Sie sind auf Seite 1von 15

Decision Support Systems 43 (2007) 445 459

www.elsevier.com/locate/dss

Comparison of the group-buying auction and the


fixed pricing mechanism
Jian Chen a,, Xilong Chen a , Xiping Song b
b

a
Department of Management Science, School of Economics and Management, Tsinghua University, Beijing, 100084, China
Department of Systems Engineering and Engineering Management, The Chinese University of Hong Kong, Shatin, NT, Hong Kong, China

Received 21 December 2004; received in revised form 10 August 2006; accepted 5 November 2006
Available online 29 December 2006

Abstract
With the development of electronic commerce, online auction plays an important role in the electronic market. This paper
analyzes the seller's pricing strategy with the group-buying auction (GBA), a popular form of online auction, which is designed to
aggregate the power of buyers to gain volume discounts. Based on the bidders' stochastic arrival process and optimal strategy with
independent private value model, this paper analyzes the sellers' optimal price curve of the GBA in the uniform unit cost case and
in some supply chain coordination contracts. We find that the best discount rate is zero, which implies the optimal GBA is
equivalent to the optimal fixed pricing mechanism (FPM). Then we compare the GBA with the FPMin two special cases, the
economies of scale and risk-seeking seller, and find that (1) when economies of scale are considered, the GBA outperforms the
FPM; (2) when the seller is risk-seeking, the GBA also outperforms the FPM.
2006 Elsevier B.V. All rights reserved.
Keywords: Group-buying auction; On-line auction; Fixed pricing mechanism

1. Introduction
With the development of E-business, online auction
plays an important role. Johnson et. al. [13] point out
that the online consumer auction sales in the US will
reach $65 billion by 2010, accounting for nearly onefifth of all online retail sales. The popularity of the
online auctions creates many new kinds of price mechanisms, where the consumers participate more and more
in the price-setting process. The group-buying auction
(GBA) is one of them. As a homogeneous multi-unit
auction, the GBA has many users on the sites such as
Corresponding author. Tel.: +86 10 62789896; fax: +86 10
62785876.
E-mail address: jchen@tsinghua.edu.cn (J. Chen).
0167-9236/$ - see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.dss.2006.11.002

LetsBuyIt.com and Ewinwin.com. In contrast to the


traditional auction, where bidders compete against one
another to be the winner with the highest price, the
GBA enables bidders to aggregate to offer a lower price
at which they all win [11].
The GBA was once regarded as a promising novel
auction mechanism, e.g., the GBA websites Mercata.com
and Mobshop.com were once widely recognized as
international market leaders. Partly because they misused
the mechanism in the B2C market, Mercata closed down
in January 2001 [6], and Mobshop changed its strategic
direction to the B2B market [5]. Since then, many groupbuying websites have failed or reoriented themselves and
have given up the mechanism. However, some (e.g.,
LetsBuyIt.com and Ewinwin.com) still use the GBA as
their pricing mechanism and have passed the winter of the

446

J. Chen et al. / Decision Support Systems 43 (2007) 445459

E-commerce. The GBA is based on the idea that globally


locate, encourage and enable all buyers wishing to
purchase a particular product or service within a given
time frame to join forces in a buying group formed
specifically to accomplish the desired purchase [11]. If
this idea does work, why do so many websites fail?
Comparing to the traditional FPM, in which scenario can
the GBA bring us more profit? These questions motivate
us to study the GBA with a theoretical analysis.
The study on the GBA is rare. Most existing papers
on the GBA is experimental. They use the data from the
GBA websites to study the customer behaviors. Kauffman and Wang [14,15] analyze the changes in the
number of orders for Mobshop-listed products over
various periods, and find the positive participation
externality effect, the price drop effect, and the
ending effect. These experimental studies on consumer behavior are very useful for building the analytical
models to study the GBA. Anand and Aron [1] devise an
analytical model to study the GBA. Based on the assumptions that the revenue function is derived from a
deterministic demand curve, they compare the fixed
pricing mechanism (i.e., FPM) with the GBA in different scenarios, e.g., uncertain demand regime and
economies of scale. Using the simple analytic model,
they finds that the GBA outperforms the FPM when 1)
the seller faces uncertain market with two possible
intersecting demand regimes and 2) the seller sets the
price vector before production in combination with scale
economies. With Anand and Aron's results, we know
that the value of the GBA depends on the nature of the
uncertainty about the demand regimes. If the sellers do
know about in which demand regime they are operating,
they are almost always better off by running a postedprice market. So, it is critical for the seller to know about
the advantage of the GBA over the FPM before using it.
Although similar to Anand and Aron's paper we also
focus on the conditions that favor the GBA, we deal
different scenarios.
First, the demand assumption is different in the
following ways.
i) Different from the deterministic demand curve in
Anand and Aron's paper, the demand in our
model is based on the rational strategy of the
bidders in the GBA. According to Gupta and
Bapna [9], the customers in the online auctions
become involved in the price-setting process and
their bidding strategy becomes very important to
determine the trading price. Hence, studying the
GBA based on the customers' behavior becomes
more practical. Our paper based on Chen et. al.'s

paper [4], which builds a dynamic game model for


the GBA and studies the bidders' optimal strategy.
It proves that the mechanism is incentive compatible for bidders under the IPV (independent
private values) assumption.
ii) We consider a different demand uncertainty case.
In Anand and Aron's paper, they consider the
seller operates in two different demand regimes. In
each demand regime, the demand is deterministic.
In reality, even if we are clear of the customer
segmentation (demand regime), the visibility in
volatile B2C markets is sharply limited because so
many different variables are in play. According to
Sull [21], in volatile markets, many variables are
individually uncertain and they interact with one
another to create unexpected outcomes, the randomness in the demand is inevitable. Hence, we
consider the randomness in the regime. Pinker et
al. [19] point out that for the online auction, the
bidding process is an important issue. Our model
introduces the bidding process, which simulates
the randomness in the demand regime and the
bidders' strategy in the GBA.
Second, we study the GBA in the case when the risk
seeking seller is considered. Astebro [2] points out that
risk-seeking is one of several plausible reasons why so
many inventors proceed to develop their inventions while
only a small fraction can reasonably expect to earn positive
returns on their efforts. Risk seeking also applies to the
E-business world. Many websites care more about the
possible explosive expanding market as Ebay encounters
than the expected profit. Hence the resulting insights on
the risk seeking sellers can be of value to the sellers whose
objective is not only on the expected revenue.
This paper considers a situation that one auctioneer
uses the GBA to sell products to the individual online
consumers. As what happens in the online auction, in
our paper, the consumers with different values to the
objects may arrive to the auction with a stochastic
process. They make their decision according to their
rational strategy. The auctioneers may face uncertain
demand, supply chain contracts, economies of scale and
risk seeking scenario. The objective of our paper is to
answer the following questions: Can the GBA bring
more profit than the fixed pricing mechanism (FPM)? In
what scenarios does the GBA perform better?
The rest of this paper is organized as follows. In
Section 2, we explain the GBA and the FPM. Section 3
describes the problem that we study. In Section 4, the
equivalence of the optimal GBA and FPM is proved.
Section 5 studies the GBA in various scenarios. In

J. Chen et al. / Decision Support Systems 43 (2007) 445459

Section 6, we summarize the paper and highlight


directions for future study.
2. The group-buying auction and the fixed pricing
mechanism
Although there exist plenty of studies on traditional
auctions (see Klemperer (1999) [16] for surveys of the
auction literature), these results cannot be used in the
GBA directly because the GBA has the following
properties. 1) It is more often used in the B2C and B2B
markets. Hence, we should pay more attention to its
multi-unit property. 2) Unlike traditional auctions, in
which the number of bidders is fixed, in the GBA the
number is random and the seller does not know how
many bidders will take part. 3) It is widely used by
different companies whose risk attitudes are quite
different. Hence, we should not only consider the riskneutral profit-maximizing seller, as do most studies of
the traditional auction. Paying more attention to the
different objectives will be more practical.
We list the main notation in Appendix F for
reference. A typical GBA includes two rounds, the
offer round and the bidding round. The auction period T
and the number of objects N are exogenously given. The
seller acts first in the offer round, and determines the
price curve P = ( p1, p2,, pN), p1 p2 pN, where
pi denotes the price if the sold quantity is i. Then the
bidding round begins at time 0. In this round, the buyers
arrive with a stochastic process. The arriving buyer
determines whether and how much to bid. Once the
buyer bids, he is not permitted to change the bid. The
auction will end when either of the following two cases
occurs: 1) there are N bids; or 2) the time reaches T. To
distinguish the buyers who bid and the buyers who do
not bid, we call all of the buyers who come to the GBA
before T as the potential buyers and the buyers who bid
in the auction as the bidders. It should be noted that only
the potential buyers who come before the auction ends
have the opportunity to bid because the auction may end
before T. Although the number of potential buyers is
independent of the price curve P, because the buyers'
bidding strategy is related to the price setting, the
number of the bidders is related to the price level. N and
P are common knowledge. Only discrete bids are
allowed, i.e., bidders can only bid pi, i [1, N].
According to the mechanism of the GBA, the number
of the winners and the trading price in the GBA can be
determine by Rule I, which is equivalent to Chen et. al.
(2002)'s paper [4]:
Rule I: If the bidding vector is B =(b1, b2,, bn), nN,
then the trading price is p(B) = p q , where q

maxij

Pn
j1

447

Hbj pi zi, and where H is the heavyside

function, i.e.,

1 if
Hx
0 if

xz0
xb0

With the trading price pq, the bidders bidding higher


than or equal to pq will get the object and pay unit price pq.
Those who don't bid or bid lower than pq will get nothing
and pay nothing.
Rule I is illustrated in Fig. 1. Let the demand curve
denote the number of bids bidding higher than or equal to
the price. The demand curve and the price curve's right
point of intersection, K( p, q), denotes the trading price
and the number of winners. In Fig. 1, K's vertical coordinate, p, is the trading price and horizontal coordinate, q, is
the number of winners. The bidders bidding higher than or
equal to p will get one object with unit price p each. The
bidders bidding lower than p will get nothing and pay
nothing.
The FPM can be described as a special case of the GBA
with the horizontal price curve, i.e., P = ( p, , p). The
bidders get the products when and only when they pay the
unit price p.
3. Problem description
The GBA can be considered as a two-stage game
model: the offer stage for the seller and the bidding stage
for the bidders. The bidders' strategies in the bidding stage
have been studied by Chen et al. [4], based on which our
study emphasizes the seller's optimal strategy in the offer
stage in various scenarios. The argument is based on the
following assumptions.
Assumption 1. The independent private values model
applies, i.e. each potential buyer gives the object a
valuation, and the valuations are statistically independent.

Fig. 1. The trading price in the GBA.

448

J. Chen et al. / Decision Support Systems 43 (2007) 445459

Assumption 2. The potential buyers' valuations are in


monetary units, and the potential buyers are risk neutral.
Assumption 3. The potential buyers are symmetric, i.e.
all potential buyers draw their valuations from the same
probability distribution. Let F() denote the cdf of the
distribution and f() the pdf respectively.
Assumption 4. There is no collusion between potential
buyers.
Assumption 5. The potential buyers' arrival times are
statistically independent and have no correlation with
their values to the object and bids.
Assumption 6. Each potential buyer's demand is 1.
Segev et al. [20] point out that in online auction, approximately 76% of the purchases made were for one item.
Assumption 7. When a potential buyer decides not to
bid, he silently drops out forever, i.e., he leaves the
auction without record and never returns to it.
Assumption 8. The interval between the bidder's
arrival and bidding is omitted, i.e., the arrival time
equals the bidding time.
Assumption 9. The buyer chooses to buy the object when
there is no difference between buying and not buying.
Assumption 10. The buyers are individual rational.
With these assumptions, Chen et al. [4] prove that
strategy S is weakly dominant.
Strategy S: If a potential buyer's valuation of the
object is, , then he bids b = (), where
8
< p1 ; mzp1
hmu pj ; pj1 Nmzpj ; ja2; N 
1
:
0; mbpN
Here bidding 0 means that the bidder doesn't bid and
drops out forever. Strategy S implies that the bidder
should drop out when his value is lower than the lowest
price level pN. Otherwise, he should bid the price level
which is just below his valuation to the object, .
Because strategy S is the buyers' weakly dominant
strategy, it is rational for the bidders to take it. It is trivial
that in the FPM, if a buyer's valuation of the object is no
less than p, he will bid p and get one object with
payment p. Otherwise he will not buy the object.
According to rule I and strategy S, the sold quantity
q(Vn,P) is
qVn ; P maxij

n
X
r1

Hhmr pi zi

where Vn = (1, 2,, n) denotes n potential buyers'


valuations of the object, where vi denotes potential buyer
i's valuation, and the profit of the seller is q(Vn,P)
( pq(Vn,P) c), where pq(Vn,P) denotes the trading price.
In both the GBA and the FPM, the seller knows the
buyers' arrival process distribution and their valuation
distribution. The seller's objective is to set the optimal
price curve P to maximize his utility. In different
scenarios, the seller may have different utility
functions, e.g., the expected profit, the expected profit
with economies of scale, and the risk-seeking utility,
which will be studied in the following sections.
4. The seller's expected profit
In this section, we suppose that the seller is riskneutral and his objective is to maximize the expected
profit. If a seller sells k identical products at price p, and
the unit cost is c, then his profit is k ( p c). The seller
faces a trade-off between losing a sale because of a high
price and losing the customer's surplus because of a low
price.
If there are n potential buyers, the seller's expected
profit n(P) is:

pn P

N
X

Prq r; n; Pd rd pr c

r0

where Prq(r, n, P) denotes the probability that there are


r sold units with n potential buyers in the auction,
0 r N.
Lemma 1. Any given P, n, p, n (L(p,N)) n(P),
where L p; k p; p; N ; p denotes a k-dimensional
|{z}
k

vector with k identical elements, p.


Lemma 1 shows that if the seller knows exactly how
many potential buyers are in the auction, then the
optimal price curve is horizontal. In traditional auctions,
the number of bidders is open to the seller. Hence, this
lemma implies that using the GBA in a traditional
manner will not bring the seller more profit. However, in
an online GBA, potential buyers come to the auction in a
stochastic process. Hence, the seller does not know
exactly how many buyers will come. In the GBA,
according to the total probability
P formula, the sellers'
expected profit is pT P l
n0 PrA T ; nd pn P,
where PrA(T, n)denotes the probability that there are n
potential buyers in GBA with period T.

J. Chen et al. / Decision Support Systems 43 (2007) 445459

449

Theorem 1. P, T (P) T (L(p ,N)), where


p* arg max pT L p; N :

Table 1
Comparison of expected revenues under the different price curves

Theorem 1 implies even when the potential buyers'


arrival process is stochastic and not known to the seller,
the GBA cannot outperform the FPM either. Because the
GBA can mimic an FPM by setting a horizontal price
curve, the GBA is at least as good as the FPM. Hence,
the optimal GBA and the optimal FPM are equivalent.
Why can't the GBA utilize the discount to bring the
seller more profit? According to Chen et al. [4], the
GBA cannot motivate the bidders bidding higher. And
also, according to Assumption 6, because the demand
for each bidder is only one, the GBA cannot motivate
the bidders bidding more. Can the GBA play as a price
discrimination mechanism based on the number of bids?
Because the GBA asks for a higher unit price when the
market is worse (less sold quantity) and asks for a lower
unit price when the market is better (more sold quantity),
this mechanism cannot be an excellent price discrimination mechanism intuitively. It follows that if the
website only considers maximizing its expected profit, it
should not use a GBA, because it cannot bring more
profit. It should be noted that if c = 0, then this theorem
implies that should the seller's objective be to maximize
their revenue, the GBA is still not a good mechanism.
We now use a numerical example to illustrate the
result that the discount of the GBA cannot bring the
seller more profit than the FPM. Suppose that the
bidders' arrival process is a Poisson one with arrival rate
= 1/3. The auction period is T = 100. The quantity of
available objects is N = 40. The distribution of the
bidders' valuations is a normal distribution with mean
= 6 and standard variance = 2. The products' unit
cost c is 0.
In the Poisson arrival process case, we have

kth Element in price curve P (1 k N )

Expected revenue

pk = p = 4.67
pk = 4.67 0.05 (k 1)
pk = 5.6 0.1 (k 1)
pk = 5.6 0.02 (k 1)
pk = 4.67 0.05 (k 1)1.1
pk = 5.8 0.1 (k 1)1.1
pk = 5.8 0.02 (k 1)1.1
pk = 4.94 0.05 (k 1)0.9
pk = 5.5 0.1 (k 1)0.9
pk = 5.5 0.02 (k1)0.9

116.4
95.8
76.8
113.7
77.1
42.28
96.2
92.4
87.2
96.3

pT L p; N
p N ekT 1F p

N 1
X
N kkT 1Fpk
k0

R l
R l
where Cn; z z t n1 et dt; Cn 0 t n1 et dt,
we can get that p = 4.67437.
Table 1 shows that although different families of
price curves lead to different revenues for the seller,
none of them outmatch the optimal price level q. This
result confirms Theorem 1, i.e. under the assumptions
the optimal price curve is horizontal in the GBA.
This conclusion may partly explain why some
websites with the GBA failed in the B2C e-market.
Products commonly sold in this market are CDs,
cameras, and other electronic products, the cost
structures of which are similar to this model. Hence,
those websites that use the GBA with a positive discount
rate can hardly outperform their competitors.
Coordination plays an important role in the supply
chain management, where the sellers, called the retailer
in the Cachon et al.'s paper [3], may face different profit
functions [3].
We suppose that if there are r sold units, where
0 r N the seller's profit is
pr r; pr rapr bcm gcf

k!
4

By solving
ApT L p; N
Ap

CN 1; 1FpkT
N d 1
CN 1
kT CN ; 1F pkT

1F ppd f p 0;
CN
5

where cf is the fixed cost, c is the variable cost and , ,


and [0,1] are the divison ratio of the unit revenue,
variable cost, and fixed cost respectively for the seller. For
example, implies the division ratio of the unit revenue
between the seller and his upstream supplier is :(1 ).
The seller's expected profit with this kind of linear
profit function is
pc P

l
X
n0

PrA T ; n

N
X

Prq r; n; Pd rd apr bcgcf

r0

7
This profit function can describe the profit of the
sellers in many typic supply chain contracts.

450

J. Chen et al. / Decision Support Systems 43 (2007) 445459

For example, in revenue-sharing contract, the seller's


profit can be expressed as

pP /SN ; Pd pSN ;P cR wd N
l
N
X
X

PrA T ; n
Prq r; n; Pd rd /pr
n0

r0

cR wd N

arrival process is a Poisson process with arrival rate .


The price curve is P p1 ; N ; p1; p2 ; p2 ; N :. The cost
|{z}

where 0 u 1 is the seller's share of revenue, S(N, P)


is the sold quantity with price vector P and total order
quantity N, cR is the unit cost for the seller and w is the
wholesale price. Suppose the order quantity N is
determined before selling, the order cost is fixed when
determine the optimal price vector.
If we set a /; b 0; g cR c wN , which implies
f
that the seller will get u share of the revenue and pay the
fixed cost caused in the selling side, i.e., (cR + w)N,
Eq. (7) is the same as Eq. (8), which implies that the
revenue-sharing contract is a special case of the contract
considered in the form of Eq. (6).

function is C = (c1,c2,l), where c1 N c2 are the cost levels


and l is threshold to get the economies of scale. That is if
the sold units are less than or equal to l, the unit cost is
c1; if the sold units are more than l, the economies of
scale are achieved and the unit cost for all products is c2.
The seller's expected profit with economies of
scale is
pG;E P

y0

5. The GBA in other scenarios


5.1. Scenario 1: economies of scale
Demand aggregation is at the core of the GBA. By
bringing together as many bidders as possible, websites can negotiate lower prices with merchant partners
or manufacturers (e.g., Letsbuyit.com, Ewinwin.com).
For GBA websites, more sold units imply less unit
cost, i.e., economies of scale are important. In this
section we introduce the economies of scale into our
model when we compare the GBA with the FPM. Here
we use the all units quantity discounts schedule model. Suppose that a seller's utility function is (r, pr) =
r (pr cr), where r is the number of the sold units, pr
is the final trading price, and cr is the unit cost when
the volume is r.
For simplification, we only consider the price curve
with two price levels and assume that the supply can
always fulfill the demand. Suppose that the bidders'

x0

l
X
kT k ekT

k!
kl1
y
k
X X
zpk; x; y yd p2 c2


Theorem 2. P, C (P) C (L(p, N)), where p is the


optimal fixed price when contract (in the form of Eq. (6))
is considered.
Theorem 2 extends the result of Theorem 1. With
Theorem 2, it is sufficient to say that the GBA cannot
outperform the FPM with linear profit function. Hence,
with the supply chain coordination, if the sellers face
this kind of linear profit function, (e.g., revenue sharing
contracts, buy back contracts,) they should not use the
GBA as a pricing mechanism.

y
l
k
X
X
kT k ekT X
zpk; x; y
d
k!
y0 x0
k0
l
X
kT k ekT
xd p1 c1
k!
kl1
y
l
X X
zpk; x; yxd p1 c1


yl1 x0

where

zpk; x; y

k!1F p1 x F p1 F p2 yx F p2 ky
x! yx!ky!

With some algebraic transactions, the expected profit


can be formulated as follows,
pG;E P kT 1F p1 p1 c1 dw p2
kT 1F p2 p2 c2 1w p2

10

where w( p) = (l, T(1 F(p))) / (l) (0,1)) and


R l
R l
Cn; z z t n1 et dt; Cn 0 t n1 et dt.
If the seller posts a fixed price, p instead, it can be
deduced from Eq. (4) that the seller's expected profit is
pF;E p kT 1F p pc1 d w p
kT 1F p pc2 1w p

11

By optimizing Eqs. (5) and (6), we get the optimal


GBA price curve and the optimal FPM, with which we
can compare the expected profit of the GBA and that of
the FPM.

J. Chen et al. / Decision Support Systems 43 (2007) 445459

Theorem 3. max pG;E PN max pF;E p


P

Theorem 3's implication is interesting. Different


from the former scenario without economies of scale,
the GBA now strictly outperforms the FPM. This means
that economies of scale are important for GBA websites.
Because the GBA automatically asks for a higher price
when the unit cost is higher and a lower one when
the unit cost is lower, it can outperform the fixed
mechanism intuitively.
To illustrate this result, we provide a numerical example. Suppose that the potential buyers' arrival process
is a Poisson process with = 10; their valuations are
drawn from the same uniform distribution in interval
[0,1], T = 7, l = 10, 20, 30, c1 = 0.8, and c2 = 0.7, 0.6,
respectively. With Mathematica 4.0, by searching the
maximal point in Eqs. (10) and (11) with the above
parameters, we get the optimal price and maximal profit,
which are shown in Table 2.
As these numbers suggest, the GBA outperforms the
FPM when we consider economies of scale, which
confirms Theorem 3. And also we can clearly observe
that the improvement of the seller's profitability in the
GBA depends on the expected demand (T) and the
threshold (l). Comparing to the expected demand, a too
long or too short threshold provides a less improvement
than do the medium one. Because the GBA outperforms
the FPM through its adaptability, i.e., the GBA
automatically asks for a higher price when the unit
cost is higher and a lower one when the unit cost is
lower. A too long or too short threshold implies the
uncertainty of the unit cost is low, i.e., we are clear if we
can use the economies of scale or not. A medium
threshold implies the uncertainty of the unit cost is high,
i.e., we don't know if we can use the economies of scale
or not before the sales realized. Hence, in this scenario,
the adaptability of the GBA values higher.

451

If the seller can obtain volume discount from the


supplier, then the seller should choose the GBA as the
pricing mechanism because this choice will bring more
profit. However, as Kauffman and Wang [15] shows,
many GBA websites in B2C markets cannot achieve the
critical mass, i.e. which implies that the threshold is too
long for the demand, which ultimately induces their
failure. To solve this problem, trying to attract more consumers in different regions through the Internet is an
effective way for sellers to expand their consumer groups.
And it is also a good idea to sign with the supplier some
contract through which the GBA websites can get the
discount with a small threshold.
5.2. Scenario 2: risk seeking seller
Another focus of the networked economy is riskseeking. The players' different risk attitudes play an
important role on the application of the auction. In the
expanding E-business environment, there exists a fairly
large group of risk-seeking persons, who believe that no
venture no gain. As Gourville [8] points out, especially for
the sellers of the new products, gains have a greater impact
on them than similarly sized losses. Hence, though they
may not have sufficient sale scale in the past, they still take
a risk and hope to get it at some point in the future.
Quadratic utility function is popular used in the literatures
to study different risk attitudes [12]. In this paper, we also
use a simple quadratic utility function to model the riskseeking sellers, where we assume that the seller's utility
function is r2pr, instead of the revenue being rpr, where r
is the number of sold units and pr is the trading price. We
only consider the two-level price curve and suppose that
the supply always fulfills the demand. We also suppose
that the bidders' arrival process is a Poisson process with
arrival rate . Hence, the utility of the seller in GBA with
price curve P p1 ; N ; p1 ; p2 ; p2 ; N : is
|{z}
l

Table 2
Comparison of GBA and FPM with economies of scale
c2

p1, p2

G,E

F,E

l = 10
0.7
0.6

0.900, 0.817
0.900, 0.775

1.356
2.656

0.815
0.771

1.270
2.604

pG;R P

l X
k
z
X
X
kTk ekT
k!
k1 z1 y1

l = 20
0.7
0.6

0.900, 0.770
0.900, 0.702

0.783
1.562

0.900
0.900

0.700
0.700

l = 30
0.7
0.6

0.900, 0.754
0.900, 0.662

0.701
0.790

0.900
0.900

0.700
0.700

k!1F p1 y F p1 F p2 zy F p2 kz 2
y p1
y!zy!kz!

l X
k
z
X
X
kTk ekT
k!
k1 zl1 y1

k!1Fp1 y F p1 F p2 zy F p2 kz 2
y p1
y!zy!kz!
l
X
kT z 1F p2 z kT 1Fp2 2
e

z p2
12
z!
zl1

452

J. Chen et al. / Decision Support Systems 43 (2007) 445459

With some algebraic transactions, the expected utility


is formulated as,

pG;R P h22 h2 p2

Cl1; h2
Cl1
Cl; h2
h1 p1 h2 p2
;
Cl

h21 p1 h22 p2

13

where hi t(1 P(qi)), i = 1, 2


When the seller uses the FPM, the expected utility is

pF;R p kt1F pd kt1F p 1d p

14

where p is the fixed price.


With these formulae, we can now compare the
expected utility of the GBA and that of the FPM. Let
Paarg maxP pG;R P.
Theorem 4. For the optimal price curve P


p
1 ; N ; p1; p2 ; p2 ; N :, where l N 0, p1 N p2 .
|{z}
l

Theorem 4 implies that for the risk-seeking sellers,


the optimal price curve in the GBA is no longer horizontal. Because the GBA can mimic the FPM by
setting a horizontal price curve, this theorem also
infers that the optimal GBA will outperform the FPM
for the risk-seeking sellers. Because the optimal price
mechanism is to find out the best tradeoff between
the sold quantity and the unit price and comparing
with the risk neutral sellers, in the marginal utility
point of view, the risk-seeking one weighs more and
more on selling one more product than on getting a
higher unit price (the power 2 on r in the objective
function r 2 p r shows a heavier weight on r when r
becomes larger.). The GBA automatically adjusts the
tradeoff between the sold quantity and unit price that
can definitely benefit the seller. Different from the
economies of scale case where exists an exogenous
threshold l to get the discount, here, l is endogenetic,
yet how to set a suitable l in the optimal GBA is our
future work.

6. Conclusions and future studies


In this paper, we study the GBA mechanism in
three scenarios. First we consider the seller's expected
profit when he is risk-neutral. By comparing the GBA
with the FPM, we find that the optimal GBA is
equivalent to the optimal FPM. This result also applies
to the seller with linear profit division contracts with
his suppliers. Second, we introduce economies of
scale into our model, and find that the GBA will
outperform the FPM in this scenario. This result
implies that GBA is more suitable for selling the goods
that produce a learning effect; it also infers that there is
a critical mass for the GBA. If the GBA websites can
never achieve a critical mass of units sold no matter
the market is good or bad, then they can hardly outperform their competitors. Finally, we consider the
GBA in the e-economy context, where do exist a large
amount of risk seeking sellers. We find that the GBA
outperforms the FPM in this scenario, which implies
that the GBA is a suitable mechanism in the expanding
E-business environment, where the seller is riskseeking.
There is still much work to do. It has been pointed
out that the GBA, in contrast to the ordinary sale
mechanism, can attract buyers with different objectives [7]. Because these buyers may take different
actions, asymmetric buyers will have to be considered.
This paper models the GBA in the B2C markets,
where the individual consumers buy the products for
their own use. When we want to extend the GBA to the
B2B world, we should pay more attention to the
common value model or an affiliate value model [18].
And also, as we point out, achieving more consumers
is a key factor for the GBA websites to get the discount
from the supplier. Collusion, cooperation of the buyers, may make the existing consumers to invite more
friends to join the GBA. Hence considering the cooperation in the GBA is useful. Almost all auction
forms are susceptible to collusion, but degrees of
incentives vary [10] and collusion may heavily affect
the efficiency of the auction [17]. Because the GBA
may use collusion to attract more consumers and more
participants benefit the seller and the buyers both, introducing collusion to the GBA may lead to some
interesting results.
Acknowledgments
This work was supported in part by the National
Science Foundation of China under Grant No. 70321001,
NSFC/RCG Joint Research Grant 79910161987.

J. Chen et al. / Decision Support Systems 43 (2007) 445459

453

Appendix A. Proof of lemma 1


Before proving this lemma, we need first prove some lemmas and a corollary for technical reasons.
Notation. (V n, P) q(V n,P) ( p q(V n ,P ) c), which denotes the seller's profit when the price curve P and n
potential buyers' valuations Vn are given.
.
Lemma A1. For any given n and any price curve P pk1 ; pk1 ; N ; pk1 ; pk ; N ; pN , let P V pk ; pk ; N ; pk ; pk1 ; N ;
|{z}
|{z}
k

k1

pN , if u( pk,c) u( pk1,c), where u( p,c) = (1 F(p))(p c) denotes the >unit expected profit, then n(P) n(P).
Let us discuss the cases according to r, r = q(Vn,P).
C1) In the case that r [0, k): because pk 1 pk, it follows that q(Vn, P) = j r.
C j d 1F pk1 j d F pk1 F pk rj
PrqVn ; P jjqVn ; P V r r
1F pk r
r
X
EjVn ; PjqVn ; P V r
id pk1 cd PrqVn ; P ijqVn ; P V r
i1

pk1 cd

So

r
X

id Cri d 1F pk1 i d F pk1 F pk rj

i0

1F pk r

pk1 cd rd 1F pk1
1F pk

EjVn ; PjqVn ; P V r pu pk1 ; c

V1
EjVn ; P VjqVn ; P V r
pu pk ; c

Hence, E((Vn, P)|q(Vn, P) = r) E((Vn,P)|q(Vn, P) = r), r[0, k).


C2) In the case that r k: With the definition of q(), q(Vn, P) = r.
EjVn ; PjqVn ; P V rV EjVn ; P VjqVn ; P V r; rzk
Integrating C1) and C2),
pn P V

N
X

PrqVn ; P V rd EjVn ; P VjqVn ; P V rz

r0

N
X

PrqVn ; P V rd EjVn ; PjqVn ; P V

r0

r pn P:

.
Lemma A2. For any given n and price curve P pk1 ; pk1 ; N ; pk1 ; pk ; N ; pN , k bN, let P V pk1 ; pk1 ; N ; pk1 ;
|{z}
|{z}
k1

pk1 ; N ; pN , if u( p k, c) b u ( p k1, c), then n (P)N n (P).

Proof. Let us discuss the cases according to r, r =q(Vn,P).


C1) In the case that r [0, k):
Because the price elements in P are not lower than those in P, r q(Vn,P).
With Rule I, there are r bidders whose values are not lower than qk1, which implies that q(Vn,P) r.
Thus, q(Vn, P) = r = q(Vn,P).
Hence, E((Vn, P)|q(Vn, P) = r) E((Vn,P)|q(Vn,P) = r).
C2) In the case that r = k:
PrqVn ; P V jjqVn ; P r

Crj d 1F pk1 j d F pk1 F pk rj


:
1F pk r

454

J. Chen et al. / Decision Support Systems 43 (2007) 445459

EjVn ; P VjqVn ; P r

r
X

jd pk1 d PrqVn ; P V jjqVn ; P r

j0
r
X
j
pk1 cd
jd Crj d 1F pk1 d F pk1 F pk rj
j0

1F pk r
EjVn ; PjqVn ; P r
pu pk ; c

b1.
It follows that
EjVn ; P VjqVn ; P r pu pk1 ; c

pk1 cd rd 1F pk1
:
1F pk

Hence, E((Vn,P)|q(Vn,P)=r) N E((Vn,P)|q(Vn,P) =r).


C3) In the case that r N k: with the definition of q(), q(Vn, P) = q(Vn, P) = r.
Hence E((Vn,P)|q(Vn,P) =r) =E((Vn,P)|q(Vn,P) =r).
Integrating C1), C2) and C3), with total probability formula,
N
X
PrqVn ; P rd EjVn ; P VjqVn ; P r
pn P V
r0

N
X

PrqVn ; P rd EjVn ; PjqVn ; P r pn P:

r0

Corollary A1. For any given n and price curve P, let B pxN 1 ; N ; pxN 1 ; pN , where pxN 1 aarg maxpj
|{z}
pu pj ; c; 1VjVN 1; pn PVpn B:
N 1
.
Proof.
Lemmas A1 and A2 imply k b N, n(P) En(P), where P pk1 ; pk1 ; N ; pk1 ; pk ; N ; pN and P V
|{z}
k1

px ; px ; N ; px; pk1 ; N ; pN with xaarg maxj pu pj ; c; j k or k 1. It follows that


|{z}
k

pn PVpn px2 ; px2 ; N ; pN Vpn px3 ; px3 ; px3 ; N ; pN V N Vpn pxN1 ; N ; pxN 1 ; pN pn B;
|{z}
|{z}
|{z}
2

N 1

Where pxN 1 aarg maxpj pu pj ; c; 1VjVN 1.

.
Lemma
A3. For any given n and price curve B pN 1 ; N ; pN1 ; pN , if u(pN,c) b u(pN1,c), let B = L(pN1,N), n
|{z}
(B) n (B).
N1
Proof. Let us discuss the cases according to r, r =q(Vn, B).
C1) In the case that r [0, N), similar with Lemma A2 C1).
EjVn ; B VjqVn ; B r EjVn ; BjqVn ; B r

C2) In the case that r = N:


This implies that there are at least N potential buyers who value the objects at no less than qN. Because the price
elements in B are not lower than those in B, q(Vn,B) q(Vn,B).
Suppose that there are i(i N) potential buyers who value the objects at no less than qN, which is denoted by AR = i, then
N 1
X
PrfqVn ; B V jjAR igd jd pN1 c PrfqVn ; B V N jAR ig
EjVn ; B VjAR i
j0
N 1
X

 N d pN 1 c
i
X

jd pN 1 cd Cij 1F pN 1 F pN 1 F pN ij

j0

1F pN i
j

N d pN 1 cd Cij 1F pN 1 F pN1 FpN ij

jN

1FpN i

J. Chen et al. / Decision Support Systems 43 (2007) 445459


i
X

i) If akaN ; i;
i
X

455

pN 1 cdCij 1F pN 1 j F pN 1 F pN ij

jk

N pN c, then

1F pN i
j

N d pN 1 cd Cij 1F pN 1 F pN 1 F pN ij

jk

NN d pN c:

1F pN i
Hence, E((Vn,B)|AR = i)
N 1
X
j0

i
X

1F pN i
i
X

N d pN 1 cd Cij 1F pN 1 F pN 1 F pN ij

jk

N d pN 1 cd Cij 1F pN 1 F pN 1 F pN ij

jk

1F pN i
N N d pN c EjVn ; BjAR i:
i
X

ii) If 8kaN ; i;

N d pN 1 cd Cij 1F pN 1 F pN 1 F pN ij

jN

1F pN i

X k1

jd pN 1 cd Cij 1F pN 1 j F pN 1 F pN ij

1F pN i

pN1 cd Cij 1F pN 1 j F pN 1 F pN ij

jk

V pN c, then

1F pN i

j
i
i
X
X
pN 1 cd Cij d 1F pN 1 d F pN1 F pN ij
ViN d pN c:
1F pN i
kN 1 jk

Because
j
i
i
X
X
pN 1 cd C ji d 1F pN 1 d F pN1 F pN ij
1F pN i
kN 1 jk
i
X
pN 1 cd jN d C ji d 1F pN 1 j d F pN 1 F pN ij
;
1F pN i
jN 1

Hence,
i
X
pN1 cd jN d Cij d 1F pN 1 j d F pN 1 F pN ij
ViN d pN c
1F pN i
jN 1
i
X

As

jd pN1 cd Cij d 1F pN1 j d F pN 1 F pN ij

j0

1F pN

A1

id pN 1 cd 1F pN 1
and
1F pN 1

pu pN 1 ; c pN 1 c1F pN1 z pN c1F pN pu pN ; c;

i
X

jd pN 1 cd Cij d 1F pN 1 j d F pN 1 F pN ij

j0

1F pN i

Nid pN c

A2

456

J. Chen et al. / Decision Support Systems 43 (2007) 445459

With in Eqs. (A1) and (A2), it can be deducted that


N
X

jd pN 1 cdC ji d1F pN 1 j dF pN 1 F pN ij

i
X

N d pN 1 cdC ji d1F pN 1 j dF pN 1 F pN ij

jN 1

j1

1F pN i

NN d pN c
i.e., E((Vn, B)|AR = i) N E((Vn,B)|AR = i).
Thus, in both cases
EjVn ; B VjAR iNEjVn ; BjAR i;
Because q(Vn,B) = N if and only if i N, s.t. AR = i,
EjVn ; B VjqVn ; B N

l
X
iN

l
X

iN

Pr

fAR igEqVn ; L pN 1 ; N jAR i


PrfqVn ; B N g

PrfAR igEqVn ; BjAR i


PrfqVn ; B N g

EjVn ; BjqVn ; B N ;

Hence,
EjVn ; BVjqVn ; B N zEjVn ; BjqVn ; B N

Thus, with total probability formula


pn B
V

N
X
r0
N
X

PrfqVn ; B rgd EjVn ; BjqVn ; B r


PrfqVn ; B rgd EjVn ; B VjqVn ; B r pn B V:

r0

Proof of Lemma 1. According to Corollary A1, let B pxN1 ; N ; pxN 1 ; pN , where pxN1 aarg maxpj pu pj ; c; 1V
|{z}
jVN 1; pn PVpn B. Hence,
N 1
1) if u( pN,c) u ( pN1,c), then n(L( pN,N)) n(B) n(P) with Lemma A1,
2) if u ( pN, c) b u ( pN1, c), then n (L( pN1, N)) n (B) n (P) with Lemma A3.
Appendix B. Proof of Theorem 1
With Lemma 1, for any given P there exists p, which
is irrelated to n, s.t., for any n, n(L( p, N)) n(P).
With total probability formula,
pT P
V

l
X
n0
l
X
n0

Suppose that p aarg maxp pT Lp; N i.e., for any


p, T (L( p,N )) T(L( p,N )).
It follows that T (P) T(L( p,N )).

Appendix C. Proof of Theorem 2


PrA T ; nd pn P
PrA T ; nd pn L p; N pT L p; N :

The proof is very similar to the proofs of


Lemma 1 and Theorem 1. The only difference is to
define u( p, c) = (1 F( p)) (p c) instead of u( p,c) =
(1 F( p)) ( p c).

J. Chen et al. / Decision Support Systems 43 (2007) 445459

Appendix D. Proof of Theorem 3


Suppose that paarg maxp pF;E p. It follows that

457

With some algebraic transactions, the expected utility


of the seller in the GBA can be formulated as follows,

c2 1w pf p1F p

Cl1; kT 1F p2
pF;R p2
Cl1
Cl1; kT 1F p2
 1
Cl1

c1 c2 w V p 1F p

I p1 I p2

pG;R P pF;R p1
dpF;E p=dppp 0 kT pf p c1 w p

Because w( p)= T eT (1F( p)) (T (1 F ( p ))) l1


f ( p) / (l ) 0,
1F pc1 c2 w V p 1F ppf p
c1 w p
c2 1w p f pz0
Suppose that p
1 arg maxp pu p; c1 . It follows that
u( p,c1) / p|p=p1 = 0
Hence,
Apu p; c1 =Apjpp 1F ppf p c1 f p
1F ppf p c1 w p
1w p f pN1F ppf p
c1 w p c2 1w p f p
1F pc1 c2 w V pz0
Apu p; c1 =Apjpp
1
So p b p1.
It follows that

Because

AI p
Ap

pp

pF;R pkTf p pG;R V pkT 1F p 1

kT 1F p 12
pF;R pkTf p

N0
kT 1F p 12

ApG;R P
Ap1

pF;E p max pF;E p

Define I( p) = F,R( p) / (T(1 F( p)) + 1)

p1 p;p2 p

AI p1 kT 1F p2 l1 kT 1F p2
e

l1!
A p1
p1 p; p2 p
kT 1F p2 l1 ekT 1F p2 AI p1
A p1
l1!

p1 p; p2 p

N0

p1 p; p2 p

NkT 1F p pc1 w p
1F p pc2 1w p

Suppose that p arg max pF;R p



l1; kT 1F p2
=Ap2
pF;R p1 pF;R p2 A C
Cl1


Cl1; kT1F p2
pF;R V p2 1
Cl1
!
kT 1F p2 l1 ekT 1F p2
I p1 I p2 A
=Ap2
l1!

kT 1F p
1 p1 c1 wp
1F p pc2 1w p

Appendix E. Proof of Theorem 4

ApF;R p1 Cl1; kT 1F p2
A p1
Cl1

ApG;R P
Ap2

It follows that maxP pG;E PN maxp pF; E p.

It follows that

max pG;E Pz p
1 ; N ; p1 ; p; N
|{z}
P

kT 1F p2 l1 kT1F p2
e
:
l1!

kT1F p2 l1 ekT1F p2
I V p2 g
l1!
p1 p; p2 p

kT 1F p2 l1 ekT 1F p2
I V p2
l1!
p1 p; p2 p

kT 1F pl1 ekT 1F p

l1!

I V pb0

458

J. Chen et al. / Decision Support Systems 43 (2007) 445459

Hence, there exists a feasible ascent direction at the


point ( p, p,) for G,R(P)
It follows that

F,R (P) Denotes the seller's utility in the FPM with the
price p when the seller is risk seeking.

p1 Np2

References

Appendix F. The notation


N
The number of objects to be sold in the GBA
T
The auction time
P = ( p1, p2, ..., pN), p1 p2 pN The price curve
in the GBA
L p; k p; p; N ; p Denotes a k-dimensional vector
|{z}
k

with k Identical elements, p.


Vn = (v1, v2, , vn) Denotes n potential buyers' valuations
iof the object, where vi denotes potential buyer i's
valuation.
PrA (T, n) Denotes the probability that there are n
ipotential buyers in GBA with period T
c
Is the unit cost of the auctioned products in the
scenario without economies of scale.
C = (c1,c2;l) Where c1 N c2 are the cost levels and l is
ithreshold to get the economies of scale. That is if
the sold units are less than or equal to l, the unit
cost is c1; if the sold units are more than l, the
economies of scale are achieved and the unit cost
for all products is c2.
Given the price curve P and posted price p
q(Vn, P) Denotes the sold quantity in the auction with n
potential buyers, Vn,
Prq (r, n, P) Denotes the probability that there are r
sold units with n potential buyers in the
auction, 0 r N
u ( p, c) = (1F( p))( p c) Denotes the unit expected
iprofit, where for any given c, u (p, c)is
unimodal in p.
n (P) Denotes the seller's expected profit with n
potential buyers in the auction.
T (P) Denotes the seller's expected profit in the GBA
with auction time T
C (P) Denotes the seller's expected profit in the GBA
with some coordination contracts.
G,E (P) Denotes the seller's expected profit in the GBA
with economies of scale where the cost function is
C.
F,E (P) Denotes the seller's expected profit in the FPM
with price p when the cost function is C
G,R (P) Denotes the seller's utility in the GBA when
the seller is risk seeking.

[1] S.K. Anand, R. Aron, Group-buying on the Web: a comparison


of price discovery mechanisms, Management Science 49 (2003)
15461562.
[2] T. Astebro, The return to independent invention: evidence of
unrealistic optimism, risk seeking or skewness loving? Economic
Journal 113 (2003) 226239.
[3] G.P. Cachon, M.A. Lariviere, Supply chain coordination with
revenue sharing contracts: strengths and limitations, Management Science 51 (2005) 3044.
[4] J. Chen, X. Chen, X. Song, Bidder's strategy under groupbuying auction on the Internet, IEEE Transactions on Systems,
Man and Cybernetics. Part A. Systems and Humans 32 (2002)
680690.
[5] D. Clark, Mobshop, a pioneer in group-buying on the web,
discontinues consumer service, Wall Street Journal (2001)
(Available on the Internet at www.mobshop.com/ar011501d).
[6] J. Cook, Venture capital: Where Mercata Led, Consumers Were
Unwilling to Follow. Seattle Post-Intelligencer, Available on the
Internet at http://seattlep-i.nwsource.com/business/vc122.shtml
2001.
[7] B. Gottlieb, Does Group-Shopping Work? The Economics of
Mercata and Mobshop, http://www.slate.lycos.com/Features/
groupshop/groupshop.asp 2000.
[8] J.T. Gourville, Eager sellers and stony buers: understanding the
psychology of new-product adoption, Harvard Business Review
(June 2005) 98106.
[9] A. Gupta, R. Bapna, Online auctions: a closer look. Handbook of
electronic commerce in business and society, CRC Press, Boca
Raton (FL), 2001.
[10] K. Hendricks, R.H. Porter, An empirical study of an auction with
asymmetric information, American Economic Review 78 (1988)
865883.
[11] T.V. Horn, N. Gustafsson, Demand Aggregation through Online
Buying Groups, U.S. Patent 6047266, 2000.
[12] J.M. Jia, J.S. Dyer, A standard measure of risk and risk-value
models, Management Science 42 (1996) 16911705.
[13] A.C. Johnson, B. Tesch, A Forecast and analysis of US auction
sales to consumers, Forrester Research, 2005.
[14] R.J. Kauffman, B. Wang, New buyers' arrival under dynamic
pricing market microstructure: the case of group-buying
discounts on the Internet, Journal of Management Information
Systems 18 (2002) 157188.
[15] R.J. Kauffman, B. Wang, Bid Together, Buy together: On the
Efficacy of Group-Buying Business Models in Internet-Based
Selling, Handbook of Electronic Commerce in Business and
Society, CRC Press, Boca Raton, FL, 2002.
[16] P. Klemperer, Auction theory: a guide to the literature, Journal of
Economic Surveys 13 (1999) 227286.
[17] P. Klemperer, Collusion and Predation in Auction Markets,
Nuffield College, Oxford University, England, 2001 www.nuff.
ox.ac.uk/economics/people/klemperer.htm.
[18] P.R. Milgrom, R.J. Weber, A theory of auctions and competitive
bidding, Econometrica 50 (1982) 10891122.
[19] J.E. Pinker, A. Seidmann, Y. Vakrat, Managing online auctions:
current business and research issues, Management Science 49 (2003)
14571484.

J. Chen et al. / Decision Support Systems 43 (2007) 445459


[20] A. Segev, C. Beam, J. Shanthikumar, Optimal design of Internetbased auctions, Information Technology and Management 2 (2001)
121163.
[21] D.N. Sull, Strategy as active waiting, Harvard Business Review
(September 2005) 120131.

Jian Chen received the B.Sc. degree in Electrical Engineering from


Tsinghua University, Beijing, China, in 1983, and the M.Sc. and the
Ph.D. degree both in Systems Engineering from the same University in
1986 and 1989, respectively.
He is a Professor and Chairman of Management Science Department,
Director of Research Center for Contemporary Management, Tsinghua
University. Dr. Chen has over 100 journal papers and has been a
principal investigator for over 30 grants or research contracts with
National Science Foundation of China, governmental organizations
and companies. His main research interests include modeling and
control of complex systems, decision support systems and information
systems, forecast and optimization techniques, supply chain management, E-commerce.
He is a senior member of IEEE and a member of INFORMS. He serves
as Chairman of the Service Systems and Organizations Technical
Committee of IEEE Systems, Man and Cybernetics Society, vice
president of China Society for Optimization and Overall Planning, a
member of the Standing Committee of Systems Engineering Society of
China, China Information Industry Association and Decision Science
Society of China. He is the recipient of Science and Technology
Progress Awards of Beijing Municipal Government, 2000 and 2001;
the Outstanding Contribution Award of IEEE Systems, Man and
Cybernetics Society, 1996; Science and Technology Progress Award of
the State Educational Commission, 1994.
He is the editor of the Journal of Systems Science and Systems
Engineering, an associate editor of IEEE Transactions on Systems,
Man and Cybernetics: Part A and IEEE Transactions on Systems,
Man and Cybernetics: Part C, and serves on the Editorial Board of
Systems Research and Behavioral Science, International Journal of
Information Technology and Decision Making and International
Journal of Electronic Business. He is the Secretary General of the
1996 IEEE International Conference on Systems, Man and Cybernetics in Beijing, co-chair of the IPC of the 3rd (1998) and 4th (2003)
International Conferences on Systems Science and Systems Engineering, and Chair of the 1st Asian eBiz Workshop in 2001, co-chair of the
International Conference on Service Systems and Service Management in 2004, Chair of the 4th International Conference on Electronic
Business in 2004.

459

Xilong Chen received the PhD. Degree from Tsinghua University,


Beijing, China, in 2004. He is currently a doctoral student at North
Carolina State University, USA.

Xiping Song received the M.Sc. Degree from Tsinghua University,


Beijing, China, in 2003. He is currently a doctoral student at the
Chinese University of Hongkong, Hongkong, China.

Das könnte Ihnen auch gefallen