- Understanding the causal mechanisms at work in the behavior of human beings and their economic systems.
- Developing models that have strong predictive power but are also explanatory. I.e. I want to know the reason that there is a causal connection, not just that there is one. In this sense, I am more interested in theories and mechanisms that are descriptive and predictive rather than normative.
- Econometric and statistical methods, particularly causal inference, bayesian statistics, and statistical learning methods. Use of bayesian statistics has expanded rapidly since the early 2000s,
- Prediction, especially of human behavior in situations of scarcity (see Sendhil Mullainathan’s work and Daniel Kahneman’s Thinking, Fast and Slow).
- How tech like ubiquitous data (gathered through wearables or observation) can be used in concert with the above topics to answer economic questions.
- Develop theory and tools to apply machine learning techniques to improve predictive abilities. A potential example is something like CausalImpact.
- As suggested by AtheyImbensXXXX, we need new cross-validation methods optimized for causal inference.
- What is the robustness of these measures?
Problems and Opportunities in Economics
- Maximization is generally not possible because the agent doesn’t have control over all the variables, so some mixture of maximization, minimization, satistficing, etc… takes place, which is not adequately described by theory (Mortgenstern1979).
- Revealed Preferences Theory is wrong because preferences can be transitive. What is a theory of choice that allows for preference transitivity? (Mortgenstern1979).
- Predictive ability of models is poor. Models are often abstracted from reality (AtheyImbensXXXX).
Stephen Intille’s PhD advice page is a great resource.