Abstract
In this project, we theoretically incorporate and experimentally investigate adding an externality into a multi-armed bandit problem. The externality also evolves through a bandit process, in that it is initially unknown but learnable through experience. In a typical multi-armed bandit problem, an individual trades off immediate reward against information about future rewards, where rewards directly benefit the individual. By adding an externality, the individual must conduct a separate trade-off for the affected party, alongside their own rewards according to their other-regarding preferences. We vary the type of externality by treatment: it either affects the payoff of another experimental subject, or the amount of carbon offsetting credits purchased through a certified retailer. We also include treatments with no externality, as well as treatments where the externality affects the individual’s own payoffs, in order to investigate the effects of additional complexity from the externality bandit process.