Professional Documents
Culture Documents
Inventory Dependent Demand
Inventory Dependent Demand
Operations Research
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.
INFORMS is the largest professional society in the world for professionals in the fields of operations research, management
science, and analytics.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
OPERATIONS RESEARCH
Vol. 62, No. 5, September–October 2014, pp. 1077–1094
ISSN 0030-364X (print) ISSN 1526-5463 (online) http://dx.doi.org/10.1287/opre.2014.1306
© 2014 INFORMS
We analyze a finite horizon periodic review joint pricing and inventory management model for a firm that replenishes and
sells a product under the scarcity effect of inventory. The demand distribution in each period depends negatively on the
sales price and customer-accessible inventory level at the beginning of the period. The firm can withhold or dispose of
its on-hand inventory to deal with the scarcity effect. We show that a customer-accessible-inventory-dependent order-up-
to/dispose-down-to/display-up-to list-price policy is optimal. Moreover, the optimal order-up-to/display-up-to and list-price
levels are decreasing in the customer-accessible inventory level. When the scarcity effect of inventory is sufficiently strong,
the firm should display no positive inventory and deliberately make every customer wait. The analysis of two important
special cases wherein the firm cannot withhold (or dispose of) inventory delivers sharper insights showing that the inventory-
dependent demand drives both optimal prices and order-up-to levels down. In addition, we demonstrate that an increase
in the operational flexibility (e.g., a higher salvage value or the inventory withholding opportunity) mitigates the demand
loss caused by high excess inventory and increases the optimal order-up-to levels and sales prices. We also generalize our
model by incorporating responsive inventory reallocation after demand realizes. Finally, we perform extensive numerical
studies to demonstrate that both the profit loss of ignoring the scarcity effect and the value of dynamic pricing under the
scarcity effect are significant.
Subject classifications: joint pricing and inventory management; inventory-dependent demand.
Area of review: Operations and Supply Chains.
History: Received January 2013; revisions received September 2013, March 2014; accepted June 2014. Published online
in Articles in Advance August 20, 2014.
product, thus creating a “hot product.” Empirical results on In this paper, we study the dynamic pricing and inventory
the scarcity effect of inventory on demand in the automo- management model under the scarcity effect of inventory.
bile industry can also be found in, e.g., Balachander et al. The stochastic demand is modeled as a decreasing func-
(2009) and Cachon et al. (2013). tion of the sales price and the customer-accessible inven-
Second, a low inventory level spreads a sense of urgency tory level at the beginning of each decision epoch. Unmet
among customers that soon the product will be sold out demand is fully backlogged to the next period. The wait
and potential buyers will be put on a wait list. Such back- lists observed or spread through word-of-mouth success-
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
logging risk motivates customers to make an immediate fully signal the high quality and popularity of the prod-
purchase instead of searching for better options. A high uct and attract more customers (see, e.g., Brown 2001
inventory, however, grants customers the luxury of waiting and Dye 2000). From the strategic perspective, joint pric-
and searching, thus lowering the current demand. A simi- ing and inventory decisions effectively deliver information
lar mechanism also drives the search behavior that a low about the quality and popularity of the product. Specif-
inventory of one product type discourages a customer from ically, pricing flexibility induces more strategic customer
searching for better types (Cachon et al. 2008). Knowingly behavior (e.g., waiting for potential price discount), which
limiting the availability of a product, the retailer can induce further strengthens the scarcity effect of inventory because
“buying frenzies” among uninformed customers and set a customers may anticipate price changes based on current
higher price (DeGraba 1995). inventory (see, e.g., Liu and van Ryzin 2008).
Third, as shown in pricing and revenue management We develop a unified joint price and inventory man-
literature (e.g., Federgruen and Heching 1999, Gallego agement model that incorporates both inventory withhold-
and van Ryzin 1994), retailers increase their sales prices ing and inventory disposal to address the scarcity effect.
when inventories are low. Therefore, customers infer from Under the inventory withholding policy, the firm displays
a low inventory level that it is unreasonable to expect a only part of its inventory and withholds the rest in a ware-
lower price and decide to purchase the item immediately house not observable by customers; this induces higher
(see, e.g., Aviv and Pazgal 2008). On the other hand, a potential demand. Analogously, with inventory disposal,
high inventory level suggests that the sales price will be the firm can dispose of unnecessary excess inventory that
more likely to decrease; this encourages customers to wait has some salvage value. Both inventory withholding and
before buying. Carefully making use of this mechanism, disposal may incur a cost. We show that a customer-
the retailer can enjoy the benefits of inducing customers to accessible-inventory-dependent order-up-to/dispose-down-
purchase early at high prices (Liu and van Ryzin 2008). to/display-up-to list-price policy is optimal. Moreover, the
A similar idea has also been adopted in the advance sell- order-up-to/display-up-to and list-price levels are decreas-
ing literature (e.g., Xie and Shugan 2001), which shows ing at the customer-accessible inventory level. When the
that firms may limit their capacity for advance selling to scarcity effect of inventory is sufficiently strong, the firm
credibly signal their pricing strategy to customers. should display no positive inventory so that every customer
Along with the rich theoretical and empirical justifica- must wait before getting the product. In this case, the strong
tions of the scarcity effect of inventory, practitioners have scarcity effect creates more opportunities than risks, so the
extensively adopted this idea in their marketing strategies. firm can proactively take advantage of it and induce more
Dye (2000) and Brown (2001) document that the scarcity demand by making customers wait (e.g., the BMW mar-
strategy, in which product supply is deliberately limited, keting strategy).
has become a basic tactic for marketers to promote sales. When it is too costly to withhold or dispose of inven-
An increasing number of automobile manufacturers cre- tory, the unified model is reduced to the model without
ate significant levels of scarcity and make a long list of inventory withholding or the model without inventory dis-
hard-to-get car models over time (see Balachander et al. posal, both of which deliver sharper insights. In the model
2009). Although facing thousands of customers on the wait without inventory withholding/disposal, we show that the
lists, none of the manufacturers rushed to accelerate pro- inventory-dependent demand increases the overstocking
duction (Wall Street Journal 1999). Likewise, Maynard risk and, thus, lowers optimal sales prices and order-up-
(2006) shows that BMW promotes its Mini Cooper line to levels. With higher operational flexibility (a higher sal-
by limiting supply and letting potential owners wait for, vage value or the inventory withholding opportunity), how-
on average, two and half months for the models. The lim- ever, the firm addresses the scarcity effect of inventory
ited distribution strategy has accelerated demand for the more effectively and, hence, increases its sales prices and
Mini Cooper since its reintroduction in the U.S. market. A order-up-to/display-up-to levels. In short, inventory dis-
similar promotional strategy is also used in the electron- posal/withholding benefits the firm by enhancing its opera-
ics market, especially at the introduction stage of a new tional flexibility and agility.
product generation; fans have been excited by the long wait We also generalize the unified model by incorporating
for Sony Play Stations (Retailing Today 2000), Nintendo responsive inventory reallocation, which allows the firm
Game Boys (Wall Street Journal 1989), and Apple iPads 2 to reallocate (with cost) its inventory between display and
(Sherman 2010). warehouse after demand is realized.. In this case, the firm
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS 1079
can keep a low inventory and better hedge against risks of a comprehensive review. Demand dependence on inven-
demand uncertainty and the scarcity effect of inventory. tory is usually modeled in two ways in the literature, i.e.,
We perform extensive numerical studies to demonstrate (1) potential demand is increasing in the inventory level
(a) the robustness of our analytical results, (b) the impact after replenishment, and (2) potential demand is decreas-
of the scarcity effect on the profitability of the firm, and ing in the inventory level before replenishment (i.e., leftover
(c) the value of dynamic pricing under the scarcity effect inventory from the previous period).
of inventory. Our numerical results show that the analytical The first approach to model inventory-dependent demand
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
characterizations of the optimal policies in our model are assumes that demand increases with inventory (the bill-
robust and hold in all of our numerical experiments. Both board effect). Gerchak and Wang (1994) study a periodic
the profit loss of ignoring the scarcity effect and the value review inventory model in which the random demand in
of dynamic pricing under the scarcity effect are signifi- each period is increasing in the inventory level after replen-
cant, and increase the intensity of the scarcity effect and/or ishment. Dana and Petruzzi (2001) consider a single-period
demand variability. The reasons are: (1) the scarcity effect newsvendor model where demand is decreasing in price
decreases future demand and magnifies future demand vari- and positively correlated with inventory level. Several other
ability; and (2) dynamic pricing allows the firm to induce operations management and marketing papers also assume
higher future demand and dampen future demand variabil- that demand depends on the instantaneous (after replen-
ity. In addition, a longer planning horizon increases the ishment) inventory level, in particular via the shelf-space
impact of the scarcity effect and decreases the value of effect. We refer interested readers to, e.g., Wang and Ger-
dynamic pricing. chak (2001 and 2002), Balakrishnan et al. (2004 and 2008),
We summarize the main contributions of this paper as Martńez-de-Albéniz and Roels (2011), Baron et al. (2011)
follows: (1) To our knowledge, we are the first to study and Chen et al. (2012).
joint pricing and inventory management under the scarcity The other effect of inventory on demand, as discussed
effect of inventory. We characterize the optimal policy in in §1, is the scarcity effect. That is, high leftover inven-
a general unified model and generalize our results to the tory (i.e., inventory at the beginning of the period before
model with responsive inventory reallocation. (2) We ana- replenishment) negatively influences the potential demand.
lyze the impact of the scarcity effect of inventory on the In the psychological commodity theory literature, Brock
firm’s optimal price and inventory policies and study the (1968) argues that supply scarcity increases the attractive-
effect of operational flexibilities on the firm’s optimal deci- ness of a product; this has been tested by numerous exper-
sions under the scarcity effect. (3) We identify the rationale iments in, e.g., Worchel et al. (1975) and van Herpen et al.
behind the phenomenon that firms with intense scarcity (2009). Stock and Balachander (2005) and Veeraraghavan
effect deliberately make their customers wait before get- and Debo (2009) use game theoretic models to show that
ting the product. (4) We numerically study the profit loss the firm can use the scarcity strategy to signal the high
of ignoring the scarcity effect and the value of dynamic quality of a product. Aviv and Pazgal (2008), among oth-
pricing under this effect. ers, demonstrate that customers may strategically wait for
The rest of the paper is organized as follows. In §2, price discounts when observing a high inventory. Liu and
we position this paper in the related literature. Section 3 van Ryzin (2008) propose an effective pricing scheme to
presents the basic formulation, notations, and assumptions induce customers to make early purchases under a revenue
of our model. In Section 4, we propose and analyze the management framework. The idea that supply condition can
unified model. Section 5 discusses additional results and signal potential pricing strategy and product quality has
insights in two important cases, i.e., the model without also been adopted in the advance selling literature (e.g., Xie
inventory withholding and the model without inventory dis- and Shugan 2001 and Yu et al. 2014). Balachander et al.
posal. Section 6 generalizes the unified model to the model (2009) and Cachon et al. (2013) conduct empirical stud-
with responsive reallocation. Section 7 reports our numeri- ies to show that the scarcity effect of inventory on demand
cal findings. We conclude by summarizing our findings and prevails in the automobile industry.
discussing a possible extension in §8. All proofs are rel- To our knowledge, Sapra et al. (2010) is the only paper
egated to the online appendix (available as supplemental in the inventory management literature that incorporates the
material at http://dx.doi.org/10.1287/opre.2014.1306). scarcity effect of inventory (called the “wait-list effect” in
that paper) and assumes that potential demand is a decreas-
ing function of leftover inventory. They show the optimal-
2. Literature Review ity of understocking and propose the inventory withholding
This paper is mainly related to two lines of research in the strategy. Our paper generalizes Sapra et al. (2010) in the
literature, i.e., (1) inventory management with inventory- following aspects: (1) We introduce a unified model that
dependent demand, and (2) optimal joint pricing and inven- encompasses dynamic pricing, inventory withholding, and
tory policy. inventory disposal, and explicitly captures the interaction
There is a large body of literature on inventory-dependent between price, inventory and demand. In particular, we ana-
demand. We refer interested readers to Urban (2005) for lytically show the impact of inventory-dependent demand
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
1080 Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS
on the firm’s pricing policy, whereas Sapra et al. (2010) do Petruzzi (2001) and Balakrishnan et al. (2008) have stud-
not allow price adjustment during the planning horizon and ied the joint pricing and inventory control problem with
numerically test the improvement of inventory-withholding inventory-dependent demand. However, both papers con-
policy under different price elasticities of demand. We also sider a single period model where demand is increasing in
numerically show that the value of dynamic pricing under the available inventory after replenishment.
the scarcity effect of inventory is significant and increases
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
with the scarcity effect intensity and/or demand variability. 3. Model Formulation
(2) Because of the endogenous pricing decision introduced
to the dynamic program, analysis of our model is more We specify our unified model, notations, and assumptions
involved and requires a different approach. (3) Two special in this section. Consider a firm facing random demand and
cases of our unified model (i.e., the model without inven- periodically making pricing and inventory decisions in a
tory withholding and the model without inventory disposal) T -period planning horizon, labeled backwards as 8T 1 T −11
demonstrate that inventory withholding and inventory dis- 0 0 0 1 19. The firm stores its on-hand inventory in two loca-
posal help mitigate the overage risk of inventory-dependent tions, one with customer-accessible inventory to satisfy and
demand. (4) In addition to the understocking and inven- stimulate demand, and the other as a warehouse to withhold
tory withholding policy proposed in Sapra et al. (2010), inventory that is unobservable to customers. The firm can
our model suggests three other strategies to dampen the replenish or dispose of inventory; it can also reallocate its
on-hand inventory between the customer-accessible storage
negative effect of inventory-dependent demand: (a) price
and the warehouse. If the firm places an order, the replen-
reduction, (b) inventory disposal, and (c) responsive inven-
ished inventory is delivered to the warehouse after which
tory reallocation. (5) We show that when the scarcity effect
the firm decides how much inventory to reallocate to the
of inventory is sufficiently strong, the firm should display
customer-accessible storage. On the other hand, if the firm
no positive inventory and let customers wait. To summa-
disposes of its on-hand inventory, it first ships inventory, if
rize, this paper generalizes the model in Sapra et al. (2010)
any, from the customer-accessible storage to the warehouse,
and strengthens its results and insights.
and then chooses the disposal quantity.
There is an extensive literature on dynamic pricing and
In each period, the sequence of events unfolds as fol-
inventory control under general stochastic demand. Fed-
lows: At the beginning of each period, the firm reviews its
ergruen and Heching (1999) study the inventory system
total and customer-accessible leftover inventories from last
in a periodic review model where the firm faces price-
period, simultaneously chooses the order/disposal and real-
dependent demand in each decision period and unsatisfied
location quantities and the sales price, pays the ordering
demand is fully backlogged. A list-price order-up-to policy
and reallocation costs, and receives the disposal salvages.
is shown to be optimal. This line of literature has grown
The ordering and reallocation lead times are assumed to be
rapidly since Federgruen and Heching (1999). For exam-
zero so that the replenished and reallocated inventories are
ple, Chen and Simchi-Levi (2004a, b and 2006) analyze
received immediately. Inventory disposal is also executed
the joint pricing and inventory control problem with fixed
at once. The demand is then realized and the revenue is
ordering cost and show the optimality of (s1 S1 p) policy
collected. At the end of the decision period, the holding
for finite horizon, infinite horizon, and continuous review
and backlogging costs are paid, and the total and customer-
models. Chen et al. (2006) and Song et al. (2009) study
accessible inventories are carried over to the beginning of
the joint pricing and inventory control problem under lost the next period.
sales. In the case of a single unreliable supplier, Li and The state of the system is given by:
Zheng (2006) and Feng (2010) show that supply uncer-
tainty drives the firm to charge higher prices under ran- Ita = the starting customer-accessible inventory level before
dom yield and random capacity, respectively. Chen et al. replenishment/disposal/reallocation in period t, t =
(2011) take into consideration costly price adjustments in T 1 T − 11 0 0 0 1 1, where the superscript “a” refers to
joint pricing and inventory management. When the replen- “customer-accessible”;
ishment lead time is positive, the joint pricing and inven- It = the starting total inventory level before replenish-
tory control problem under periodic review is extremely ment/disposal/reallocation in period t, t = T 1 T − 11
difficult. Pang et al. (2012) partially characterize the struc- 0 0 0 1 1.
ture of the optimal policy. We refer interested readers to
Note that the amount of inventory the firm withholds in
Chen and Simchi-Levi (2012) for a comprehensive survey
the warehouse is It − Ita ¾ 0. We introduce the following
on joint pricing and inventory control models. The major
notation to denote the decisions of the firm:
difference between our paper and this stream of research
is that we take into account inventory-dependent demand pt = the sales price charged in period t, t = T 1 T −11 0 0 0 1 1;
and show that the scarcity effect of inventory drives the xta = the customer-accessible inventory level after replen-
firm to order less, dispose more, withhold inventory, and ishment/disposal/reallocation but before demand is
charge a lower price. To our knowledge, only Dana and realized in period t, t = T 1 T − 11 0 0 0 1 1;
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS 1081
xt = the total inventory level after replenishment/disposal/ base our analysis of the problem on the following demand
reallocation but before demand is realized in period t, form:
t = T 1 T − 11 0 0 0 1 1.
4pt 1 It 1
t 5 = 4d4pt 5 + 4Ita 55
tm +
ta 1
We assume that the price pt is bounded from above
by the maximum allowable price p̄ and from below by where Ɛ8
ta 9 = 0 and Ɛ8
tm 9 = 10 (1)
the minimum allowable price p. Without loss of gener- We assume that
t ’s are i0i0d0 random vectors with
ta sup-
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
ality, we also assume that the customer-accessible inven- ported on 6a1 ā7 and
tm supported on 6m1 m̄7 (m ¾ 0). At
tory storage capacity of the firm is Ka (0 < Ka ¶ +), least one of the two random variables (
ta and
tm ) follows a
whereas the warehouse capacity is infinite. In other words, continuous distribution (i.e., a 6= ā or m 6= m̄). This ensures
the customer-accessible inventory level after replenish- that Dt follows a nondegenerate continuous distribution
ment/disposal/reallocation cannot exceed Ka in each period, supported on the interval: 64d4pt 5 + 4Ita 55m + a1 4d4pt 5 +
i.e., xta ¶ Ka for all t = T 1 T − 11 0 0 0 1 1. Following the 4Ita 55m̄ + ā7 for any 4pt 1 Ita 5. Note that the above demand
“no-artificial wait-list” notion (see Sapra et al. 2010), model is quite general and includes as special cases sev-
we assume that the firm cannot decrease its customer- eral demand models from the existing literature. For exam-
accessible inventory level if a wait list already exists, i.e., ple, when
tm = 1 with probability 1, the demand model
xt ¾ xta ¾ min8Ita 1 09. is reduced to the additive demand model; if
ta = 0 with
We introduce the following model primitives: probability 1, it is reduced to the multiplicative demand
= discount factor of revenues and costs in future periods, model (as a generalized version of the model proposed in
0 < ¶ 1; Sapra et al. 2010); and if 4 · 5 ≡ 0, the demand model is
c = purchasing cost per unit ordered; reduced to the standard price-dependent demand model (as
s = salvage value per unit disposed; proposed in Chen and Simchi-Levi 2004a). The term d4pt 5
b = backlogging cost per unit backlogged at the end of a summarizes the impact of price on demand in period t.
period; As assumed above, d4 · 5 is strictly decreasing in pt . In
ha = holding cost per unit stocked and accessible to cus- some market where competition is fierce and the firm has
tomers at the end of a period; no pricing power, the price is exogenously fixed at p0 and
hw = holding cost per unit stocked in the warehouse at the the price induced demand is fixed at d0 = d4p0 5. The term
end of a period; 4Ita 5, which is a decreasing function of Ita , captures the
rd = unit reallocation fee from the warehouse to the scarcity effect of inventory on demand. Hereafter, we refer
customer-accessible storage; to 4 · 5 as the scarcity function, and 0 4 · 5 as the intensity
rw = unit reallocation fee from the customer-accessible of scarcity effect. The dependence of demand on inventory
storage to the warehouse. is measured by 0 4 · 5. i.e., the smaller the 0 4 · 5, the more
intensive the potential demand depends on the customer-
Without loss of generality, we assume that the following accessible inventory level. When demand is independent of
inequalities hold: inventory, 4Ita 5 ≡ 0 for all customer-accessible inventory
b > 41 − 54rd + c5: the backlogging penalty is higher level Ita . Note that our demand model generalizes the model
than the saving from delaying an in Sapra et al. (2010) in the sense that our model also cap-
order to the next period, so that the tures the impact of endogenous sales price on demand.
firm will not backlog all of its Since d4 · 5 is strictly decreasing in pt , we assume p4dt 5
demand; to be its strictly decreasing inverse. For the convenience
c > s: unit procurement cost dominates of our analysis, we change the decision variable from pt
to dt ∈ 6d1 d7,¯ where d = d4p̄5 and d¯ = d4p5. To avoid
the unit salvage value;
p > 4c + rd 5 + b 2 positive margin for backlogged the unrealistic case where demand becomes negative, we
demand. assume that d + 4Ka 5 ¾ 0 to ensure that Ɛ8Dt 9 = dt +
4Ita 5 ¾ 0 for any dt ∈ 6d1 d7 ¯ and Ita ¶ Ka . We impose the
Note that although we assume that the parameters and following three assumptions throughout our analysis:
demand are stationary throughout the planning horizon, the
structural results in this paper remain valid when the param- Assumption 1. The inverse demand function p4 · 5 is twice
eters and demand distributions are time-dependent. continuously differentiable and concavely decreasing in dt ,
¯ In addition, p4dt 5dt is con-
with p0 4dt 5 < 0 for dt ∈ 6d1 d7.
As discussed in §1, we assume that demand in
period t, Dt , depends negatively on the prevailing price cave in dt .
and customer-accessible inventory level at the beginning The concavity of p4dt 5dt in dt suggests the decreas-
of this period according to a general stochastic functional ing marginal revenue with respect to demand, which is a
form: Dt = 4pt 1 Ita 1
t 51 where
t is a random term with standard assumption in joint pricing and inventory manage-
a known continuous distribution and a connected support. ment literature, see, e.g., Chen and Simchi-Levi (2004a),
( · 1 · 1
t ) is a twice continuously differentiable function Li and Zheng (2006) and Pang et al. (2012). For a more
strictly decreasing in pt and decreasing in Ita for any
t . We comprehensive discussion on decreasing marginal revenue
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
1082 Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS
assumptions, see Ziya et al. (2004). The concavity of p4 · 5 Condition (3) is complicated and somewhat difficult to
implies that the demand is more price sensitive when sales understand. Hence, we give the following simpler necessary
prices are higher. This is also a common assumption in the condition for Assumption 3 to hold:
literature; see, e.g., Federgruen and Heching (1999).
As Sapra et al. (2010), we also assume that demand Lemma 2. If R4 · 1 · 5 is jointly concave on its domain, then
is concavely decreasing in the customer-accessible leftover we have:
inventory. (a) For any Ita such that 00 4Ita 5 = 0, 0 4Ita 5 = 0 as well.
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
3.1. Discussions on Assumption 3 is jointly concave in 4dt 1 Ita 5 for dt ∈ 6d1 d7 ¯ and Ita ¶ Ka .
00
(b) Suppose that p 4 · 5 6= 0 for any dt ∈ 6d1 d7. ¯ For any
Assumption 3 is essential to show the analytical results in
this paper. We first characterize the necessary and sufficient scarcity function 4 · 5, there exists a threshold ∗ < +,
condition for Assumption 3: such that, for any ¾ ∗ , with ˆ 4 · 5 2= 4 · 5 + ,
R̂ 4dt 1 Ita 5 2= 4p4dt 5 − b − 4c + rd 554dt + ˆ 4Ita 55 is jointly
Lemma 1. R(dt 1 Ita ) is jointly concave in (dt 1 Ita ) on its ¯ and Ita ¶ Ka .
concave in 4dt 1 Ita 5 for dt ∈ 6d1 d7
domain if and only if
4p00 4dt 54dt +4Ita 55+2p0 4dt 554p4dt 5−b −4c +rd 55 00 4Ita 5 Lemma 3 demonstrates that, as long as the condition
characterized in Lemma 2(b) on the scarcity functions,
¾ 4p0 4dt 5 0 4Ita 552 1 ¯ and Ita ¶ Ka 0 (3)
for all dt ∈ 6d1 d7 4 · 5, is satisfied, R4 · 1 · 5 is jointly concave on its domain
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS 1083
if (a) the sales price of the product, p4 · 5, is sufficiently high customer-accessible leftover inventory not only incurs
high relative to the inverse of price sensitivity, p0 4 · 5; a high holding cost but also suppresses potential demand.
or (b) the price is not linear in demand, and the scarcity Both inventory withholding and inventory disposal policies
effect driven demand, 4 · 5, is sufficiently high relative to enable the firm to strategically maintain a low customer-
the scarcity intensity, 0 4 · 5. These sufficient conditions accessible inventory, so as to induce high potential demand
have a clear economic interpretation: The price elastic- and mitigate the overstocking risk. Hence, we incorporate
ity of demand (i.e., 4 d dt /dt 5/4 d pt /pt 5) is sufficiently inventory withholding and inventory disposal into our uni-
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
= 4p4dt 5−4c +rd 5−b54dt +4Ita 55−4c −s54xt −It 5− When the firm disposes of its on-hand inventory, x̃ta 4Ita 5 is
the optimal display-up-to inventory level and x̃t 4Ita 5 is the
−4rd +rw 54xta −Ita 5− −4hw +41−5c5xt
optimal dispose-down-to inventory level, whereas d˜t 4Ita 5 is
+4hw +b −41−5rd 5xta the optimal expected price-induced demand. The following
lemma establishes the properties of the two optimizers:
+Ɛ −4ha +b54xta −4dt +4Ita 55
tm −
ta 5+
Lemma 4. For each t = T 1T −1100011, the following state-
+6Vt−1 4xta −4dt +4Ita 55
tm −
ta 1xt
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
ments hold:
−4dt +4Ita 55
tm −
ta 5−rd 4xta −4dt +4Ita 55
tm −
ta 5 (a) Jt 4xta 1xt 1dt 1Ita 1It 5 is jointly concave and continu-
ously differentiable in (xta 1xt 1dt 1Ita 1It ) except for a set
−c4xt −4dt +4Ita 55
tm −
ta 57
of measure zero; for any fixed (Ita 1It ), Jt 4·1 ·1 ·1Ita 1It 5 is
= R4dt 1Ita 5−4xt −It 5− −4rd +rw 54xta −Ita 5− −xt +xta strictly jointly concave in (xta 1xt 1dt ).
(b) Vt 4Ita 1It 5 is jointly concave and continuously dif-
+Ɛ Gt 4xta −4dt +4Ita 55
tm −
ta 1xt
ferentiable in (Ita 1It ), whereas Vt 4Ita 1It 5−rd Ita −cIt is
−4dt +4Ita 55
tm −
ta 5 1
(5) decreasing in Ita and It .
Lemma 4 proves that the objective function in each
period is jointly concave and almost everywhere differ-
where Gt 4x1y5 2 = −4b +ha 5x+ +4Vt−1 4x1y5−rd x −cy51 entiable and the value function is jointly concave and
continuously differentiable. Moreover, the second half of
2 = c −s = the unit loss of inventory
Lemma 4(b) implies that the normalized value function,
disposal1 (6)
Vt 4Ita 1It 5−rd Ita −cIt , is decreasing in both the customer-
2 = hw +41−5c = the unit cost of accessible inventory level Ita and the total inventory level It ,
replenishing and holding inventory which generalizes Proposition 5.1 in Sapra et al. (2010).
in the warehouse1 We note that the joint concavity of R4·1 ·5 on its entire
domain is necessary to prove that the objective functions
2 = hw +b −41−5rd
Jt 4·1 ·1 ·1Ita 1It 5 and that the value functions Vt 4·1 ·5 are
= the unit saving of reallocating jointly concave, which is essential to analytically estab-
warehouse inventory to the lish other structural results in our paper. We can easily
customer-accessible storage0 find examples in which R4·1 ·5 fails to be jointly concave
(e.g., 4·5 contains a linear and strictly decreasing piece)
We use 4xta∗ 4Ita 1It 51xt∗ 4Ita 1It 51dt∗ 4Ita 1It 55 to denote the and leads to nonconcave Jt 4·1 ·1 ·1Ita 1It 5s and Vt 4·1 ·5s. In
maximizer in (4), which stands for the optimal policy in this case, we are unable to analytically show the struc-
period t, with customer-accessible inventory level Ita and tural results in our paper (e.g., Theorem 1 and Theorem 3).
total inventory level It . To characterize the structure of In §7.1, we numerically test whether the structure of the
the optimal inventory replenishment/disposal/reallocation optimal policy characterized in our theoretical model still
and pricing policies, we define the following optimizers: holds. With the help of Lemma 4, we characterize the struc-
4xta 4Ita 51dt 4Ita 55 and 4x̃ta 4Ita 51 x̃t 4Ita 51 d˜t 4Ita 55. Let tural properties of the optimal policy in the unified model
as follows:
4xta 4Ita 51dt 4Ita 55 2= R4dt 1Ita 5+xta
argmax Theorem 1. For t = T 1T −1100011, the following state-
xta ∈6min8Ita 1091Ka 71dt ∈6d1 d7
¯
ments hold:
+Ɛ6Gt 4xta −4p4dt 51Ita 1
t 51xta (a) xta 4Ita 5 ¶ x̃t 4Ita 5. Moreover, let
−4p4dt 51Ita 1
t 557 1
(7) qt∗ 4Ita 1It 5 2= xt∗ 4Ita 1It 5−It
where 2= b −41−54c +rd 5 > 00 (8) denote the optimal order/disposal quantity and we have:
a a
Note that xta 4Ita 5 is the optimal order-up-to inventory level, > 0 if It < xt 4It 51
if the firm procures positive inventory and displays all of qt∗ 4Ita 1It 5 = 0 if xta 4Ita 5 ¶ It ¶ x̃t 4Ita 51
its on-hand inventory to customers, whereas dt 4Ita 5 is the
< 0 otherwise1
optimal expected price-induced demand in this case. Let
i.e., it is optimal to order if and only if It < xta 4Ita 5 and to
4x̃ta 4Ita 51 x̃t 4Ita 51 d˜t 4Ita 55 dispose if and only if It > x̃t 4Ita 5.
(b) If It < xta 4Ita 5,
2= argmax R4dt 1Ita 5+4 −5xt
4xta 1xt 1dt 5∈F 4Ita 5 xta∗ 4Ita 1It 5 = xt∗ 4Ita 1It 5 = xta 4Ita 51 dt∗ 4Ita 1It 5 = dt 4Ita 51
−4rd +rw 54xta −Ita 5− +xta
i.e., it is optimal to order and display up to xta 4Ita 5 and
+Ɛ8Gt 4xta −4p4dt 51Ita 1
t 51xt −4p4dt 51Ita 1
t 559
0 (9) charge a list price p4dt 4Ita 55.
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS 1085
(c) If It > x̃t 4Ita 5, strategies. As suggested in Theorem 1, the firm needs
to price the product in accordance with the customer-
4xta∗ 4Ita 1It 51xt∗ 4Ita 1It 51dt∗ 4Ita 1It 55 = 4x̃ta 4Ita 51 x̃t 4Ita 51 d˜t 4Ita 551 accessible inventory level so as to better control the scarcity
effect of demand. Theorem 1 also shows that when the total
i.e., it is optimal to dispose the total inventory level down inventory is high, the firm should withhold and dispose of
to x̃t 4Ita 5, display x̃ta 4Ita 5, and charge a list price p4d˜t 4Ita 55. its on-hand inventory, which saves holding costs and miti-
(d) If It ∈ 6xta 4Ita 51 x̃t 4Ita 57, xt∗ 4Ita 1It 5 = It , i.e., it is opti- gates the risk of suppressing potential demand. On the other
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
mal to keep the total inventory level. hand, the opportunity to redisplay the withheld inventory in
(e) xta 4Ita 5 is continuously decreasing in Ita , whereas the warehouse enables the firm to satisfy potential demand
dt 4Ita 5 is continuously increasing in Ita . without discouraging it. In short, combining dynamic pric-
Theorem 1 generalizes Proposition 5 in Sapra et al. ing, inventory withholding, and inventory disposal policies
(2010) by characterizing the structure of the optimal pol- helps the firm better match supply and demand and greatly
icy in our unified model. We show that a customer- enhances its profitability.
accessible inventory-dependent order-up-to/dispose-down- We now analyze how the model primitives influence the
to/display-up-to list-price policy is optimal. The optimal firm’s optimal operational decisions, such as inventory dis-
policy is characterized by two thresholds, i.e., the order- posal, inventory withholding, and inventory display.
ing threshold xta 4Ita 5 and the disposal threshold x̃t 4Ita 5,
Theorem 2. The following statements hold:
both of which depend on the customer-accessible inven-
(a) If hw ¾ c −s, x̃t 4Ita 5 = x̃ta 4Ita 5 for any t = T 1T −11
tory level, Ita . If the total inventory level, It , is below the
00011.
ordering threshold, i.e., It < xta 4Ita 5, the firm should order-
(b) There exists an s∗ < c, such that, if s ¶ s∗ , x̃t 4Ita 5 =
up-to this threshold, display all of its on-hand inventory
+ for any Ita ¶ Ka and t = T 1T −1100011.
to customers, and charge a customer-accessible inventory-
(c) If inf Ita <Ka 0 4Ita 5 ¾ −M, for some M < +, there
dependent list-price p4dt 4Ita 55. If the total inventory level
exists an r∗ < +, such that, if rw ¾ r∗ , x̃ta 4Ita 5 ¾ Ita , for
is higher than the disposal threshold, i.e., It > x̃t 4Ita 5,
any Ita ¶ Ka and t = T 1T −1100011. On the other hand, if
the firm should dispose-down-to this threshold, display
inf Ita <Ka 0 4Ita 5 = −, for any rw > 0, there exists a thresh-
part of its on-hand inventory, x̃ta 4Ita 5, to customers, and
old It∗ 4rw 5 < Ka , such that, if Ita ¾ It∗ 4rw 5, x̃ta 4Ita 5 < Ita , for
charge a customer-accessible inventory-dependent list-price
p4d˜t 4Ita 55. If the total inventory level is between the above any t = T 1T −1100012.
two thresholds, i.e., It ∈ 6xta 4Ita 51 x̃t 4Ita 57, the firm should (d) Let < 1, and D̄ 2= sup8ã2 P 4Dt ¾ ã5 ¾ 9, i.e., the
keep its total net inventory and display part of it to cus- probability that the demand in period t exceeds D̄ is
tomers. In particular, Theorem 1(b) implies that if it is smaller than , regardless of the policy the firm employs. If
optimal to order, the firm should not withhold anything.
Order-and-withhold policy is dominated by displaying the 4p −b −4c +rd 5+m541−5 0 4−D̄5
same amount of inventory to customers but not order- +4rd +rw +5 ¶ 01 (10)
ing the inventory that will be withheld (so no inventory
will be withheld). This is intuitive, because the marginal then xta∗ 4Ita 1It 5 ¶ 0 for any Ita ¶ Ka , It , and t = T 1T −11
cost of order-and-withhold is at least c +hw (procure- 00011.
ment cost and holding cost in the warehouse), while the
marginal benefit of inventory withholding is at most c Theorem 2(a) shows that, when the warehouse holding
(saving from the purchasing cost in the next period). More- cost is sufficiently high (hw ¾ c −s), the firm should dis-
over, part (e) of Theorem 1 demonstrates that as the play all of its on-hand inventory to customers. Part (b)
excess customer-accessible inventory level increases, lower demonstrates that, when inventory disposal is sufficiently
demand is induced and the firm has a greater incentive to costly (s ¶ s∗ ), the firm would rather not dispose any of its
turn it over, both of which give rise to lower optimal order- inventory, regardless of its total inventory level. When the
up-to levels and optimal sales prices. condition in part (a) [part (b)] holds, the unified model is
The firm’s excess inventory generally has three impacts reduced to the model without inventory withholding [inven-
on the performance of the system: (1) satisfying future tory disposal]. This generates additional insights and is
demand, (2) incurring holding costs, and (3) induc- thoroughly discussed in §5.1 [§5.2]. Theorem 2(c) reveals
ing/suppressing potential demand, the first with positive that the optimal inventory reallocation balances the trade-
marginal value and the other two with negative marginal off between saving the current reallocation cost and stim-
values. Hence, after normalizing the first effect (Vt 4Ita 1It 5− ulating future demand. More specifically, if the intensity
rd Ita −cIt ), the value-to-go function of the firm is decreas- of scarcity effect is bounded, the firm should not real-
ing in its customer-accessible inventory level and total locate its inventory from the customer-accessible storage
inventory level. To better address the intertwined trade-off to the warehouse, as long as the reallocation fee is suffi-
between these three effects, the firm can adopt dynamic ciently high. Otherwise (i.e., the intensity of scarcity effect
pricing, inventory withholding, and inventory disposal is unbounded), the firm should always withhold part of
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
1086 Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS
its inventory in the warehouse, if the excess customer- Because no inventory is withheld in the warehouse in this
accessible inventory level is high enough. model, Ita = It , there is no need to record the total inventory
Theorem 2(d) shows that when the demand-stimulating level It . Therefore, the state space dimension is reduced to
effect/scarcity effect of inventory is sufficiently strong one. Similarly, we will not incur the warehouse inventory
(characterized by (10)), the backlogging cost incurred by holding cost (hw ), the redisplay cost (rd ), and the withhold-
the wait list is dominated by the revenue generated by the ing cost (rw ) in this model. The superscript “s” refers to
scarcity effect. Therefore, the firm should not display any “single location storage.”
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
positive inventory, and every customer is placed on a wait Without loss of generality, we assume that the excess
list before receiving the product. This analytical result justi- inventory in the last period (period 1) is discarded without
fies the marketing strategy adopted by, e.g., BMW, in which any salvage value, i.e., V0s 4I0a 5 = 0, for any I0a ¶ Ka . The
the availability of the Mini Cooper is intentionally limited value functions satisfy the following recursive scheme:
and more customers are attracted by its wait list.
Vts 4Ita 5 = cIta + max Jts 4xta 1dt 1Ita 51
4xta 1dt 5∈F s 4Ita 5
5.1. Without Inventory Withholding +Ɛ6Gst 4xta −4p4dt 51Ita 1
t 5571
In some circumstances, the firm cannot store its inven- where Rs 4dt 1Ita 5 2 = 4p4dt 5−b −c54dt +4Ita 551
tory in the warehouse. Such storage may be too costly Gst 4y5 2 = −4b +ha 5y + +6Vt−1
s
4y5−cy71
or transportation too inconvenient. For instance, car deal-
ers usually display all of their automobiles because with- s 2 = b −41−5c1 (11)
holding and redisplaying the inventory is too costly and
and is defined in (6). Note that, under Assumption 3,
inconvenient. In this subsection, we confine our analysis
Rs 4dt 1Ita 5 = R4dt 1Ita 5+rd 4dt +4Ita 55 is jointly concave
to the model without inventory withholding. In this model,
on its domain.
because no inventory is stored in the warehouse, the state
As a corollary to Theorem 1, the optimal policy in
space dimension is reduced to one. This reduction offers
the model without inventory withholding is an inventory-
new results and sharper insights into how the inventory-
dependent order-up-to/dispose-down-to list-price policy, as
dependent demand influences the firm’s optimal decisions.
shown below:
More specifically, we demonstrate that the scarcity effect of
inventory increases the overstocking risk and, thus, drives Theorem 3. Consider a model without inventory withhold-
the firm to set a lower order-up-to level and charge a lower ing. For each t = T 1T −1100011, the following statements
sales price. On the other hand, when the firm has higher hold:
disposal flexibility (i.e., a higher salvage value), it can more (a) gts 4xta 1dt 1Ita 5 2= Ɛ6Gst 4xta −4p4dt 51Ita 1
t 557 is jointly
easily mitigate such overage risk by disposing of its sur- concave and continuously differentiable in 4xta 1dt 1Ita 5 if
plus inventory. We show that the firm with a higher salvage xta 6= Ita ; for any fixed Ita , gts 4·1 ·1Ita 5 is strictly concave.
value sets higher order-up-to levels and sales prices. (b) Vts 4Ita 5 is concave in Ita . Vts 4Ita 5−cIta is decreasing
To formulate the planning problem as a dynamic pro- and continuously differentiable in Ita .
gram, let: (c) Jts 4·1·1Ita 5 is strictly concave for any fixed Ita , and
there exists a unique 4xts∗ 4Ita 51dts∗ 4Ita 55 such that
Vts 4Ita 5 = the maximum expected discounted profits in peri-
ods t1t −1100011, when starting period t with a 4xts∗ 4Ita 51dts∗ 4Ita 55 = argmax Jts 4xta 1dt 1Ita 50
customer-accessible inventory level Ita . 4xt 1dt 5∈F s 4Ita 5
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS 1087
(d) Let qts∗ 4Ita 5 = xts∗ 4Ita 5−Ita denote the optimal order/ (b) Assume that 0 4Ita 5 ¶ ˆ 0 4Ita 5 for all Ita ¶ Ka and that
disposal quantity. There exist two threshold inventory levels limIta →− 4Ita 5 = limIta →− 4I
ˆ ta 5 = 0 . Let p = p̄ = p0 and
ItH and ItL (ItL < ItH ), such that, d0 = d4p0 5. We have ItL ¶ IˆtL , ItH ¶ IˆtH and xts∗ 4Ita 5 ¶ x̂ts∗ 4Ita 5
a L for all Ita ¶ Ka .
> 0 if It < It 1
qts∗ 4Ita 5 = 0 if ItL ¶ Ita ¶ ItH 1 As a generalization of Theorem 3.2 in Sapra et al. (2010)
to the model with dynamic pricing and inventory disposal,
< 0 otherwise,
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
Theorem 4 shows that the firm should understock and
i.e., the firm should order if its inventory level Ita is less underprice the product under the scarcity effect of inven-
than the lower threshold ItL , dispose if it is more than tory. In Theorem 4, we need the additive demand assump-
the higher threshold ItH , and not order or dispose if it is tion, i.e.,
tm = 1 almost surely. The additive demand model
between the two thresholds. is widely applied in the joint pricing and inventory control
(e) If Ita < ItL or Ita > ItH , the optimal order-up-to/ literature (see, e.g., Li and Zheng 2006, Feng 2010 and
dispose-down-to level xts∗ 4Ita 5 is decreasing in Ita . If ItL ¶ Pang et al. 2012), primarily because it enhances techni-
Ita ¶ ItH , the optimal inventory after replenishment/disposal cal tractability and facilitates analysis. To show Theorem 4
is increasing in Ita . and other comparisons between the optimizers in different
(f) The optimal price-induced-demand dts∗ 4Ita 5 is models (Theorems 5–7 below), we need to iteratively estab-
increasing in Ita . lish comparisons between the derivatives of value functions.
Theorem 3 implies that, when the firm cannot withhold The additive demand form is necessary to link the mono-
its on-hand inventory, the optimal policy is to order when tonicity relationship between optimizers and that between
the customer-accessible inventory level is low (below ItL ), derivatives. All results in this paper, except Theorems 4–7,
to dispose of it when it is high (above ItH ), and to make hold for the more general demand form introduced in (1).
no adjustments when it is between the two thresholds. The Efficiently disposing of surplus inventory protects the
optimal order-up-to/dispose-down-to and list-price levels firm from the demand-suppressing effect of inventory. As
are customer-accessible-inventory-dependent. As shown in the salvage value increases, the cost of inventory disposal
Theorem 3, when the customer-accessible inventory level is decreases, and the firm has greater disposal flexibility. We
higher, order-up-to/dispose-down-to levels and sales prices characterize the impact of the salvage value on the optimal
are lower because a high customer-accessible inventory pricing and inventory decisions in the following theorem:
level suppresses potential demand and the firm has a strong Theorem 5. Consider a model without inventory withhold-
incentive to turn it over. ing. For any t = T 1T −1100011, assume that the demand is
We now analyze how the scarcity effect of inventory of additive form (i.e.,
tm = 1 with probability 1), and s < ŝ.
impacts optimal pricing and inventory policies. Compared (a) ¡Ita V̂ts 4Ita 5 ¾ ¡Ita Vts 4Ita 5.
with the model in which demand is independent of inven- (b) IˆtL ¾ ItL .
tory, when potential demand is negatively correlated with (c) x̂ts∗ 4Ita 5 ¾ xts∗ 4Ita 5 and, hence, q̂ts∗ 4Ita 5 ¾ qts∗ 4Ita 5 for
customer-accessible leftover inventory levels, the marginal all Ita ¶ IˆtH .
value of on-hand inventory decreases and the firm suffers (d) dˆts∗ 4Ita 5 ¶ dts∗ 4Ita 5.
from the demand reduction caused by a high inventory
level. As a result, the firm should order less/dispose of Theorem 5(a) shows that the marginal value of on-
more to mitigate the additional overstocking risk caused by hand inventory increases in the salvage value. Parts (b)–(d)
the scarcity effect of inventory. At the same time, to bet- demonstrate that with a higher salvage value, the firm should
ter catch the sales opportunity, it is optimal to underprice set higher ordering thresholds, order-up-to levels, and sales
the product so as to attract more customers. Moreover, in a prices. Recall from Theorem 4 that the inventory-dependent
market where the firm has little power to set the sales price, demand strengthens overstocking risk by suppressing poten-
we show a sharper result: With a more intensive scarcity tial demand so that optimal order-up-to/disposal-down-to
effect, the firm should keep a lower inventory level after levels and optimal sales prices are lower in the model with
replenishment/disposal. The following theorem formalizes inventory-dependent demand than those in the model with
these intuitions. inventory-independent demand. On the other hand, The-
orem 5 demonstrates that increased operational flexibility
Theorem 4. Consider a model without inventory withhold-
ing. Assume that Dt = 4dt 1Ita 1
t 5 and D̂t = 4d ˆ t 1I a 1
t 5 with (i.e., a higher salvage value) mitigates the demand loss
t
inventory dependent term 4It 5 and 4I a a
ˆ t 5, respectively. We driven by a high customer-accessible inventory level. Hence,
also assume that the demand is of additive form (i.e.,
tm = 1 with higher disposal flexibility, the firm can set higher order-
with probability 1). The following statements hold: up-to levels and sales prices to achieve higher profits.
(a) Assume that 4I ˆ ta 5 = 0 = limx→− 4x5 for all Ita ¶
Ka , i.e., D̂t does not depend on the customer-accessible 5.2. Without Inventory Disposal
inventory level. We show that ItL ¶ IˆtL , ItH ¶ IˆtH , xts∗ 4Ita 5 ¶ The model without inventory disposal applies to the cases
x̂ts∗ 4Ita 5 and dts∗ 4Ita 5 ¾ dˆts∗ 4Ita 5 for all Ita ¶ Ka . wherein the inventory is too expensive or too inconvenient
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
1088 Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS
to dispose of. For example, in the automobile industry, 6. Responsive Inventory Reallocation
it is too costly to dispose of unsold cars of the last In our previous analysis, we assume that the firm can with-
year model. In other industries, such as chemical engi- hold and redisplay inventory only at the beginning of the
neering, products are often so environmentally unfriendly decision epoch before the demand is realized. We now relax
that they cannot be disposed of arbitrarily. The model this assumption by allowing the firm to responsively real-
without inventory disposal has a simpler optimal policy locate its on-hand inventory after the demand realization.
structure (customer-accessible-inventory-dependent order-
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
Theorem 7. Consider a model without inventory disposal. −rw 4xta −Ita +Dt 5− −bxta− −ha xta+
For any t = T 1T −1100011, assume that the demand is
−hw 4xt −xta −Dt 5+Vt−1
r
4xta 1xt −Dt 5
of additive form (i.e.,
tm = 1 with probability 1), rd =
rw = 0 (i.e., reallocation is cost free). If It = Ita , we
Hence, the optimal value functions satisfy the following
have xta 4Ita 5 ¾ xts∗ 4Ita 5 for Ita ¶ max8Ita 2 xta 4Ita 5 ¾ Ita 9, and
recursive scheme:
dt∗ 4Ita 1It 5 ¶ dts∗ 4Ita 5 for Ita ¶ Ka .
Note that Theorem 6 needs the assumption that inven- Vtr 4Ita 1It 5
tory reallocation is cost free (rd = rw = 0); this assumption
is necessary to reduce the state space dimension in its = maxr p4dt 5Ɛ84p4dt 51Ita 1
t 59
4xt 1dt 5∈F 4Ita 5
proof. We also assume rd = rw = 0 for Theorem 7, mainly
for expositional convenience. The results still hold under −c4xt −It 5+ +s4xt −It 5−
the general condition that rd 1rw ¾ 0.
To summarize, inventory withholding and inventory dis- +ƐDt a
max
a
8−rd 4xta −Ita +Dt 5+
min801It −Dt 9¶xt ¶min8Ka 1xt −Dt 9
posal have similar strategic implications in addressing
inventory-dependent demand. The firm uses these strategies −rw 4xta −Ita +Dt 5− −bxta− −ha xta+
to hedge against the overage risk caused by the scarcity
effect of inventory and stimulate more potential demand. −hw 4xt −xta −Dt 5+Vt−1
r
4xta 1xt −Dt 59 1
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS 1089
¯ Fol-
where F r 4Ita 5 2= 84xt 1dt 5 2 xt ¾ min8Ita 1091dt ∈ 6d1 d79. order if the total inventory level is low (It < xtr 4Ita 5), to dis-
lowing the algebraic manipulation similar to that in Equa- pose of inventory if it is high (It > x̃tr 4Ita 5), and to maintain
tion (5), we have: the starting inventory level otherwise. Compared with The-
orem 1, which characterizes optimal policy in the unified
Vtr 4Ita 1It 5 model, Theorem 8 demonstrates that it is possible that the
firm orders and withholds some inventory under the opti-
= rd Ita +cIt + maxr R4dt 1Ita 5+rd 4dt +4Ita 55 mal responsive inventory reallocation policy, because, in
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
Comparing the value functions (12) and (4), it is clear 7. Numerical Studies
that by postponing the reallocation decision until after This section reports a set of numerical studies that (a) verify
demand realization, the firm achieves a higher expected the robustness of our analytical results when Assumption 3
total profit. In the following theorem, we characterize the does not hold; (b) quantify the profit loss of ignoring the
optimal inventory replenishment/disposal/reallocation and scarcity effect of inventory when making the pricing and
pricing policy in the model with responsive inventory inventory decisions; and (c) quantitatively evaluate the ben-
reallocation: efit of dynamic pricing in the presence of the scarcity effect.
Theorem 8. The following statements hold for t = T 1T −11 Our numerical results demonstrate that (1) the structural
00011: results developed in our theoretical model are robust and
(a) Vtr 4Ita 1It 5 is jointly concave and continuously differ- hold for a large set of nonconcave R4·1 ·5 functions; (2) the
entiable in (Ita 1It ), whereas the normalized value function impact of the scarcity effect is significant and is higher
Vtr 4Ita 1It 5−rd Ita −cIt is decreasing in Ita and It . when the scarcity intensity, demand variability, and/or plan-
(b) For any given xt and realized Dt , vtr 4yta Ita 1xt 1Dt 5 2= ning horizon length increase; and (3) the value of dynamic
−4rd +rw 54yta −Ita 5− +yta +Grt 4yta −Dt 1xt −Dt 5 is con- pricing under the scarcity effect is significant and higher
cave in yta . Therefore, the optimal customer-accessible- under higher scarcity intensity, demand variability, and/or
inventory level is: shorter planning horizon.
Throughout our numerical studies, we assume that the
xtra∗ 4Ita 1xt 1Dt 5 firm can neither withhold nor dispose of its on-hand inven-
tory for two reasons: (a) to have a clear illustration of the
= argmax 8vtr 4yta Ita 1xt 1Dt 59−Dt 0 optimal policy structure in a model where Assumption 3
min8Dt 1Ita 9¶yta ¶min8xt 1Ka +Dt 9
does not hold; and (b) to single out and highlight the impact
(c) There exist two customer-accessible inventory-level- of the focal operational elements (i.e., the scarcity effect
dependent thresholds, xtr 4Ita 5 and x̃tr 4Ita 5 (xtr 4Ita 5 ¶ x̃tr 4Ita 5), of inventory and the dynamic pricing strategy). We also
such that it is optimal to order up to xtr 4Ita 5 if and only assume that the demand in each period is of the additive
if It < xtr 4Ita 5, to dispose down to x̃tr 4Ita 5, if and only if form, i.e.,
tm = 1 almost surely and Dt = dt +4Ita 5+
ta .
It > x̃tr 4Ita 5, and to keep the total inventory level otherwise. Let 8
ta 9Tt=1 follow i0i0d0 normal distributions with mean 0
Moreover, there exist two customer-accessible-inventory- and standard deviation . The inverse demand function is
level-dependent sales prices, p4dtr 4Ita 55 and p4d˜tr 4Ita 55, linear with slope −1, i.e., p4dt 5 = p0 −dt . We set the dis-
such that it is optimal to charge a sales price p4dtr 4Ita 55 count factor = 0095, the unit holding cost h = 1, and the
if It ¶ xtr 4Ita 5, and to charge a sales price p4d˜tr 4Ita 55 if It ¾ unit backlogging cost b = 10.
x̃tr 4Ita 5.
Theorem 8(a) proves the joint concavity and continu- 7.1. Optimal Policy Structure with Nonconcave
ous differentiability of the optimal value functions. Part (b) R4·1 ·5 Functions
shows that, in each period, the optimal reallocation policy In this subsection, we numerically examine whether the
is obtained by solving a one-dimensional convex optimiza- structural results in our theoretical model are robust when
tion after the demand is realized. Consistent with Theo- Assumption 3 does not hold, i.e., R4·1 ·5 is not jointly con-
rem 1, part (c) of Theorem 8 proves that it is optimal to cave. We have performed extensive numerical experiments
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
1090 Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS
to test the robustness of our analytical results. In all our Figure 2. (Color online) Optimal price-induced demand.
numerical experiments, although Assumption 3 is violated,
12
the characterizations of the optimal policy by our theoreti- With scarcity effect
cal analysis (i.e., Theorem 1, Theorem 3, and Theorem 4) Without scarcity effect
10
ing in the starting inventory level; (b) the optimal sales
price [price-induced demand] is decreasing [increasing] in
9
the starting inventory level; and (c) compared to an inven-
tory system without the scarcity effect, the firm with the
scarcity effect sets lower order-up-to levels and lower sales 8
16 V∗
Our numerical experiments are conducted under the follow-
15
ing values of parameters: p0 = 21, 0 = 4, c = 4, = 00351
00410045100510055, = 11213, 6d1 d7 ¯ = 661127, Ka = 18, and
14
t = 5110.
13
Figures 3–4 summarize the results of our numerical study
on the impact of the scarcity effect on the firm’s prof-
12
itability. Our results show that, when the scarcity effect
With scarcity effect
Without scarcity effect is ignored, all numerical experiments exhibit a significant
11 profit loss, which is at least 16041% and can be as high
–10 –5 0 5 10 15 20 as 64052%. Moreover, the impact of the scarcity effect is
Customer accessible inventory level increasing in the scarcity intensity, demand variability, and
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS 1091
Figure 3. (Color online) Value of scarcity 2 t = 5. et al. (2010), which shows that the profit loss of ignor-
ing the scarcity effect is decreasing in demand variability.
0.7
In their experiments, the potential demand is convexly
=1
=2 decreasing in the leftover inventory level, so higher demand
0.6 =3 variability increases the expected potential demand and,
thus, the firm’s profitability under the scarcity effect.
0.5
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
Optimality loss
=1 0.10
=1
0.6 =2 =2
=3 0.09 =3
Value of dynamic pricing
0.08
0.5
Optimality loss
0.07
0.06
0.4
0.05
0.3 0.04
0.03
0.2 0.02
0.01
0.1 0
0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.30 0.35 0.40 0.45 0.50 0.55 0.60
value value
Yang and Zhang: Pricing and Inventory Management Under Inventory-Dependent Demand
1092 Operations Research 62(5), pp. 1077–1094, © 2014 INFORMS
0.08
customer-accessible inventory level. We propose a unified
0.07
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.
Gallego G, van Ryzin G (1994) Optimal dynamic pricing of invento- Verhallen TM, Robben HS (1994) Scarcity and preference: An experiment
ries with stochastic demands over finite horizons. Management Sci. on unavailibility and product evalation. J. Econom. Psych. 15(2):
40(8):999–1020. 215–331.
Gerchak Y, Wang Y (1994) Periodic-review inventory models with Wall Street Journal (1989) Nintendo goes portable, stores go gaga (Octo-
inventory-level-dependent demand. Naval Res. Logist. 41(1):99–116. ber 4), B1.
Li Q, Zheng S (2006) Joint inventory replenish and pricing control for Wall Street Journal (1999) Beating the wait list for the hottest wheels
systems with uncertain yield and demand. Oper. Res. 54(4):696–705. (June 4), W1.
Liu Q, van Ryzin GJ (2008) Strategic capacity rationing to induce early
Downloaded from informs.org by [128.252.111.87] on 21 February 2015, at 08:53 . For personal use only, all rights reserved.