Joyce (1984) reports the results of experiments with a Walrasian tâtonnement auction that show that the mechanism is stable, exhibits strong convergence properties, and generates efficiencies that average better than 97%. He also found that when subjects could see part of the order flow (excess demand), prices tended to be lower. His experiments consisted of a stationary environment where subjects were provided single unit supply and demand functions.We assess the robustness of his results in a more general multiunit per subject setting; and we systematically investigate the effect of various rules about order flow information and message restriction rules on the performance of the Walrasian mechanism.

Our experiments are motivated by several considerations.

1. When there are both buyers and sellers in the market, each of which has one unit to buy or sell, the only Nash equilibria of the Walrasian tâtonnement mechanism are those that support the competitive equilibrium outcome. Furthermore, a Walrasian tâtonnement process can be designed that has a dominant strategy equilibrium where each participant reveals value or cost (see McAfee, 1992). The design imposes constraints on participant messages; specifically, at the announced price at t, if excess demand is positive (negative), any seller (buyer) not registering a sell (buy) order at t cannot register an order at time t + 1. Without this improvement rule, the dominant strategy equilibrium outcome no longer exists.But even with this improvement rule, the dominant strategy revelation property does not hold when demands and supplies are multiunit, since a participant may influence price without being entirely out of the market.