Abstract (eng)
Traditional oligopoly theory predicts production quantities at the Nash equilibrium level when
firms have perfect information. A learning algorithm which allows firms to converge to the joint-profit
maximizing outcome, even if they have only very limited information, is provided by Huck
and Oechssler (2000). On the other hand, an agent-based model with a very similar learning
structure results in the Cournot solution, i.e., the Nash equilibrium level (Kimbrough and Murphy,
2009). It is a-priori unclear which exact ingredient of the two models causes the strikingly different
results. I identify the key difference between the models by introducing step-wise changes to the
agent-based model and verifying the results in each step. The findings suggest that the key
difference in the models is strict rigidity, which is absent in the model of Kimbrough and Murphy.
In their model, firms rather explore than optimize in their attempt to maximize profits, involving
a strong random component. This makes simultaneous downward movements very unlikely, which
would be beneficial for both firms. In the model of Huck and Oechssler, firms maintain their
direction of movement as long as this is beneficial for them. According to these results, inclusion
of a random component in a model might lead to a drastically different outcome.