European Journal of Operational Research 125 (2000) 588±601

www.elsevier.com/locate/dsw

Theory and Methodology

The e€ects of in¯ation and time-value of money on an economic

order quantity model with a random product life cycle
Ilkyeong Moon *, Suyeon Lee
Department of Industrial Engineering, Pusan National University, Pusan 609-735, South Korea
Received 1 November 1997; accepted 1 February 1999

Abstract

For several decades, the Economic Order Quantity (EOQ) model and its variations have received much attention
from researchers. Recently, there has been an investigation into an EOQ model incorporating a random product life
cycle and the concept of time-value of money. This paper extends the previous research in several areas. First, we
investigate the impact of in¯ation on the choice of replenishment quantities. Second, the unit cost, which has been
inadvertently omitted in the previous research, is included in the objective function to properly model the problem.
Third, we consider the normal distribution as a product life cycle in addition to the exponential distribution. Fourth, we
develop a simulation model which can be used for any probability distribution. Ó 2000 Elsevier Science B.V. All rights
reserved.

Keywords: Inventory; In¯ation; Time-value of money

1. Introduction

The Economic Order Quantity (EOQ) model initiated by Harris [13] has been extended many ways to
improve the practicality of the model. See, for example, [3,4], etc. One of the important extensions is to
properly recognize the time-value of money in determining the optimal order quantity. Trippi and Lewin
[21] have adopted the Discounted Cash-Flows (DCF) approach for the analysis of the basic EOQ model.
Kim et al. [14] extend Trippi and Lewin's work by applying the DCF approach to various inventory
systems. Chung [5] has studied the DCF approach for the analysis of the basic EOQ model in the presence
of the trade credit. Recently, Haneveld and Teunter [11] apply the DCF approach to the basic EOQ model
with a Poisson demand process.

*
Corresponding author. Tel.: +82-51-510-2451; fax: +82-51-512-7603.
E-mail address: [email protected] (I. Moon).

0377-2217/00/$ - see front matter Ó 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 7 - 2 2 1 7 ( 9 9 ) 0 0 2 7 0 - 2
 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601 589

Gurnani [9] has applied the DCF approach to the ®nite planning horizon EOQ model in which the
planning horizon is a given constant. Gurnani [10] has claimed that an in®nite planning horizon does not
exist in real life, and a ®nite horizon inventory model (accounting for the time-value of money) is theo-
retically superior and of greater practical utility. Chung and Kim [7] prove that Gurnani's [9] model is
essentially identical to an in®nite planning horizon model since the planning horizon is assumed to be a
given constant. As pointed out by Gurnani [9], an in®nite planning horizon occurs rarely because costs are
likely to vary disproportionately and product speci®cations and design are prone to change, abandonment
or substitution by another product due to rapid technological development. This phenomenon can be
observed frequently in high-technology product markets. Chung and Kim [7] have also suggested that the
assumption of an in®nite planning horizon is not realistic, and called for a new model which relaxes this
assumption. Moon and Yun [16] answered this request by developing a ®nite planning horizon EOQ model
where the planning horizon is a random variable following an exponential distribution. The unit cost,
which does not a€ect the replenishment quantity (or cycle length) in the in®nite planning horizon model,
was not included in their model. However, the unit cost does a€ect the replenishment quantity in the ®nite
planning horizon model, as will be demonstrated later.
One of the assumptions in most derivations of the inventory model has been a negligible level of in-
¯ation. Unfortunately, many countries have recently been confronted with ¯uctuating in¯ation rates that
often have been far from negligible [20]. Silver et al. [20] investigate the impact of in¯ation on the choice of
replenishment quantities in the basic EOQ model. There have been several studies which consider various
in¯ationary situations. Bose et al. [2] have investigated an EOQ model for deteriorating items with linear
time-dependent demand rate and shortages under in¯ation. Hariga and Ben-Daya [12] have developed
time-varying lot-sizing models with linear trend in demand, taking into account the e€ects of in¯ation and
time value of money.
The objectives of this study are fourfold. Firstly, we include the unit cost, which has been inadvertently
omitted in [16], in the model to properly derive the optimal order quantity. Secondly, the impact of in¯ation
on the choice of order quantities is investigated. Thirdly, we consider the normal distribution as a product
life cycle in addition to the exponential distribution. Fourthly, a simulation model is developed to be used
for any probability distribution.
In Section 2 we introduce the notation and mathematically formulate the problem. Analytical results for
both exponential and normal distributions are developed. In Section 3, we perform some computational
results to show the cost savings compared with the solution of Moon and Yun [16]. A simulation model
which can be used for any probability distribution is presented in Section 4 to complement the analytical
work. The paper concludes with Section 5 which includes some possible research problems.

2. Analytical modeling

The following notation will be used:

Q the order quantity

T the cycle length
p the product life cycle (random variable)
f …p† the probability density function of p
D the demand rate per year
S the ordering cost per order
c the unit cost
h the inventory carrying cost per unit per year ( ic where i is the inventory carrying charge rate)
590 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601

a the discount rate representing the time-value of money

f the in¯ation rate

We assume that a > f , that is the interest rate is larger than the in¯ation rate, which is a practical as-
sumption. As pointed out by Seo and Kim [18], the true optimal order quantities may vary over the cycle
except for the exponential distribution case which possesses a memoryless property. However, we assume
that all future replenishments will be of the same size as the current replenishment to obtain a tractable
analytical result. As claimed by Silver et al. [20], this assumption is not as serious as it would ®rst appear. In
particular, at the time of the next replenishment we would recompute a new value for that replenishment,
re¯ecting any changes in cost (and other) parameters that have taken place in the interim. In addition to
this, we employ discounting so that assumptions about future lot sizes should not have an appreciable e€ect
on the current decision.
Present value of cash ¯ows for the ®rst cycle: The present value of cash ¯ows for the ®rst cycle is
Z T
S ‡ cQ ‡ h …Q ÿ tD† eÿat eft dt: …1†
0

Eq. (1) can be simpli®ed as

 
hD 1 ÿ…aÿf †T
S ‡ cDT ‡ T‡ …e ÿ 1† : …2†
…a ÿ f † …a ÿ f †

Present value of cash ¯ows up to the beginning of the last cycle: If we assume that the planning horizon
p fully accommodates ®rst k cycles, and ends during …k ‡ 1†th cycle, then the present value of the total
ordering costs and inventory carrying costs up to the beginning of …k ‡ 1†th cycle is

X   kÿ1
k
hD eÿ…aÿf †T ÿ 1 X
…S ‡ cDT † eÿiaT eifT ‡ T‡ eÿiaT eifT : …3†
iˆ0
…a ÿ f † …a ÿ f † iˆ0

Eq. (3) can be simpli®ed as

    
1 ÿ eÿ…aÿf †…k‡1†T hD eÿ…aÿf †T ÿ 1 1 ÿ eÿ…aÿf †kT
…S ‡ cDT † ‡ T‡ : …4†
1 ÿ eÿ…aÿf †T …a ÿ f † …a ÿ f † 1 ÿ eÿ…aÿf †T

Present value of inventory carrying cost during the last cycle: The present value of inventory carrying cost
during the last cycle, i.e. …k ‡ 1†th cycle can be obtained as follows:
Z pÿkT
h eÿ…aÿf †kT …Q ÿ tD†eÿat eft dt: …5†
0

Eq. (5) can be simpli®ed as

   
hD ÿ…aÿf †kT 1 hD ÿ…aÿf †p 1
e Tÿ ‡ e ÿ …k ‡ 1†T ‡ p : …6†
aÿf …a ÿ f † aÿf …a ÿ f †

Expected present value of total inventory carrying and ordering costs: Since the planning horizon p has a
p.d.f. f …p†, the expected present value of total inventory carrying and ordering costs, say C…T †, is
 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601 591

1 Z
X …k‡1†T
C…T † ˆ ‰…4† ‡ …6†Šf …p† dp
kˆ0 kT

Z
S ‡ cDT X 1
 ÿ…aÿf †…k‡1†T
 …k‡1†T
ˆ 1 ÿ e f …p† dp
1 ÿ eÿ…aÿf †T kˆ0 kT

  1 Z
hD eÿ…aÿf †T ÿ 1 X  ÿ…aÿf †kT
 …k‡1†T
‡ T‡ 1ÿe f …p† dp
…a ÿ f †…1 ÿ eÿ…aÿf †T † aÿf kˆ0 kT

 X
1 Z …k‡1†T
hD 1
‡ Tÿ eÿ…aÿf †kT f …p† dp
aÿf aÿf kˆ0 kT

1 Z …k‡1†T  
hD X 1
‡ ÿ …k ‡ 1†T ‡ p eÿ…aÿf †p f …p† dp: …7†
a ÿ f kˆ0 kT aÿf

2.1. Exponential distribution case

Masters [15] develops a basic EOQ model in which the product life cycle follows an exponential dis-
tribution. He shows that the traditional approach of in¯ating the inventory carrying charge rate is ap-
propriate for items subject to sudden obsolescence. Even though the time-value of money is not considered
in his model, we show that the approximation is quite good in the computational section later. If the
planning horizon p follows an exponential distribution with parameter k, we can obtain an exact optimal
cycle length using the following propositions.

Proposition 1.
 
S ‡ cDT hD eÿ…aÿf ‡k†T ‡ …a ÿ f †T ÿ 1 hkD‰2…a ÿ f † ‡ kŠ
C…T † ˆ ‡ 2
‡ 2 2
1ÿe ÿ…aÿf ‡k†T
…a ÿ f † ‰1 ÿ e ÿ…aÿf ‡k†T Š …a ÿ f † …a ÿ f ‡ k†
hkDT
ÿ : …8†
…a ÿ f †…a ÿ f ‡ k†‰1 ÿ eÿ…aÿf ‡k†T Š

Proof. See Appendix A.

Proposition 2.
(a) C…T † has a unique local minimum on ‰0; 1†.
(b) The optimal cycle length, T  , satis®es

2 
…a ÿ f ‡ k† eÿ…aÿf ‡k†T cD
ˆ :
i‰1 ÿ eÿ…aÿf ‡k†T  ‰1 ‡ T …a ÿ f ‡ k†ŠŠ ‡ ‰1 ÿ eÿ…aÿf ‡k†T  Š…a ÿ f ‡ k† ÿ T  …a ÿ f ‡ k†2 eÿ…aÿf ‡k†T  S
…9†
592 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601

Proof. (a) The ®rst derivative of C…T †, say C 0 …T †, is as follows:

 
0 hD 1 ÿ eÿ…aÿf ‡k†T ‰1 ‡ T …a ÿ f ‡ k†Š
C …T † ˆ 2
…a ÿ f ‡ k†‰1 ÿ eÿ…aÿf ‡k†T Š
 
…S ‡ cDT †…a ÿ f ‡ k†2 eÿ…aÿf ‡k†T ÿ cD 1 ÿ eÿ…aÿf ‡k†T …a ÿ f ‡ k†
ÿ :
…a ÿ f ‡ k†‰1 ÿ eÿ…aÿf ‡k†T Š2

C 0 …T † has the same sign as the numerator of C 0 …T †, say f …T †:

  2
f …T † ˆ hD 1 ÿ eÿ…aÿf ‡k†T ‰1 ‡ T …a ÿ f ‡ k†Š ÿ …S ‡ cDT †…a ÿ f ‡ k† eÿ…aÿf ‡k†T
 
‡ cD…a ÿ f ‡ k† 1 ÿ eÿ…aÿf ‡k†T :

f …T † is a strictly increasing function since

f 0 …T † ˆ hDT …a ÿ f ‡ k†2 eÿ…aÿf ‡k†T ‡ …S ‡ cDT †…a ÿ f ‡ k†3 eÿ…aÿf ‡k†T > 0:

Since f …0† ˆ ÿS…a ÿ f ‡ k†2 < 0, and f …1† ˆ hD ‡ cD…a ÿ f ‡ k† > 0, there exists a unique local mini-
mum on ‰0; 1†.
(b) From C…0† ˆ C…1† ˆ 1 and the result of (a), the optimum cycle length T  satis®es C 0 …T † ˆ 0. The
equation C 0 …T  † ˆ 0 can be rewritten as Eq. (9). 

Remark 1. It is known that the unit cost c does not a€ect the optimal order quantity (or optimal cycle
length) in the in®nite planning horizon EOQ model. However, it is clear that the unit cost c a€ects the
optimal order quantity (or optimal cycle length) em from Eq. (9). Consequently, we can obtain a better
solution than that of Moon and Yun [16]. We will illustrate this using numerical examples later.

Corollary 1. (a)
2
…a ÿ f ‡ k† eÿ…aÿf ‡k†T
I…T † ˆ
i‰1 ÿ eÿ…aÿf ‡k†T ‰1 ‡ T …a ÿ f ‡ k†ŠŠ ‡ ‰1 ÿ eÿ…aÿf ‡k†T Š…a ÿ f ‡ k† ÿ T …a ÿ f ‡ k†2 eÿ…aÿf ‡k†T

is strictly decreasing in T.
(b) The optimal cycle length increases (decreases) as ordering cost increases (decreases). It decreases
(increases) as unit cost and/or demand rate increases (decreases).

Proof. (a) Let the left-hand side of Eq. (9) as I…T †. It is clear that the numerator of I…T † strictly decreases as
T increases. The denominator of I…T † is strictly increasing in T since its ®rst derivative is …i ‡ a
2
ÿf ‡ k†…a ÿ f ‡ k† T eÿ…aÿf ‡k†T > 0 for all T > 0. Consequently, I…T † is a strictly decreasing function of T.
(b) The optimal cycle length is an intersection of I…T †, which is strictly decreasing, and cD=S. The results
follow directly from the behavior of the optimal cycle length as cD=S changes. 

From Proposition 2 and Corollary 1, the optimal cycle length can be easily obtained by the following
algorithm which requires only a line search. Then, an optimal lot size is computed using Q ˆ DT  .

Algorithm
p
Step 0. Start from T ˆ 2S=cD…i ‡ k ÿ f † (an optimal cycle length for the basic EOQ modi®ed for
in¯ation and with the carrying charge rate approximately adjusted for the risk of obsolescence).
Tlow ˆ 0; Thigh  T :
 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601 593

Step 1. Compute
2
…a ÿ f ‡ k† eÿ…aÿf ‡k†T
I…T † ˆ :
i‰1 ÿ eÿ…aÿf ‡k†T ‰1 ‡ T …a ÿ f ‡ k†ŠŠ ‡ ‰1 ÿ eÿ…aÿf ‡k†T Š…a ÿ f ‡ k† ÿ T …a ÿ f ‡ k†2 eÿ…aÿf ‡k†T

Step 2.
If I…T † > cD=S, increase T and go to Step 1 (e.g. Tlow ˆ T , T ˆ …Tlow ‡ Thigh †=2.)
If I…T † < cD=S, decrease T and go to Step 1 (e.g. Thigh ˆ T , T ˆ …Tlow ‡ Thigh †=2.)
If I…T † ˆ cD=S, stop. T  ˆ T and Q ˆ DT  .

Remark 2. Note that we can develop a more sophisticated search algorithm similar to that of Chung and
Lin [6]. However, it only takes less than a second to ®nd an optimal solution on an IBM Pentium II
personal computer with 266 MHz clock speed. Consequently, the above algorithm based on a simple line
search is sucient enough to be used in practice.

2.2. Normal distribution case

If the planning horizon p follows a normal distribution with mean l and variance r2 , C…T † can be
represented as follows:

Proposition 3.
1 
X     
1 ÿ eÿ…aÿf †…k‡1†T hD 1 1 ÿ eÿ…aÿf †kT
C…T † ˆ …S ‡ cTD† ‡ T‡ …eÿ…aÿf †T ÿ 1†
kˆ0
1 ÿ eÿ…aÿf †T aÿf aÿf 1 ÿ eÿ…aÿf †T

     
hD eÿ…aÿf †kT 1 …k ‡ 1†T ÿ l kT ÿ l
‡ Tÿ U ÿU
…a ÿ f † aÿf r r

     
hD …aÿf †2 r2 ÿ2…aÿf †l kT ÿ l ‡ …a ÿ f †r2 …k ‡ 1†T ÿ l ‡ …a ÿ f †r2
‡ e 2 r / ÿ/
aÿf r r

   
1 …k ‡ 1†T ÿ l ‡ …a ÿ f †r2
‡ ÿ …k ‡ 1†T ‡ l ÿ …a ÿ f †r2 U
aÿf r

 
kT ÿ l ‡ …a ÿ f †r2
ÿU ;
r

where / and U denote the probability density function and the cumulative distribution function of the standard
normal distribution, respectively.

Proof. First, we derive the following equations:

Z …k‡1†T    
…k ‡ 1†T ÿ l kT ÿ l
f …p† dp ˆ U ÿU ;
kT r r
594 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601

Z …k‡1†T   
2 2 …k ‡ 1†T ÿ l ‡ …a ÿ f †r2
eÿ…aÿf †p f …p† dp ˆ e‰…aÿf † r ÿ2…aÿf †lŠ=2
U
kT r
 2

kT ÿ l ‡ …a ÿ f †r
ÿU ;
r

Z …k‡1†T
p eÿ…aÿf †p f …p† dp
kT
     
kT ÿ l ‡ …a ÿ f †r2
‰…aÿf †2 r2 ÿ2…aÿf †lŠ=2 …k ‡ 1†T ÿ l ‡ …a ÿ f †r2
ˆe r / ÿ/
r r
    
ÿ 2

…k ‡ 1†T ÿ l ‡ …a ÿ f †r 2
kT ÿ l ‡ …a ÿ f †r2
‡ l ÿ …a ÿ f †r U ÿU :
r r

By substituting (4) and (6) into (7), and simplifying the expression using the above equations, we can obtain
C…T †. 

Since we cannot obtain an explicit expression for C…T †, we use an approximation scheme as follows: The
idea is to obtain an upper bound on k for a given T so that the summation in C…T † can be computed. If we
use
 
l ‡ 3:1r
…10†
T

where bxc denotes a largest integer which is less than or equal to x, then the probability that p is larger than
kT is less than 0.1%. Using this upper bound on k, we can compute C…T †. Thus, we can ®nd an approximate
optimal solution by varying T and compare C…T † values.

Remark 3. We may use another approximation scheme. The expressions in C…T † become smaller and
smaller for larger k beyond …k ‡ 1†T ˆ l. A bound on k can be determined on the residual of C…T †. For
example, we can sum up the expressions in C…T † until the present value of the kth cycle becomes less than,
say $D. The smaller the value of D we use, the greater the precision and the greater the computational time.
In the next section, we will compare the two approximation schemes.

3. Computational results

Let the cycle length determined from the basic EOQ model modi®ed for in¯ation be T D1 (see Chapter 5
in [20]).
s
2S
T D1 ˆ : …11†
cD…i ÿ f †

Masters [15] shows that the following reorder interval is optimal when we ignore the time-value of money
and the product life cycle is a random variable which follows an exponential distribution with parameter k:
s
2S
T ˆ : …12†
cD…i ‡ k†
 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601 595

If we modify (12) for in¯ationary situation, we obtain

s
D2 2S
T ˆ : …13†
cD…i ‡ k ÿ f †

Denote the cycle length determined from the algorithm of Moon and Yun [16] to be T 0 . It means that T 0 is
an optimal cycle length ignoring the e€ects of in¯ation and the unit cost. Let T  be the optimal cycle length
from our algorithm in which in¯ation and the unit cost have been properly considered. The cost penalties
associated with using T D1 (obtained from the basic EOQ model ignoring the time-value of money), T D2
(modi®ed from the formula in [15]), and T 0 (that ignores in¯ation and the unit cost) can be found from the
following equations, respectively:

C…T D1 † ÿ C…T  †
 100;
C…T  †

C…T D2 † ÿ C…T  †
 100;
C…T  †

C…T 0 † ÿ C…T  †
 100:
C…T  †

We solve several examples to see the cost savings of using the proper value of cycle length in the presence of
in¯ation.
Case 1 shows the basic example. If the carrying charge rate is increased from 0.3 to 0.45 while all
other parameters are ®xed, then an optimum cycle length is decreased to 0.0966 yr for the exponential
distribution (see Case 2 of Table 1) and 0.1127 for the normal distribution (see Case 2 of Table 2). If the

Table 1
Computational results for main examples (exponential distribution case)
Case 1a Case 2b Case 3c Case 4d Case 5e Case 6f
D1
Cycle T 0.2236 0.1690 0.3162 0.2582 0.1581 0.2236
length T D2 0.1195 0.1085 0.1690 0.1054 0.0845 0.0913
T0 0.1788 0.1465 0.2507 0.1788 0.1272 0.1761
T 0.1043 0.0966 0.1469 0.0905 0.0740 0.0832
Expected C…T D1 † $18,779 $18,689 $19,692 $27,997 $36,295 $10,801
total C…T D2 † $18,296 $18,420 $18,993 $26,881 $35,623 $10,205
present cost C…T 0 † $18,523 $18,562 $19,310 $27,311 $35,946 $10,532
C…T  † $18,281 $18,408 $18,670 $26,859 $35,603 $10,200
C…T D1 †ÿC…T  †
Cost saving C…T  †
2.72% 1.53% 3.81% 4.24% 1.94% 5.90%
C…T D2 †ÿC…T  †
C…T  †
0.08% 0.06% 0.12% 0.08% 0.06% 0.05%
C…T 0 †ÿC…T  †
C…T  †
1.32% 0.83% 1.79% 1.68% 0.96% 3.25%
a
Case 1. Computational results for D ˆ 1000 units=yr, S ˆ $50, c ˆ $10=unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 0:5/yr.
b
Case 2. Computational results for D ˆ 1000 units=yr, S ˆ $50, c ˆ $10=unit, i ˆ 0:45$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 0:5/yr.
c
Case 3. Computational results for D ˆ 1000 units=yr, S ˆ $100, c ˆ $10=unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 0:5/yr.
d
Case 4. Computational results for D ˆ 1000 units=yr, S ˆ $50, c ˆ $15=unit, i ˆ 0:2$=$/yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 0:5/yr.
e
Case 5. Computational results for D ˆ 2000 units=yr, S ˆ $50, c ˆ $10=unit, i ˆ 0:3$=$/yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 0:5/yr.
f
Case 6. Computational results for D ˆ 1000 units=yr, S ˆ $50, c ˆ $10=unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 1/yr.
596 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601

Table 2
Computational results for main examples (normal distribution case)
Case 1a Case 2b Case 3c Case 4d Case 5e
Cycle T D1 0.2236 0.1690 0.3162 0.2582 0.1581
length T0 0.1821 0.1480 0.2536 0.1821 0.1291
T 0.1291 0.1127 0.1821 0.1127 0.0899

Expected C…T D1 † $35,571 $35,690 $36,845 $52,778 $69,365

total C…T 0 † $35,312 $35,542 $36,444 $52,059 $69,013
present cost C…T  † $35,150 $35,447 $36,232 $51,743 $68,783
C…T D1 †ÿC…T  †
Cost saving C…T  †
1.20% 0.69% 1.69% 2.00% 0.85%
C…T 0 †ÿC…T  †
C…T  †
0.46% 0.27% 0.59% 0.61% 0.33%
a
Case 1. Computational results for D ˆ 1000 units/yr, S ˆ $50, c ˆ $10/unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, l ˆ 4 yr,
r2 ˆ 1 yr2 .
b
Case 2. Computational results for D ˆ 1000 units/yr, S ˆ $50, c ˆ $10/unit, i ˆ 0:45$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, l ˆ 4 yr,
r2 ˆ 1 yr2 .
c
Case 3. Computational results for D ˆ 1000 units/yr, S ˆ $100, c ˆ $10/unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, l ˆ 4 yr,
r2 ˆ 1 yr2 .
d
Case 4. Computational results for D ˆ 1000 units/yr, S ˆ $50, c ˆ $15/unit, i ˆ 0:2$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, l ˆ 4 yr,
r2 ˆ 1 yr2 .
e
Case 5. Computational results for D ˆ 2000 units/yr, S ˆ $50, c ˆ $10/unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, l ˆ 4 yr,
r2 ˆ 1 yr2 .

ordering cost is increased from $50 to $100 while all other parameters are ®xed, then an optimum cycle
length is increased to 0.1469 yr for the exponential distribution (see Case 3 of Table 1) and 0.1821 yr for
the normal distribution (see Case 3 of Table 2). If the unit cost is increased from $10 to $15 and the
carrying charge rate is reduced to 0.2 (in order to keep h ®xed) while all other parameters are ®xed, then
an optimum cycle length is decreased to 0.0905 yr for the exponential distribution (see Case 4 of Table 1)
and 0.1127 yr for the normal distribution (see Case 4 of Table 2). If the demand rate is increased from
1000 to 2000 units/yr while all other parameters are ®xed, then an optimum cycle length is decreased to
0.0740 yr for the exponential distribution (see Case 5 of Table 1) and 0.0899 yr for the normal distri-
bution (see Case 5 of Table 2). If the mean of life cycle is reduced to 1 yr while all other parameters are
®xed, then an optimum cycle length is decreased to 0.0832 yr for the exponential distribution (see Case 6
of Table 1). All the behaviors of the optimal cycle length conform to the result of Corollary 1. In ad-
dition, it has been shown that T D2 works extremely well for exponential distribution which justi®es the
work of Masters [15].
To compare the two approximation schemes in Remark 3, we solve Case 1 in Table 2 under the each
approximation scheme. Under the scheme in (10), T  ˆ 0:1291 with the objective value of $35,150. The
summation has been computed up to 50 cycles. Meanwhile, if we sum up until the present value of the kth
cycle becomes less than $1, we get T  ˆ 0:1281 with the objective value of $35,615. The summation has been
computed up to 62 cycles.
We solve supplementary examples to compare the optimal cycle lengths when we use the same mean
(that is, k ˆ 1 for the exponential distribution and l ˆ 1 for the normal distribution) for the two distri-
butions (see Tables 3 and 4). We conjecture that the optimal cycle length for exponential distribution is
shorter than that for normal distribution. This conjecture is based on the fact that the exponential dis-
tribution is skewed to left. Among four out of ®ve cases, the optimal cycle length for exponential distri-
bution is shorter than that of normal distribution. If we use the same variance in addition to the same mean,
we think that the optimal cycle length for exponential distribution will always be shorter than that for the
normal distribution.
 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601 597

Table 3
Computational results for supplementary examples (exponential distribution case)
Case 1a Case 2b Case 3c Case 4d Case 5e
Cycle T D1 0.2236 0.1690 0.3162 0.2582 0.1581
length T D2 0.0913 0.0861 0.1291 0.0778 0.0645
T0 0.1761 0.1448 0.2455 0.1761 0.1259
T 0.0832 0.0792 0.1170 0.0707 0.0591
Expected C…T D1 † $10,801 $10,613 $11,560 $16,235 $20,566
total C…T D2 † $10,205 $10,260 $10,687 $14,945 $19,743
present cost C…T 0 † $10,532 $10,478 $11,146 $15,532 $20,207
C…T  † $10,200 $10,256 $10,679 $14,940 $19,737
C…T D1 †ÿC…T  †
Cost saving C…T  †
5.90% 3.48% 8.25% 8.67% 4.20%
C…T D2 †ÿC…T  †
C…T  †
0.05% 0.04% 0.07% 0.04% 0.03%
C…T 0 †ÿC…T  †
C…T  †
3.25% 2.16% 4.37% 3.97% 2.39%
a
Case 1. Computational results for D ˆ 1000 units/yr, S ˆ $50, c ˆ $10/unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 1/yr.
b
Case 2. Computational results for D ˆ 1000 units/yr, S ˆ $50, c ˆ $10/unit, i ˆ 0:45$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 1/yr.
c
Case 3. Computational results for D ˆ 1000 units/yr, S ˆ $100, c ˆ $10/unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 1/yr.
d
Case 4. Computational results for D ˆ 1000 units/yr, S ˆ $50, c ˆ $15/unit, i ˆ 0:2$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 1/yr.
e
Case 5. Computational results for D ˆ 2000 units/yr, S ˆ $50, c ˆ $10/unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, k ˆ 1/yr.

Table 4
Computational results for supplementary examples (normal distribution case)
Case 1a Case 2b Case 3c Case 4d Case 5e
D1
Cycle T 0.2236 0.1690 0.3162 0.2582 0.1581
length T0 0.1799 0.1420 0.2453 0.1799 0.1285
T 0.0871 0.0818 0.1227 0.0730 0.0600
Expected C…T D1 † $11,103 $10,925 $11,811 $16,646 $21,210
total C…T 0 † $10,854 $10,785 $11,421 $16,007 $20,872
present cost C…T  † $10,544 $10,604 $11,021 $15,452 $20,425
C…T D1 †ÿC…T  †
Cost saving C…T  †
5.30% 3.03% 7.17% 7.73% 3.85%
C…T 0 †ÿC…T  †
C…T  †
2.94% 1.71% 3.63% 3.59% 2.19%
a
Case 1. Computational results for D ˆ 1000 units/yr, S ˆ $50, c ˆ $10/unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, l ˆ 1 yr,
r2 ˆ 0:3 yr2 .
b
Case 2. Computational results for D ˆ 1000 units/yr, S ˆ $50, c ˆ $10/unit, i ˆ 0:45$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, l ˆ 1 yr,
r2 ˆ 0:3 yr2 .
c
Case 3. Computational results for D ˆ 1000 units/yr, S ˆ $100, c ˆ $10/unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, l ˆ 1 yr,
r2 ˆ 0:3 yr2 .
d
Case 4. Computational results for D ˆ 1000 units/yr, S ˆ $50, c ˆ $15/unit, i ˆ 0:2$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, l ˆ 1 yr,
r2 ˆ 0:3 yr2 .
e
Case 5. Computational results for D ˆ 2000 units/yr, S ˆ $50, c ˆ $10/unit, i ˆ 0:3$=$=yr, a ˆ 0:2=yr, f ˆ 0:1=yr, l ˆ 1 yr,
r2 ˆ 0:3 yr2 .

4. Simulation modeling

To complement the analytical modeling in Section 2, we develop a simulation model which can be used
for any distribution. The simulation program has been coded using GAUSS [1]. The outline of the sim-
ulation algorithm is as follows:
598 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601

Simulation algorithm
Step 1. Start from a T.
Repeat Steps 2±4 number of replications given, say 500.
Step 2. Generate a random product life cycle p which follows a speci®c distribution.
Step 3. Compute k ˆ bp=T c. This is the number of cycles for the product life cycle using T.
Step 4. Compute C…T † using Eqs. (4) and (6) with the k value computed in Step 3.
Step 5. Compute the average C…T † for a given T. Go to Step 2 after increasing T by D, say 0.0001 yr.
Repeat until T reaches a high value which is not appropriate for reorder interval, say 1 yr.

To verify whether the above simulation algorithm works well, we ®rst run it with exponential and normal
distributions for case 1 in Section 3. The detailed results are shown in Table 5. We set the number of
replications to 500, and D ˆ 0:0001 yr. The solutions from the simulation are very close to the analytical
solutions. The small deviations might be caused from slight deviations of the mean and variance of
the generated random variables from the speci®ed mean and variance. The mean and variance of the
500 random product life cycle generated for the exponential distribution are 2.0161 yr and 4.0433 yr2 ,
respectively. Note that the mean and the variance have been set to 2 yr and 4 yr2 , respectively.
In many cases, obsolescence may arise in a more orderly fashion, perhaps due to a more regular product
life cycle or due to a form of item phase-out or planned obsolescence [15]. Masters [15] suggests that the
product life cycle might better be represented by another distribution such as the lognormal. In Table 6, we
summarize the simulation results for normal, lognormal, and gamma under the same mean and variance.
The data used for Cases 1±3 in Table 2 have been used for this simulation. As expected, the optimal cycle
lengths for lognormal and gamma are shorter than that for normal distribution. This result is due to the
skewness of the lognormal and gamma distributions. Let T F be the optimal cycle length when the true
distribution is F (for example, N ˆ Normal, L ˆ Lognormal, G ˆ Gamma), and C F …T F † be the corre-
sponding objective value. Suppose we use T N (optimal cycle length for normal distribution) when the true
distribution is Lognormal, the associated cost penalty is as follows:

Table 5
Comparison of simulation results with analytical results
Distribution Analytical result Simulation result
T C…T  † T C…T  †
Exponential 0.1043 $18,281 0.1040 $18,387
Normal 0.1291 $35,150 0.1301 $34,822

Table 6
Simulation results for several di€erent distributions
Case Distribution TF C F …T F † C F …T N † C F …T N †ÿC F …T F †
C F …T F †

1 Normal 0.1301 $34,822 ÿ ÿ

Lognormal 0.1257 $35,511 $35,524 0.04%
Gamma 0.1285 $35,079 $35,136 0.16%
2 Normal 0.1127 $35,514 ÿ ÿ
Lognormal 0.1086 $35,992 $36,009 0.05%
Gamma 0.1113 $35,482 $35,520 0.11%
3 Normal 0.1829 $35,728 ÿ ÿ
Lognormal 0.1788 $35,611 $35,651 0.11%
Gamma 0.1799 $36,438 $36,495 0.16%
 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601 599

C L …T N † ÿ C L …T L †
:
C L …T L †

For all the cases we have tested, these values are very small (see Table 6). Thus, it can be an additional
justi®cation for using normal distribution as a product life cycle.

5. Concluding remarks

As pointed out by Silver [19] in his review, if the quantitative models are to be more useful as aids for
managerial decision-making, they must represent and formulate more realistic problems. This paper has
been motivated by the need for such problem formulation, and we have investigated the e€ect of in¯ation
and time-value of money in an EOQ model with a random product life cycle. It is worthwhile to note that
the optimal cycle length may be very insensitive to in¯ation in the presence of discounts because the
breakpoint value is often the best order quantity over a wide range of parameter values [8].
Park and Son [17] have applied the DCF approach to four classical inventory models including a basic
EOQ model, an Economic Production Quantity (EPQ) model, an EOQ model with shortages, and an EPQ
model with shortages. In this study, we have only obtained the results for a basic EOQ model. Product life
cycles have been assumed to be random variables which follow any appropriate distribution. Analytical
results for an exponential distribution and a normal distribution have been presented. A simulation model
which can be used for any distribution has been developed to complement the analytical result. We did not
present analytical formulations for other distributions such as Gamma, Lognormal, Weibull, etc.

Acknowledgements

The authors are very grateful to the constructive and thorough comments raised by three anonymous
referees, Dr. Kyungkeun Lee, Dr. Wonyoung Yun, and Sherry Oh. Especially, Remark 3 and the simu-
lation model have been suggested by a referee. The authors also thank Dr. Edward Silver for providing
reference [15]. A part of this research was conducted while Ilkyeong Moon held a visiting position in the
Faculty of Management at the University of Calgary while on sabbatical leave. This research has been
supported by the 11th IBRD fund through its grant of research equipment to young professors sponsored
by the Ministry of Education in South Korea.

Appendix A. Proof of Proposition 1

Z
S ‡ cDT X 1
 ÿ…aÿf †…k‡1†T
 …k‡1†T ÿkp
C…T † ˆ 1ÿe k e dp
1 ÿ eÿ…aÿf †T kˆ0 kT

  1 Z
hD eÿ…aÿf †T ÿ 1 X  ÿ…aÿf †kT
 …k‡1†T ÿkp
‡ T ‡ 1 ÿ e k e dp
…a ÿ f †…1 ÿ eÿ…aÿf †T † aÿf kˆ0 kT

 X
1 Z …k‡1†T
hD 1 ÿ…aÿf †kT
‡ Tÿ e k eÿkp dp
aÿf aÿf kˆ0 kT
600 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601

 X
1 Z …k‡1†T
hD 1
‡ ÿT eÿ…aÿf †p k eÿkp dp
aÿf aÿf kˆ0 kT

X
1 Z …k‡1†T
hD
‡ …p ÿ kT †eÿ…aÿf †p k eÿkp dp:
aÿf kˆ0 kT

C…T † can be represented as follows after integrating by parts.

S ‡ cDT X 1
  
C…T † ˆ 1 ÿ eÿ…aÿf †…k‡1†T eÿkkT ÿ eÿk…k‡1†T
1ÿe ÿ…aÿf †T
kˆ0
  1
hD eÿ…aÿf †T ÿ 1 X   
‡ T‡ 1 ÿ eÿ…aÿf †kT eÿkkT ÿ eÿk…k‡1†T
…a ÿ f †…1 ÿ e ÿ…aÿf †T † aÿf kˆ0
 X 1
hD 1  
‡ Tÿ eÿ…aÿf †kT eÿkkT ÿ eÿk…k‡1†T
aÿf a ÿ f kˆ0
 
hkD 1 ÿ
X 1
‡ ÿ T 1 ÿ eÿ…aÿf ‡k†T eÿ…aÿf ‡k†kT
…a ÿ f †…a ÿ f ‡ k† a ÿ f kˆ0
 X 1
hkD 1 ÿ eÿ…aÿf ‡k†T
‡ ÿ T eÿ…aÿf ‡k†T eÿ…aÿf ‡k†kT : …A:1†
…a ÿ f †…a ÿ f ‡ k† aÿf ‡k kˆ0

We can simplify Eq. (A.1) using the following equations,

X
1
 ÿkkT 
e ÿ eÿk…k‡1†T ˆ 1;
kˆ0

X
1
1
eÿ…aÿf ‡k†kT ˆ ;
kˆ0
1ÿ eÿ…aÿf ‡k†T

X
1
  1 ÿ eÿkT
eÿ…aÿf †kT eÿkkT ÿ eÿk…k‡1†T ˆ :
kˆ0
1 ÿ eÿ…aÿf ‡k†T

Consequently,
 
S ‡ cDT hD eÿ…aÿf ‡k†T ‡ …a ÿ f †T ÿ 1 hkD‰2…a ÿ f † ‡ kŠ
C…T † ˆ ‡ 2
‡ 2 2
1ÿe ÿ…aÿf ‡k†T
…a ÿ f † ‰1 ÿ eÿ…aÿf ‡k†T Š …a ÿ f † …a ÿ f ‡ k†
hkDT
ÿ : 
…a ÿ f †…a ÿ f ‡ k†‰1 ÿ eÿ…aÿf ‡k†T Š

References

[1] Aptech Systems, GAUSS System version 3.2.32, Aptech Systems, Washington, 1997.
[2] S. Bose, A. Goswami, K.S. Chaudhuri, An EOQ model for deteriorating items with linear time-dependent demand rate and
shortages under in¯ation and time discounting, Journal of the Operational Research Society 46 (1995) 771±782.
[3] M.L. Carlson, J.L. Rousseau, EOQ under date-terms supplier credit, Journal of the Operational Research Society 40 (1989)
451±460.
 I. Moon, S. Lee / European Journal of Operational Research 125 (2000) 588±601 601

[4] T.C. Cheng, An economic order quantity model with demand-dependent unit production cost and imperfect production process,
IIE Transactions 23 (1991) 23±28.
[5] K.H. Chung, Inventory control and trade credit revisited, Journal of the Operational Research Society 40 (1989) 495±498.
[6] K.J. Chung, S.D. Lin, A note on the optimal cycle length with a random planning horizon, The Engineering Economist 40 (1995)
385±392.
[7] K.H. Chung, Y.H. Kim, Economic analysis of inventory systems: A rejoinder, The Engineering Economist 35 (1989) 75±80.
[8] N. Gaither, The e€ects of in¯ation and fuel scarcity upon inventory policies, Production and Inventory Management 22 (1981) 37±
48.
[9] C. Gurnani, Economic analysis of inventory systems, International Journal of Production Research 21 (1983) 261±277.
[10] C. Gurnani, Economic analysis of inventory systems a reply, International Journal of Production Research 23 (1985) 771±772.
[11] W.K. Haneveld, R.H. Teunter, E€ects of discounting and demand rate variability on the EOQ, International Journal of
Production Economics 54 (1998) 173±192.
[12] M.A. Hariga, M. Ben-Daya, Optimal time varying lot-sizing models under in¯ationary conditions, European Journal of
Operational Research 89 (1996) 313±325.
[13] F.W. Harris, How many parts to make at once, Factory 10 (1913) 135±136.
[14] Y.H. Kim, G.C. Philippatos, K.H. Chung, Evaluating investments in inventory systems: A net present value framework, The
Engineering Economist 31 (1986) 119±136.
[15] J. Masters, A note on the e€ect of sudden obsolescence on the optimal lot size, Decision Sciences 22 (1991) 1180±1186.
[16] I. Moon, W. Yun, An economic order quantity model with a random planning horizon, The Engineering Economist 39 (1993) 77±
86.
[17] C.S. Park, Y.K. Son, The e€ect of discounting on inventory lot sizing models, Engineering Costs and Production Economics 16
(1989) 35±48.
[18] I. Seo, Y. Kim, An optimal policy in a ®nite horizon inventory model with discounted costs, Working Paper, Korea Advanced
Institute of Science and Technology, Taejon, 1994.
[19] E. Silver, Modeling in support of continuous improvements towards achieving world class operations, Perspectives in Operations
Management, Kluwer Academic Publishers, Hingham, MA, 1993, pp. 23±44.
[20] E.A. Silver, D. Pyke, R. Peterson, Inventory Management and Production Planning and Scheduling, 3rd ed., Wiley, New York,
1998.
[21] R.R. Trippi, D.E. Lewin, A present value formulation of the classical EOQ problem, Decision Sciences 5 (1974) 30±35.