Stability of the truth-telling strategy in multi-unit option allocation auctions: laboratory experimentation

Computational mechanism design has become very popular among computer scientists. Lots of mechanism/protocol have been designed and proposed; however, virtually none of them has been widely used so far, including the more traditional Vickrey-Clarke-Groves mechanism. One major reason why these new protocols have not been used to date is that people do not trust howwell a new protocol might perform in a real-world setting. In the real-world setting, there exist not only computational agents, but also real humans. To fill the gap between theory and the real world, we need to examine the performance of a new protocol when it is used among real humans and/or computational agents so that we can use the obtained knowledge to design a better protocol that can achieve desirable properties. In this paper, we focus on laboratory experimentation using real humans to observe how bidders will behave in an environment that includes both real humans and computational agents. This paper examines the performance of a multi-unit auction protocol, called the OPtion allocation (OP) protocol [3]. The protocol is the first non-trivial open ascendingprice auction protocol that generalizes the Ausubel auction [1] in terms of the types of value functions. Furthermore, we compare the OP with a uniform-price (UP) auction protocol, which is a representative conventional auction. Note that we examine the sealed-bid version of the UP and OP as the first step toward the analysis of open auctions in this paper. The experimental results lead to our proposing segmentation, which protocol a seller should adopt according to the aim of his business. Furthermore, we observe lots of overbidding behavior that makes the performance of our proto-