Information Technology Reference
In-Depth Information
Ta b l e 2 .
A breakdown of the percentages of CPU and non-CPU components ordered in response
to requests sent on days 0, 1, and 2+, with respect to the total quantity ordered by each agent
Agent
CPU 0 CPU 1 CPU 2+ Other 0 Other 1 Other 2+
FreeAgent
0.61
0.05
0.34
0
0.05
0.95
Mr.UMBC
0.57
0.16
0.27
0.29
0.47
0.24
UMTac-04
0.85
0.15
0.00
0.39
0.60
0.01
Botticelli
0
0
1
0
0
1
Deep Maize
0.90
0.01
0.09
0.46
0.02
0.52
SouthamptonSCM
0.81
0
0.19
0.54
0.38
0.07
3.3
A Blocking Strategy in TAC-04/SCM
The overall levels of day-0 procurement observed in TAC-04/SCM were similar to those
observed in 2003, so it is perhaps not surprising that a blocking strategy again proved
pivotal in the final round. However, the specific preemptive tactic employed by
Deep
Maize
in 2003 was no longer useful due to changes in the supplier's pricing formula.
The
Deep Maize
preemptive strategy relied on accepting partial fulfillment offers at
low prices to purchase some cheap components on day 0, while still blocking some of
the opponent's requests. In the 2004 rules these partial offers have very high prices, so a
blocking agent must either pay high prices or try to purchase components later, despite
a bad reputation with suppliers.
Blocking and purchasing later is a somewhat risky strategy, but
FreeAgent
used this
tactic in the final round.
5
The strategy had a novel twist that mitigated some of the
risks:
FreeAgent
only blocked requests for non-CPU components and purchased large
quantities of CPU components on day 0 along with the other agents. This is signifi-
cant because CPUs cost much more than any of the other components (on average, the
CPU represents half of the total cost of components for a PC). The strategy locked in
relatively low prices for the components with the highest base prices, but risked pay-
ing relatively higher prices for the cheaper components in order to disrupt the other
agents' procurement strategies. Table 2 illustrates the effect of this blocking strategy
on the distribution of purchases over the game for each agent. Most of the agents, in-
cluding
FreeAgent
, procured the majority of their CPU components on day 0 to take
advantage of the low prices.
FreeAgent
used its requests for “other” (non-CPU) com-
ponents on day 0 to block opponents' requests. Consequently, it ordered no compo-
nents of these types on day 0, and the quantities purchased by the other agents were
reduced.
One of the crucial disparities between the agents was how they reacted to the new
market environment created by
FreeAgent
's blocking strategy.
Mr.UMBC
,
UMTac-
04
,and
SouthamptonSCM
had backup strategies that procured large quantities of
components again on the next simulation day. Since most of the supplier capacity was
uncommitted at this point, the agents secured reasonably low prices for these backup
5
Mr.UMBC
also submitted large blocking requests for some types non-CPU components. How-
ever, these RFQs were relatively low in the priority ordering assigned by the agent and likely
did not come into play.
