Archive | economics RSS for this section

Econometrics, economic theory and reduced forms

The following post is a slightly altered version of a column I’ve written for our student newspaper (see page 8 of the November 2011 issue of Nekst). In Tilburg, we have a separate department of econometrics, which I am a member of.

Wikipedia says econometrics is the discipline that “aim[s] to give content to economic relations”, citing the New Palgrave dictionary in economics. A sub-discipline of it is theoretical econometrics, which is concerned with statistical properties of econometric procedures, among other things. When people say econometrics in the Netherlands, this is what they mean. But actually, econometrics is much broader.

The most prestigious journal of our field is Econometrica, and when you visit the website looking for the “aims and scope” of it, then you will find that it “promotes studies that aim at the unification of the theoretical-quantitative and the empirical-quantitative approach to economic problems and that are penetrated by constructive and rigorous thinking” (emphasis added; you find a similar statement for the Journal of Econometrics). Interestingly, the ultimate goal is given here, namely to study economic problems. Studying statistical properties of estimators is part of it, but ultimately we want to answer questions that are interesting in the eyes of economists.

This brings me to the general approach in economics. Most economists study economic problems by means of models. An economic model is an abstraction of the world, which is made to focus on just a few aspects of human behavior and interaction. It is phrased in terms of assumptions, and one can think of it as a tale. We tell a story, and thereby hopefully learn about the big picture. Econometrics is of course concerned with the empirical side to this.

When you read an applied article in the Quarterly Journal of Economics, then you will often see that people exploit, in one way or another, exogenous variation to learn something about the average causal effect this variation has on some outcome. They then call this a reduced form approach, because they don’t estimate what the mechanism is through which something has an effect on something else. But they then move on and interpret their findings, having particular mechanisms or a class of models in mind. For example, when interpreting findings on unemployment duration, they will interpret their findings against the background of job search theory. Traditionally, there is a heated debate between followers of this approach, and followers of the so-called structural approach to econometrics. The structural approach is to spell out an economic model, and to estimate parameters of that model. Having the estimated parameters in hand, one can simulate how people would react to a policy change such as a tax reform that has not taken place. In a recent issue of the Journal of Econometrics, Michael Keane propagates the latter, in a somewhat provocative article.

What may seem confusing in this context is that every structural model has  so-called “reduced forms”, which one gets by solving the model and expressing some of the variables as functions of the other variables. Under the right assumptions, this yields an equation that is linear in parameters, and that can also be estimated. Then, one estimates structural parameters from a reduced form equation. But these are often not the same reduced forms people have in mind when they read the Quarterly Journal of Economics. So, one way to confuse them is to ask what the equation they are looking at is a reduced form of (but many of the authors will actually have a good answer to this question). In any case, Tilburg does not take sides in this debate. There are people who pursue the reduced-form approach, and people who pursue the structural approach, also within our department. The latter have recently formed a group, and you can check their website in case you are curious. In our seminar series, you can see the whole variety of things econometrics has to offer, and you are of course welcome to attend.


Starting to work on a Ph.D. thesis

I would also like to use this Blog to share some advice to Ph.D. students. Let’s start with advice for Ph.D. students who have just completed the first two years of courses (small aside: In some European countries, the first two years of a Ph.D. program are referred to as the “Research Master” or “MPhil” Phase and after that one starts in year one; in the U.S. and other countries, one would call this year three).

So, it’s August now and you are about to get started with the thesis phase. I would suggest that you first think about where you want to be by the end of the Ph.D. Of course, you want to have a degree, but you also want to have a job. And the idea of a Ph.D. program is to prepare you for a job as an academic. For that reason, I find it helpful to already now read Cawley’s guide on the academic job market, available at the AEA job market website. There, you will also find other interesting papers that look at this question from a different angle.

The most important thing to realize is that when you want to have a job on September 1, then you need a single-authored job market paper in November of the previous year.

Next, I suggest you do some more preparatory reading that is related to the actual activity of doing research. I personally liked the book “How to write a lot“, because it does not only help you to overcome the writer’s block, but also gives you advice on how to organize your day so that you are as productive as possible. I’ve also heard good things about the book “Writing your dissertation in 15 minutes a day“.

It’s also time to acquire some software skills. I suggest to look into LyX for text processing, programs to organize your bibliography files (such as JabRef) and to have a look at R, as more and more people seem to use it, especially when they deal with big data sets. Moreover, get your IT environment in good shape. Buy a tablet, install all necessary apps on it (most importantly a PDF reader with annotation function, such as Goodreader). And come up with a good file structure on your hard disk so that you find things. I personally have everything in my Dropbox. This makes it easy to share folders and to synchronize files between your devices. And you automatically back up your files. You can also use any other such service, such as the SURFdrive here at Tilburg University.

Generally, get everything out of the way so that you can get started. All this can be done in one week. Or even less time.

Next, start thinking about research ideas. An idea is a question you would like to answer and that you think one can answer (it doesn’t matter if you don’t know how exactly). Most research ideas won’t work out, but some of them will. Compose a list of your 10 best ideas. Then, take the most promising one and ask yourself how the ideal data set would look like with which you could answer the question. Think about the ideal natural experiment you would like to exploit. And do a google (scholar) search on the topic. Don’t be disappointed if somebody else has already (tried to) answer the question. Instead, ask yourself whether you think it’s possible to give a much better answer.

Actually, you should actually start thinking about ideas already in the second year, before picking an adviser and starting the thesis phase. From then onward, think about new ideas all the time. Discuss them with your classmates and also with faculty. More junior faculty are usually very approachable, while the senior ones are more busy. But you should also talk to the more senior ones about your work. Try to make an appointment if necessary. This could be part of the process of looking for a good match for your adviser(s). Refine your list by getting rid of the less promising ideas and replacing them with more promising ones. This list will be helpful when somebody asks you the important question “What are you working on?”. And many people will ask that question.

Talking to others about will probably lead to a joint project at some point. Sometimes, it is very helpful to start a research project with a junior faculty member, preferably if he or she will likely be on your dissertation committee or your advisor. This will foster learning by doing, which will help you for your future projects. At the beginning you will think that you have a lot of stupid questions and make a lot of stupid mistakes. But you will learn that this is just part of the thought process that (sometimes) leads to insights that are all but stupid. And this is what counts.

But will all this, keep in mind that your most important goal is to have a single-authored job market paper at the end of your dissertation phase. Start talking to your advisor about this already after a few months.

Join a group such as the structural econometrics group in Tilburg. Here, ideas are discussed almost every week. Check the seminar schedules and make a plan which seminars to attend in the upcoming semester.

Start reading. In general. Make it a habit to browse through the recent issues of the top 5 journals. Learn how great work looks like. Learn how to get the main idea of a paper without spending days reading it.

Also read the original versions of the classic articles. For example, Arrow’s classic piece on health economics. Or McFadden’s classic articles on travel demand (he has a very nice website with lots of linked articles). This can be very inspiring because the original articles are often very clear and explicit, and in some sense easier to understand than more recent textbook treatments of the material. You can also browse through Train’s book, for an introduction on using simulation techniques to incorporate random-coefficients into your empirical model. Also find out about recent important developments, for example by reading Varian’s article on big data.

I assume that you already know which field you want to be working in, because you have made a conscious choice after having attended comprehensive field courses and you’re already paired up with an adviser. If you still have the feeling that you are not so sure what the current topics are that are being discussed in the field, then go back to the overview articles you have been referred to in the field courses. For empirical industrial organization, for instance, I would suggest you read the article by Ackerberg, Benkard, Berry and Pakes, among others of course.

This brings me back to Cawley’s article. Go talk to your supervisor about the plan for the next months, ask him which chapters or articles you should read, and start thinking about multiple projects at the same time. But only work on one or two in the beginning. Nevertheless, keep in mind that you should update the list with the 10 research ideas. One of them will be your job market paper, but it’ll be a while until you know which ones.

Let the journey begin!

Recommended reading

I bumped into a nice blog by Kevin Bryan, a PhD student at Kellogg/Northwestern when I looked at the website of my colleague Sebastian Ebert. Recommended to those who are interested in reading about academic research. What I like is that Bryan makes a selection and discusses the respective contributions, so it’s not like subscribing to yet another newsletter from journal XYZ. And of course things make even more sense when they are put into perspective.

The effects of more transparency on the behavior of doctors, and sellers on eBay

A recent piece by The Economist suggests that medical doctors treat patients better when they in turn have the possibility to share their experiences online. This is exactly what my co-authors Christian Lambertz, Konrad Stahl and I find in a recent paper we have written on the effects of market transparency on eBay (we finished revising it last week). We argue and show that improving the design of the reputation mechanism on eBay allowed buyers to share negative experiences without the fear of retaliation, and therefore seller behavior improved. The design was improved by changing the main rating system. We compare seller behavior after that change to the one before. What we find particularly nice and what we exploit in our paper is that there was a second rating system in place in which buyers evaluated seller, and which was actually not changed at the same time.