Robust Identification in Repeated Games: An Empirical Approach to Algorithmic Competition
Cristina Gualdani (Queen Mary University of London)
Paper
Abstract:
We study the identification of economic primitives when algorithms make repeated strategic choices. While theory and simulations suggest that repeated-game equilibrium-like behavior often emerges in these settings (e.g., Calvano et al., 2020), a wide range of outcomes can emerge, depending on typically unobserved features of algorithms and the environment. We develop an incomplete model to characterize bounds on cost parameters without assumptions on equilibrium selection or algorithmic design. Through simulations of algorithmic play, we show our bounds are valid across different algorithm specifications. Moreover, bounds can be informative especially when leveraging demand variation across markets.