A “Quantal Regret” Method for Structural Econometrics in Repeated Games

We suggest a general method for inferring players’ values from their actions in repeated games. The method extends and improves upon the recent suggestion of (Nekipelov et al., EC 2015) and is based on the assumption that players are more likely to exhibit sequences of actions that have lower regret.

We evaluate this “quantal regret” method on two different datasets from experiments of repeated games with controlled player values: those of (Selten and Chmura, AER 2008) on a variety of two-player 2×2 games and our own experiment on ad-auctions (Noti et al., WWW 2014). We find that the quantal-regret method is consistently and significantly more precise than either “classic” econometric methods that are based on Nash equilibria, or the “min-regret” method of (Nekipelov et al., EC 2015).