Improving E-Trade Auction Volume by Consortium - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 113

Table 8. Summary results for trading volumes with and without consortium

Matched Volumes

Case
Method

Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7 Case 8 Case 9 Case 10

Direct Match 100 170 260 140 410 340 400 180 140 230

Consortium 800 240 350 280 410 520 450 470 300 490

concluding r eMarks

This chapter considers a two-tier online auction supply chain system in which buyer agents propose
transactions that are matched by seller agents, or are otherwise referred to a community of seller
agents, who will match the buyer demand in terms of volume. Price is fixed and is not a consideration
in the matching process. Alternately, the seller agent proposes a volume that is matched individually
by a buyer, or by the community of buyer agents by means of collaboration. It is also possible to have
trades matched and completed by a consortium of seller agents matching the volume of a consortium
of buyer agents, or vice versa. The hypothesis is that collaboration between the same class of agents,
and the formation of consortiums within the respective agent communities, offers greater opportuni-
ties for matching trading volumes and increases the total trading volume. Experimental results with
random generation of individual trading volumes for buyers and sellers, and volume matching of trades
irrespective of price, shows that collaboration and consortium formation is usually an improvement in
trading volumes, with a major improvement in some cases of the volume of trade, over the traditional
model of non-cooperation and competition.
An implementation is possible with JADE ( Java Agent Development Framework ) 1 . JADE is imple-
mented in Java, and can be controlled by a remote GUI. It is compliant with FIPA 2 specifications, and
can be implemented on multiple machines on multiple operating systems. The current JADE version is
3.4 (released March 14, 2006). JADE is free software from Telecom Italia, as a FIPA compliant agent
platform with a reduced footprint, and with compatibility with a mobile Java environment. An alternate
modeling and implementation framework is MADKIT (Multi-Agent Development Kit 3 ) available as a free
download. This is a modular and scalable multi-agent platform in Java, built on the AGR (Agent/Group/
Role) organizational model, i.e. agents are situated in groups and play roles. MADKIT allows simulations
of agent performance, heterogeneity in agent architectures and communication languages, as well as a
number of customizations. Like JADE, MADKIT is a peer-to-peer mechanism with full facilities full
facilities for launching, displaying, developing, and monitoring agents and agent organizations.
The real direction of future work consists of extending the model to include other considerations
important in modeling auction trades by agents. The patterns of agent trading follow the patterns of
auction trading in general. In auction trading practice, one of the primary considerations is price. A
seller agent makes an offer at a certain price. If the offer is not matched at the quantity and price it must
be carried over to the next period at a cost to the seller agent. The seller agent may have to dispose
of the offer at a reduced price in a future period to avoid carryover costs. Similar conditions apply to