As with all my handwritten notes, this has the usual disclaimer: these posts are just so I can use nice indexed search to find my notes, which are sufficient for me to recall talks and papers but probably not much use to anyone else. Slides here.
Game theory allows us to answer qualitative questions regarding the behavior of strategic agents. For instance, given a particular game, what will the strategies of players be, will the equilibrium be efficient, how many equilibria are there? The answers can depend on multiple features of the game such as the structure of payoffs, information and beliefs of players. In practice, even if we know that a certain interaction between people or firms can be modeled by a game, we usually know little about the primitive structure of this game, i.e. we don’t know the exact payoffs from different actions, equilibrium selection mechanism or private information of players. However, we can have the data that shows what were the actions of players in a particular game.
In this tutorial, we will look at some modern empirical techniques that allow us to recover the “primitive” structure of the game from the data on its past realizations. We will talk about static games of complete and incomplete information and learn fast and efficient methods for modeling these games using simple statistical software. We will study some tools that will allow us to empirically discriminate between different game models using the data. During the discussion of the games of incomplete information we will also consider auctions as a special case. We will learn how to recover unobserved valuations of bidders from their bids in auctions which are not necessarily truthful. We will also look at some examples where we can use such estimates for predicting welfare and revenue when we make changes to the auction mechanism.