Three minimal market institutions with human and algorithmic agents: Theory and experimental evidence

We define and examine the performance of three minimal market games (sell-all, buy-sell, and double auction) in laboratory relative to the predictions of theory. Unlike open or partial equilibrium settings of most other experiments, these closed exchange economies have limited amounts of cash to facilitate transactions, and include feedback. The experiment reveals that (1) the competitive general equilibrium (CGE) and noncooperative (NCE) models serve as reasonable anchors to locate most but not all the observed outcomes of the three market mechanisms; (2) outcomes tend to get closer to CGE predictions as the number of players increases; (3) prices and allocations in double auctions with full feedback reveal persistent deviations from CGE predictions; (4) the outcome paths from the three market mechanisms exhibit significant and persistent differences among them; (5) since the dynamics of markets populated by profit-motivated human subjects is at least partially captured in markets with simple algorithmic traders, the importance of market structures in determining their outcomes is reinforced; and (6) none of the three markets necessarily dominates the others across the six measures of performance. The results suggest that abstracting away from all institutional details does not help understand dynamic aspects of market behavior and that inclusion of mechanism differences into theory may enhance our understanding of important aspects of markets.