Modern econometrics literature tells us little about effect of monetary policy tightening
Modern econometrics literature focuses on shocks to interest rates, not changes in interest rates. So it is hard to know what are the effects of such changes on the economy.
John Cochrane has great new post on empirics of effects of monetary policy shocks, in which he partly touches up on a topic I was meaning to write about: measuring of effect of monetary policy tightening. John’s blog post is really a critical discussion of the empirical literature that tries to establish the effect of interest rates on inflation. There is many great points in that discussion[1], and I don’t aim anywhere close to his ambition in terms of topic. Instead, there is a more specific point I wish to make: modern econometrics literature is not really studying the effects of monetary policy tightening, it is studying the effect of monetary shocks. And that is a very different thing.
Overall changes vs. unexplained changes
John starts with explaining that all that the empirical literature is trying to answer is question “What happens to the economy if the Fed raises interest rates unexpectedly, for no particular reason at all?”. At first glance this seems like a reasonable empirical question to ask, especially if you have academic background. But only at first glance. When you start thinking about it more, it becomes much less clear whether this question is the question we should be asking.
To appreciate this, notice how this question is being answered. As John points out, basically the literature takes some model which decomposes federal funds rate into function of some economic variables and a disturbance:
interest rate = (#) output + (#) inflation + (#) other variables + disturbance.
The point is that what is studied is the effect of (unexpected) movement in the disturbance, which is what we call monetary policy shocks. In other words, what is studied is the effect of change in monetary policy after accounting for reaction to the economic variables included in the model. This is different from studying the effect of change in monetary policy (not after accounting for…).
And this distinction is important. I think the key question right now is “What will be the effect of monetary policy tightening on inflation and output going forward?”. If you would use the approach of “after accounting for reaction to…” then you would focus on the disturbances, i.e. the part not explained by other macro variables. How do these disturbances look over last 2 years? Unsurprisingly, if one uses the standard Taylor rule, the disturbances are hugely negative, which really just reflects that policy rates have failed to keep up with inflation during 2021 and 2022:
So if one would identify monetary policy with these disturbances then one would conclude that monetary policy has been stimulative over last two years.[2]
The problem is that this is not what normal people mean by the effect of monetary policy tightening on inflation and output. Rather than answering what is the effect of unexplained increase in interest rates, the question people typically want to answer is what is the effect of overall increase in interest rates from their initial values to their current values. In other words, we are not interested only in the effect of the shocks, but also in the increase in reaction to other economic variables!
This is something that I struggled with when I for the first time encountered modern literature on monetary policy effects during my student years. Not knowing much, I thought that the literature is trying to answer the questions I had in mind, like “What was the effect of the 2004-2006 tightening on output and inflation (and house prices etc.)?”. I was surprised that most of the literature does not study such effects of changes in policy rates. Eventually, when I learnt more, it all made sense why, but I still feel that this is somewhat unappreciated thing both in academia and outside of it.
Solution via benchmarking
Are the questions really that different? Or is it possible that the distinction does not matter that much? In some way it does not, because in a sense it is a matter of benchmarking. Is your benchmark the value of monetary policy rates implied by your model (e.g. the Taylor rule as in chart above)? Or is your benchmark the level of policy rates before policy intervention?[3]
This was brought home for me by chart from ECB’s economic bulletin, which tried to answer the question of effect of the current tightening on output and inflation[4]:
Notice that hidden in the footnote below the chart is the key aspect of the exercise: the effects are calculated as difference from benchmark that assumes that rates followed path expected in December 2021 (which was almost flat). In one way this is logical: if you want to know by how much output and inflation are (and will be) lower thanks to the totality of the action of the ECB, you need to compare forecast in absence of the action and forecast with the action. But on the other hand, most of the action is a reaction to high inflation, so it is a question whether that constitutes “action” by central bank or not. Effectively, there is a choice one needs to be made regarding what is monetary policy action and what is just reflection of other stuff.
But even the use of different benchmarks does not solve our problem completely. Consider a standard approach that estimates effect of interest rate shock on output, say something like 0.5% per 1 percentage point increase in interest rates. In principle you could conclude that 4 percentage points increase in interest rates leads to output being lower by 2%. But to arrive at such conclusion you need to be willing to assume that unexpected and unexplained change in policy rates has the same effects as expected and explained change. And that is a pretty strong assumption to make, especially because there is no empirical literature that tries to demonstrate this (and neither it is very sound theoretically). So you are left with conclusion that the vast modern econometric literature either does not address at all, or at best addresses only very indirectly, what I would consider the key macroeconomic policy question.
Answering question we can answered or answering question that we are actually interest in?
At this point you might wonder why the hack is modern econometrics answering different question, and not the question most are interested in? There is of course a very good reason: Only by studying shocks can we really avoid the pitfalls of endogeneity caused by reverse causality and simultaneity. So the question is chosen in order for us to be able to get a correct answer. Getting a correct answer to the question of “what is the effect of the totality of increase in interest rates?” is pretty much hopeless endeavor due to the endogeneity problems, I believe. So in a sense, this whole literature is a nice example of distinction between answering questions we can answer, and answering question that we are actually interest in. Ideally these two coincide, but in reality (especially for macroeconomics) they often do not. And sadly academia is overly focused on the former, and I am afraid that for the most part it does not even realize it.
[1] This should not be a surprise: I think that John has unapparelled ability to take some extensive economic literature, distill the key points and put them in digestible text. Or at least I, personally, despite having a PhD in economics, learn a lot from John’s posts. This post is actually not a great example of this unique ability, given its length and complexity (and bit of a rambling nature). Much better example was his post from last month on heterogenous-agent models.
[2] Indeed, some argue this is the case, but then you are left with the puzzle why did inflation come down and growth was relatively subdued if monetary policy was super stimulative. I actually argued on Twitter that the fact that many people criticized or even mocked central banks for claiming that they will defeat inflation with negative real interest rates is best argument for last years’ inflation actually being transitory, but that is different discussion.
[3] This is closely related to John’s discussion of shocks vs rule changes: a rule change is in a sense a change in benchmark.
[4] This graph is also nice illustration of Johns skepticism about our knowledge. Just notice the huge differences in the answers provided by three state-of-the-art models used by a central bank, with cumulative effect on level of output between 5% and 13%. That is huge variation. This lack of consensus – and the fact that ECB clearly does not consider either model’s answer superior to other models’ answer – really speaks volumes about the uncertainty we face even about the most basic question.
Moreover, I would add that the three answers individually and together do not seem to pass a basic smell test, as they suggest that in absence of the tightening the eurozone economy would be going through unprecedented boom. Hard to believe that…