that is that the economy is engaging in rent dissipation and that’s why there is universal upsetness.
Backing this up is the fact that only thing that pisses off people as much as recessions are wars, and we generally think wars are bad precisely because they are a giant exercise in rent dissipation.
Just checked the UMN website, and it turns out I probably took my advanced economics classes with Sims and Sargent as the instructors in the 1970s. LOL !!!
Karl Smith: I take your point, but I don’t think your argument is convincing.
When we fight a recession, we might very well be trading off current benefits against future costs. The fact that everyone currently alive favors that tradeoff does not imply that everyone affected favors that tradeoff.
Of course, as you say, there might be rents being dissipated, and an opportunity for a Pareto improvement because of that. But I don’t see how we can know that from IS-LM, which can’t talk about utility, can’t talk about surpluses, and therefore can’t talk about rents.
I’m with Harold: models that explain how the world works can INDIRECTLY address utility. In an old Slate article on global warming, Steven Landsburg listed a set of questions that must be answered to calculate the optimal response to global warming:
“How much does human activity affect the climate? This is actually a whole menu of questions: What can we expect given the current level of carbon emissions? What if we cut those emissions by half? By two-thirds? And so on. These are questions for physical scientists, not economists or politicians.”
Climate models don’t address utility either. Would a climate model that “can’t decide whether the US economy is in a state of Nirvana” be useless for policy evaluation? No, because it can be combined with models that DO include utility to address the welfare issues of climate change. (Utility models specify the objective function, climate models indirectly specify the constraint function by giving prices for things like fossil fuel consumption).
Why doesn’t the same logic apply to IS-LM? If, for the sake of argument, IS-LM describes the employment response to increasing aggregate demand, isn’t this useful in the same way climate models are useful? Can’t this knowledge be combined with utility models specifying how much people value employment versus inflation, taxes, ect, to produce policy conclusions?
@Karl_Smith: I mean you could just drop in “SUM { U(Ct) } for all t” and you would be done. That is to say that people like consumption and dislike the lack of it.
Okay, but help me out here: How does such a utility model handle the uncertainty individuals have, which drives their desire to hold liquid goods like cash? After all, the very reason people aren’t robotically upping their consumption of goods purchased with dollars is that they _don’t know_ whether a given purchase maximizes discounted present utility, and therefore prefer to keep their options open.
In a sense, holding liquid goods is like a insurance policy against supply shocks. Should you then count the imputed insurance premiums as part of consumption, or leave the benefit of hoarding out of the social welfare function entirely?
It just seems like you’re sidestepping the entire issue of what causes liquidity preference, leading you to believe people are made better off when forced to spend until they have no “safety cushion”, simply because it means “more consumption”.
I think the extremists on this issue are those who would throw out a perfectly good and useful model because it does not meet their standards of perfection with regard to modelling intertemporal decisions. No one is saying IS-LM doesn’t have faults–all models do, that is why they are models.
And BTW, Robert Mundell won the Nobel prize for using the IS-LM model to explain exchange regimes.
Would it be possible to invent a set of utility functions and then derive the IS/LM? After all, there seem to be utility functions implicit in the thinking of those who talk about the IS/LM.. “liquidity preference…” etc.
However, mostly I see the IS/LM presented as a semi-formal way to understand what’s happening now – low employment, but low bond interest rates in spite of increases in govt debt. The effects of low employment are fairly obvious – families suffering, and long-term damage to productivity. To quibble about utility functions must seem a bit like the captain of the Titanic saying “Ah, but exactly how far ahead is this purported iceberg?”
In the long run, economists can go back and forth in economic journals and fine-tune their utility functions vis-a-vis the IS/LM. In the meantime, the IS/LM tells you the response of the economy to various types of govt intervention, that a massive sudden spending program on infrastructure would boost GDP and employment – but there’s no time to fine-tune the models first – and it’s obvious to most laypeople that boosting GDP and employment would be a good thing.
Agreed, IS/LM doesn’t tell you that boosting GDP/employment is good. However, if it accurately predicts the response of the economy, it’s a more useful tool to policy-makers than a model with utility functions that gives wrong answers.
Better to grope in the dark than search for the keys under the wrong streetlight.
Karl Smith:
Let me add this: If your objective function is SUM(U(Ct)), and you have no explicit cash-in-advance constraint (or any other non-utility based reason for holding money), then nobody holds money, so IS-LM can’t be right.
So I think you are just plain wrong to say that you can just tack this utility function onto IS-LM.
The term “Liquidity Preference” indicates that people find utility in liquidity (ie, holding money).
What happens if you define
U(m,b) = m * b^i,
where i is the interest rate? Can the LM curve be derived?
here, m is the amount of cash I hold, and b is the dollar value of my bond holdings.
I think the issue is that to some, whether boosting employment is obviously a good thing depends on how “massive” the required stimulus is. E.g. is reducing the UR to 8% by election day worth $500 billion, or $2 trillion, or $5 trillion? (BobbyJay: identifying that a trade-off exists isn’t the same as providing a useful framework for prioritizing it.) Clearly people are bringing their implicit views of utility (read political economy) to the table, which is why a quaint little model seems to turn into a catfight. Always better to make all the assumptions explicit, rather than jump to calling people silly or mean. To me Cowen’s point about stocks vs. flows of wealth seems like an important one, and I wish I had a stronger sense that people are really “cross-checking their conclusions with something intertemporal” as PK describes.
@iceman : no need to speculate about that.
Normally, employment/GDP is pushed up by government spending and down by interest rate increases, and interest rates increase as the government spends. However, the IS/LM informs us now that the marginal rate of increase in interest rates per extra dollar borrowed is currently 0%/$.
Now, imagine the government borrows an extra dollar and spends it.
The above shows us that the rate of increase in GDP/Employment per dollar spent is currently positive – interest rates don’t change, so there’s no negative impact on GDP from the borrowing, while the spending directly increases GDP by $1, with possible knock-on effects depending on how the $1 was spent.
Also, we know from the bond market that the government (not the politicians who compose it), the marginal cost of government medium-term borrowing is, currently, about 0. They’ll pay no interest on the debt.
If the government “wants” GDP to rise, clearly the strategy of borrowing the dollar and spending it has positive utility for the government, and continues to do so until
* bond prices start to fall, equivalently,
* the “natural” real interest rate becomes positive again, equivalently,
* we get out of the “liquidity trap”
Individuals and corporations likewise “want” employment/GDP to rise, interest rates to stay low, and gain positive utility if the government borrows and spends, at least until we exit the liquidity trap.
The IS/LM model doesn’t talk about inflation, but more complex models do – and they predict that inflation is approximately 0 in the liquidity trap, so considering inflation doesn’t change the conclusions above.
However, if the government, public and industry all have positive utility from government borrowing and spending, why, then, is the government not borrowing and spending??
Either
* the model is wrong – but it (and its cousins) have made very good predictions about bond prices and inflation
* at least some powerful people are irrational (in the economic sense), or…
* Krugman’s right, the GOP just wants to squish the economy until the election, after which they’ll all rediscover their inner Keynesian.
@iceman :
In other articles on liquidity trap economics, Krugman makes the case that
* government stimulus, in the liquidity trap, should be (approximately) sufficient to restore full employment
* overstimulating the economy is less of a problem than understimulating it, because traditional interest rate policy starts working again in the former case.
Errata “Krugman’s right, the GOP just wants to squish the economy”
Actually, I suspect Krugman didn’t actually say that, although he does vehemently wonder why they support policies that have that effect.
@Keshav Srinivasan
“Steve, this may be a silly question, but doesn’t the existence of demand imply the existence of utility? That is to say, by looking at a demand curve, aren’t you implicitly looking at how much people want the product or service demanded? So can’t you construct some measure of social welfare just by looking at aggregate demand?”
In the words of Chris Sims, there is no such thing as aggregate demand.
Demand cannot be aggregated, because the demand for different but equally priced baskets of goods can be different. It doesn’t only matter how much people want, it also matters what they want.
As a side note, we can only aggregate supply because of Says Law. If there is no general glut, then we can price things at the price they are exchanged. However, if there is a general glut, then the “excess supply” would never be sold anyway. There is no meaningful “price” for these excess goods so we can’t directly compare them to other goods and aggregate them together. This is one of the big problems I have with Keynesian models. They reject Says Law, but in order to make any sense at all, they require Says Law.
@Mike H.
“the model is wrong – but it (and its cousins) have made very good predictions about bond prices and inflation”
The IS-LM model has enough free parameters that its “predictions” are largely meaningless. It doesn’t predict so much as it is just naturally able to fit any data presented to it. As such, its not even a theory.
Steve needs to get on that Nobel nominating committee.
@Mike H: “there’s no negative impact on GDP from the borrowing”
Well that’s kinda the assumption at issue isn’t it? Understood that drawing a flat LM curve neatly provides for no-cost activist policy. But folks like new Nobelist Sargent have suggested even in the SR expectations can matter (even if you’re not “all in” on Lucas or Barro). One might argue that we’re currently at a pretty high watermark for such concerns to impact behavior. And when we eventually repay the debt out of future GDP something gets crowded out (hence the term “smoothing”), particularly if real resources were consumed in the process. So it still seems fair to ask just how much stimulus we’re looking at to “restore” employment, and “as much as it takes” isn’t a very satisfying answer to those reluctant to assume away all intertemporal trade-offs. I know Krugman believes the risks of “modest” inflation are overrated… which implies there are others who think differently, maybe out of a healthy skepticism over our ability to always achieve the proverbial soft landing.
I have no doubt a policy of borrowing and spending has positive utility for the government. But it is certainly not true that all individuals (or corporations) “want interest rates to stay low” — yet another pesky trade-off here for all would-be savers and investors (perhaps even a generational bias?).
@Mike H: “there’s no negative impact on GDP from the borrowing”
@iceman: Well that’s kinda the assumption at issue isn’t it?
No, it’s a conclusion from the model.
The question is whether or not to believe the model, and this doesn’t have to be an assumption either. Just look at all the models that made accurate predictions over the past 5 years, and pick the simplest one. You’ll find yourself with a Keynesian model.
These models imply correctly that there is no downside, and only upside, to having the government borrow heavily now and spend on infrastructure projects.
Your point that there are both borrowers and savers is taken, thank you. However, if the marginal change in interest rates wrt govt borrowing/spending is as close to 0 as IS/LM (and the data over the past 3 years) predicts, no matter how many people want interest rates to go up or down, we can neglect this term.
Let me write it out again :
utility = utility from change in GDP + utility from change in interest rates + utility from change in inflation.
if the govt spends and borrows in a liquidity trap (ie, now), GDP goes up, inflation remains unchanged, interest rates remain unchanged. Therefore,
* utility from change in GDP is positive
* utility from change in interest rates is 0, since they don’t change
* utility from change in inflation is 0, since it doesn’t change
Therefore,
total utility = positive.
….unless you want to argue that people *like* recessions.
Appreciate the exchange. Of course an “all in” Austrian might say we *need* recessions…but here are a few final thoughts you might find more relevant:
– The model doesn’t specify the shape of the LM curve; if you assume it’s flat your conclusions follow (within the “myopic” construct of that model).
– The main prediction I can recall is that $900 billion in stimulus would lower the UR to 8% (like 2 years ago), which wasn’t too accurate. This causes some to wonder why the normal fiscal policy levers might be pushing on a string, e.g. if the extra-ordinary level of national debt and/or “stock” decline in housing wealth may be impacting behavior. I recently read a research piece suggesting that this time lower-income people might not have the highest MPC because they were hardest hit by the housing bust and most focused on balance sheet repair.
– The normal concerns over crowding out or inflation are a little more forward-looking so the jury’s still out.
– The low interest rates are not purely exogenous but partly the result of activist (monetary) policy.
To try to tie this back to Landsburg’s initial point, it seems to me the reason a critique of an ostensibly utility-free model could cause such a brouhaha is that people are either implicitly bringing their own utility preferences to the table, or assuming a liquidity trap obviates the need for evaluating utility trade-offs at the moment. You may be focused specifically on the current cycle, however Krugman and DeLong are making a more general case which suggests the former is also at work. This would help to explain an otherwise curious observation: 1) PK and BD say one should use the simplest model that captures the essence of what you’re after; 2) Cowen says IS/LM adds variables which can lead the intuition astray (other models can derive “the most important points” without the “extra baggage”); 3) PK and BD blast this application of the principle as a step backward and even “tribal” (you knew there had to be an ad hominem swipe in there somewhere).
@iceman : you don’t have to “assume” the curve is horizontal.
Naturally, it generally slopes upward, since as money supply/income increases, people buy more securities, and interest rates rise.
However, it can’t maintain that same slope as money supply/income decreases, since it can’t cross the axis. If nominal interest rates reach zero, there’s no preference for bonds over cash. If nominal interest rates are negative, people will not buy bonds with cash.
Therefore, the LM curve must have a region where it’s almost horizontal.
You also don’t have to “assume” that the economy is currently in that region – that’s evident from the data on interest rates vs govt borrowing since 2008.
The main prediction of Keynesian models about the stimulus would gave been that “this level of stimulus will improve the economy by x% over what it otherwise would have been”. Any specific figure, eg 8%, for predicted unemployment rate (BTW – I didn’t see that figure – do you have a source?) would relied on accurate projections of the unemployment rate without the stimulus.
@Mike H: Still out there? I think the administration’s claim that the stimulus would lower the UR from 10% to 8% was pretty widely reported at the time. So not a positive data point, and while they can always say trends were worse than they realized, at some point any valid theory/model has to provide for some way of being disproven. On the simple correlation between govt borrowing and interest rates, I think that’s so confounded by other factors — e.g. it was likely negative during the 1980s – that historically it’s not been too useful for proving or disproving just about anything. What are the “accurate predictions over the last 5 years” you’re referring to? Models that foresaw an unprecedented collapse in the housing market? I thought the world had changed dramatically over the last 3 years alone.
No one disputes that within the construct of IS/LM, if/where LM is flat as you describe, doing something that (successfully) pushes out the IS curve could increase output without raising interest rates. To you that seems to have obviously “correct” real-world implications for current policy. But again, the tiff Landsburg alluded to suggests that some think that framework may overlook factors that could help explain why, at least to some, the latest stimulus wasn’t more effective in practice (and why therefore doing more might not be a panacea). E.g., again if the initial debt/deficit position is sufficiently bad, the impact on expectations could offset short-term spending efforts; or in terms of the model, the IS curve might not be as easy to push out as you thought. (Or can the IS curve be pretty flat too?) From there the dispute seems to be over whether this means we need to add more variables, or if at least at the moment a “simpler” approach might yield better intuition and prescriptions. Seems like a legitimate debate to me.
I would also be a little concerned about taking a position that seems to suggest that any environment of low interest rates whatever the source (e.g. aggressive Fed policy irrespective of any underlying private hoarding?) means we can increase spending costlessly just because IS/LM tells us so. (Note I’m not totally knocking the model as never being useful as a first approximation; once upon a time I used it as the basis for a thesis about budget vs. trade deficits.)