Information Technology Reference
In-Depth Information
2.
Hotel auctions
. There are two hotels at TACtown, namely Shoreline Shanties (SS)
and Tampa Towers (TT), with TT being the nicer hotel and each hotel having 16
rooms available over 4 different days. Thus, there are 8 different hotel auctions
(given the 2 hotels and rooms being available for 4 different days). Hotel rooms are
traded in 16th-price multi-unit English auctions, whereby the 16 highest bidders are
allocated a room for a particular day in a particular hotel, and at the end of every
minute except the last minute, a hotel auction randomly closes, and the 16th and
17th highest price of each hotel auction that is still open is published.
3.
Entertainment auctions
.There are three types of entertainment in TACtown, namely
a museum, an amusement park and a crocodile park, and 12 different entertain-
ment auctions. At the beginning of the game, each agent is randomly allocated 12
entertainment tickets tradeable in the different multi-unit CDAs which clear con-
tinuously and close at the end of the game.
Given this background on the TAC environment, our objective is to design a trading
strategy for an autonomous software agent participating in such a game. We develop
the strategy by using the IKB framework, adopting the multi-layered approach. We
now describe the strategy within the different layer prescribed by the IKB.
5.1
The Behavioural Layer
The issues associated with the bidding behaviour can be summarised as follows: (1)
What item to bid for? (2) How much to bid for? (3) When to bid?
Definition 6.
Optimal Plan.
The optimal plan is the set of travel packages, for 8 dif-
ferent clients, that would yield the maximum profit, given the clients' preferences (that
determine the utility of the package) and the cost of the packages.
Definition 7.
Marginal Profit
12
.
The marginal profit of a hotel room (in a particular
hotel on a particular day) is the decrease in the agent's total profit if it fails to acquire
that room. Thus, the marginal profit of a hotel room that is not required in the optimal
plan is 0. The marginal profit is defined only for the more competitive hotel auctions.
Our strategy uses the history of market information,
H
(
p
M
(
t
k−
1
)) without any external
information. The publicly viewable market state
p
M
(
t
k
) is a subset of bids and asks in
all open auctions as well as the clearing price of all the closed auctions. Only the 16th
and the 17th highest bid are visible for hotel auctions, and the outstanding bid and ask
for the entertainment auctions.
We address the first issue by considering the
optimal plan
(see Definition 6). Thus,
the agent always bids for the set of items (flight tickets, hotel rooms and entertainment
tickets) required for the optimal plan, querying the optimal plan from the KL every 10
seconds. As a hotel auction closes every 60 seconds, the set of items available to the
agent is further constrained and the optimal plan has to be recalculated. We address the
other issues by considering the different auction formats.
First, we consider the 8 flight auctions. Given the manner in which the flight prices
update, it is possible to predict the trend of the price update. Such a trend is queried
12
The marginal profit described here is similar in essence to the marginal value used in [3].
