Choosing a topic

You can find most of the information from this page in a convenient pdf here.

My expertise: I specialize in empirical or computational research and I am strongest there. I am interested in empirical industrial organization.

Choosing a topic: This is the most important problem. Know that it is ok to have a general interest but not necessarily a specific setting that you want to apply it to. There are, however, some general guidelines.

1. Empirical or not: If you want to work with real data, then the choice of topic is all about the data. Of course there are some datasets that can be used to study many different questions, but for many datasets it is somewhat limited what you can do. Here’s an excellent resource for finding papers that can be replicated, which typically means that there is both data and code accompanying: uni-goettingen.de.
2. Know thy self: Find out what constitutes a good project for you. You must be brutally honest about your objectives – you only deceive yourself otherwise. Look at the Goals section below and find out what matters to you: focus on this when choosing a topic.

Goals

When writing a thesis, there are some general tradeoffs to consider which determine the best path to follow. For instance, it may sometimes be possible to choose a simpler question to minimize the risk of getting stuck at the cost of a ceiling on how fancy the project can get.

People differ quite a lot in terms of what matters to them. Therefore, rank the following objectives according to your own preferences.

Task: Rank the following goals according to your preferences.

• The grade: Geting it as high as possible. Some feel uncomfortable admitting that the grade is what matters. But there is no shame in identifying what motivates you - this is how you ensure that you will apply yourself fully and fulfill your potential.
• The process: A predictable, straightforard process where you constantly know where you are and what the next tasks will be, and how much work is remaining.
• The topic: The topic can matter for a variety of reasons:
  • Real-world relevance: Working on a problem, which matters to someone in the real world. For instance, solving a problem faced by a firm or public entity right now.
  • Personal relevance: Working on a topic that you personally care about. For instance a cause (like saving the environment), a pass-time activity (like a sport or game you enjoy), or an industry you have worked in.
  • Mastery: Getting the sense of truly understanding something at a very deep level, or building/making something that you are proud of.
  • Job prospects: Learning skills that will make you employable in a specific job type later, which will impress a specific type of employer, or which may be useful in making your own startup.

Examples

• A high grade: Focus on a project that allows you to use a technically sophisticated tool that is (slightly) beyond what gets taught in the more advanced courses (in a meaningful way). If working empirically, make sure the dataset is easy to use, since the marginal benefits to the grade from raw data cleaning work start to diminish quickly. A mathematically or computationally sophisticated method typically involves a lot of hard but predictable work and will look flashy regardless of whether it makes a huge difference in practice.
• A good process: Focus on an existing paper and re-trace their steps. For empirical projects, finding a paper that is published in a great journal and has replication data is ideal. Then for the first 80-90% of your thesis work, your goal is simply to replicate everything they have done, while you blindly trust that at the end of all this work, you will have thought of a brilliant extension / alternative specification / alteration of the key outcome / different policy counterfactual, which will set you apart.

Using AI to choose a topic

Here’s my suggestion for a prompt to use in ChatGPT or similar:

You're a professor in economics and I'm an economics student about to choose the topic for my thesis. Interview me about how I value the tradeoffs between the following three criteria: 
- a high grade,
- a predictable process, 
- the topic (writing about something that matters to me/society/etc.).

Next, ask me about topics in an effort to zoom in. First, 
- Empirical thesis: in that case, the dataset determines what can be done, 
- Not empirical: then the field is more open, but that can make it harder to be concrete. 

Finally, we need to narrow down the topic. Maybe it's a specific industry, a method from a course you have had, a specific policy question, etc. Or maybe we need to just brainstorm some ideas, or to ask the real-world supervisor for some examples of past theses to get inspiration from.

The overall goal is to arrive at a good research question. The research question should incorporate an important economic tradeoff. It needs to be broad enough to accommodate the necessary changes that arise along the path, but at the same time be as concrete as possible. 

For empirical projects, the dataset and available variables largely determine the research question, at least broadly speaking. Good sources can be replication datasets from published papers, or public datasets from government agencies. Replication exercises are good for the process but may limit the potential for originality. Originality is not crucial for a bachelor's or master's thesis, but for some students it may be important to feel that they are doing their own thing. If not, then replication ensures a more predictable process.

Let's have a dialectic approach to reach clarity about my direction forward.

Of course, the goal is not to outsource the thinking or the supervising to an AI. But my experience is that too many initial meetings start out very vague and unstructured, which makes it hard to get anywhere. Having a structured conversation with an AI first can help you get clarity about what you want, and at the very least to become aware about what questions you have for me, the real-world supervisor (although of course, some of this text is also written by an AI… maybe all? Surely not…).