| The subject of the first chapter is a class of games called departure-time choice (DTC) games, which is played by a population of commuters who choose their departure times in order to minimize cost of delay due to congestion. We define DTC games with a set of assumptions that that capture some common features of a dynamic congestion game: departures at one point in time have negative externalities on those at later times, and congestion externalities diminish with time distance. The main insight of this paper is that DTC games with absolutely continuous departure patterns turn out to be instances of stable games. Results on convergence of equilibrium has been established for stable games under a list of evolutionary dynamics in a finite-dimensional strategy space. We contribute to this literature by extending the convergence result of stable games to a infinite-dimensional strategy space with the logic dynamic and show that the logit equilibrium of a stable game not only exists, but is unique and globally asymptotically stable.;The second chapter studies the emergence of two-period cycles as a result of information lags in congestion games. A population of agents chooses between two alternatives, with the objective being to land on the less congested choice. However, agents cannot obtain real-time congestion information. Responding sensitively to information with lag in congestion games leads to overreaction in the next period, which initiates a two-period cycle. The cycling phenomenon can be observed in examples such as fluctuating traffic congestion patterns as well as price cycles in agriculture commodity markets. In a heterogeneous population, we show that whether the system dynamic diverges to cycles depends on the information lags and on the average of response sensitivities of the heterogeneous population. More interestingly, when agents endogenously choose their response sensitivity, we present the irreconcilability between agents' incentive to avoid oscillation and to pick the less congested choice. Lastly, the possibility of a predictor with perfect foresight is considered. |
