Judgement Calls and Microfoundation Tricks

Noah Smith has a really nice piece about when a microfounded model does or does not violate the Lucas critique. (See also this useful post from Bruegel.) Noah suggests that this comes down to a judgement call, which in turn introduced a potential ideological bias. I want to elaborate, and suggest another bias that may result: a bias towards simplicity. However I also want to suggest a further bias that potentially undercuts the methodological rationale behind microfoundations.

The key idea behind the Lucas critique was that models should be derived from ’deep’ parameters, like agents preferences or technological parameters. These were parameters that could reasonably be described as independent of the way monetary policy was conducted. The target of the Lucas critique was models where expectations formation was implicit in the model’s equations: even if you only half believed in rational expectations, changes in how monetary policy was done would change how expectations were formed, and therefore change those equations.

Noah argues that whether a parameter is independent of policy is essentially a judgement – our evidence base is not good enough to show us one way or another. Where you have judgement, various biases, including ideological views, can get in. I think this is right, but I also suspect the point will not bother most macroeconomists too much. They are – rightly or wrongly – fairly happy with treating preference parameters as exogenous, whereas treating expectations processes as independent of policy seems clearly problematic. (I appeal here to what most macroeconomists will think, and not what is right. In a recent post, for example, I argue that people’s preferences over which party to vote for are pretty malleable.)

However, once you go beyond the very simple RBC type models, the range of deep parameters extends beyond preferences and technology. To take the obvious example, if you want to have something useful to say about monetary policy, you need sticky prices, and these are usually microfounded in terms of Calvo contracts. The deep parameter in Calvo contracts is the probability that a firm’s price will change each period. Is this parameter independent of monetary policy?

The paper by Chari et al to which Noah refers puts the same point in a slightly different way. If the parameters of the model are not deep (independent), then the implied shocks to the model will not be ‘structural’ i.e. identifiable and independent of policy. They look at the shock processes typically included in New Keynesian models, and split them into two groups: potentially structural shocks, which include technology shocks, and dubiously structural shocks, which include mark-up shocks.

How do Chari et al decide which of these two categories shocks should be classified in? Noah would say judgement, whereas the authors would say microeconomic evidence. However this is not a debate I want to get into, interesting though it is. Instead I want to agree with Chari et al: the shocks in New Keynesian models are pretty dubious, and their deep parameters, like the Calvo parameter, are not obviously invariant to policy.

So why do New Keynesian models contain problematic features like Calvo contracts? Calvo contracts are a ‘trick’, by which I mean a device that allows you to get sticky prices into a model in a reasonably tractable way. Doing this job ‘properly’ might involve adding menu costs into the model, but this quickly gets intractable. So Calvo contracts are a trick that acts ‘as if’ firms were faced by menu costs. But whether this trick works – whether Calvo contracts really do mimic what an otherwise intractable model with menu costs would show – is inevitably a judgement call.

Because these judgement calls are problematic, there is a bias towards avoiding them by keeping the model simple. Here Chari et al are explicit. “One tradition, which we prefer, is to keep the model very simple, keep the number of parameters small and well-motivated by micro facts, and put up with the reality that such a model neither can nor should fit most aspects of the data. Such a model can still be very useful in clarifying how to think about policy.”

Suppose we do not follow this tradition, and instead attempt to explain more aspects of the data by building models that incorporate dubious judgement calls. I think we then have to recognise that these judgement calls will be influenced not just by the microeconomic evidence, or ideology as Noah suggests, but also on the need to have models that explain the real world. That is I believe a quite reasonable thing to do, but as Chari et al point out it does mean potentially compromising the internally consistency of the model (and therefore its immunity from the Lucas critique). As I have argued at length elsewhere (article, working paper), microfounded models have become dominant because they have let the evidence influence model structure through a back door. Individual equations may no longer be selected by directly confronting the data, but the data has influenced the judgemental calls involved in the microfoundations.