I like to criticize financies, financial regulators and fresh water economists. I should defend something for once. It is easy to criticize. So I ask to what extent has the IS-LM-Phillips curve model been proven false. Also what useful insights can one gain from the IS-LM-Phillips curve model.
A long long trip down memory lane after the jump.
Update: I honestly just found this article. Click and hope you have access to Jstor.
Warning. I have really nothing to say which Krugman hasn’t written already at his blog. I realized this after I had typed for a long long time and had proudly finished my little essay.
I am provoked by John Cochrane who said
“It’s not part of what anybody has taught graduate students since the 1960s,” Cochrane said. “They are fairy tales that have been proved false. It is very comforting in times of stress to go back to the fairy tales we heard as children but it doesn’t make them less false.”
Just before passing on, I note that Cochrane’s first claim is false. I was exposed in graduate school to an IS-LM model in 1985 (by prof. Larry Summers). Also in 1985 I was a research assistant working on a research paper based on a modified IS-LM model. The authors were Larry Summers and N. Gregory Mankiw. Cochrane’s intellectual history is incorrect. Also, he seems to assume that the fact that something is removed from graduate economics programs is strong evidence that it has been proved false. I think that changes in macroeconomic research have a lot to do with intellectual fashion and little to do with evidence. Thus I am interested in the evidence that the IS-LM-Phillips curve model is false. I think that Cochrane has not thought about which of the 3 blocks has been proven false and about which has recently been assumed to be a useful approximation to reality by, for example, Larry Summers.
To define terms, I assert that a logically consistent theory can be proved false only by facts. There is no doubt that the IS-LM model is ad hoc as in developed during the great depression as part of an effort to understand what was going on. The assumptions at the base of the model were, roughly, those needed so that one could have a depression and so that capitalism could be saved by a jump start. I recall a story I once heard from Stuart Holland — a student of Hicks — about a tutorial with John Hicks where Hicks opened the discussion by asking “What’s wrong with the IS-LM model” As student was shocked and noted that Hicks was the author of the IS-LM model. Hicks replied that it was “a doodle.” Not a model taken seriously by the doodler. A model taken very seriously right now. Something funny happened between 1937 and 2009.
OK starting at the end, the equation which dominated the macroeconomic policy debate in the 60s was the Phillips curve, which, as the name suggests, is not a doodle drawn by John Hicks. It came much later (1961 I think) and was the empirical observation that over roughly 100 years a simple relationship between unemployment and inflation held in UK data.
It was clear at the time that this was not a social law — clearly Germany did not stay on a Phillips curve during the hyperinflation. Milton Friedman pointed out that the pattern made no sense unless one assumed that expected inflation was constant and not affected by actual inflation. He predicted that if there were steady inflation, then the Phillips curve would shift out. He was right.
Hard core old Keynesians shifted to an expectations augmented Phillips curve with adaptive expectations. Lucas, among others, noted that the assumption of adaptive expectations was no more plausible than the assumption of constant expected inflation. In 1973, he predicted that the expectations augmented Phillips curve would shift out. Then there was an oil shock. The debate shifted from pro and anti Phillips to one between Muth supply function afficionados and “New Keynesians” who attempted to base the claim that there were nominal rigidities on analysis of optimizing behavior.
The Phillips curve ceased to be a topic of much academic research. Adaptive expectations augmented Phillips curves contintued to be the basis of Macroeconomic policy. The debate between academic macroeconomists ceased to be relevant to policy makers. The hope was that this was a transitional problem. The new models were very new — examples more that possibly realistic approximations to the economy. The hope was that with more work on the foundations new valid useful models would be developed in around 30 years. Actually this hope was expressed in around 1978. The new useful models are not on the horizon. This is not my topic today.
For my purposes, the point is that the work of Phillips Sr became an example of what not to do (in contrast the work of his son P.C.B. Phillips became an example of how to set the record for most papers published in “Econometrica” but that sure isn’t my topic today). Phillips Sr was discussed in history of thought and taught to undergraduates who, it was assumed, couldn’t handle the math of the new models. Further work on the Phillips curve by academics was devoted to getting nice illustrations for undergraduate text books (hey macroeconomists respond to economic incentives and the economic incentives say “write an undergraduate textbook with nice illustrations”). Embarrassingly, the history of thought and make a nice picture version of the Phillips curve, officially called “The Splitting Model” fit the data. Thus an ad hoc model first presented in an undergraduate text book fit the data out of sample. This caused some amusement but had no effect on the academic debate on macroeconomics.
It also has no role in the current policy debate. I think the reason is simple — the old argument against Keynesian stimulus was “that will just cause inflation.” It has been noticed that deflation is a disaster. It is trivial to understand why deflation is a disaster under the assumption that economic agents have rational expectations and markets clear and whatever you want. The implied inflation forecast from the market for TIPPS and regular treasury bills is zero. Just causing inflation would be pretty good right now.
Now the failure of the Phillips curve can’t imply that the IS-LM model is proved false. It just means that one of the inputs to the IS-LM model is the price level and the model only gives a prediction about real GNP if the price level is determined by something else (say the interaction of IS_LM and Phillips curve) or in the very special case in which a change in the price level other things equal implies zero change in GNP.
The price level appears in the LM curve which gives money demand as a function of nominal GNP and the nominal interest rate. The LM curve is a hypothesis about money balances, GNP and the nominal interest rate. It has been overwhelmingly rejected by the data. Worked great up until 1981, then totally failed. Efforts were made to modify the money demand equation so that it fit the new data. Of course it is possible in year t to fit money demand up until year t. Predictions for t+1 were way off. This held for t going from 1981 to 1986 or 1986 then economist gave up.
At least one Undergraduate Macroeconomics textbook doesn’t even mention the LM curve. Monetary policy is discussed via the assumption that the FED controls the federal funds rate somehow (don’t worry your little heads as to how just note they declare a target and they hit the target). So the model becomes the IS_BB model (BB stands for Ben Bernanke and I sure am not going to talk about the open economy).
Now the fact is that policy makers (especially including BB) rely on the IS-BB adaptive expectations augmented Phillips curve model. If it is a fairy tail which has been proved wrong, we have been counting on it continuously and not just when panicked.
There has been a very broad consensus that it is better to stabilize using monetary policy than fiscal policy. This does not imply that economists agreed that fiscal stimulus doesn’t work. It does not imply that economists agree that the IS part of the model is dead wrong. I have no doubt that it had rather a lot to do with perceptions of the relative ability of Paul Volker and Alan Greenspan compared to Ronald Reagan, Newt Gingrich and George W. Bush Jr. I mean there were polite non ad hominem explanations for the choice, but I don’t believe them.
Now monetary policy is pedal to the metal. The target interest rate is effectively zero. None of the arguments which supported the consensus that monetary stabilization policy is superior to fiscal stabilization policy has any relevance. Ben Bernanke is real smart. He says a fiscal stimulus is needed.
Given that the safe short term nominal interest rate can’t fall any further, Hicks’s doodle says we either have to shift the IS curve to the right (higher GNP for a given real interest rate) or cause the inflation rate to increase (lower real interest rate for given nominal interest rate). Phillips and Muth would agree that they way to do either is what is called fiscal stimulus. All that is needed is that fiscal stimulus increases nominal aggregate demand.
OK so what about criticisms of the IS curve. They aren’t very relevant to the current debate among academic macroeconomists either. One is obvious, the model doesn’t keep track of the federal debt. Needless to say, people haven’t forgotten about the debt. Most economists agree that it would be better if it were lower (even the former proponents of starving the beast have noticed that debt doesn’t reduce the beast’s appetite).
Another is that the old Keynesian model of consumption really really makes no sense and doesn’t fit the data. All existing models which fit the data suggest that national debt will depress consumption and people forecast higher taxes in the future. This can, in the extreme case called Ricardian equivalence, imply that tax cuts have no effect on nominal aggregate demand. It does not imply that increased public spending has no effect on aggregate demand. Krugman presented a model with Ricardian equivalence which implies a public spending muliplier of 1.
Becker and Murphy (a Nobel laureate and the economist whose intelligence has never been rated, in my presence, below tied for first) argue that a multiplier lower than one is more plausible. As far as I can tell, their argument is valid exactly to the extent in which private sector employment can be reduced in 2 ways
1) there are job vacancies — a firm which is trying to hire a worker hasn’t found a suitable worker yet
2) firms don’t post vacancies (start trying to hire a worker) because of the expected cost of finding one.
I don’t see how they get to 50% crowding out that way. I’d be interested in a survey of employers asking how important those 2 factors are right now.
It definitely is *not* enough that increased public employment will drive up private sector nominal wages. That does not crowd out private sector employment except by driving up the nominal interest rate.
In any case, models which imply Ricardian equivalence are rejected by the data. For example, cost of living increases in Social Security old age pensions are associated with increased aggregate consumption. This is pretty much a direct test of the claim that taxes and transfers don’t affect consumption. The increases are common so the estimates are fairly precise. Models with Ricardian equivalence have been popular for a while. However, the idea that current income affects consumption more than expectable discounted future income has not been proved false. If anything it has been proved true. The debate continues (the debates always continue) but the null required for Ricardian equivalence is usually rejected by papers written by people on both sides (the only exception I know of was a regression pretty much of consumption growth on consumption growth times an almost constant and other things which found an insignificant coefficient on the other things).
Here, as usual, there is something which fresh water economists think is proven to be true on the grounds that there might be something wrong with the apparent proof that it is false.
Now this doesn’t mean that the simple model of consumption in the IS-LM model is valid, but it does mean that a model which fits the facts (say the model due to Campbell and Mankiw) has pretty much the same implications for fiscal policy.
OK so I said closed economy so I have one topic left — investment. In the original IS-LM model, investment depended on GNP and the real interest rate. In the data it is well fit by the flexible accelerator which just says that it increases in GDP growth and decreases in the real interest rate. There are modern micro based models of investment. However, they don’t fit the data as well as the flexible accelerator. The dependence of investment on GNP increases multipliers. This does not depend on irrationality, nominal rigidity (at least the accelerator part) or Ricardian non equivalence.
The dependence on the real interest rate often makes the fiscal multiplier zero (the FED can prevent the Treasury from stimulating the economy if it so chooses) but given current monetary policy it implies that increased nominal demand which causes increased inflation (reduced deflation) will cause increased real GNP.
My trip down memory lane causes me to conclude that facts not known to Hicks as he was doodling do not reduce the relevance of his doodle to policy. The only point is that policy makers who are worried about the deficit should be told that spending gives more stimulus bang for the deficit buck than tax cuts. This was true according to Hicks and further evidence on consumption suggests that it is more true than Hicks would have guessed.
I admit to the so patient reader that I have no criticisms of Krugman and nothing original to add to what he said. In fact, the post is maybe a contribution to the history of really recent thought as, topic by topic, I might help people understand why Krugman believes what he believes.
Update: However, I don’t mix Indian, Japanese, and Roman metaphors “Tobin was never a guru in the way Milton Friedman was; he never had legions of Samurai ready to spring to the defense of his theories,”