Monday, December 4, 2000 - 1:30pm - 2:30pm
Suzhou Huang (Ford Motor Company)
We study a class of dynamic pricing duopoly games that model the type of environment in which e-commerce will be carried out in not so distant future. Under Markov settings these games can be solved via backward induction. The equilibrium structure is found to display very complex patterns when parameters of the model are varied, due to bifurcation phenomena in the discrete map induced by backward induction. However, it is possible to define an effective but simpler dynamics that retains the optimality of the original game in the long run. We further show that this effective dynamics can be sustained by steady self-confirming equilibria. Our results (1) set limits on what learning algorithms based on Markov assumptions can obtain and (2) imply that learning in this kind of games should not be focused on the exact reaction functions, but rather on achieving optimal net present values with the realized time series of prices.