We have presented a hierarchy of pricing schedules for the complex space of electronic goods and studied their performance using analytical methods for the steady state and simulations to determine their dynamic behavior. We find that most of the improvement in schedules comes with the move from one-parameter to two-parameter schedules. More complex schedules can provide some additional profit at equilibrium, but it is not clear that this small increase is balanced by the longer time needed to learn these schedules. The simpler schedules are also appealing from a more practical point of view; they are easier for a consumer to understand, and consequently it is easier for the consumer to participate. All of these results are based on the assumption that consumers will always act to maximize their surplus. Our continuing research questions the effect of consumers who are unable to determine how many goods to buy (because the schedule is too complex), or who act strategically to exploit the fact that the producer is learning.
Another area of future research involves the introduction of multiple producers or changing consumers. In this case, the optimal set of prices will change over time, and both the pricing schedule and the producer's learning algorithm will need to be able to cope with these changes.
A third area of future research includes the introduction of consumer modeling into the learning process. If a producer has access to some information about the consumer's valuations, it may be able to learn the optimal set of prices more quickly and therefore increase its profits. The question remains open as to what sorts of consumer knowledge are realistic for a producer to be able to infer, given that real-world producers seem to need large amounts of data to do a reasonable job of modeling consumers.