Information Technology Reference
In-Depth Information
which, as described in Subsection 3.2, is computed from the market information. With
the relevant knowledge of the market and its goals, the agent
i
forms a decision based
on its strategy
S
i
and interacts with the market through actions
SA
i
. The goal of an
agent's strategy is typically profit-maximisation, with the more sophisticated strategies
considering both short-term and long-term risk. The formulation of the strategy usually
depends on such goals and the market protocols.
Given this insight, we categorise the different behavioural properties of the strategy
into different levels. In more detail, we distinguish those strategies in terms of the type
of information (in Equation 1) that is used,
i.e.
, whether they use a history of market
information or not, and, where they consider external information or not.
1.
No History
(ignores
H
(
p
M
(
t
k−
1
)) from Equation 1). Such reactive strategies
make myopic decisions based only on current market conditions,
p
M
(
t
k
).The
myopic nature of these strategies imply a lower workload on the KL since they
require less information to sense and process. Reactive strategies usually exploit
the more complex bidding behaviour of competing strategies and thus require less
computational resources to strategise. One example of such a strategy is the
eSnipe
strategy
5
which is frequently used on Ebay to submit an offer to buy near the end
of the auction.
2.
History
(considers
H
(
p
M
(
t
k−
1
) in Equation 1). We further subdivide those strate-
gies that use a history of market information as being predictive or not (i.e. whether
they predict
or not). The non-predictive strategies typ-
ically use
H
(
p
M
(
t
k−
1
)) to estimate
p
M
(
t
k
).
(a)
Non-predictive
: The non-predictive strategy is typically belief-based and forms
a decision based on some belief of
the current market conditions
. The agent's
belief is computed from the history of market information in the KL, and usu-
ally represents the belief that a particular action will benefit the agent in the
market (e.g. an offer to buy that is accepted). Given its belief over a set of
actions, the agent then determines the best action over the short or long term.
(b)
Predictive
: A strategy makes a prediction about the market state in order to
adapt to it. Now, because future market conditions (that the trading agent adapts
to) cannot be known
apriori
, the adaptive strategy typically makes some pre-
diction using the history of market information. The KL is required to keep
track of how the market (knowledge) is changing to predict the future mar-
ket, while the BL uses this knowledge about the market dynamics to improve
its response in the market. Being adaptive is particularly important in situa-
tions where the environment is subject to significant changes. By tracking such
changes and adapting its behaviour accordingly, the agent aims to remain com-
petitive in changing market conditions.
3.
No External Information
(ignores
ext
1
,...,ext
n
in Equation 1). In this case,
the strategy does not consider any signals external to the market (e.g. the falling
market price of a good affecting the client's preferences for another type of good
in an auction). However, the agent can choose whether or not to use the (internal)
information (e.g. the e-Snipe strategy uses the internal market information, while
the ZI Strategy [10] in the CDA does not make use of any market information).
{
p
M
(
t
k
+1
)
,p
M
(
t
k
+2
)
,...
}
5
www.esnipe.com
