Robust control theory extends rational expectations models to acknowledge fear of model misspecification. Here is a paragraph from the first chapter that convinces me just how fundamental this is for economists to address:

Rational expectations models presume that decision makers know the correct model, a probability distribution over sequences of outcomes. One way to justify this assumption is to appeal to adaptive theories of learning that endow agents with very long histories of data and allow a Law of Large Numbers to do its work. But after observing a short time series, a statistical learning process will typically leave agents undecided among members of a set of models, perhaps indexed by parameters that the data have not yet pinned down well. This observation is the starting point for the way that we use detection error probabilities to discipline the amount of model uncertainty that a decision maker fears...(Hansen and Sargent Section 1.4.5).Robust control theory began to be developed by control theorists and applied mathematicians in the late 1970s for application to a variety of engineering and physical problems. Model misspecification is probably a much greater problem in economics, where there is much less consensus about how the system works and what kinds of policies are effective. The engineering and physics applications don't include discounting, or at least not in the way that economists do. They also don't include multiple agents in a game theory context. Hansen and Sargent add both discounting and multiple agent settings to robust control theory in this text, greatly facilitating its application to economics.

As I start this study, there are a few things I am wondering:

- Positive or normative? Does robustness describe what policymakers do, or what they should do? In other words, by understanding robust control theory, will I be better able to understand why policymakers made the decisions they made, or what they should have done/should do differently?
- Can robust control theory incorporate policymakers' motives that may not be simply social welfare maximization? Reputation concerns, power struggles, etc. may lead to purposeful "misspecification."
- Is there some concept like "retrospecification"? That's just a term I'm making up, but perhaps the policy justification comes after the policy in some cases?
- Does robust control theory apply to institutional design and regulatory design in a similar way as it applies to policymaking?
- Misspecification implies that there is a "true" model. Is there?

## No comments:

## Post a Comment