Information Technology Reference
In-Depth Information
strategies within it. Secondly, we consider a more complex scenario, the Travel Game
of the International Trading Agent Competition (TAC) where an agent has to strategise
in multiple simultaneous auctions of different formats. In both cases, we employed our
IKB model successfully.
The remainder of this paper is structured as follows. We review related work in the
field in Section 2. Section 3 outlines the IKB model, which is then applied to our trading
market examples in Sections 4 and 5. Section 6 concludes.
2
Related Work
Much work has been carried out on abstracting the design of electronic markets [13,16].
However, this work tends to emphasise the methodologies for designing the markets
themselves or on proposing new market infrastructures [2,19]. The systematic design
of strategies for agents operating in these markets has, in general, been considered to a
lesser extent. In this latter vein, however, Vetsikas
et al.
[20] proposed a methodology
for deciding the strategy of bidding agents participating in simultaneous auctions. Their
methodology decomposes the problem into sub-problems that are solved by
partial
or
intermediate
strategies and then they advocate the use of rigorous experimentation
to evaluate those strategies to determine the best overall one across all the different
auctions. However, their methodology is very much tailored to simultaneous auctions
in general and the TAC in particular [22]. Thus, it cannot readily be generalised to other
auction formats or other market mechanisms. Furthermore, other approaches, including
[2,8], look at the strategic behaviour of agents. However, they avoid issues related to the
information and knowledge management aspects of designing trading agents (focusing
instead mainly on the strategic behaviour of the strategy).
3TheIKBMod l
In this section, we detail the main components that the designer of a trading agent strat-
egy should pay attention to. In so doing, we develop a framework for designing strate-
gies in trading markets. In our model, we have a market
regulated by its predefined
protocol. The collection of variables representing the dynamics of the system at time
t
k
(where
k
indexes changes in the market) is represented by the state variable
p
M
(
t
k
).
Within this market, there is a set of trading agents,
M
, that approach the market through
a set of actions which are determined by their strategies. In order to formulate its best
strategy, an agent
ideally
needs to know which state it is currently in (agent state), the
market state and the actions it can take.
I
Definition 1.
Agent's State.
An agent
i
's state,
p
i
(
t
k
)
, at time
t
k
is a collection of
variables describing its resources (computational and economic) and privately known
preferences.
Definition 2.
Market State.
The market state,
p
M
(
t
k
)
, at time
t
k
is a collection of
variables describing all the (public and private) attributes of the market.
