Information Technology Reference
In-Depth Information
Ta b l e 1 .
Average scores and breakdowns from the TAC-04/SCM final round (in millions of dol-
lars). Margin is the raw difference between revenue and supply costs.
Agent
Score Revenue Supplies Margin Storage Penalty Interest
FreeAgent
10.28
99.06
-80.94
18.12
-7.14
0
-0.61
Mr.UMBC
8.65
94.14
-76.49
17.65
-8.39
0
-0.61
UMTac-04
6.52
83.59
-67.21
16.37
-8.97
0
-0.88
Botticelli
0.44
25.83
-23.74
2.09
-0.39
-1.16
0
Deep Maize
-5.12
56.24
-48.32
7.92
-9.59
-2.49
-0.95
SouthamptonSCM -10.41
71.37
-69.25
2.12
-11.34
0
-1.20
2
2004 Final Round Results
The 2004 competition started with qualifying and seeding rounds lasting two weeks
each; these rounds were used primarily for development and testing. The top 24 agents
participated in a three day tournament at AAMAS-04. A quarterfinal round eliminated
12 agents and a semi-final round eliminated 6 more. The surviving 6 agents played 14
games in the final round. Here we focus on analyzing the results of the final round, as
it represents direct competition between the strongest agents. In Table 1 we see that
the top three agents were very close, with scores in a narrow range of $4M. These
agents had much higher raw margins than the bottom three agents, and generally higher
transaction volume. Among the top three, small differences in storage costs and raw
margins determined the ordering.
SouthamptonSCM
had supply volume comparable
to the top finishers, but high storage costs and low revenue indicate possible sales prob-
lems.
Botticelli
and
Deep Maize
both transacted substantially less volume than the top
finishers.
3
Agent Procurement in Supplier Markets
3.1
A History of Early Procurement
Understanding the market for PC components TAC-04/SCM is aided by discussion of
results from the first competition. During the early rounds of the 2003 competition
agent designers noticed that there were strong incentives to procure large quantities of
components on the very first day of the simulation, day 0.
1
By the end of the seed-
ing rounds most agents were making very large component purchases at the start of the
game, before any information about customer demand was available. In games with low
demand this could lead to large losses for all agents, as in one semi-final heat where all
agents purchased aggressively and finished with negative average profits. In the other
semi-final heat and the final round,
Deep Maize
surprised the field with a novel
pre-
emptive
strategy that blocked the other agents from making large day-0 purchases.
2
1
Available supplier capacity was at a maximum, and prices were at a minimum.
2
Essentially, the agent requested most of the supplier capacity for the entire game and the sup-
pliers reserved it until the next day. Any requests considered after this request generated useless
offers. The agent accepted a partial quantity from this request.
