Information Technology Reference
In-Depth Information
Ta b l e 1 .
Analysis of five CDA strategies under the IKB model
ZI
ZIP
Kaplan
GD
RB
Information
Limit
Limit price and
Limit price and
Limit price and
Limit price and
Layer
price
transaction price and
Outstanding
history of bid/ask
transaction price
Current bid/ask and
bid/ask
and transaction price
and limit price
current profit margin
Knowledge
None
Competitive profit
Measures for
Belief that bid/ask
Target price based
Layer
margin, success
heuristics
will be accepted
on estimate of CE
of trade
price, risk factor
Behavioural Random
History,
No history,
History,
History,
Layer
predictive
non-predictive
non-predictive
predictive
- The Zero-Intelligence Plus (ZIP) Strategy [4]:
This is a predictive strategy that
uses the history of market information to predict the future market condition and
adapt to it. It learns the profit margin of agent
i
to remain competitive given the
changing market conditions. The IL collects
bid
(
t
k
),
ask
(
t
k
) and
price
(
t
k
) (as
instructed by the KL). The IL forwards this data, as well as the agent's profit margin
(private information in its IL), to the KL. That knowledge is then used in the BL to
predict the future market and adapt its profit margin,
μ
i
, to it. The BL then submits
A
i
=
<bid
i
|
ask
i
,silent>
,where
bid
i
or
ask
i
=(1+
μ
)
v
n
i
(
t
k
)
,i
.
- The Kaplan Strategy [7]:
This is a non-predictive strategy that makes a decision
based only on simple heuristics, and ignores the history of market information. The
IL collects the outstanding bid and ask (
bid
(
t
k
) and
ask
(
t
k
) respectively) from the
MS. Thereafter, using this information from the IL, the KL calculates the measures
that are used in the heuristic rules of Kaplan's BL [7]. These rules determine what
action,
ask
i
,silent>
, the agent
i
submits in the market.
- The GD Strategy [9]:
This is a non-predictive strategy that uses a history of market
information. The BL decides on an action,
<bid
i
|
A
i
=
<bid
i
|
ask
i
,silent>
, by solving a risk-
neutral utility maximisation problem involving a belief that a bid or an ask at a par-
ticular value will be successful in the market, and its limit price,
v
n
i
(
t
k
)
,i
. Thus, the
BL instructs the KL that it requires such knowledge. The KL then defines the Infor-
mation Filter (see figure 1), so that relevant information, namely the history of bids,
asks and transaction prices (
H
(
bid
(
t
k−
1
)),
H
(
ask
(
t
k−
1
)) and
H
(
price
(
t
k−
1
)) re-
spectively) are filtered to the IL. That information, along with the agent's limit price
is passed to the KL. The KL can then compute the belief and passes it, along with
the limit price, to the BL.
- The Risk-based (RB) Strategy [21]:
This strategy is predictive and uses a history
of market information. Furthermore, the RB has a more complex behaviour than the
ZIP. The intrinsic parameter of the strategy, which is updated in response to chang-
ing market conditions, is the risk factor associated with the current good to buy or
sell. The IL is instructed (by the KL) to record
bid
(
t
k
) and
ask
(
t
k
) and a history of
transaction prices,
H
(
price
(
t
k−
1
)). The KL then uses
H
(
price
(
t
k−
1
)) to estimate
the competitive equilibrium price
9
and then a target price (which the agent con-
9
The competitive equilibrium is a price at which transaction prices are expected to converge to
as given by the classical micro-economic theory [15].
