Blanchard on DSGE

Olivier Blanchard, former director of the IMF’s research department, has written a short critical piece about DSGE models. Forget all the econblog reaction that essentially says he has been too kind: DSGE completely dominates academic macroeconomics, and there is no way that all these academics are going to suddenly decide this research programme is a waste of time. (I happen to think Blanchard is right that it isn’t a waste of time.) What is at issue is not the existence of DSGE models, but their hegemony.

One of Blanchard’s recommendations is that DSGE “has to become less imperialistic. Or, perhaps more fairly, the profession (and again, this is a note to the editors of the major journals) must realize that different model types are needed for different tasks.” The most important part of that sentence is the bit in brackets. He talks about a distinction between fully microfounded models and ‘policy models’. The latter used to be called Structural Econometric Models (SEMs), and they are the type of model that Lucas and Sargent famously attacked.

These SEMs have survived as the core model used in many important policy institutions (except for the Bank of England) for good reason, but DSGE trained academics have followed Lucas and Sargent as viewing these as not ‘proper macroeconomics’. Their reasoning is simply wrong, as I discuss here. As Blanchard notes, it is the editors of top journals that need to realise this, and stop insisting that all aggregate models have to be microfounded. The moment they allow space for eclecticism, then academics will be able to choose which methods they use.

Blanchard has one other ‘note for editors’ remark, and it also gets to the heart of the problem with today’s macroeconomics. He writes “Not every discussion of a new mechanism should be required to come with a complete general equilibrium closure.” The example he discusses, and which I have also used in this context, is consumption. DSGE modellers have of course often departed from the simple Euler equation, but I suspect the ways they have done this (rule of thumb consumers, habits) reflect analytical convenience rather than realism.

What sometimes seems to be missing in macro nowadays is a connection between people working on partial equilibrium analysis (like consumption) and general equilibrium modellers. Top journal editors’ preference for the latter means that the former is less highly valued. In my view this has already had important costs. I argue that the failure to take seriously the strong evidence about the importance of changes in credit availability for consumption played an important part in the inability of macroeconomics to adequately model the response to the financial crisis (for more discussion see here and here). Even if you do not accept that, the failure of most DSGE models to include any kind of precautionary saving behaviour does not seem right when DSGE has a monopoly in ‘proper modelling’. [1]

Criticism of the DSGE hegemony from those outside economics, from macroeconomists who are not part of it, or even from economic policymakers has had little impact on those all important journal editors up until now. Perhaps similar comments from one of the best macroeconomists in the world might.

[1] I discuss the reasons why this may have occurred in relation to Chris Carroll’s work here.

Reform and revolution in macroeconomics

Mainly for economists

Paul Romer has a few recent posts (start here, most recent here) where he tries to examine why the saltwater/freshwater divide in macroeconomics happened. A theme is that this cannot all be put down to New Classical economists wanting a revolution, and that a defensive/dismissive attitude from the traditional Keynesian status quo also had a lot to do with it.

I will leave others to discuss what Solow said or intended (see for example Robert Waldmann). However I have no doubt that many among the then Keynesian status quo did react in a defensive and dismissive way. They were, after all, on incredibly weak ground. That ground was not large econometric macromodels, but one single equation: the traditional Phillips curve. This had inflation at time t depending on expectations of inflation at time t, and the deviation of unemployment/output from its natural rate. Add rational expectations to that and you show that deviations from the natural rate are random, and Keynesian economics becomes irrelevant. As a result, too many Keynesian macroeconomists saw rational expectations (and therefore all things New Classical) as an existential threat, and reacted to that threat by attempting to rubbish rational expectations, rather than questioning the traditional Phillips curve. As a result, the status quo lost. [1]

We now know this defeat was temporary, because New Keynesians came along with their version of the Phillips curve and we got a new ‘synthesis’. But that took time, and you can describe what happened in the time in between in two ways. You could say that the New Classicals always had the goal of overthrowing (rather than improving) Keynesian economics, thought that they had succeeded, and simply ignored New Keynesian economics as a result. Or you could say that the initially unyielding reaction of traditional Keynesians created an adversarial way of doing things whose persistence Paul both deplores and is trying to explain. (I have no particular expertise on which story is nearer the truth. I went with the first in this post, but I’m happy to be persuaded by Paul and others that I was wrong.) In either case the idea is that if there had been more reform rather than revolution, things might have gone better for macroeconomics.

The point I want to discuss here is not about Keynesian economics, but about even more fundamental things: how evidence is treated in macroeconomics. You can think of the New Classical counter revolution as having two strands. The first involves Keynesian economics, and is the one everyone likes to talk about. But the second was perhaps even more important, at least to how academic macroeconomics is done. This was the microfoundations revolution, that brought us first RBC models and then DSGE models. As Paul writes:

“Lucas and Sargent were right in 1978 when they said that there was something wrong, fatally wrong, with large macro simulation models. Academic work on these models collapsed.”

The question I want to raise is whether for this strand as well, reform rather than revolution might have been better for macroeconomics.

First two points on the quote above from Paul. Of course not many academics worked directly on large macro simulation models at the time, but what a large number did do was either time series econometric work on individual equations that could be fed into these models, or analyse small aggregate models whose equations were not microfounded, but instead justified by an eclectic mix of theory and empirics. That work within academia did largely come to a halt, and was replaced by microfounded modelling.

Second, Lucas and Sargent’s critique was fatal in the sense of what academics subsequently did (and how they regarded these econometric simulation models), although they got a lot of help from Sims (1980). But it was not fatal in a more general sense. As Brad DeLong points out, these econometric simulation models survived both in the private and public sectors (in the US Fed, for example, or the UK OBR). In the UK they survived within the academic sector until the latter 1990s when academics helped kill them off.

I am not suggesting for one minute that these models are an adequate substitute for DSGE modelling. There is no doubt in my mind that DSGE modelling is a good way of doing macro theory, and I have learnt a lot from doing it myself. It is also obvious that there was a lot wrong with large econometric models in the 1970s. My question is whether it was right for academics to reject them completely, and much more importantly avoid the econometric work that academics once did that fed into them.

It is hard to get academic macroeconomists trained since the 1980s to address this question, because they have been taught that these models and techniques are fatally flawed because of the Lucas critique and identification problems. But DSGE models as a guide for policy are also fatally flawed because they are too simple. The unique property that DSGE models have is internal consistency. Take a DSGE model, and alter a few equations so that they fit the data much better, and you have what could be called a structural econometric model. It is internally inconsistent, but because it fits the data better it may be a better guide for policy.

What happened in the UK in the 1980s and 1990s is that structural econometric models evolved to minimise Lucas critique problems by incorporating rational expectations (and other New Classical ideas as well), and time series econometrics improved to deal with identification issues. If you like, you can say that structural econometric models became more like DSGE models, but where internal consistency was sacrificed when it proved clearly incompatible with the data.

These points are very difficult to get across to those brought up to believe that structural econometric models of the old fashioned kind are obsolete, and fatally flawed in a more fundamental sense. You will often be told that to forecast you can either use a DSGE model or some kind of (virtually) atheoretical VAR, or that policymakers have no alternative when doing policy analysis than to use a DSGE model. Both statements are simply wrong.

There is a deep irony here. At a time when academics doing other kinds of economics have done less theory and become more empirical, macroeconomics has gone in the opposite direction, adopting wholesale a methodology that prioritised the internal theoretical consistency of models above their ability to track the data. An alternative - where DSGE modelling informed and was informed by more traditional ways of doing macroeconomics - was possible, but the New Classical and microfoundations revolution cast that possibility aside.

Did this matter? Were there costs to this strand of the New Classical revolution?

Here is one answer. While it is nonsense to suggest that DSGE models cannot incorporate the financial sector or a financial crisis, academics tend to avoid addressing why some of the multitude of work now going on did not occur before the financial crisis. It is sometimes suggested that before the crisis there was no cause to do so. This is not true. Take consumption for example. Looking at the (non-filtered) time series for UK and US consumption, it is difficult to avoid attaching significant importance to the gradual evolution of credit conditions over the last two or three decades (see the references to work by Carroll and Muellbauer I give in this post). If this kind of work had received greater attention (which structural econometric modellers would almost certainly have done), that would have focused minds on why credit conditions changed, which in turn would have addressed issues involving the interaction between the real and financial sectors. If that had been done, macroeconomics might have been better prepared to examine the impact of the financial crisis.

It is not just Keynesian economics where reform rather than revolution might have been more productive as a consequence of Lucas and Sargent, 1979.

[1] The point is not whether expectations are generally rational or not. It is that any business cycle theory that depends on irrational inflation expectations appears improbable. Do we really believe business cycles would disappear if only inflation expectations were rational? PhDs of the 1970s and 1980s understood that, which is why most of them rejected the traditional Keynesian position. Also, as Paul Krugman points out, many Keynesian economists were happy to incorporate New Classical ideas. 

US savings behaviour, and empirical research strategies

In this post I want to look at a paper by Chris Carroll, Jiri Slacalek and Martin Sommer for two reasons. The first is for what the paper tells us about US consumption behaviour, and potentially consumption behaviour in any advanced economy. The second thing I want to use it for is as an example of different ways of doing empirical research in a microfoundations world.

The mainstay of modern macroeconomics is the consumption Euler equation, where consumption is proportional to the sum of financial wealth and human wealth, where human wealth is the discounted present value of future labour income. This model implies consumption aims to smooth out erratic movements in income through borrowing and saving. In this model periods of high saving can reflect periods of temporarily higher income, or temporarily high real interest rates. Adaptations of this model that are commonplace are to assume that some proportion of consumers are liquidity constrained, and therefore consume all their income, or that consumption is subject to ‘habits’, which generates additional inertia. This model with or without these adaptations is not very helpful in explaining why savings rose sharply in the Great Recession.

Rather more worrying is that this model is not very good at explaining US savings behaviour before the Great Recession either. As I noted here, US savings rates fell steadily for about twenty years from the early 1980. You might think that explaining such a large and important trend would be a sine qua non of any consumption function routinely used in macromodels, but you would be wrong. Consistency with the data is not the admissibility criteria for a microfounded macromodel.

The Carroll et al paper finds two explanations for the pre-recession trend and the increase in savings during the recession. The first is easier credit conditions, and the second is employment uncertainty. The mechanism through which both work is precautionary savings. If the risk increases that your income will fall sharply because you will lose your job, you need to build up some capital to act as a buffer. The easier credit is to obtain, the less precautionary savings you need.

The reason why precautionary savings represents a significant departure from the basic Euler equation model is intuitive. If you want to hold a certain amount of precautionary savings, you have a target for wealth. A wealth target pulls in the opposite direction to consumption smoothing. If you have a one-off increase in income, consumption smoothing says you should consume it very gradually, perhaps only consuming the interest. The marginal propensity to consume that extra income is tiny. But this leaves wealth higher for a very long time. If you have a wealth target, your marginal propensity to consume that additional income will be larger, perhaps a lot larger.

Now for the methodology part. These empirical results are in sections 3 and 4 of their paper. They call their empirical results in section 4  ‘reduced form’, because they come from a regression relating saving to wealth, credit constraints and expected unemployment. However the authors feel that this is not enough. In section 2 they discuss a structural theoretical model. Because modelling labour income uncertainty is very difficult, their microfounded model assumes that once someone becomes unemployed, they become unemployed forever. Section 5 then estimates this structural model.

The authors describe a number of reasons why directly estimating the structural model may be better than estimating the reduced form. But in order to get their structural model they have to make the highly unrealistic assumption noted above. The reduced form, on the other hand, does not have this assumption imposed on it. So I do not think we can say that the results in Section 5 are more or less interesting than those in Section 4, which is why both are interesting, and why both are included in the paper. There does not seem to be any compelling reason to elevate one above the other.

OK, a last  - perhaps wild - pair of questions. Is it the case that, compared to a few decades ago, there are far fewer papers in the top journals that simply try and explain historical time series for a single key macro aggregate (like consumption or saving)? If that is the case, is this due to the difficulties in getting microfounded models to fit, or something else?  

Data, Theory and Central Bank Models

As the Bank of England finally publishes [1] its new core model COMPASS, I have been thinking more about what the central bank’s core model should look like. (This post from Noah Smith I think reacts indirectly to the same event.) Here I will not talk in terms of the specification of individual equations, but the methodology on which the model is based.

In 2003, Adrian Pagan produced a report on modelling and forecasting at the Bank of England. It included the following diagram.

One interpretation is that the Bank has a fixed amount of resources available, and so this curve is a production possibility frontier. Although Pagan did not do so, we could also think of policymakers as having conventional preferences over these two goods: some balance between on the one hand knowing a forecast or policy is based on historical evidence and on the other that it makes sense in terms of how we think people behave.

I think there are two groups of macroeconomists who will feel that this diagram is nonsense. The first I might call idealists. They will argue that in any normal science data and theory go together – there is no trade-off. The second group, which is I will call purists, will recognise two points on this curve (DSGE and VARs), but deny that there is anything in between. I suspect most macroeconomists under 40 will fall into this group.

The purists cannot deny that it is possible to construct hybrid models that are an eclectic mix of some more informal theory and rather more estimation that DSGE models involve, but they will deny that they make any sense as models. They will argue that a model is either theoretically coherent or it is not – we cannot have degrees of theoretical coherence. In terms of theory, there are either DSGE models, or (almost certainly) incorrect models.

At the time Pagan wrote his report, the Bank had a hybrid model of sorts, but it was in the process of constructing BEQM, which was a combination of a DSGE core and a much more data based periphery. (I had a small role in the construction of both BEQM and its predecessor: I describe the structure of BEQM here.) It has now moved to COMPASS, which is a much more straightforward DSGE construct. However judgements can be imposed on COMPASS, reflecting a vast range of other information in the Bank’s suite of models, as well as inputs from more informal reasoning.

The existence of a suite of models that can help fashion judgements imposed on COMPASS may guard against large errors, but the type of model used as the core means of producing forecasts and policy advice remains significant. Unlike the idealists I recognise that there is a choice between data and theory coherence in social science, and unlike the purists I believe hybrid models are a valid alternative to DSGE models and VARs. So I think there is an optimum point on this frontier, and my guess is that DSGE models are not it. The basic reason I believe this reflects the need policymakers have to adjust reasonably quickly to new data and ideas, and I have argued this case in a previous post. 

Yet I suspect it will take a long time before central banks recognise this, because most macroeconomists are taught that such hybrid models are simply wrong. If you are one of those economists, probably the best way I can persuade you that this position is misguided is to read this paper from Chris Carroll. [2] It discusses Friedman’s account of the permanent income hypothesis (PIH). For many years graduate students have been taught that while PIH was a precursor to the intertemporal model that forms the basis of modern macro, Freidman’s suggestion that the marginal propensity to consume out of transitory income might be around a third, and that permanent income was more dependent on near future expectations than simple discounting would suggest, were unfortunate reflections of the fact that he didn’t do the optimisation problem formally.

Instead, Carroll suggests that the PIH may be reasonable approximation to how an optimising consumer might behave as they anticipate the inevitable credit constraints that come with old age. There is also a lot of empirical evidence that consumers do indeed quickly consume something like a third of unexpected temporary income shocks. In other words Friedman’s mix of theoretical ideas and empirical evidence would have done rather better at forecasting consumption than anything microfounded that has supplanted it. If that can be true for consumption, it could be true for every other macroeconomic relationship, and therefore a complete macromodel.

[1] A short rant about the attitude of the Bank of England to the outside world. COMPASS has been used for almost two years, yet it has only just been published. I have complained about this before, so let me just say this. To some senior officials at the Bank, this kind of lag makes perfect sense: let’s make sure the model is fit for purpose before exposing it to outside scrutiny. Although this may be optimal in terms of avoiding Bank embarrassment and hassle, it is obviously not optimal in terms of social welfare. The more people who look at the model, the sooner any problems may be uncovered.  I am now rather in favour of delegation in macroeconomics, but delegation must be accompanied by the maximum possible openness and accountability.

[2] This recently published interview covers some related themes, and is also well worth reading (HT Tim Taylor).

Handling complexity within microfoundations macro

In a previous post I looked at a paper by Carroll which suggested that the aggregate consumption function proposed by Friedman looked rather better than more modern intertemporal consumption theory might suggest, once you took the issue of precautionary saving seriously. The trouble was that to show this you had to run computer simulations, because the problem of income uncertainty was mathematically intractable. So how do you put the results of this finding into a microfounded model?

While I want to use the consumption and income uncertainty issue as an example of a more general problem, the example itself is very important. For a start, income uncertainty can change, and we have some evidence that its impact could be large. In addition, allowing for precautionary savings could make it a lot easier to understand important issues, like the role of liquidity constraints or balance sheet recessions.

I want to look at three responses to this kind of complexity, which I will call denial, computation and tricks. Denial is straightforward, but it is hardly a solution. I mention it only because I think that it is what often happens in practice when similar issues of complexity arise. I have called this elsewhere the streetlight problem, and suggested why it might have had unfortunate consequences in advancing our understanding of consumption and the recent recession.

Computation involves embracing not only the implications of the precautionary savings results, but also the methods used to obtain them as well. Instead of using computer simulations to investigate a particular partial equilibrium problem (how to optimally plan for income uncertainty), we put lots of similar problems together and use the same techniques to investigate general equilibrium macro issues, like optimal monetary policy.

This preserves the internal consistency of microfounded analysis. For example, we could obtain the optimal consumption plan for the consumer facing a particular parameterisation of income uncertainty. The central bank would then do its thing, which might include altering that income uncertainty. We then recompute the optimal consumption plan, and so on, until we get to a consistent solution.

We already have plenty of papers where optimal policy is not derived analytically but through simulation.(1) However these papers typically include microfounded equations for the model of the economy (the consumption function etc). The extension I am talking about here, in its purest form, is where nothing is analytically derived. Instead the ingredients are set out (objectives, constraints etc), and (aside from any technical details about computation) the numerical results are presented – there are no equations representing the behaviour of the aggregate economy.

I have no doubt that this approach represents a useful exercise, if robustness is investigated appropriately. Some of the very interesting comments to my earlier post did raise the question of verification, but while that is certainly an issue, I do not see it as a critical problem. But could this ever become the main way we do macroeconomics? In particular, if results from these kinds of black box exercises were not understandable in terms of simpler models or basic intuition, would we be prepared to accept them? I suspect they would be a complement to other forms of modelling rather than a replacement, and I think Nick Rowe agrees, but I may be wrong. It would be interesting to look at the experience in other fields, like Computable General Equilibrium models in international trade for example.

The third way forward is to find a microfoundations 'trick'. By this I mean a set up which can be solved analytically, but at the cost of realism or generality. Recently Carroll has done just that for precautionary saving, in a paper with Patrick Toche. In that model a representative consumer works, has some probability of becoming unemployed (the income uncertainty), and once unemployed can never be employed again until they die. The authors suggest that this set-up can capture a good deal of the behaviour that comes out of the computer simulations that Carroll discussed in his earlier paper.

I think Calvo contracts are a similar kind of trick. No one believes that firms plan on the basis that the probability of their prices changing is immutable, just as everyone knows that one spell of unemployment does not mean that you will never work again. In both cases they are a device that allows you to capture a feature of the real world in a tractable way.

However, these tricks do come at a cost, which is how certain we can be of their internal consistency. If we derive a labour supply and consumption function from the same intertemporal optimisation problem, we know these two equations are consistent with each other. We can mathematically prove it. Furthermore, we are content that the underlying parameters of that problem (impatience, the utility function) are independent of other parts of the model, like monetary policy. Now Noah Smith is right that this contentment is a judgement call, but it is a familiar call. With tricks like Calvo contracts, we cannot be that confident. This is something I hope to elaborate on in a subsequent post. 

This is not to suggest that these tricks are not useful – I have used Calvo contracts countless times. I think the model in Carroll and Toche is neat. It is instead to suggest that the methodological ground on which these models stand is rather shakier as a result of these tricks. We can never write ‘I can prove the model is internally consistent’, but just ‘I have some reasons for believing the model may be internally consistent’. Invariance to the Lucas critique becomes a much bigger judgement call.

There is another option that is implicit in Carroll’s original paper, but perhaps not a microfoundations option. We use computer simulations of the kind he presents to justify an aggregate consumption function of the kind Friedman suggested. Aggregate equations would be microfounded in this sense (there need be no reference to aggregate data), but they would not be formally (mathematically) derived. Now the big disadvantage of this approach is that there is no procedure to ensure the aggregate model is internally consistent. However, it might be much more understandable than the computation approach (we could see and potentially manipulate the equations of the aggregate model), and it could be much more realistic than using some trick. I would like to add it as a fourth possible justification for starting macro analysis with an aggregate model, where aggregate equations were justified by references to papers that simulated optimal consumer behaviour.  

(1) Simulation analysis can make use of mathematically derived first order conditions, so the distinction here is not black and white. There are probably two aspects to the distinction that are important for the point at hand, generality and transparency of analysis, with perhaps the latter being more important. My own thoughts on this are not as clear as I would like.