This paper discusses the use of a computational learning approach based on a leader-follower multiagent framework in the study of regulation of restructured electricity markets. In a leader-follower multiagent system (LFMAS), a leader (regulator) determines an appropriate incentive, which motivates a set of self-interested followers (the generators, in this case) to act such that some measure of overall performance is maximized. In the computational learning approach presented, models of followers as well as the leader incorporate reinforcement learning, allowing the exploration of outcomes with different incentives, and also the learning of 'optimal' incentive given some measure of desired overall performance. The approach is demonstrated in studying the effect of price caps on the outcome of electricity auctions (uniform and discriminatory) in oligopoly settings for which analytical treatments do not exist.