Learning in Cournot Oligopoly: An Experiment

Subject-specific performance information can worsen the tragedy of the commons: Experimental evidence

“…The 'best response' process defined in (12) yields a Markov chain over the state space , whose convergence to a stable equilibrium (the NE profile) cannot be assured globally. 8 It is well known that the introduction of inertia 9 stabilizes the best response dynamics, as shown by Huck et al (1999), whose result can be readily applied to our linear CPR setting. Note that in our experimental design we did not introduce any inertia.…”

“…9 Inertia is introduced by assuming that in round with independent probability subject will stick to her previous effort instead of following the best response. Huck et al (1999) demonstrated that the resulting Markov process converges globally in finite time to the NE for any .…”

“…First, it constitutes a reciprocating strategy in the sense of 'giving as good as you get' (equivalent to a tit-for-tat strategy). Second, as discussed by Huck et al (1999), the rule seems reasonable because subjects may assume that while individual choices may be wrong, the 'average choice' cannot be 'too wrong'. Further reasons for applying this rule can be a preference for cautious behavior and a taste of conformity.…”

“…Note that any symmetric state is stable under the dynamics in (16). Hence, Vega-Redondo introduced some perturbation to the process, whose interpretation is that subjects either experiment with new strategies or make mistakes 11 from time to time.…”

Rage Against the Machines: How Subjects Learn to Play Against Computers

“…Figure 4(a) reveals that the subject simply got lucky. 14 It was a first-time player in the no-history setting, i.e., a player with very little information about the game. The reinforcement algorithm locked in at very low quantities in the range of 10 and the subject roughly played a best response to that, which resulted in an average profit of 2117.…”