Int. J.

Production Economics 65 (2000) 141}171

Optimal procurement policies under price-dependent demand

Hakan Polatoglu *, Izzet Sahin

Frank Sawyer School of Management, Suwolk University, Boston, MA 02108, USA

School of Business Administration, University of Wisconsin-Milwaukee, Milwaukee, WI 53201, USA
Received 19 June 1998; accepted 17 September 1998

Abstract
We study a periodic-review inventory model where, in addition to the procurement quantity, price is also a decision
variable. We develop a model where demand in each period is a random variable having a price- and, possibly,
period-dependent probability distribution, with the expected demand decreasing in price. The model includes price limits
and "xed ordering costs in addition to unit procurement holding and shortage costs. We study the optimal policies which
jointly maximize the discounted expected pro"t over a "nite planning horizon. We characterize the form of the optimal
procurement policy under a general price}demand relationship and give a su$cient condition for it to be (s , S ) type. We
L L
also discuss some special cases and extensions to the basic model, including the in"nite horizon problem.  2000
Elsevier Science B.V. All rights reserved.
Keywords: Pricing; Inventory; (s, S) policy

Notation
n
N
i
L
q
L
p
L
Pl
P

c
r
h
v
K
L

period index (n"1 corresponds to the last period)

total number of periods in the planning horizon
beginning inventory level before ordering in period n
beginning inventory level after ordering in period n
price in period n
price #oor
price ceiling
unit procurement cost
unit shortage cost
unit holding cost
unit salvage value at the end of the planning horizon
"xed ordering cost in period n

* Corresponding author. Protek Computer Systems Inc., Perpa Ticaret Merkezi, Elektrokent, Kat 13, No 1981, Okmeydani, 80270
Istanbul, Turkey. Tel:. #90 (212) 210-1730; fax: #90 (212) 210-1765.
E-mail: Hakan.Polatoglu@protek.com.tr (H. Polatoglu)
0925-5273/00/$ } see front matter  2000 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 5 - 5 2 7 3 ( 9 8 ) 0 0 2 4 0 - 0

142

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

a
X(p)
XM (p)
(p)
;(p)
f (x; p)
F(x; p)
p (q )
L L
M (i , p , q )
L L L L

periodic discount factor

random demand when price is p
expected demand as a function of price ("E[X(p)])
lower bound on random demand, 0)(p))X(p)
upper bound on random demand, X(p));(p)(R
probability density function of X(p)
probability distribution of X(p)
optimal price when the inventory level is q in period n
L
n-period (periods n}1) pseudo-pro"t function (pro"t excluding the e!ect of the "xed ordering
cost)
M
M (i , p , q ) expected value of M (i , p , q )
L L L L
L L L L
M
M (i , q )
n-period expected pseudo pro"t when the optimal value of p is implemented in period n, i.e.,
L L L
L

M (i , p (q ), q )
M
M (i , q ),M
L L L
L L L L L
PM (i , p )
n-period expected pro"t when the optimal value of q is implemented in period n
L L
L
PM (i )
n-period expected pro"t when the optimal values of q and p are implemented in period n
L
L
L
d(z)
Heavyside function, d(z)"0 for z)0 and d(z)"1 for z'0
[z]>
positive value function, [z]>"zd(z)

1. Introduction
Under increased competition, inventory-based businesses are forced to better coordinate their procurement and marketing decisions, to avoid carrying excessive stock when sales are low or shortages when they
are high. An e!ective means of such coordination is to conduct the inventory control and pricing decisions
jointly. The main task in doing so is to determine the optimal inventory policy, given the price}demand
relationship that is expected to prevail in the market place in the short term.
In addressing the above issue, we are concerned in this paper with a single-item, periodic-review inventory
system where the vendor, who enjoys a degree of monopoly power in the market place, is in a position to
in#uence demand by its pricing decisions. It is thus confronted with simultaneous pricing and procurement
quantity decisions, which would jointly maximize the present value of expected pro"t over a planning horizon.
We will assume that the ordering policy does not change the demand pattern during the planning horizon.
Ordering cost, shortage cost and temporal increases in the procurement cost are among the important
factors that force the decision maker to carry inventories. Holding cost has the opposite e!ect. Thus, in the
absence of pricing, the main decision problem is to determine the optimal inventory levels to strike a balance
between these opposing factors, under a given demand forecast, cost structure, and a set of operating
conditions.
There are critical di!erences between a "xed-price, periodic-review inventory model [1] and a model that
also includes pricing decisions. The economical interpretations of these di!erences relate to the various ways
that price plays into the decision problem. First, price is a decision variable that determines the revenue per
unit sold. Second, price is a factor that in#uences the demand, thus the period-ending inventory levels. In
addition, when backlogging occurs, apart from the amount of backorders, the backlogging policy must also
account for the price that applies to backorders and the timing of revenue collection. For instance,
backorders could be sold at the current price in advance, or at a future price set at the time of delivery.
Therefore, when the decision maker confronts the pricing and inventory decisions simultaneously,
in addition to the two opposing cost-related e!ects that we noted above, price-related factors must also
be taken into account. Mainly, there will be a trade-o! between the high-price low-demand and low-price
high-demand scenarios, in terms of discounted total revenue. This revenue trade-o!, however, will not be

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

143

independent of the above described cost trade-o!, since the demand levels (i.e., inventory levels) will be
a!ected by the pricing decisions. Unfortunately, this complex relationship between the cost and revenue
trade-o!s does not allow the model to simplify into separate pricing and procurement decisions.
Various versions of the procurement-and-pricing model has been studied in the literature. In his pioneering work, Whitin [2] proposed a link between price theory and inventory control. He noted that such
a model would have stronger managerial implications, as compared to "xed-price models. Later, Mills [3,4]
and Karlin and Carr [5] established a conceptual framework for a general inventory model, where price is
a decision variable. Subsequent studies concentrated mostly on the characterization of the optimal solution
for some special cases of the general model [6}13]. Thomas [14] provided some numerical examples to
demonstrate the nature of the decision problem under the presence of "xed ordering costs. More recently,
Gallego and Van Ryzin [15] studied the continuous-time version of the problem, and Petruzzi [16]
investigated the approximate solutions under a learning approach. Also, under a dynamic model with
Bayesian learning about the demand distribution, Subrahmanyan and Shoemaker [17] studied a number of
numerical examples which provide insights about the sensitivity of optimal prices and inventory levels to
changes in the unit procurement cost, price elasticity of demand, form of the demand distribution and the
expected demand function.
The existing analytical models impose rather restrictive assumptions on the form of the expected demand
function (e.g., concavity assumptions in [9,11,12]), demand distribution (e.g., multiplicative demand in [9],
additive demand in [9,11]), or the cost structure (e.g., parameter restrictions and no "xed ordering cost in
[11,10]) in analyzing the optimal procurement policy. In this paper, by relaxing some of the more limiting
modeling assumptions, we seek to characterize the fundamental properties of the optimal procurement policy
and the resulting expected pro"t in a more general setting (i.e., general demand distribution, linear cost
structure with the addition of set-up cost). We also address the sensitivity of the optimal procurement policy
to the underlying price-dependent demand uncertainty.
In what follows, we describe the price}demand relationship in Section 2, develop the basic model in
Section 3, and characterize its solution in Section 4. The rest of the paper is devoted to special cases and
extensions. We consider the in"nite-horizon problem in Section 5, the model with no "xed ordering costs in
Section 6, the non-stationary extensions of the basic model in Section 7, and the special case with deterministic demand in Section 8. Section 9 is devoted to some concluding remarks. Sections 4 and 5 include
numerical examples. Proofs are given in the appendix.

2. Demand uncertainty
It has been a common practice in demand modeling to express random demand as a combination of an
expected demand function, which exhibits some form of price-dependency, and a random term, which is
price-independent. This approach conveniently isolates the e!ects of price and uncertainty, while retaining
mathematical tractibility. However, it has some shortcomings.
Under the additive model [3,5,9,11], we have X(p)"XM (p)#e where XM (p) is a decreasing function of p and
e is a random variable with E[e]"0. The additive model can also be interpreted as a homoscedastic
regression model where XM (p) represents the regression function and e is the error term. One implication of
this model is that while the expected demand is a function of the price, the demand variance is priceindependent (demand distribution shifts as price varies). Also, the model allows for negative demand, unless
the price values are bounded from above.
The multiplicative model [5,6,8] is another commonly used form where X(p)"eXM (p) with E[e]"1. This
model implies the restriction that the demand equals to the product of its expected value and a random term.
As a result, the coe$cient of variation of demand equals that of e, which is constant in price, and demand
variance decreases at a rate faster than the expected value and it approaches to zero at high prices.

144

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

In addition to these, there are models that represent random demand by a mixture of additive and
multiplicative terms [10,12,13], such as X(p)"XM (p)a (e)#a (e), or X(p)"a (p)#ea (p), where a and





a are di!erentiable functions.

Additional simpli"cation can be achieved by assuming that the price}demand relationship is perfectly
predictable. This leads to the deterministic model, X(p)"XM (p), which serves as a "rst-order approximation,
and which has been utilized as a benchmark in the literature [3,5,9,12]. It has been reported in these studies
that the form of demand uncertainty, whether it is additive or multiplicative, plays a critical role in
determining optimal prices. Optimal prices are lower than their deterministic counterparts under the additive
demand model, while they are found to be greater under the multiplicative demand model. Additional
"ndings about the form of demand uncertainty, as it impacts optimal prices and inventory levels, are reported
in [18] p. 617.
In this study, we represent the demand by a continuous random variable, distributed over the range
[(p), ;(p)] with a known density function f (x; p). and ; are di!erentiable functions which represent the
bounds on demand, where 0)(p))X(p));(p)(R for all p. For convenience, we "rst assume stationarity, whereby the demand density is given by f in all periods. We then show in Section 8 that the results we
obtain for the "nite-horizon problem can be extended to non-stationary (period-dependent) demand
distributions.
There is empirical evidence that consumers react to price in purchasing. Other things being equal, a lower
price facilitates demand in a `faira market. Therefore, it is natural to assume that the probability that demand
is less than a given level x, F(x; p)"P[X(p))x], is a non-decreasing function of price. That is,
*F(x; p)
*0 x3((p), ;(p)).
*p

(1)

It follows from Eq. (1) that F(x; p ))F(x; p ) for p (p . Thus, price induces a stochastic ordering of




demand distributions.
The expected demand is given by
3N

XM (p)"
xf (x; p) dx" [1!F(x; p)] dx.

(2)

*N

It also follows from Eqs. (1) and (2) that XM (p) is a decreasing function of p:

*F(x; p)
dXM (p)
*F(x; p)
*0N
"!
dx(0.
*p
dp
*p


3. Mathematical model and assumptions

Consider a periodic-review inventory system with N periods where the decision variables are q and p . The
L
L
review periods are linked by period-ending inventory levels such that the leftovers are transferred fully to the
next period and shortages are lost. That is, i "[q !X(p )]> for 0)n)N!1, and i *0 is a given
L
L>
L>
,
parameter. At the beginning of a period, the vendor decides how much to order (q !i ) and what price to
L
L
charge until the next decision point. There is a "xed cost of ordering (K), but no cost is assumed for pricing.
Therefore, it is for the vendor's bene"t to reconsider pricing at each decision epoch.
We assume that the vendor has full information about the inventory and procurement costs, and the
demand distributions in all periods of the planning horizon. At the beginning of a review period, say period n,

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

145

given the inventory position, the vendor is to determine the procurement quantity and the price to maximize
the expected n-period pro"t which represents the expected value of the sum of the current period's pro"t and
the discounted optimal expected pro"t to be obtained during the remaining periods.
Let PLl and PL be the price #oor and the price ceiling, respectively, in period n, such that the interval

[PLl, PL ] represents the range of allowable prices. Then, the sequences +Pl , Pl ,2, P,l , and +P, P,2, P,,

 

establish limiting-price pro"les during the planning horizon. In regulated markets, these pro"les represent
the temporal rules of regulation. Also, in a case where a manufacturer might issue price limits for its dealers
(price-maintenance), or provide `suggested retail pricesa, the allowable price ranges could be represented by
the limiting-price pro"les.
In the basic model, without loss of generality, we assume uniform pro"les where the feasible price values in
any period are between Pl and P . If there is no price regulation or other constraint on prices, then the price

limits may be taken as 0 and p , respectively, where p , referred to as the `null pricea in the literature, is the


highest possible price level at which there will be no demand, that is, P[X(p))0]"1 for all p*p .

Allowing p "R leads to detailed technical considerations (see [8,9]) which we choose to avoid in this paper.

We assume that inventory costs are proportional to period-ending inventory levels. The cost and discountfactor parameters may be allowed to change from period to period. However, as in the case of the demand
distribution, we keep them time-invariant for convenience in developing the basic model (see Section 8 for
non-stationary extensions of the basic model). We take P 'c so that it is possible to make pro"t. In

addition, the leftovers at the end of the planning horizon are assumed to be salvaged at a discount price, that
is, v)c.
Under these assumptions, the n-period optimal expected pro"t, as a function of i , is obtained from
L
PM (i )"max+PM (i , p ): p 3[Pl, P ],,
L L
L L L L

and
M (i , p , q )!Kd(q !i ): q *i ,,
PM (i , p )"max+M
L L L
L L L L
L
L L
L
where M
M is the expected value of the n-period pseudo-pro,t function which is de"ned for n*1 as
L
aPM (0)# p q !r(X(p )!q ),
0)q )X(p ),
L\
L L
L
L
L
L
M (i , p , q )"!c(q !i )#

L L L L
L
L
(q !X(p ))# p X(p )!h(q !X(p )), X(p ))q ,
aPM
L\ L
L
L
L
L
L
L
L
with PM (i )"v[i ]>. This leads to
 

M
M (i , p , q )"(p #r!c)q !rXM (p )! (p #r#h)H(p , q )#ci # aPM (0)[1!F(q ; p )]
L L L L
L
L
L
L
L L
L
L\
L L
OL\*NL
PM (x) f (q !x; p ) dx,
(3)
#a
L\
L
L

where H(p , q )"E[[q !X(p )]>] is the expected value of leftovers at the end of period n. It is seen that
L L
L
L
H(p , q ) is a convex increasing function in q and an increasing function in p ; its additional properties are
L L
L
L
given in [19].
Therefore, the overall decision problem is described by

M (i , q )!Kd(q !i ): q *i ,
(4)
PM (i )"max+M
L L
L L L
L
L L
L
M
M (i , q )"max+M
M (i , p , q ): p 3[Pl, P ],"M
M (i , p (q ), q ),
(5)
L L L
L L L L L

L L L L L
for all n*1, where p (q ) represents the optimal price for a given q in period n. The objective is to determine
L L
L
the optimal decision variables p and q for all n, which jointly maximize PM for a given i . Note that if
L
L
,
,

PM (i ))0 then the optimal decision would be not to do business.

L ,

146

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

4. Model solution
In order to determine the optimal procurement policy for period n, we need to characterize the form of
M
M (i , q ). First, we note from Eqs. (3) and (5) that M
M satis"es
L L L
L
M
M (i , q )"MM (0, q )#ci ,
L L L
L
L
L
so that MM is additively separable in i and q . Hence, it su$ces to characterize the form of M
M (0, q ). For that
L
L
L
L
L
purpose, we need to solve the maximization problem:

M
M (0, q )"max (p #r!c)q !rXM (p )! (p #r#h)H(p , q )# aPM (0)[1!F(q ; p )]
L
L
L
L
L
L
L L
L\
L L
OL\*NL
PM (x) f (q !x; p ) dx: p 3[Pl, P ] .
#a
(6)
L\
L
L
L


It turns out that the solution methods that have been developed for the "xed price (i.e., Pl"P ) versions of

the model [20}23] are not directly applicable for the solution of the problem at hand. Instead, we will follow
a new solution approach which follows the inductive setting described in [21].

4.1. Form of the optimal procurement policy

The dynamic programming problem represented by Eq. (6) is worked out in the appendix. The results are
summarized below in a theorem and a corollary after the following de"nitions.
De5nition. Let G(i, p, q)"(p#r!c)q!rXM (p)!(p#r#h!ac)H(p, q)#ci, and G(0, q)"max+G(0, p, q):
p3[Pl, P ],: then, we de"ne the critical inventory levels 0)sK(s (Sl(sl(S (s(S as shown in
E
E
E
E
E

E
E
Fig. 1.
Note that the generic function G represents the expected pro"t in a single-period, lost-sales model where
the unit salvage value is c (i.e., the leftovers are returned to the supplier at the original cost). It will play
a critical role in the characterization of MM , and in establishing bounds on the optimal control parameters.
L
The following theorem describes the form of M
M in general:
L
Theorem 1. here exist an even integer k and critical inventory levels, s(s(2(sIL\(S (sIL, de,ned by
L
L
L
L
L
L
S "argmax+M
M (0, q): 0)q(R,,
L
L
s"min+q: M
M (0, q)*M
M (0, S )!K,,
L
L
L
L
sH "min+q: q'sH\, M
M (0, q)"M
M (0, S )!K, for j"2, 3,2, k ,
L
L
L
L
L
L
such that if M
M (0, q ) is unimodal, then for n*2:


(a) M
M (0, q )"G(0, q )#aMM (0, s ), q 3[0, s ],
L
L
L
L\
L\
L
L\
(b) M
M (0, q ))G(0, q )#aMM (0, s )#aK, q 3[s ,R),
L
L
L
L\
L\
L
L\
(c) M
M (0, q )*G(0, q )#aM
M (0, s ), q 3[s , sIL\ ],
L
L
L
L\
L\
L
L\ L\
(d) q 3[S , sIL\ ], M
M (0, q )*M
M (0, q )!aK, for any q 3(q ,R),
L
E L\
L
L
L
L
L
L
(e) M
M (0, 0)(R,
L
M
M (0, q )"!R.
(f ) lim
OL L
L

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

147

Fig. 1. De"nitions of the critical inventory levels under G(0, q).

Corollary 1. For n*2, sK(s, sIL\(sl, s(sIL, and Sl(S (S; also, if s (s, then s (s.
E
L L
E E
L
E
L
E
E

E
L
Regarding the condition of unimodality of M
M (0, q ) in Theorem 1, there are analytical di$culties


in proving this property, except for some special cases [3,5,8,12,19,24]. For instance, it is reported in

[19] that M
M (0, q ) is unimodal if (1) demand is deterministic, (2) demand is additive, e has a uniform


distribution and XM (p) is linear, or (3) demand is multiplicative, e has an exponential distribution and
XM (p) is linear. The discussions of unimodality and su$cient conditions can be found in the above-cited
references.
Since G is a special case of M
M (with v"c), under the unimodality assumption, G is also a unimodal

function with a maximizer at S and a reorder point at s ((S ). Thus, it follows from (a) and Corollary
E
E
E

1 (s (S ) that MM (0, q ) is an increasing function of q on [0, s ]. Furthermore, (a) implies that the
L
L
L
L\
L\
E
optimal pricing decision, p (q ), for q 3[0, s ] is given by argmax+G(0, p, q ): p3[Pl, P ],, the maximizing
L L
L
L\
L

price under the generic single-period model. (Note that if q )s , there will be an order placed in period
L
L\
n!1; hence, the inventory carried over to period n!1 would worth ac, and in this case the pricing decision
is based only on G.)
It is implied by (b) and (c) that MM (0, q ) is bounded by two unimodal functions which are at most aK
L
L
apart on [s , sIL\ ]. Hence, (b) and (c) establish bounds on M
M (0, q ) based on the value of the optimal
L\ L\
L
L
(n!1)-period pro"t, provided that an order takes place in period n!1. In this case, the contribution of the

current period's pro"t is limited to between G (0, q ) and G (0, q )!aK.

L
L

148

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

Fig. 2. A general n-period expected pseudo-pro"t function, MM (0, q ) vs. q , for n*2 where k "6.
L
L
L
L

It follows from (a) that M

M (0, 0)"G(0, 0)#aM
M (0, s ); hence, properties (b) and (c) can be rewritten
L
L\
L\

independent of M
M
(0, s ) as
L\
L\
(b) M
M (0, q )!M
M (0, 0))G(0, q )!G(0, 0)#aK, q 3[s ,R),
(7)
L
L
L
L
L
L\

(c) M
M (0, q )!M
M (0, 0)*G (0, q )!G (0, 0), q 3[s , sIL\ ].
L
L
L
L
L
L\ L\
Thus, under the optimal solution, the increase in expected n-period pro"t, upon raising the stock level from
0 to q , must be greater than or equal to the change in one-period expected pro"t (with v"c) under a similar
L
stock movement (i.e., G(0, q )!G(0, 0)). This increase, however, is limited by G(0, q )!G(0, 0)#aK.
L
L
Property (b) extends the upper bound function G(0, q )!aM
M (0, s )#aK over [sIL\ , R). On the
L\
L\
L\
L

other hand, it follows from (d) that M

M (0, q ) is aK-decreasing ([22]) over [S , sIL\ ], and the functional value
L
L
E L\

of M
M , evaluated at any point beyond sIL\ , is strictly below the global maximum of M
M (0, S ).
L
L\
L
L
A general expected n-period pseudo-pro"t function, as characterized by Theorem 1, is depicted in Fig. 2. It
is seen that there could be more than one order-up-to level, and for each order-up-to level there could be
more than one reorder interval. This characterization is weaker than that in the "xed-price model, where it
can be shown that in each period a single optimal reorder-point, order-up-to-level policy exists [21].
The possible presence of multiple order-up-to levels poses serious di$culties in characterizing and
computing the optimal procurement policies. These di$culties, however, can be circumvented if the
beginning inventory level before ordering in an arbitrary period n is less than or equal to sIL. Then, since
L
M
M (0, 0) is "nite and M
M (0, q ) tends to decline as q tends to in"nity, the optimal procurement policy can be
L
L
L
L
characterized as follows provided that i )sIL:
L
L
S if i 3O ,

L
L
q " L
(8)
L
i
if i 3OM ,
L
L
L
for all n, where O , the `reorder regiona, is the union of reorder intervals, given by
L
[0, s]6[s, s]626[sIL\, sIL\] and OM is the complement of O with respect to [0, sIL]. Note that under
L
L L
L
L
L
L
L
the unimodality condition imposed on M
M , we have k "2 and q is determined by an (s , S ) policy.



 
In implementing the above optimal procurement policy, we need to compute sIL for each period, which can
L
be a computational burden. Instead, a stronger but more useful condition can be established by referring to
Corollary 1, where we have s)sIL. That is, we replace the condition of the above optimal policy (i )sIL) by
E
L
L
L

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

149

i )s which involves the computation of s once and in advance by using the problem primitives. Also, it is
L
E
E
seen that the condition i )sIL is satis"ed if S (s for all n and i )S .
L
L
L
E
,
,
The fundamental di!erence between the (s , S ) policies de"ned for the "xed-price models [20,21] and the
L L
policy we de"ne in Eq. (8) is that, at each decision epoch, the vendor has to make a pricing decision whether
or not it decides to place an order. Thus, the vendor needs to determine the optimal price, p (i ), at the
L L
beginning of each review period, given the observed value of i . This requires the preparation of price tables
L
for the vendor to read the optimal prices from. These tables can be prepared simultaneously during the
computation of reorder points and the order-up-to level in each period.

4.2. A suzcient condition for a single reorder point

Properties (b) and (c) in Theorem 1 indicate that M
M (0, q ) is aK-increasing over [s , sl]. Theoretically,
L
L
L\ E
this allows for more than one reorder point. However, operating a system under multiple reorder points
could be a burden in practice. Determination of the critical levels s, s,2, sIL for all periods with su$cient
L L
L
accuracy could be di$cult. It is, therefore, important to know under what conditions there exists a single
reorder point (i.e., k "2).
L
It follows from property (a) in Theorem 1 that M
M (0, q ) is increasing over [0, s ]; thus, if s)s , that
L
L
L\
L
L\

is if M
M (0, s )*M
M (0, S )!K, then k "2 and s is the only reorder point. However, this is not
L
L\
L
L
L
L
a convenient su$cient condition, since, under a general demand distribution, the value of s , or
L\
M
M (0, s ), is not analytically measurable with su$cient accuracy.
L
L\

l
Another possibility is to show that M
M (0, q ) is increasing over [s , s ], implying k "2 under Theorem
L
L
L\ E
L
1. (sIL\)sl for all n; see Corollary 1.) To this end, we assume that MM (0, q ) is increasing over
E
L\
L\
L l

[s , s ], and investigate the su$cient conditions under which M

M (0, q ))MM (0, q ) for s )q
L\ E
L
L
L
L
L\
L
(q )sl with arbitrary q and q . This leads to the following result:
L
E
L
L
Corollary 2. MM (0, q ) is non-decreasing in q over [s , sl] for n*1, that is there exists a single reorder point
L
L
L
L\ E
(k "2), if sl)S and for all q 3[sK, sl]:
E

L
E E
L
p #r!c
L
F(q ; p ))
, p 3+p : p#h!ac'0, Pl)p)P ,.
L L
L

p #r#h!ac
L

(9)

The RHS in Eq. (9) is a concave increasing function of p , and F(q ; p ) is assumed to be a non-decreasing
L
L L
function of p for all q (see Section 2).
L
L
In view of Eq. (9), we can make the following observations. If the vendor administers a higher price, he is
able to increase F(q ; p ), the likelihood that there will be no shortage at level q , to a desired level, while the
L L
L
relative weight of unit underage cost rises (relative weight of unit overage cost declines). Hence, the vendor is
inclined to increase the stock level to above q and this continues until a break-even point is reached. This is
L
veri"ed by the fact that *G(0, p, q)/*q*0 under Eq. (9). Thus, based on the demand distribution and unit
costs that prevail in the market, Eq. (9) re#ects the vendor's capability to increase the expected current period
pro"t by increasing the stock level to above sl. In other words, Eq. (9) attributes a higher monopoly power to
E
the vendor than the vendor would have in its absence. In this interpretation, we utilized the myopic notion
that the higher the monopoly power, the less responsive is the demand to changes in price. That is,
probability of satisfying the demand fully, F(q ; p ), does not respond strongly to an increase in price, since
L L
the increase in F(q ; p ) is limited by the RHS in Eq. (9).
L L
Under Theorem 1 and Corollary 2, the optimal procurement policy is de"ned by
q"i #(S !i )d(s!i ),
L
L
L
L L
L

(10)

150

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

provided that i )s is satis"ed for n'1. It also follows that sl)S for n*1, such that sl is a lower bound
L
E
E
L
E
on order-up-to levels.
Note that in order to have k "2, M
M (0, q ) need not be increasing over [s , sl]. There could be other
L
L
L
L\ E
conditions leading to the same result. Our experience with a number of numerical examples suggests that
cases with k '2 would be rare in practice.
L
4.3. Suzcient conditions for the optimality of (s , S ) policies
L L
If, in property (d), the range for q were established as [S ,R), then M
M (0, q ) would be characterized as
L
E
L
L
a K-decreasing function over [S ,R), S would be the only order-up-to level, and there would be no need for
E
L
the condition i )sIL in Eq. (8) or Eq. (10). This is precisely the case in the development of the optimality of
L
L
(s , S ) policies under the "xed-price (Pl"P ) model [21].
L L

In the following corollary, we investigate the su$cient conditions under which MM assumes a desirable
L
shape over [S ,R) to yield a single order-up-to level.
E
Corollary 3. M
M (0, q )*MM (0, q )!aK for all q and q with S )q (q , if s )S and
L
L
L
L
L
L
E
L
L

E
(p #r#h!ac)F(q ; p )*p #r!c, p 3[Pl, P ], for n*1.
L
L L
L
L


(11)

Note that for large q values, F(q; p) tends to 1 (particularly, F(q; p)"1 when q*;(p)), and this complies
well with condition Eq. (11). That is, M
M (0, q ) is K-decreasing in the limit as q tends to in"nity. Also, it is
L
L
L
intuitive that at considerably large inventory levels, the vendor would tend to reduce the price substantially,
even below the unit procurement cost, in order to deplete inventories. In this regard, it is seen in Eq. (11) that
the condition tends to hold when c and h get larger and r gets smaller (it holds at all q levels when
p#r!c)0)p#r#h!ac, that is, ac!h!r)p)c!r). In general, the condition tends to hold
when c and h get larger or r gets smaller.
Thus, under Theorem 1 and Corollary 3, the optimal procurement policy is characterized by

S if i 3O ,
L
L
q" L
L
i
if i 3OM or i 'sIL.
L
L
L
L
L
Also, under both Corollaries 2 and 3, it follows from Theorem 1 that k "2 for all n and q is determined by
L
L
an (s , S ) policy: q"i #(S !i )d(s!i ).
L L
L
L
L
L L
L
4.4. Numerical example
To demonstrate our "ndings, we consider a 5-period problem de"ned by the base parameter set
c"0.5, r"1.5, h"0.4, v"0.1, K"5, a"0.95; the expected demand function XM (p)"ae\@N with a"150
and b"0.2; the price limits Pl"0.1 and P "10; and the additive-uniform demand distribution

F(x; p)"0.5(x!XM (p)#j)/j for !j)x!XM (p))j with j"20. (j"0 corresponds to the deterministic
demand case.) For sensitivity analyses, we used multiple parameter values as h"0.02, 0.1, 0.4;
K"1, 5, 10; b"0.05, 0.1, 0.2; and j"0, 5, 10, 20.
Using a computer program, we computed the expected pseudo-pro"t functions M
M , M
M ,2, M
M over the



q range of [0, 250] with a step size of 1. The resulting optimal values of the control parameters and the
L
critical functional values are shown in Tables 1 and 2. Also, the pseudo-pro"t functions computed for
j"0, K"10, the deterministic demand case, and j"10, K"10 are plotted in Figs. 3 and 4.

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

151

Table 1
Critical inventory levels, prices and pseudo-pro"t values under the 5-period example problem
K

s
L

S
L

M
M (0, S )
L
L

p (S )
L L

10

20

0.2

0.4

10

20

0.1

0.4

10

20

0.05

0.4

10

20

0.2

0.1

10

20

0.2

0.02

20

0.2

0.4

20

0.2

0.4

10

10

0.2

0.4

10

0.2

0.4

10

0.2

0.4

1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5

49.6
51.1
51.1
51.1
51.1
64.2
65.3
65.2
65.2
65.2
100.0
100.9
100.9
100.9
100.9
50.8
52.4
53.2
52.4
53.1
51.1
56.2
56.8
55.6
55.5
54.2
55.7
55.7
55.7
55.7
60.5
62.1
62.1
62.1
62.1
43.0
43.7
43.7
43.7
43.7
39.7
40.0
40.0
40.0
40.0
36.5
36.5
36.5
36.5
36.5

65.7
67.5
67.5
67.5
67.5
72.5
73.6
73.6
73.6
73.6
108.3
109.3
109.3
109.3
109.3
67.1
110.4
125.8
125.4
125.0
67.5
119.9
161.4
196.3
133.0
65.7
67.5
67.5
67.5
67.5
65.7
67.5
67.5
67.5
67.5
57.8
58.7
58.7
58.7
58.7
53.9
54.2
54.2
54.2
54.2
49.9
49.9
49.9
49.9
49.9

235.3
455.8
665.2
864.1
1053.1
509.2
990.5
1447.7
1882.0
2294.6
849.3
1654.8
2420.1
3147.1
3837.7
240.3
466.1
681.5
885.3
1079.7
241.7
472.7
692.8
900.7
1098.1
235.3
460.5
674.4
877.7
1070.7
235.3
464.3
681.9
888.5
1084.9
242.5
466.7
679.7
882.0
1074.2
246.1
472.7
688.0
892.6
1086.9
249.7
477.3
693.6
899.1
1094.3

5.5
5.5
5.5
5.5
5.5
10.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.6
5.5
5.5
5.5
5.5
5.6
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5

152

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

Table 2
Critical inventory levels that are obtained under the generic pseudo-pro"t function
K

sK
E

s
E

Sl
E

sl
E

S
E

s
E

S
E

G(0, S )
E

10
10
10
10
10
5
1
10
10

20
20
20
20
20
20
20
10
5

0.2
0.1
0.05
0.2
0.2
0.2
0.2
0.2
0.2

0.4
0.4
0.4
0.1
0.02
0.4
0.4
0.4
0.4

45.0
62.0
97.8
46.2
46.5
51.3
60.1
38.2
34.9

51.1
65.3
101.1
52.4
52.7
55.7
62.1
43.7
40.1

51.5
65.5
101.3
52.8
53.1
56.0
62.3
44.1
40.4

63.7
71.8
107.6
65.3
65.7
64.8
66.3
55.2
51.0

67.5
73.7
109.5
69.2
69.7
67.5
67.5
58.7
54.3

71.5
75.7
111.5
74.2
81.1
70.3
68.8
62.3
57.7

93.2
96.8
132.6
146.2
281.1
82.1
73.0
83.9
79.2

241.7
516.0
856.1
247.2
248.8
241.7
241.7
245.7
247.7

We observe that the optimal control values are quite sensitive to the price senstivity of demand (note that
b is the price elasticity multiplier of the expected demand). The more sensitive the expected demand to price,
the less is the expected pro"t. As for the holding cost, smaller h values tend to yield considerably higher
order-up-to levels.
Comparing the critical inventory levels listed in Tables 1 and 2, we see that the conditions given in
Corollary 1 hold in each case. We also note that S )s in all cases except h"0.1 and h"0.02, for which the
L
E
reorder-region order-up-to level policy will be optimal.
Table 1 shows that the optimal prices evaluated at respective order-up-to levels are the same in all periods.
Consequently, if an order is to be placed in each period, then the optimal prices in successive periods will be
constant. We also observe that demand uncertainty has considerable impact on the parameters of the
optimal procurement policy. As the demand variance is decreased (i.e., j is decreased), the order-up-to and
reorder levels both tend to be lower, and the vendor foresees higher expected pro"ts.
On the other hand, on comparing the deterministic and probabilistic pseudo-pro"t curves in Fig. 3, we
"nd that the deterministic values are greater than the corresponding probabilistic values, especially, about
the q mid-range, which includes respective order-up-to levels.
L
In Fig. 4, we plot M
M (0, q )!MM (0, 0) vs. q together with the boundary functions. It is seen that the
L
L
L
L
properties listed in Theorem 1 and Corollary 1 hold. The boundary functions established in Eq. (7) also
satisfy the conditions (b) and (c). We also observe that the pseudo-pro"t curves exhibit considerable slope
changes across the q mid-range.
L
The best price trajectories p (q ), evaluated for n"1, 2,2, 5, are plotted in Fig. 5. The "gure shows that
L L
the optimal pricing decision at a given inventory level is not necessarily trivial. We observe that p (q ) is
L L
generally decreasing in q , but it can exhibit mild increases, or stagnate (become `stickya) over various
L
q ranges where it does not respond to changes in stock levels. Also, on comparing the deterministic and
L
probabilistic optimal prices, we observe that it is not necessarily true that the deterministic prices are lower
than the probabilistic prices or vice versa (cf. [4,5]).
Intuitively, price should be lower at higher stock levels, so as to facilitate higher demand to deplete the
excess inventories. However, the vendor would still tend to increase the price for two basic reasons: to shrink
the demand so that there will be leftovers for future periods, or to realize a higher pro"t per unit sold. If the
"xed ordering cost in a future period is to be saved by satisfying that period's demand by carying a part of the
current inventory forward, then both of these reasons will prevail. Once the total cost of inventory holding
and the di!erential cost of procurement breaks even with the savings in the "xed ordering cost, however, it
would not be pro"table to consider carrying stock into future periods. As n gets larger, there are more periods
ahead to consider; thus, there could be a recurrence of this break-even behaviour at higher q levels. This can
L
be seen in Fig. 5 by comparing p (q ) with p (q ). The passage from one pricing regime to another, at
 
 
a break-even point, need not be smooth and there could be sudden changes in p (q ).
L L

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

153

Fig. 3. Expected pseudo-pro"t curves, MM (0, q ) vs. q , that are obtained for the 5-period example problem in Section 5.4
L
L
L
(c"0.5, r"1.5, h"0.02, b"0.2, K"10, j"10.0). The thin curves represent the deterministic case where j"0.

154

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

Fig. 4. Combined representation of the curves in Fig. 3. The thin curves represent MM (0, q )!MM (0, 0) vs. q for n*1 and the thick
L
L
L
L
curves represent the boundary functions G(0, q )!G(0, 0) and G(0, q )!G(0, 0)#K.
L
L

5. In5nite-horizon model
In this section, we study the in"nite-horizon version of the model. We drop the period indices from the
notation and rewrite the expected pseudo-pro"t function from Eq. (6) as
O\*N
M (0, s)#a
[MM (0, x)!M
M (0, s)]> f (q!x; p) dx: p3[Pl, P ]
M
M (0, q)"max G(0, p, q)#aM



O\*NO
"G(0, p , q)#aM
M (0, s)#a
[M
M (0, x)!M
M (0, s)]> f (q!x; p ) dx,
(12)
O
O

where p ,p(q), the maximizing price at q.
O
The solution for M
M (0, q) involves the integral in Eq. (12) which represents the convolution of the density
function with a "ltering function based on M
M (0, q) (i.e., the function MM (0, q) is "ltered over the level

M
M (0, s)). Because of this, we need to solve the integral Eq. (12) recursively. In each stage of the solution,
M
M (0, q) is established over a non-overlapping interval, and the solution obtained is employed in the next
stage.

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

Fig. 5. Optimal price trajectories that are obtained for the 5-period example problem shown in Fig. 3.

155

156

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

It follows from Eq. (12) that for 0)q)s we have M

M (0, q)"G(0, q)#aM
M (0, s), which is an

increasing function of q over [0, s] where s(S . Evaluating M

M (0, q) at s, we obtain
E
M
M (0, s)"G(0, s)/(1!a), which implies
a
M
M (0, q)"G(0, q)#
G(0, s),
'
1!a

(13)

where the index in roman numeral denotes the stage number. Successive stages I, II, III, 2 correspond to
intervals [0, s], [s, s],[s, s],2, respectively.
In the next stage, we have s)q)s for which the e!ective integration range starts at s in Eq. (12) and
we have
O\*NO

[M
M (0, x)!M
M (0, s)] f (q!x; p ) dx
M
M (0, q)"G(0, p , q)#aMM (0, s)# a
O
''
O
''

Q
O\*NO

"G(0, p , q)#a[1!F(q!s; p )]M

M (0, s)#a
(14)
M
M (0, x) f (q!x; p ) dx,
O
O
''
O
Q
which is a renewal equation [25] with solution

a
M
M (0, q)"max G(0, p, q)#G(0, s)
!R (q!s; p)
''
?
1!a




O\*N
#
G(0, p, x) dR (q!x; p): p3[Pl, P ] ,
?


Q
where R (x; p)"  aGFG(x; p) and FG is the i-fold convolution of F with itself. (Derivation is given in the
?
G
appendix.)
Having established M
M (0, q) over [0, s], we check whether s(s where s is a lower bound on sI (see
E
E
Corollary 1). If s)s, then k"2 and we stop. In this case, there will be one reorder point. If s(s,
E
E
however, then we need to consider the q range beyond s and pass to the next stage.
Evaluating Eq. (12) for s)q)s, we obtain

Q

M (0, x)!M
M (0, s)] f (q!x; p )
M
M (0, q)"G(0, p , q)#aMM (0, s)# a [M
'''
O
''
O

Q
Q
a
"G(0, p , q)#
G(0, s)[1#F(q!s; p )!F(q!s; p )]#a M
M (0, x) f (q!x; p ) dx.
O
O
O
''
O
1!a

Q
M .
Since M
M inside the integral is obtained earlier, it can be substituted in the above to solve for M
''
'''
The fourth stage will involve another renewal equation. For s)q)s, we have

Q

M
M (0, q)"G(0, p , q)#aM
M (0, s)# a [M
M (0, x)!MM (0, s)] f (q!x; p )
'4
O
''
O

Q
O\*NO
#a
[M
M (0, x)!M
M (0, s)] f (q!x; p )
'4
O

Q

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

157

a
"G(0, p , q)#
G(0, s)[1#F(q!s; p )!F(q!s; p )! F(q!s; p )]
O
O
O
O
1!a
Q

O\*NO
#a M
M (0, x) f (q!x; p ) dx# a
M
M (0, x) f (q!x; p ) dx,
''
O
'4
O


Q
Q

which is a renewal equation on MM . The solution for this equation can be obtained by following a similar
'4
procedure to the one used in solving for M
M (0, q).
''
If s)s, then we stop. Otherwise, we follow the same solution process to establish M
M in the next two
E
stages, and check the stopping criterion.
This method works in principle. However, there is a computational di$culty due to the fact that critical
inventory levels s, s, s,2 are unknown in advance. They have to be computed simultaneously as we
establish M
M numerically.
One approach is to start with the assumption that s"s . Based on this initial value of s, M
M can be
E

established over [0, s] and the value of s can be found by incrementing the value of q until M
M (0, q) equals
''
M
M (0, s). Note that s will be the "rst point greater than s satisfying the above equality. Likewise, the
''
process can be continued to include s and s and so on. Once the stopping condition (s)sI) is satis"ed,
E
using the current solution of M
M , s can be updated. If it is tolerably close to its previous value, the procedure

can be terminated. Otherwise, based on its updated value, a new M

M will be established over [0, sI].
It is beyond the scope of this paper to investigate the convergence of the numerical method proposed
above. To demonstrate its merits, however, we use it to obtain the solution for an in"nite-horizon problem
de"ned by the parameter set c"0.5, r"0.25, h"0.3, v"0.1, K"8, a"0.7; the expected demand function XM (p)"150 exp(!0.5p); the price limits Pl"0.1 and P "4; and the multiplicative-exponential demand

distribution F(x; p)"1!exp(!x/XM (p)), x*0, under which the discounted renewal function is given by
R (x; p)"a[1!exp(!(1!a)x/XM (p))]/(1!a).
?
We obtained the solution in 10 iterations. The successive s values were 28.20, 35.25, 31.26, 33.40, 32.15,
32.87, 32.44, 32.70, 32.55 and 32.64 until we reached a tolerance of 0.1 units. In this example problem, it was
found that k"2; hence, it was not necessary to pass beyond the second stage in computations.
In order to demonstrate the transition from M
M to M
M , we solved a 15-period problem under the

same problem data, and compared the resultant expected pseudo-pro"t functions with the function
we obtained by solving the in"nite horizon problem. Fig. 6 depicts M
M (0, q ) functions for n"1, 2,2, 15
L
L

and M
M , which become indistinguishable from the pseudo-pro"t functions for large n, as predicted by the
theory.

6. No-5xed-ordering-cost
In order to establish a link between this paper and the earlier studies reported in the literature, here we take
a brief look at the model with K"0.
Following a similar inductive approach to the one we used previously, we suppose that M
M (0, q ) is



unimodal and PM
(i ) is determined by a single-parameter policy:
L\ L\
M
M (0,
L\
PM (i )"ci #
L\ L\
L\
M
M (0,
L\

S ), i )S ,
L\
L\
L\
i ), S (i .
L\
L\
L\

158

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

Fig. 6. Expected pseudo-pro"t curves, M

M (0, q ) vs. q , that are evaluated for the 15-period problem speci"ed in Section 6. The curve at
L
L
L
the top represents the solution for the in"nite horizon model with the same parameter set. (It coincides with the expected pseudo-pro"t
curves at large n.)

Under this setting, it follows from Eq. (6) that

M
M (0, q )"max G(0, p , q )#aM
M (0, S )
L
L
L L
L\
L\
OL\*NL
M (0, S )] f (q !x; p ) dx: p 3[Pl, P ] .
#a
[M
M (0, x)! M
(15)
L\
L\
L\
L
L
L

1L\
It can be seen that for q )S , the integral drops out and we have
L
L\
M
M (0, q )"G(0, q )#aM
M (0, S ),
L
L
L
L\
L\
where G(0, q ) is a unimodal function. For S (q , however, we note that the integral in Eq. (15) is
L
L\
L
negative-valued. Based on these observations, we establish M
M (0, q ) in two alternative ways, as shown in
L
L
Fig. 7. Considering Eq. (15) and Fig. 7, we discover that

min+S , S ,)S )S ,
(16)
E L\
L
E
for n*2, and the con"guration shown in Fig. 7a can only occur for n"2.
The recursive ordering in Eq. (16) implies that if S )S , then S "S for all n*2. It is clear that if v"c,
E

L
E
then M
M ,G, and S "S by de"nition. But, it is more plausible, in general, that v(c, in which case S will


E

be di!erent from S .
E

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

159

Fig. 7. Construction of M
M (0, q ) vs. q for K"0; dashed curves represent G(0, q ) for: (a) S (S , and (b) S (S .
L
L
L
L
E
L\
L\
E

Intuitively, when the price is "xed, S must be greater than S . Having the option of salvaging the leftovers
E

at a higher return, the decision maker will be encouraged to procure up to higher levels. But, under the
impact of pricing, there could be a trade-o! between small-quantity high-price and large-quantity low-price
scenarios. Demand uncertainty at a given level will then be contingent on the pricing decision. For this
reason, under a general price-demand relationship, it is theoretically possible that S (S .
E

Fig. 7b shows that the expected pro"t level will be higher if S
shifts towards S and the contribution of
L\
E
G is more than the contribution of the negative valued integral in Eq. (15). Thus, overcoming the transient
behaviour represented by Eq. (16), there will be a propensity, under the optimal solution, for S to approach
L
S as the number of periods is increased.
E
Su$cient conditions for the optimality of a single-critical-number policy are given in [11]. In principle,
these conditions can be utilized to verify the form of the optimal procurement policy, but they are di$cult to
interpret in terms of their managerial implications. Under some restrictions, Thowsen [11] shows that the
single-parameter policy will be optimal if the demand is additive with a PF (Po& lya frequency function of type

2) distributed random term and a linear expected demand function. Also, su$cient conditions for the
existence and uniqueness of the optimal decision variables are given in [9,10]. The characterization
portrayed in these studies are based on taking derivatives. The drawback of that approach is that it can not
be immediately extended to the model with "xed ordering costs, where di!erentiability is restricted to
sub-regions.

160

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

7. Non-stationary extensions
In addition to being dependent on price, it is possible that the demand distribution exhibits periodic shifts
during the planning horizon due to seasonalities or changing market conditions. The natural way of
incorporating such time-variability in our model would be to represent the demand in period n by a random
variable X (p ) with distribution F (x; p ). The generic function, now denoted by G , and the critical
L L
L
L
L
parameters de"ned for G, sK (2(S , would also be period-dependent.
L EL
EL
By virtue of the fact that the demand distribution depends on price and the price can be set at the
beginning of each period, extensions to time-dependent demand are immediate. It can be seen, going through
the proofs in the appendix, that this does not alter the arguments and derivations, provided that G(0, q ) is
L
L
unimodal and the condition s )sIL\ holds for all n'1. This condition parallels the "xed-price version of
L\
EL
the non-stationary demand model [21,23] where it is assumed that S )S
for n'1. It can also be seen
EL
EL\
that if S )s , then k "2.
L\
L
EL
Similarly, we can let the cost parameters and the discount factor depend on the period (i.e.,
c , r , h , K , a ), without any di$culty, provided that K *a K
(a "1), and M
M is unimodal for
L L L
L L
L
L\ L\ 


n'1 (which implies unimodality of G ).

L
8. Deterministic demand
Under deterministic demand, X (p)"XM (p) for all n. For technical reasons, we will assume that X (p) is
L
L
L
o(1/p) as pPR and as pP0>. Under this assumption, the riskless revenue function R (p)"pX (p) starts
L
L
growing at 0 when p"0, reaches a maximum level, and declines down to 0 as p tends to in"nity. We further
assume that R (p) is a pseudoconcave function of p on (0,R) so that there is a unique breakeven point
L
between low-price-high-demand and high-price-low-demand situations in terms of revenue. It is shown in
[19] that if X (p) is a convex or concave decreasing function, then R (p) is pseudoconcave on (0,R).
L
L
A detailed discussion of the properties of the riskles revenue function and related functions can be seen in
[7,19].
Using X (p)"XM (p) in Eq. (3) we obtain
L
L
M
M (i , p , q )"G (i , p , q )# a+PM ([q !X (p )]>)! c[q !X (p )]>,,
L L L L
L L L L
L\ L
L L
L
L L
where
G (i , p , q )"(p #r!c)q !rX (p )! (p #r#h!ac)[q !X (p )]># ci .
L L L L
L
L
L L
L
L
L L
L

(17)

This representation leads to:

Corollary 4. ;nder deterministic demand, MM (0, q ) is quasiconcave, and M
M (0, q ) is quasi-K-concave for


L
L
n*2, where:
(i) M
M (0, q ) is increasing over q 3[0, X (PL)], n*2, where PL is the maximizer of (p !c)X (p ) over
L
L
L
L A
A
L
L L
[Pl, P ].

M (0, s ), MM (0, q ))G (0, q )#aMM (0, s )#aK, for q '
(ii) M
M (0, q )*G (0, q )#aM
L
L
L\
L
L\
L\
L
L
L\
L
L\
L\
L
X (PL) and n*2.
L A
(iii) he optimal inventory level is determined by the (s , S ) policy: q"i #(S !i )d(s !i ).
L L
L
L
L
L L
L
This result was also obtained by Thomas [26], through a price-dependent lot-sizing approach. He
demonstrated that there exists a positive integer m, where m#1 is referred to as the planning horizon, such

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

161

that for all n, S "X (pH)#X (pH )#2#X (pH ), where pH

is the maximizer of
L
L L
L\ L\
L\K L\K
L\H
(p !e )X (p ), with
L\H
L\H L\H L\H



H\
e " c#h 1# (a a 2a )
L\H
L\ L\
L\I
I



(a a 2a ),
L\ L\
L\H

for 1)j)m and pH"PL.

L
A
Thus, since e
'e *c for 1)j)m, it follows from Theorem A.1 in the appendix that
L\H>
L\H
and pH "PL\H for 1)j)m; in particular, if X ( ) )"X ( ) ) for 1)j)m, then
L\H
A
L
L\H
)pH )pH )2)pH .
L\
L\
L\K

pH"PL
L
A
pH"PL
L
A

9. Conclusions
We established that a reorder-region, order-up-to-level procurement policy is optimal for the decision
problem of interest in this paper, given that M
M (0, q ) is unimodal, and i )s for n*2. Under this policy,


L
E
the vendor administers the best price in each period, whether or not an order is placed. We have also
provided upper and lower bounds on the reorder and order-up-to levels, which should expedite the
computational work, and su$cient conditions under which a reorder-point, order-up-to level policy is
optimal.
Under the special case with deterministic demand, we have shown that the optimal ordering policy is
determined by an (s, S) type policy, provided that the revenue function R(p) is unimodal. For the in"nite
horizon stationary model, we have provided a solution method that could be used to obtain the optimal
solution numerically.
Furthermore, in the absence of a "xed ordering cost, the order-up-to levels are bounded from above by S ,
E
and they are ordered in a non-decreasing fashion in n (see Eq. (16)). With K'0, this ordering does not hold,
however, and S is not necessarily an upper bound for order-up-to levels. For instance, we have obtained
E
S (S (S (S (S (S for the example problem with h"0.1 (see Tables 1 and 2).

E




One of the contributions of the paper is the solution method which is based on a non-derivative approach.
Through this approach, we were able to develop distribution-independent and non-stationary results. The
existing solution methods predominantly utilize the derivative approach, and impose restrictive assumptions
on the demand distribution, the expected demand function, or the parameter set.
In terms of additional research that could follow the present e!ort, one interesting area is approximations.
As a stationary approximation, we formulated the in"nite-horizon model in Section 5, and proposed an
algorithm for it, based on the recursive solution of a sequence of renewal equations. It would be interesting to
study the underlying convergence problems, both in terms of the proposed algorithm and the optimal control
parameter values.
We have demonstrated that the optimal pricing trajectories, p (q ), n*1, can be complicated enough to
L L
make the evaluation and administration of the optimal pricing strategy impractical. A simpler strategy,
though not optimal, might be desirable in applications. One possibility is to approximate the best price by
utilizing the single-period model characterized by G . That is, p (q )"argmax+GH(0, p, q ): p3[Pl, P ],, for
L
E L
L
L

n'1. The function M
M (0, p (q ), q ), as an approximation for M
M (0, q ), would then be used recursively for
L
E L L
L
L
the determination of M
M
. This myopic approximation to the pricing policy worked very well in a prelimiL>
nary investigation of its performance. We feel that a detailed study of its virtues and limitations is justi"ed. It
would also be interesting to investigate optimal procurement policies under this and other suboptimal
pricing strategies.

162

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

Appendix A
Proof of Theorem 1. Under the unimodality assumption about M
M (0, q ), it is su$cient to show, inductively,



that M
M (0, q ) satis"es conditions (a)}(f), given that MM
(0, q ) does, since using the same line of
L
L
L\
L\
arguments in the proof, it can be trivially shown that M
M (0, q ) satis"es these conditions also. In what


L\
follows, we have q )sI unless speci"ed otherwise.
L\
L
De"ning the `reorder regiona O
for period n!1 by
L\

O "+q: M
M
(0, q))M
M
(0, S )!K, 0)q)sIL\\,
L\
L\
L\
L\
L\
"[0, s ]6[s , s ]626[sIL\\, sIL\\],
L\
L\ L\
L\
L\
M (0, s ) and
it follows under the inductive assumption about M
M that PM (0)"M
L\
L\
L\
L\

(0, s ), i 3O ,
M
M

L\
L\
L\
L\
PM
(i )"ci #
(A.1)
L\ L\
L\
M
M (0, i ), i 3OM
,
L\
L\
L\
L\
where the region OM
is complementary to region O
with respect to [0, sIL\ ]. Note that O
is the
L\
L\
L\
L\
union of k /2 reorder intervals.
L\
We assume that K is not exceedingly large so that s '0. This assumption does not have any critical
L\
in#uence on the results; it only decreases the number of terms to carry in the analysis and simpli"es the
mathematical representation.
By considering Eq. (A.1), the integral in Eq. (3) can be rewritten as

OL\*NL




PM (x) f (q !x; p ) dx"cH(p , q )#

M
M (0, s ) f (q !x; p ) dx
L\
L
L
L L
L\
L\
L
L
5
VZ-L\  OL\*NL

M
M (0, x) f (q !x; p ) dx
L\
L
L

[M
M (0, x)!M
M (0, s )] f (q !x; p ) dx
L\
L\
L\
L
L

VZ-M L\5  OL\*NL

"cH(p , q )#MM (0, s )F(q ; p )

L\
L\
L L
L L
#

VZ-M L\5  OL\*NL

"cH(p , q )#MM (0, s )F(q ; p )

L L
L\
L\
L L
OL\*NL
#
[M
M (0, x)!M
M (0, s )]>f (q !x; p ) dx,
L\
L\
L\
L
L

where the integral with respect to x3O 5[0, q !(p )] represents the sum of integrals over the
L\
L
L
composite x range O 5[0, q !(p )]. Same is true for OM
5[0, q !(p )].
L\
L
L
L\
L
L
Using this result in Eq. (3) we obtain

M
M (i , p , q )"(p #r!c)q !rXM (p )! (p #r#h!ac)H(p , q )# ci #aM
M (0, s )
L L L L
L
L
L
L
L L
L
L\
L\
OL\*NL
M (0, s )]>f (q !x; p ) dx
#a
[M
M (0, x)! M
L\
L\
L\
L
L

OL\*NL

"G(i , p , q )#aM
M
(0, s )# a
[M
M (0, x)! MM (0, s )]>f (q !x; p ) dx,
L L L
L\
L\
L\
L\
L\
L
L


163

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

which can be substituted in Eq. (5) to obtain

M
M (0, q )"max G(0, p , q )#aM
M (0, s )
L
L
L L
L\
L\
OL\*NL
#a

[M
M (0, x)! M
M (0, s )]>f (q !x; p ) dx: p 3[Pl, P ] ,

L\
L\
L\
L
L
L

(A.2)


where it is understood that q !(p ))sIL\ , which is also implied by q )sIL\ .
L
L
L\
L
L\
It follows that for q (s the integrand in Eq. (A.2) vanishes and we have:
L
L\
M
M (0, q )"G(0, q )#aM
M (0, s ).
(A.3)
L
L
L
L\
L\

This establishes (a). Since G is a special case of M

M (with v"c), we consider G to be a unimodal function

with a maximizer at S and a reorder level s . Thus, under the inductive assumption that s (S , we
E
E
L\
E

conclude from Eq. (A.3) that M

M (0, q ) is an increasing function of q on [0, s ]. Later, we shall
L
L
L
L\
demonstrate that in fact s(S for all n*2 (see the proof of Corollary 1).
L
E
Next, we consider the q range of [s , sIL\ ]. Evaluating the maximand in Eq. (A.2) at p , where
L
L\ L\
E
p "argmax+G(0, p , q ): p 3[Pl, P ],, we obtain
E
L L L

OL\*NE

M
M (0, q )*G (0, q )#aM
M
[M
M (0, x)! MM (0, s )]>f (q !x; p ) dx.
(0, s )# a
L
L
L
L\
L\
L\
L\
L\
L
E

(A.4)

We note that for all q 3[s , sIL\ ],

L
L\ L\
OL\*NL
[M
M (0, x)!M
M (0, s )]>f (q !x; p ) dx)KF(q ; p ))K,
0)
L\
L\
L\
L
L
L L

for all p 3[Pl, P ]. Thus, it follows from Eq. (A.4) that
L

M
M (0, q )*G(0, q )#aM
M (0, s ), q 3[s , sIL\ ],
L
L
L
L\
L\
L
L\ L\
which proves (c).
For the proof of (b), we need to extend the q range beyond sIL\ ; hence, we cannot use Eq. (A.2) as
L
L\
a starting point. We rather start at Eq. (3).
Since, under the inductive assumption, PM (x))cx#M
M (0, S ) for all x3[0,R), the integral in
L\
L\
L\
Eq. (3) satis"es

OL\*NL

PM (x) f (q !x; p ) dx)cH(p , q )#MM (0, S )F(q ; p ).

L\
L
L
L L
L\
L\
L L


Therefore, it follows from Eq. (3) that
M
M (0, q ))max+G(0, p , q )#aM
M (0, s )[1!F(q ; p )]
L
L
L L
L\
L\
L L
#a[M
M (0, s )# K]F(q ; p ): p 3[Pl, P ],
L\
L\
L L L

)G(0, q )#aM
M (0, s )# aKmax+F(q ; p ): p 3[Pl, P ],,
L
L\
L\
L L L

which implies (b).

(A.5)

164

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

Under properties (b) and (c) we have

G(0, q )#aM
M (0, s ))MM (0, q ))G(0, q )#aM
M (0, s )#aK,
(A.6)
L
L\
L\
L
L
L
L\
L\

which indicates that M

M (0, q ) is bounded by two unimodal functions that are at most aK distance
L
L
(vertically) apart on q 3[s , sIL\ ].
L
L\ L\
Next, we consider the region [S , sIL\ ]. Let q 3(q ,R) be an arbitrary level for any given q 3[S , sIL\ ]. It
E L\
L
L
L
E L\
follows from Eq. (A.5) that
OL\*NL

PM (x) f (q !x; p ) dx!cH(p , q )! M

M (0, s )F(q ; p ))KF(q ; p ))K,
L\
L
L
L L
L\
L\
L L
L L

(A.7)


for all p 3[Pl, P ]. Therefore, by de"ning p with MM (0, q )"MM (0, p, q ), considering Eqs. (A.4) and (A.7),

L
L
L
L
L
and recalling that G is non-increasing on [S ,R) we can proceed to prove (d) as follows:
E
OL\*NE

M
M (0, q )*G (0, q )#aM
M
(0, s )# a
[M
M (0, x)! MM (0, s )]>f (q !x; p ) dx
L
L
L
L\
L\
L\
L\
L\
L
E

*G(0, q )#aM
M (0, s )
L
L\
L\
*G(0, p, q )#aM
M (0, s )
L
L\
L\
OL\*NY

*G(0, p, q )#aM

M
PM (x) f (q !x; p) dx! cH(p, q )
(0, s )# a
L
L\
L\
L\
L
L




!M
M (0, s )(F(q ; p)!K)
L\
L\
L

"M
M (0, q )!aK,
L
L
M (0, sIL)*MM (0, q )!aK for all
which implies that M
M (0, q ) is aK-decreasing over [S , sIL\ ], and M
L
L
E L\
L
L
L
L
q 3(sIL,R).
L L
So far, we have shown that properties (a)}(d) hold. For properties (e) and (f), we establish M
M (0, q ) both at
L
L
q "0 and as q tends to in"nity. It follows from Eq. (A.3) that
L
L
M (0, s ).
M
M (0, 0)"G(0, 0)#aM
M (0, s )"!rXM (P )#aM
L
L\
L\

L\
L\
Thus, it is clear that M
M (0, q ) has a "nite support at q "0.
L
L
L
On the other hand, it has been shown in [10] that lim
M
M (0, q )"!R. To establish a similar result

O



for all n, we let lim

M
M
(0, q )"!R, then we obtain
OL\ L\
L\

lim M
M (0, q )"max lim G(0, p , q )#aM
M (0, s )[1!F(q ; p )]
L
L
L L
L\
L\
L L
OL
OL
OL
#a
PM (q !x) f (x; p ) dx: p 3[Pl, P ] "!R.
L\ L
L
L

*NL
Thus, lim
M
M (0, q )"!R for all n. This completes the proof of Theorem 1.
OL L
L

Proof of Corollary 1. Evaluating Eq. (A.6) at q "S , we obtain

L
E
G(0, S )#aM
M (0, S ).
M (0, s ))MM (0, S ))M
E
L\
L\
L
E
L
L

165

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

Similarly, for q "S , we have

L
L
M
M (0, S ))G(0, S )#aMM (0, s )#aK)G(0, S )#aM
M (0, s )#aK.
L
L
L
L\
L\
E
L\
L\
Combining the above inequalities we get
G(0, S )#aM
(A.8)
M (0, s ))MM (0, S ))G(0, S )#aM
M (0, s )#aK.
L\
L\
L
L
L\
L\
E
E
Thus, considering the two boundary functions in Eq. (A.6) which can be represented by a pair of unimodal
functions aK distance apart, and by indicating the critical inventory levels sK(s (Sl(sl(S (s(S
E
E
E
E
E
E
E
as in Fig. 1, it can be seen that sK(s(sIL\(sl(s(sIL, and Sl(S (S for n'1. Moreover, since
E
L
L
E
E
L
E
L
E
G is an increasing function over [0, S ], it follows from the proof of properties (a)}(c) in Theorem 1 that if
E
s (s, then s (s for all n'1.
E

E
L
M (0, q ) is increasing over
Proof of Corollary 2. To initiate the inductive proof, we assume that sl)S and M
E



l

[0, s ]. Since under property (a) M

M (0, q ) is increasing over [0, s ], it follows from Eq. (A.2) that
E
L
L
L\

M (0, s )
M
M (0, q )"max G(0, p , q )#aM
L L
L
L
L\
L\
OL\QL\
#a
[M
M (0, q !x)! M
M (0, s )] f (x; p ) dx: p 3[Pl, P ] .

L\
L
L\
L\
L
L
*NL
Note that M
M (0, q !x) is increasing in q over the range of interest and [M
M (0, q !x)!M
M (0, s )]
L\
L
L
L\
L
L\
L\
is positive over the x range of [q !s , q !s ]. Thus, we can proceed as follows:
L
L\ L
L\
M (0, s )
M
M (0, q )"G(0, p (q ), q )#aM
L
L
L L L
L\
L\
OL\QL\
#a
[M
M (0, q !x)!M
M (0, s )] f (x; p (q )) dx
L\
L
L\
L\
L L
L
L
*N O 
)G(0, p (q ), q )#aM
M (0, s )
L\
L\
L L L
OL\QL\
#a
[M
M (0, q !x)! M
M (0, s )] f (x; p (q )) dx
L\
L
L\
L\
L L
*NLOL
)G(0, p (q ), q )#aM
M (0, s )
L L L
L\
L\
OL\QL\
#a
[M
M (0, q !x)! M
M (0, s )] f (x; p (q )) dx
L\
L
L\
L\
L L
*NLOL
"G(0, p (q ), q )#aM
M (0, s )# G(0, p (q ), q )!G(0, p (q ), q )
L L L
L\
L\
L L L
L L L
OL\QL\
#a
[M
M (0, q !x)! M
M (0, s )] f (x; p (q )) dx
L\
L
L\
L\
L L
*NLOL
)M
M (0, q )#G(0, p (q ), q )! G(0, p (q ), q ).
L
L
L L L
L L L
Hence, if G(0, p (q ), q ))G(0, p (q ), q ), then M
M (0, q ))M
M (0, q ) which implies that M
M (0, q ) is nonL L Ll
Ll L L
L
L
L
L
L
L
decreasing over[s , s ], where s )S .
L\ E
E
E





166

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

It follows from the de"nition of G (see Section 4.1) that G(0, p (q ), q )!G(0, p (q ), q )"
L L L
L L L
(q !q )(p (q )#r!c)!(p (q )#r#h!ac)[H(p (q ), q )!H(p (q ), q )]. Thus, we have
L
L L L
L L
L L L
L L L
G(0, p (q ), q )*G(0, p (q ), q ) 0
L L L
L L L

H(p (q ), q )!H(p (q ), q )
p (q )#r!c
L L L
L L L)
L L
,
q !q
p (q )#r#h!ac
L
L
L L

(A.9)

provided that p (q )#r#h!ac'0. In fact, it can be shown through standard derivative analysis that
L L
(0, which implies that p (q )#h!ac'0, and this in turn yields
p (q ) satis"es *G(0, p, q )/*p"
L L
L L
L
NLOL
p (q )#r#h!ac'0.
L L
Eq. (A.9) can be veri"ed exactly only if p (q ) is known for all q 3[s , sl] which is, in general, analytically
L L
L
L\ E
intractible. For this reason, we need to strengthen Eq. (A.9) in order to establish a veri"able su$cient
condition.
Since H(p , q ) is a continuous, convex increasing function in q for all p , and *H(p , q )/*q "F(q ; p ), the
L L
L
L
L L
L
L L
result in the Corollary is implied by Eq. (9) and Corollary 1.
Proof of Corollary 3. To initiate the inductive proof, we assume that s )S , which ensures that M
M (0, q ) is

E


K-decreasing over [S ,R). Then, we let S (q (q where it was determined earlier that
E
E L
L
sIL\\(S (sIL\ . Furthermore, we assume that M
M
(0, q ) is K-decreasing over [S ,R). Then, it
L\
E
L\
L\
L\
E
follows from Eqs. (3) and (5) that

M
M (0, q )"max G(0, p , q )#aM
M (0, s )
L
L
L L
L\
L\
OL\QIL\
#a
[M
M (0, q !x)!K! M
M (0,
L\
L
L\
*NL
OL\QL\
[M
M (0, q !x)!K! M
M (0,
#a
L\
L
L\
OL\QIL\
Having this representation, we can proceed as follows:

s )] f (x; p ) dx
L\
L

s )]>f (x; p ) dx: p 3[Pl, P ] .

L\
L
L


M
M (0, q )*max G(0, p , q )#aM
M (0, s )
L
L
L L
L\
L\
OL\QIL\
#a
[M
M (0, q !x)!K! M
M (0, s )] f (x; p ) dx
L\
L
L\
L\
L
L
*N 
OL\QL\
#a
[M
M (0, q !x)!K! M
M (0, s )]>f (x; p ) dx: p 3[Pl, P ]
L\
L
L\
L\
L
L


I
OL\QL\
"max+M
M (0, p , q )#G(0, p , q )! G(0, p , q )
L
L L
L L
L L




# aK[1!F(q !s ; p )]: p 3[Pl, P ],!aK

L
L\ L
L

*M
M (0, q )#G(0, p (q ), q )! G(0, p (q ), q )!aK
L
L
L L L
L L L

167

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

which implies that

M
M (0, q )*M
M (0, q )!aK,
(A.10)
L
L
L
L
if G(0, p (q ), q )!G(0, p (q ), q )*0. Note that Eq. (A.10) characterizes M
M as aK-decreasing (which
L
L L L
L L L
implies that it is K-decreasing) over [S ,R), and this leads to the desired result.
E
Derivation of M
M
II (0, q). The solution of the renewal equation Eq. (14) is ([25]):
O\*NO
M
M (0, q)"G(0, p , q)#a[1!F(q!s; p )]M
M (0, s)#
G(0, p , x)dR (q!x; p )
''
O
O
O
?
O

Q
O\*NO
#aMM (0, s)
[1!F(x!s; p )]dR (q!x; p ),
O
?
O

Q
where R is determined by
?
V
R (x; p)"aF(x; p)#a
F(x!u; p) dR (u; p).
?
?
*N

(A.11)

We evaluate R (x; p) at x"q!s and p"p ,

?
O
O\*NO
R (q!s; p )"aF(q!s; p )# a
F(x!s; p )dR (q!x; p ),
?
O
O
O
?
O

Q
which can be substituted in Eq. (A.11). Thus, we obtain

O\*NO
G(0, p , x) dR (q!x; p )
M
M (0, q)"G(0, p , q)#
O
?
O
''
O

Q
# aM
M (0, s)[1!F(q!s; p )# R (q!s; p )#F(q!s; p )! R (q!s; p )/a]
O
?
O
O
?
O
O\*NO
a
"G(0, p , q)#
G(0, p , x) dR (q!x; p )# G(0, s)
!R (q!s; p ) .
O
O
?
O
?
O
1!a
Q
This implies the desired result.




Proof of Corollary 4. Before we present the induction proof, we will establish the following property:
Property 1. a, b with a'b, we have P )P where P "argsup+(p#a)X(p): p3(0,R), and P "
?
@
?
@
argsup+(p#b)X(p): p3(0,R),.
Proof of Property 1. Let A(p)"(p#a)X(p) and B(p)"(p#b)X(p) for a'b. It is shown in [19] that if R(p)
("pX(p)) is pseudoconcave on (0,R), which is our basic assumption, then A(p) and B(p) are pseudoconcave
functions of p as well. In Fig. 8, we demonstrate A(p) and B(p) under an hypothetical X(p) function.

168

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

Fig. 8. Plots of the auxiliary functions (p#a)X(p) and (p#b)X(p) under an hypothetical X(p) function.

If P , P 3(0,R), then they must satisfy the "rst-order conditions A(P )"0 and B(P )"0. We rewrite
? @
?
@
A(p) as A(p)"B(p)#(a!b)X(p), which leads to A(p)"B(p)#(a!b)X(p). Evaluating the last equation at
p"P we get A(P )"B(P )#(a!b)X(P ), which implies
?
?
?
?
B(P )"!(a!b)X(P )'0.
(A.12)
?
?
That is, B(p) is increasing at p"P . Since B(p) is a pseudoconcave function we deduce that P )P . If P and
?
?
@
?
P are both non-interior point solutions, then P "P "0 since A(p)*B(p) for all p. If P "0 and
@
?
@
?
P 3(0,R), then P )P . Finally, if P 3(0,R), then from Eq. (A.12) we conclude that P )P . Hence,
@
?
@
?
?
@
P )P in any case.
?
@
We identify PL as the maximizer of (p !c)X (p ) over [Pl, P ]. That is, PL"min+max+PM L, Pl,, P , where
A
L
L L

A
A

PM L is the maximizer of (p !c)X (p ) over (0,R). Similarly, PL, PL and P are the maximizers of
A
L
L L
F FA
PA
(p #h!v)X (p ), (p #h!ac)X (p ) and (p !r!ac)X (p ) over [Pl, P ], respectively. Since !c)h!v,

L
L L L
L L
L
L L
!c)h!ac and !r!ac)!c, it follows from Property 1 that PL*PL, PL*PL and P *PL. That is,
A
F A
FA
PA
A
X (PL))X (PL), X (PL))X (PL ) and X (PL ))X (PL).
L A
L F
L A
L FA
L PA
L A
Now we can establish the proof of Corollary 4. It is shown in [19] that GH(0, q ) is quasiconcave with
L
L
S "X (PL). Having this in mind, assume that M
M (0, q ) satis"es (i) and (ii) in Theorem 2. Thus, there
L
L A
L\
L\
exists single reorder and order-up-to levels for period n!1. That is, k "2. Using this result, we obtain
L\
M
M (0, s ),
[q !X (p )]>(s ,
L\
L\
L
L L
L\
M
M (i , p , q )"G (i , p , q )#a
(A.13)
L L L L
L L L L
M
M (0, [q !X (p )]>), s )[q !X (p )]>.
L\
L
L L
L\
L
L L
First, we represent the n-period pseudo-pro"t function as M
M (0, q )"max+M
M (0, q ), M
M (0, q ),, where
L
L
L
L
L
L
M
M (0, q )"max+M
M (0, p , q ): Pl)p )pN ,,
L
L
L
L L
L
L
M
M (0, q )"max+M
M (0, p , q ): pN )p )P ,,
L
L
L
L L L
L

and pN is de"ned by X (pN )"max+min+q , X (Pl),, X (P ),. That is, pN "Pl if q *X (Pl), pN "P if
L
L L
L L
L 
L
L
L
L

q )X (P ); otherwise, pN is the unique solution of X (pN )"q . Note that if p 3[Pl, pN ), then q )X (p ) and
L
L 
L
L L
L
L
L
L
L L
if p 3(pN , P ], then q *X (p ). Therefore, M
M  and M
M  represent the n-period pseudo-pro"t values
L
L 
L
L L
L
L

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

169

under the best pricing policy that results in no leftovers and no shortages, respectively, at the end of period
n for a given q . In what follows, we work with these complementary subproblems to analyse M
M (0, q ) on
L
L
L
various q ranges and then combine our "ndings to complete the proof.
L
It follows from Eqs. (A.13) and (17) that, M
M (0, q )"aM
M (0, s )!cq #max+ p q !r(X (p )!q ):
L
L
L\
L\
L
L L
L L
L
Pl)p )pN ,, where the maximand is an increasing function of p . Hence, we obtain
L
L
L
MM (0, P , q ), 0)q )X (P ),
L
 L
L
L 
M
M (0, q )"
(A.14)
L
L
MM (0, q ),
X (P ))q .
L
L
L 
L
Note that MM (0, P , q ) is a linear increasing function of q , and if P PR, then X (P )P0 and we have
L
 L
L

L 
M
M (0, q )"M
M (0, q ) for q 3[0,R).
L
L
L
L
L
Next, consider the q range of [X (P ), X (PL)], where we have established that M
M (0, q ),
L
L 
L A
L
L
M
M (0, q ). We further divide M
M  as M
M (0, q )"max+M
M (0, q ), M
M (0, q ),, where
L
L
L
L
L
L
L
L
L
M
M (0, q )"max+M
M (0, p , q ): pN )p )pL ,
L
L
L
L L L
L
L
(A.15)
"aMM (0, s )#max+G (0, p , q ): pN )p )pL ,,
L\
L\
L
L L L
L
L
M
M (0, q )"max+M
M (0, p , q ): pL )p )P ,
L
L
L
L L L
L

"max+aM
M (0, q !X (p ))# G (0, p , q ): pL )p )P ,,
(A.16)
L\
L
L L
L
L L L
L

and X (pL )"max+min+q !s , X (Pl),, X (P ),. That is, pL "P if q )X (P )#s
and pL "Pl if
L 
L

L
L 
L\
L
L L
L
L\ L
X (Pl)#s )q , otherwise pL is the unique solution of X (pL )"q !s . Note that q !X (p ))s
L
L\
L
L
L L
L
L\
L
L L
L\
when p (pL , and s )q !X (p ) when pL (p .
L
L
L\
L
L L
L
L
We shall demonstrate that M
M  and M
M  are both increasing on [X (P ), X (PL)] which implies that
L
L
L 
L A
M
M  is also increasing over the same q range.
L
L
Considering the maximand in Eq. (A.15) and Fig. 8 with a"h!ac. We observe that the maximizing price
is pN for X (P ))q )X (PL ). Because, over this q range we have PL )pN and (p #h!ac)X (p ) is
L
FA
L
L
L L
L
L 
L
L FA
decreasing on [pN , P ]. For X (PL ))q )X (PL )#s , we have pN )PL )pL . Thus, the maximizer in
L 
L FA
L
L FA
L\
L
FA
L
Eq. (A.15) is PL . For X (PL )#s )q )X (Pl)#s , we have pL )PL , and the maximizer is pL . Finally,
L\
L
FA
L
FA
L FA
L\
L
L
for X (Pl)#s )q we have pL "Pl, and the function M
M (0, q ) is not de"ned over the q range of
L
L\
L
L
L
L
L
n[X (Pl)#s ,R).
L
L\
Summarizing our "ndings we have

M
M (0, q )"a(M
M
(0, S )!K )
L
L
L\
L\
L\
(pN !c)q ,
X (P ))q )X (PL ),
L
L
L 
L
L FA
# (PL #h!ac)X (PL )!(c#h!ac)q , X (PL ))q )X (PL )#s ,
FA
L FA
L
L FA
L
L FA
L\
X (PL )#s )q )X (Pl)#s .
(pL !c)q !(c#h!ac)s ,
L
L
L\
L FA
L\
L
L
L\
(A.17)

For q "X (pN ), it can be shown that (pN !c)q is a pseudoconcave function of q [19], where
L
L L
L
L
L
d
d
dpN
+(pN !c)q ," +(pN !c)X (pN ), L*0 0 pN *PL.
L
L
L
L L dq
L
A
dq
dpN
L
L
L
Note that dpN /dq (0, and increasing q means tracing p backwards in Fig. 8. Thus, (pN !c)q is increasing
L L
L
L
L
L
on [X (P ), X (PL)] and it is decreasing on [X (PL), X (Pl)]. Likewise, (pL !c)q is decreasing on
L 
L A
L A
L
L
L
[X (PL )#s , X (Pl)#s ], since pL )PL )PL over this q range. It, therefore, results from Eq. (A.17)
L FA
L\ L
L\
L
FA
A
L
that M
M  is increasing on [X (P ), X (PL)] and decreasing on [X (PL), X (Pl)#s ].
L
L 
L A
L A
L
L\

170

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

Next, we shall analyse M

M . Note that for q )X (P )#s , we have pL "P and M
M "M
M . Thus,
L
L
L 
L\
L

L
L
we need to consider M
M (0, q ) only on [X (P )#s ,R).
L
L
L 
L\
We identify q arbitrarily with X (P )#s )q (q )X (PL) to show that M
M (0, q )(
L
L 
L\
L
L
L A
L
L
M
M (0, q ). Let p be the maximizer in Eq. (A.16) which satis"es pL )p)P . Then, de"ne pL  by
L
L
L

L
q !s "X (pL  ), and p by q !X (p)"q !X (p). Thus, we have
L
L\
L L
L
L
L
L
q !X (p)"q !X (p)*s "q !X (pL  ),
L
L
L
L
L\
L
L L
which implies PL)pL  )p(p)P . Considering the order of these critical prices and observing Fig. 8 we
A
L

write
M
M (0, q )"aMM (0, q !X (p))!(c#h!ac)q # (p#h!ac)X (p)
L\
L
L
L
L
L
L
"aMM (0, q !X (p))!(c#h!ac)[q !X (p)]#(p!c)X (p)
L\
L
L
L
L
L
"aMM (0, q !X (p))!(c#h!ac)[q !X (p)]#(p!c)X (p)
L\
L
L
L
L
L
(aMM (0, q !X (p))!(c#h!ac)[q !X (p)]#(p!c)X (p)
L\
L
L
L
L
L
)max+aM
M (0, q !X (p ))! (c#h!ac)q # (p #h!ac)X (p ): pL  )p )P ,
L\
L
L L
L
L
L L L
L

"MM (0, q ),
L
L
which indicates that M
M  is an increasing function of q over [X (P )#s , X (PL)].
L
L
L 
L\ L A
We have so far shown that M
M  and M
M  are increasing in q over [X (P ), X (PL)]; thus, MM  ("M
M ) is
L
L
L
L 
L A
L
L
also increasing in q over the same range.
L

The next q
region will be [X (PL), X (PL )#s ]. We have MM
(0, q !X (p )))
L
L A
L FA
L\
L\
L
L L
M
M (0, S ) for all feasible p and q values. In addition, since PL )pL , (p #h!ac)X (p ) is decreasing
L\
L\
L
L
FA
L L
L L
over [pL , P ]. Hence, we have
L 
M
M (0, q ))max+aM
M (0, S )!(c#h!ac)q # (p #h!ac)X (p ): pL )p )P ,
L
L
L\
L\
L
L
L L L
L

)max+aM
M (0, S )!(c#h!ac)q # (p #h!ac)X (p ): pN )p )pL ,
L\
L\
L
L
L L L
L
L
"MM (0, q )#aK ,
L
L
L\
which implies that M
M (0, q ), given by Eq. (A.6), is bounded by MM (0, q ) and M
M (0, q )#aK , where
L
L
L
L
L
L
L\
M (0, q ) is an aK -decreasing function over
M
M  is decreasing (see Eq. (A.17)). Hence, M
L
L
L
L\
[X (PL), X (PL )#s ].
L A
L FA
L\
Consider now X (PL )#s )q . In this range, pL  )pL )PL . Hence, (p #h!ac)X (p ) is increasing
L FA
L\
L
L
L
FA
L
L L
over [pL  , pL ]. In addition, under the inductive assumption, MM (0, q !X (p )) is K -decreasing in q for
L L
L\
L
L L
L\
L
X (PL )#s )q with pL )p . Hence, we have
L FA
L\
L
L
L
M
M (0, q )"max+aM
M (0, q !X (p ))! (c#h!ac)q # (p #h!ac)X (p ): pL )p )P ,
L
L
L\
L
L L
L
L
L L L
L

"max+aM
M (0, q !X (p ))! (c#h!ac)q # (p #h!ac)X (p ): pL  )p )P ,
L\
L
L L
L
L
L L L
L

*max+a[M
M (0, q !X (p ))!K ]!(c#h!ac)q #(p #h!ac)X (p ): pL  )p )P ,
L\
L
L L
L\
L
L
L L L
L

'max+aM
M (0, q !X (p ))!(c#h!ac)q #(p #h!ac)X (p ): pL )p )P ,!aK
L\
L
L L
L
L
L L L
L

L\
"MM (0, q )!aK ,
L
L
L\

H. Polatoglu, I. Sahin/Int. J. Production Economics 65 (2000) 141}171

171

which implies that M

M  is aK -decreasing. Since M
M (0, q ) is decreasing over [X (PL )#s ,
L
L\
L
L
L FA
L\
X (Pl)#s ] (see Eq. (A.17)), M
M (0, q ) is aK -decreasing over [X (PL )#s ,R).
L
L\
L
L
L\
L FA
L\
Finally, to complete the continuity we need to consider q "X (PL )#s . Since MM (0, S ) and
L
L FA
L\
L\
L\
(PL #h!ac)X (PL ) are the global maximas of their respective functions, and M
M  is decreasing over
FA
L FA
L
[X (PL), X (PL )#s ], we obtain from Eq. (A.17) that
L A
L FA
L\
M
M (0, q )*M
M (0, q )"a(M
M (0, S )!K )! (c#h!ac)q # (PL #h!ac)X (PL )
L
L
L
L
L\
L\
L\
L
FA
L FA
*M
M (0, q )!aK ,
L
L
L\
for all q 'q "X (P )#s .
L
L
L FA
L\
Therefore, the proof follows by combining the results for the entire q range.
L
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]

K.J. Arrow, T. Harris, J. Marschak, Optimal inventory policy, Econometrica 19 (1951) 250}272.
T.M. Whitin, Inventory control and price theory, Management Science 2 (1955) 61}68.
E.S. Mills, Uncertainty and price theory, Quarterly Journal of Economics 73 (1959) 116}130.
E.S. Mills, Price, Output, and Inventory Policy, Wiley, New York, 1962.
S. Karlin, R.C. Carr, Prices and optimal inventory policy, in: K.J. Arrow, S. Karlin, H. Scarf (Eds.), Studies in Applied Probability
and Management Science, Stanford University Press, Stanford, CA, 1962.
A.J. Nevins, Some e!ects of uncertainty: Simulation of a model of price, The Quarterly Journal of Economics 80 (1966) 73}87.
A.L. Hempenius, Monopoly with random demand, Rotterdam University Press, Rotterdam, 1970.
E. Zabel, Monopoly and uncertainty, Review of Economic Studies 37 (1970) 205}219.
E. Zabel, Multiperiod monopoly under uncertainty, Journal of Economic Studies 5 (1972) 524}536.
E. Zabel, Price, output and inventory behavior with a general demand structure, in: The Economics of Inventory Management,
Elsevier, Amsterdam, 1988, pp. 319}332.
G.T. Thowsen, A dynamic nonstationary inventory problem for a price/quantity setting "rm, Naval Research Logistics Quarterly
22 (1975) 461}476.
L. Young, Price, inventory and the structure of uncertain demand, New Zealand Journal of Operational Research 6 (1978)
157}177.
L. Young, Uncertainty, market structure, and resource allocation, Oxford Economic Papers 46 (1979) 47}59.
L.J. Thomas, Price and production decisions with random demand, Operations Research 22 (1974) 513}518.
G. Gallego, G. Van Ryzin, Optimal dynamic pricing of inventories with stochastic demand over "nite horizons, Management
Science 40 (1994) 999}1020.
N.C. Petruzzi, Learning models for pricing and inventory control under uncertainty, Ph.D. Dissertation, Department of
Management, Purdue University, West Lafayette, IN, 1995.
S. Subrahmanyan, R. Shoemaker, Developing optimal pricing and inventory policies for retailers who face uncertain demand,
Journal of Retailing 72 (1996) 7}30.
E.L. Porteus, Stochastic inventory theory, in: D.P. Heyman, M.J. Sobel (Eds.), Handbooks in OR and MS, vol. 2, Elsevier,
North-Holland, 1990.
H. Polatoglu, Optimal order quantity and pricing decisions in single-period inventory systems, International Journal of
Production Economics 23 (1991) 175}185.
H. Scarf, The optimality of (s, S) policies in the dynamic inventory problem, in: Mathematical Methods in the Social Sciences,
Stanford University Press, Stanford, CA, 1960.
A.F. Veinott, On the optimality of (s, S) inventory policies: new conditions and a new proof, SIAM Journal of Applied
Mathematics 14 (1966) 1067}1083.
E.L. Porteus, On the optimality of generalized (s, S) policies, Management Science 17 (1971) 411}426.
M. SchaK l, On the optimality of (s, S)-policies in dynamic inventory models with "nite horizon, SIAM Journal of Applied
Mathematics 30 (1976) 528}537.
A. Lau, Hing-Ling, Lau Hon-Shiang, The newsboy problem with price-dependent demand distribution, IIE Transactions 20 (1988)
168}175.
I. Sahin, A generalization of renewal processes, Operations Research Letters 13 (1993) 259}263.
L.J. Thomas, Price-production decisions with deterministic demand, Management Science 16 (1970) 747}750.