You just couldn't resist the title, could you?
Don't worry, I'm not going to be too harsh. After all, I'm rather fond of those who practise "applied microeconometrics" - especially lightly sautéed, with a little pepper and garlic. Sorry! Sorry!
The point that I want to make is a simple one, and I'll be brief.
How many seminars have you attended where the speaker has gone through the details of a formal microeconomic model, and then proceeded to a potentially interesting empirical application? And in how many cases was there a total "disconnect" between the theoretical model and the empirical model?
Hand up! Don't be shy! Wow - that's almost everyone!
In particular, how often have you been presented with an empirical application that's based on just a reduced-form model that essentially ignores the nuances of the theoretical model?
I'm not picking on applied microeconomic papers - really, I'm not! The same thing happens with some applied macroeconomics papers too. It's just that in the micro. case, there's often a much more detailed and rich theoretical model that just lends itself to some nice structural modelling. And then all we see is a regression of the logarithm of some variable on a couple of interesting covariates, and a bunch of controls - the details of which are frequently not even reported.
It just seems such a shame when this happens. It doesn't have to be this way. Put down your favourite economics journal for a moment, and take a look at some of the terrific papers that you'll find in journals such as Journal of Applied Econometrics, or Journal of Business and Economic Statistics.
To take a specific example, let's consider the modelling of consumer demand. Nobody in their right mind would try and get away with a paper that took a specific (direct) utility function; maximized it subject to a budget constraint; derived the associated system of demand equations; and then added an application in which the logarithm of the quantity of each good was regressed in turn on the logarithms of income and all of the relative prices!
Here, the application has nothing to do with the theoretical part of the study, and it ignores the rich set of cross-equation restrictions on the parameters that are implied by that theory. Yet, this is analogous to what we see in far too many papers!
Why on earth do people do this? And why do we continue to sit there in seminars and let them get away with it?
Microeconomic theory can provide us with some really interesting and challenging empirical opportunities, so let's back to rising to those challenges instead of muttering "reduced-form model".
In short, let's put the "ECON" back into microeconometrics!
How many seminars have you attended where the speaker has gone through the details of a formal microeconomic model, and then proceeded to a potentially interesting empirical application? And in how many cases was there a total "disconnect" between the theoretical model and the empirical model?
Hand up! Don't be shy! Wow - that's almost everyone!
In particular, how often have you been presented with an empirical application that's based on just a reduced-form model that essentially ignores the nuances of the theoretical model?
I'm not picking on applied microeconomic papers - really, I'm not! The same thing happens with some applied macroeconomics papers too. It's just that in the micro. case, there's often a much more detailed and rich theoretical model that just lends itself to some nice structural modelling. And then all we see is a regression of the logarithm of some variable on a couple of interesting covariates, and a bunch of controls - the details of which are frequently not even reported.
It just seems such a shame when this happens. It doesn't have to be this way. Put down your favourite economics journal for a moment, and take a look at some of the terrific papers that you'll find in journals such as Journal of Applied Econometrics, or Journal of Business and Economic Statistics.
To take a specific example, let's consider the modelling of consumer demand. Nobody in their right mind would try and get away with a paper that took a specific (direct) utility function; maximized it subject to a budget constraint; derived the associated system of demand equations; and then added an application in which the logarithm of the quantity of each good was regressed in turn on the logarithms of income and all of the relative prices!
Here, the application has nothing to do with the theoretical part of the study, and it ignores the rich set of cross-equation restrictions on the parameters that are implied by that theory. Yet, this is analogous to what we see in far too many papers!
Why on earth do people do this? And why do we continue to sit there in seminars and let them get away with it?
Microeconomic theory can provide us with some really interesting and challenging empirical opportunities, so let's back to rising to those challenges instead of muttering "reduced-form model".
In short, let's put the "ECON" back into microeconometrics!
© 2013, David E. Giles
Can you point to a journal article that is a paragon of empirical microeconometrics?
ReplyDeleteJohn - what about Deaton & Muellbauer: http://www.jstor.org/stable/1805222?seq=8
DeleteBest post in the blogosphere I've read in a long time. I've started to think that in some fields people have completely given up on economic theory and become nothing more than applied statisticians. I love a good paper that carefully ties together an economic theory and an estimation strategy. IO people still seem to do that.
ReplyDeleteYou're right I couldn't resist the title.
ReplyDeleteI do hope you are gaining readers from down under!!
I have a question similar to John's but opposite in direction. Can you point to one or more journal articles that are "worst offenders" in this regard?
ReplyDeleteBen - I'd prefer not to point the finger at individuals. I think we all now it when we see it though.
DeleteHi, Professor, thanks for your response. While it would be helpful for those of your readers, I included, who come from a statistics as opposed to economics or econometrics background who do not "see it" and therefore do not "know it when we see it", I understand your desire not to "point fingers".
DeleteBen - thanks for the comment. Let me give you an analogy. If you look back at my post on estimating an Euler equation using GMM (http://davegiles.blogspot.ca/2013/05/estimating-euler-equation-using-gmm.html) you'll see that the problem is set up as an inter-temporal optimization problem. The solution is in the form of a specific non-linear implicit equation, whose parameters have interesting interpretations.
DeleteNow, suppose we went through the optimization, derived the Euler equation, and then said something like:
"Let's look at an empirical application of this. Clearly the theory tells us that consumption depends on the interest rate. Let's fit an OLS regression that explains Log(C(t)) in terms of r(t)."
This is the sort of thing I'm complaining about in a lot of empirical microeconomic papers. There's a total "disconnect" between the theoretical section of the paper and the empirical application. The theory section is essentially "window dressing" for a very uninspiring piece of empirical work.
Hope this helps!
DG
Hi, Professor, this is supremely helpful. Thank you. Let me push this a bit further. Suppose I, as a theory-aware empirical researcher, set up the inter-temporal optimization problem correctly and estimate the appropriate non-linear equation. Suppose I also fit the naive OLS model. Now, suppose after fitting the two models, I do some model evaluation and find that the latter model appears to fit the data better. What do I do now?
DeleteI ask this bc I often see theory-driven models being estimate but their fit never being assessed. I think you would agree that assessing fit is important.
Thank you!
Ben - I agree totally. And often there are competing theories, some of which may be "nested" (i.e., one is a restricted version of another). In part, it may come down to the motivation for estimating the model in the first place. Take your example - if the motivation is short-term forecasting, then the naive model may be your choice, especially if it seems to stand up well over sub-samples, etc. On the other hand, if your interest lies in getting a measure of the structural parameters (so you can address policy questions), then the theory-driven model would be what you need. But I'd still want it to fit the data well, and of course I'd want it to pass other model specification tests.
DeleteAs you'll have guessed by now, my concern is with theory used purely as "window dressing"! :-)
This makes great sense and this conversation has been quite illuminating. Thank you, Prof.
ReplyDeleteYou were right, I couldn't resist the title. But I also can't think of any examples of what you're talking about. I'm surprised that the applied model would be the less nuanced, since case-specific details are available.
ReplyDelete