This chapter analyzed a new market institution, the multidimensional auction. It highlighted in particular one ICT-related design issue, namely the information architecture of the market. Laboratory experiments were carried out to test for the effects of information architecture as well as the number of bidding rounds on two market performance measures, winner efficiency and optimality. Results show that an unrestricted information architecture (i.e. more information revealed about the state of competition and the bid taker’s preferences) increases the optimality and winner efficiency of the multidimensional auction. Auctioning over 4 rounds instead of 2 improves the optimality of the auction, but has no overall effect on winner efficiency. These are by and large conform expectations, although a stronger effect of the number of rounds was expected, particularly regarding winner efficiency. Perhaps the increase from 2 to 4 four rounds was too small to yield significant improvements, but this is something that future experiments over more rounds will have to resolve.
What is somewhat surprising however, is that the effects of information architecture and auctioning rounds are highly dependent on each other: the presence of strong interaction effects shows that an increase of one experimental variable only improves market performance if the other experimental variable is low. In other words, auctioning more rounds can act as a partial substitute for more feedback in a given round and vice versa.
A possible explanation for this effect may be found in the role of information in the decision processes of the individual bidders. As Koppius and van Heck (2001) argued, the revision of bids by an individual bidder from round to round seems to be consistent with the general belief-adjustment model proposed by Hogarth and Einhorn (1992). An important feature of the belief-adjustment model is that the degree of adjustment is not just dependent on whether the evidence is positive or negative, but also dependent on the previous belief against which it is evaluated. For example, information that is negative compared to the held belief will cause minor belief-adjustment if that belief is already fairly close to the (negative) evidence. If the same negative information is evaluated against a strongly held belief, belief-adjustment will be much stronger (‘the harder they come, the harder
they fall’). In the negative information case, the adjustment is proportional to the current position, whereas in the positive information case, it is inversely proportional to the current position (Hogarth and Einhorn, 1992).
Most subjects indicated after the experiment that they primarily thought of the bidding process as a search for the bid taker’s optimum, which is quite different from a single-dimensional auction where the emphasis is on beating the competition. So, if we view the bidding process in a multidimensional auction as a search for the bid taker’s optimum, the bidder’s belief corresponds to where he thinks the bid taker’s optimum is located and he is assumed to bid accordingly. Information at the end of the round can then be interpreted as positive if the highest bid is close (in the bid space) to the bidder’s bid in that round and negative if the highest bid is located far from the bidder’s bid in that round. Some bidders perceived this information role of bids quite well when they commented that it was a disadvantage to be the highest bidder in the penultimate round, particularly in the restricted information architecture, because then you had less information than the other bidders to go on in revising your bid.
As bids converge during the auction towards where bidders initially think the optimum is, the information that those bids reveal at the end of each round becomes less informative, because the revealed information is already fairly close to the position held. In other words, there will be only minor belief-revision (i.e. updating of the assumed bid taker’s optimum) and bidding is likely to remain in that region of the bid space, with auction performance remaining stable at a sub- optimal level.
Although the discussion above pertains to multidimensional auctions, this result yields an interesting speculative interpretation of what it takes to make a market in general perform well in terms of efficiency and optimality. A less restricted information architecture increases the amount of information about competitors’ private information and bid taker preferences that is revealed during the market process. The same line of reasoning applies to an increase in the number of rounds: auctioning more rounds gradually reveals more information about bid taker’s preferences and the state of competition. The interaction effect between those two implies strongly diminishing returns to revealing additional information, either through a less restricted information architecture or through revealing information over more rounds. The reason is that the additional information, despite being credible and correct, simply is not effective in changing bidding
behavior. Bidders’ belief that the current highest bid is correct may simply be too strong to be swayed by any further information. This implies that market performance will then remain at a sub-optimal level. In other words, there seems to be a phenomenon of information saturation at work in the market: beyond a certain point, more information does not improve market performance any further.
CHAPTER 6 - SYNTHESIS: A CONCEPTUAL MODEL OF