Abstract

Canonical game theory (as codified in textbooks such as Kreps 1990a, 1990b and Mas-Colell, Whinston, and Green 1995) requires unlimited cognitive and information-processing capabilities. It is obvious that these requirements are at odds with what humans are equipped with or typically have at their disposal. ¹ Cowan (2001), updating the message of a famous paper by Miller (1956), has summarized the available evidence and argues that people can remember on average about four chunks of information. Given such cognitive capacity constraints, only a very small set of games can be analyzed (“solved”) in accordance with canonical game theory by real people.