Information Technology Reference
In-Depth Information
As argued in Section 1, an agent's sensory and computational limitations imply that
it will rarely have perfect and complete information. For example an agent might not
be aware of its complete action set (i.e. an agent might believe that its action set at
time
t
k
is
A
i
⊂A
i
) or it may be unsure of which state it is in (i.e. it expresses an
uncertainty over
p
i
(
t
k
)). Thus, the agent will need to have certain heuristics in order to
guide its search for information. This information can be gathered from public, semi-
private and private sources. Public information is observable by all agents (
i ∈I
)inthe
market and includes things such as the market price in a stock exchange, the minimum
increment in an eBay auction and the number of lots of flowers on sale in a Dutch
flower auction. Semi-private information is that which is available to a subset of the
agents (
i
) and includes things such as the amount that a supplier might
require from an agent and the code to signalling actions by a bidder ring in an auction
[14]. Private information is only observable by a single agent and includes items such
as its budget or the goods it is interested in. Thus, given the required information that
the KL has requested, the agent will devote its limited resources to obtaining it. Then
having gathered the required information from the market, the agent proceeds to use
this information to infer knowledge in the KL.
∈J ⊂I
3.2
The Knowledge Layer
The Knowledge Layer connects the Information and the Behavioural layers (see Section
3.3). It infers knowledge from the information sensed by the agent and passes it to the
BL which acts upon it. In order to do so, the KL is requested by the BL as to which
knowledge to acquire. This knowledge could be, for example, the current Sharpe ratio
3
of a stock or a forecast of market price based on a particular prediction model. Based
on this and the current knowledge of the agent's state, the KL will decide upon the
information it requires and set the information filter accordingly. The KL will then use
the input from the IL to infer the appropriate knowledge which it will output to the BL.
Mirroring the IL, the KL can be segmented into knowledge about the agent's and the
market's state. The former is what the agent knows about itself. This includes knowl-
edge pertaining to its sub-goals (such as its risk attitude or the deadline by which a good
is to be delivered) and knowledge about its state
p
i
(
t
k
). The latter is what the agent
knows about the market and would include items such as the degree of competitiveness
in the market, the opponents' state and any available market indicators.
3.3
The Behavioural Layer
The Behavioural Layer represents the decision-making component of the strategy. In
this context, such strategies are targeted towards finding the optimal action
4
in the mar-
ket. However, as outlined earlier, more often than not, there is no known optimal action,
as the market is too complex and the set of actions too large to determine such an opti-
mal action analytically. Then, as there is no best strategy, a heuristic approach is taken.
Thus, the BL instructs the KL as to what knowledge it needs to gather from the market
3
The Sharpe ratio is a measure of a stock's excess return relative to its total variability [17].
4
Here, “optimal” means the agent's most profitable action, given the current market conditions.
