We develop a prior-free mechanism for an asset market that is dominant-strategy incentive compatible, ex post individually rational, constrained efficient, and asymptotically optimal—as the number of agents grows large, the designer's profit from using this mechanism approaches the profit it would optimally make if it knew the agents' type distribution at the outset. The direct implementation first identifies the agent whose value equals the Walrasian price. The second step can be described algorithmically as consisting of ascending and descending clock auctions that start from the Walrasian price, estimate virtual types, and stop eliminating trades when the estimated virtual value exceeds the estimated virtual cost. The mechanism permits partial clock auction implementation. Our approach accommodates heterogeneity among groups of traders and discrimination among these, provided heterogeneity is not too accentuated.