Information Technology Reference
In-Depth Information
Definition 3.
Strategy.
Astrategy,
, defines a mapping from the
history of the agent state
H
(
p
i
(
t
k−
1
))
and the market states
H
(
p
M
(
t
k−
1
))
, and cur-
rent agent state
p
i
(
t
k
)
and the market state
p
M
(
t
k
)
to a set of atomic actions
SA
i
=
{
S
i
,foragent
i
∈I
a
1
,a
2
,...,a
r
,...
}
,a
r
∈A
i
where
A
i
is the set of all possible actions for agent
i
at
time
t
k
.
The actions chosen by strategy
S
i
then affect the external environment such that it
causes a change in the market state. In fact, this strategy could interplay with strategies
selected by other agents,
i
, as well as some external input(s),
ext
n
,(where
n
is the
number of external signals not caused by participatory agents) so as to lead the market
to the new state:
I\
p
M
(
t
k
+1
)=
T
(
p
M
(
t
k
)
,H
(
p
M
(
t
k−
1
))
,SA
1
,...,SA
I
,ext
1
,...,ext
n
)
(1)
where
T
(
.
) is the state transfer function. From Definition 3, it is clear that in order
for an agent to know which strategy is best, it should know the complete description
and history of the states (all market information), a complete description of all actions
available to it, its preferences over the states, a model of its opponents' state, behaviour
and preferences, and the state transfer function.
In practice, however, an agent will typically not have all this information (for nu-
merous reasons, such as limited sensory capabilities, privacy of opponent's information
and limited knowledge of relevant external signals). Furthermore, an agent's limited
computational resources imply that it might not be able to keep a history of all past
interactions. Given this, there is a need for designing feasible strategies that use limited
computational and sensory resources. To this end, we advocate the following design
principle where an agent manages its limited capabilities through its Information Layer
(IL), its Knowledge Layer (KL) and its behavioural Layer (BL) (as shown in figure 1).
In more detail, the Market State (MS) contains public information (i.e. information
available
to all agents in the market) and private/semi-private information (i.e. informa-
tion
available
to one/some agents). We now provide a description of each of the layers
that pertain to the agent:
- Information Layer.
The IL contains data which the agent has extracted from the
MS and private information about its own state. This extraction is a filtering process
(which we represent as the Information Filter in figure 1) whose objectives are
defined by the KL (e.g. filtering out only transaction prices).
-KnowledgeLayer.
The KL represents the gathered
knowledge
that is aggregated
from the data in IL (e.g. bids submitted in the market). The BL queries the KL to
obtain the knowledge it requires.
-BehaviouralLayer.
The BL determines the agent's strategic behaviour by deciding
on how to use the information available to it in order to interact with the market
through a set of actions (e.g. submitting a bid). It queries the KL for the relevant
knowledge it requires (e.g the belief that a bid will be accepted in the market).
We next describe each of these layers in further detail, whilst explaining the process
through which an agent uses a plethora of raw data to select appropriate actions.
