Introduction to inventory control

Marco Bijvank
Rotterdam School of Management, Erasmus University, Burgemeester Oudlaan 50, 3062
PA, Rotterdam, The Netherlands
e-mail: mbijvank@rsm.nl

Inventory is the number of products or resources held available in stock by an organization

and can include raw materials, work-in-process, component parts, and finished products.
The inventory of manufacturers, distributors, and wholesalers is clustered in warehouses.
Retailers keep their inventory either in a warehouse or in a store accessible to customers.
Many types of inventory exist:
safety stock is the amount of inventory kept on hand to protect against uncertainties
in customers demand and supply of items. The reason to keep this type of inventory
is because demand and lead times are not always known in advance and have to be
predicted. This type of inventory is also called buffer stock.
seasonal stock is the inventory built up to anticipate on expected peaks in sales or
supply, such that the production rate can be stabilized. This is also called anticipation
stock.
cycle stock consists of the inventory waiting to be produced or transported in batches
instead of one unit at a time. Reasons for batch replenishments include economies of
scale and quantity discounts.
decoupling stock is used to decouple the output of two inter-dependent workstations
because of different processing rates, set-up times or machine breakdowns. This permits
the separation of decision making.
congestion stock results from items that share the same production equipment. Consequently items have to wait for workstations to become available and inventory is built
up.
pipeline stock includes inventory in transit between different parties of the supply chain.
This is also called work-in-progress.
Not every type of inventory is kept at all parties in the supply chain. For example,
decoupling stock and congestion stock are mainly kept in a manufacturing environment,
whereas safety stock and pipeline stock are more important in a retail environment. In order
to handle the different types of inventory, control systems have to be developed. According
to Hax and Candea [1984], an inventory control system is a coordinated set of rules and
procedures that allows for routine decisions on when and how much to order of each item in
order to meet customer demand. A replenishment policy specifies how to decide upon these
1

inventory control
```
`

```

single-item

``

multi-item

`
```
```
``

stochastic demand

deterministic demand

P
PP

PP


single-period

HH

H

constant
demand

multi-period

time-varying
demand

H

HH

H

periodic
review

continuous
review

Figure 1: A classification of inventory control systems.

two decision variables. A classification of inventory control systems is given in Figure 1 and
will be explained below.
Most inventory systems concern single-item systems and consider one type of product at
a time. In multi-item inventory systems a number of products are considered simultaneously
because of limited capacity availability, economies of scale for joint replenishments or other
reasons. The classification for single-item systems is described in the remainder of this
section. A similar classification can be made for multi-item systems and is therefore not
included in the figure. See for more details Zipkin [2000]. Furthermore, we talk about an
inventory model when it represents the inventory system. A model can include assumptions
and, therefore, it is a simplification of reality.
There is a distinction between predictable and unpredictable demand. If the demand
in future periods can be forecasted with considerable precision, it is reasonable to use an
inventory policy that assumes that all forecasts will always be accurate. This is the case in
deterministic inventory models. However, when demand cannot be predicted very well, it
becomes necessary to use a stochastic inventory model where the demand in any time period
is a random variable rather than a known constant. Models in which demand is known (or
forecasted) and constant over a planning horizon are called classical lot size models. A wellknown example is the economic order quantity (EOQ) model introduced by Harris [1913].
When demand is not constant but still predictable, the models are called dynamic lot size
models. Examples of techniques to determine the order size for such models are the WagnerWithin algorithm [Wagner and Whitin, 1958] and the Silver-Meal heuristic [Silver and Meal,
1973].
In real life demand is mostly not known in advance. Therefore, a probability distribution
is used to describe the behavior of the demand in stochastic inventory models. A special
class of inventory control systems is concerned with products which have a very limited
period before it can no longer be sold. Examples are perishables (like food and flowers) or
items with a limited useful life (like newspapers and fashion). For such inventory systems
no decision has to be made regarding the order moment when the replenishment should take

place, but only the order size has to be determined for a single time period. Such models are
called single-period inventory models or newsboy models. In multi-period inventory models
it should also be determined when replenishment orders are triggered.
Different inventory levels are considered to determine when an order has to be placed.
The on-hand inventory level is the amount of physical inventory immediately available on
the shelves in a store or warehouse to meet customer demand. The occasion when the
inventory level drops to zero is called a stock out. The demand exceeding the available
stock is referred to as excess demand. There are two ways to deal with this demand when
there is a stock out. First, if the customer is willing to wait, the excess demand is held
until the next delivery replenishes the inventory. This is called backlogging or backordering.
Alternatively, the customer may not be willing to wait. In this case excess demand is lost,
which is called lost sales. The on-hand inventory minus the backorders is called the net
inventory. A positive net inventory represents the inventory on hand whereas a negative net
inventory refers to a backlog. The inventory on order is the work-in-progress or the items
ordered but not yet delivered due to the lead time. When there are backorders, a part of the
inventory on order is already reserved to meet customer demands from the past. Therefore,
the inventory position is defined as the sum of the inventory on hand plus the inventory on
order minus the outstanding backorders (or backlog).

Replenishment Policies
How often the inventory status should be checked for replenishments is specified by the review
interval. This is the period that elapses between two consecutive times at which the stock
level is known. Two types of review systems are widely used in business and industry. Either
inventory is continuously monitored (continuous reviews) or inventory is reviewed at regular
periodic intervals (periodic reviews). The former type of control is often called transaction
reporting, since continuous surveillance is not required but only at each transaction that
changes the inventory position (e.g., demand or order delivery). Whether or not to order at
a review time is determined by the reorder level. This is the inventory position at which a
vendor is triggered to place a replenishment order in order to maintain an adequate supply
of items to accommodate current and new customers. The mathematical notation for the
reorder level equals either R or s dependent on the type of replenishment policy. It comprises
the safety stock and the quantity of stock required to meet the average demand during the
lead time plus the time until the next review moment. The lead time is the period of time
between order placement and the delivery of the order such that the order is available for
satisfying customer demands. An order size can either be fixed or variable. The type of
replenishment policies with variable order quantities are called order-up-to policies in which
the order size is such that the inventory position is increased to an order-up-to level. This
level is denoted by S. Figure 2 shows the difference between continuous and periodic reviews
when the order size is a fixed number and each customer demand equals one unit (also called
unit-sized demand ). In the continuous review case, the order is placed immediately when the
inventory position reaches the reorder level. In the periodic review case, the order placement
has to wait for the next review time after the inventory position has reached the reorder level.
Figure 3 illustrates the concept of order-up-to policies in the case where customer demands
3

(b) periodic review

inventory level

inventory level

(a) continuous review

6

R


lead time

time

review

1


lead time

time

Figure 2: The on-hand inventory level (solid line) and inventory position (dashed line) for
replenishment policies with a fixed order size under (a) continuous and (b) periodic review.
(a) continuous review

(b) periodic review

S6
inventory level

inventory level

S6

lead time

1


time

review  lead time time

-

Figure 3: The on-hand inventory level (solid line) and inventory position (dashed line) for
order-up-to replenishment policies under (a) continuous and (b) periodic review.
are not always unit sized. Notice that the delay in the actual order placement can result in
larger order sizes in case of periodic reviews as compared to continuous reviews. There can
be fixed order costs incurred with each order. When no fixed order cost is charged, there
is no incentive not to place an order at each review moment in case of variable order sizes.
Consequently, the reorder level does not play a role in order-up-to policies (i.e., it equals
S 1).
Based on these characteristics on the replenishment process we present the mathematical
notation for the different types of replenishment policies in Table 1. The letter r specifies the
length of the review interval. The review interval length for continuous review systems is zero
and is therefore omitted. Furthermore, Q stands for fixed order quantities, S denotes the
order-up-to level, whereas R and s represent the reorder level for fixed-order-size policies and
order-up-to policies, respectively . Figure 2a and Figure 2b illustrate the (R, Q) policy for
continuous and period reviews, respectively, whereas Figure 3a and Figure 3b give an example
of the (s, S) policy for continuous and period reviews, respectively. A special class within
the order-up-to policies are base-stock policies, in which the satisfied demand in between two
review times is immediately ordered at the next review time. For such policies, the reorder

order size

order moment
continuous review
periodic review
fixed
variable

(R, Q)
no fixed cost:
fixed cost:

(R, Q)

(S 1, S)
(s, S)

no fixed cost:
fixed cost:

(S 1, S)
(s, S)

Table 1: The notation for the types of replenishment policies most often applied in literature
and practice.
level s is equal to the order-up-to level minus one in case of discrete demand. When demand
is continuous, the reorder level is equal to the order-up to level. Models with a base-stock
policy are denoted as (S 1, S) policies. In continuous review models, base-stock policies
are also called one-for-one policies since every customer demand immediately triggers a new
order. Notice that the (R, Q) and (s, S) policy are identical for continuous review models if
all demand transactions are unit sized, s = R and S = R + Q. In that case, replenishments
are always made when the inventory position is exactly at the reorder level. Consequently,
the order size always equals Q or S s. See Figure 2a for an illustration.

Objective Function
To compare the performance of the different replenishment policies, the costs associated
with each controlling system has to be minimized while simultaneously meeting a desired
customer service level. There are three types of inventory costs: (1) order costs associated
with placing an order, (2) holding costs for carrying inventory until it is sold or used, and
(3) penalty costs for unfulfilled customer demand. The order cost can consist of fixed cost
for each time an order is placed and variable cost for each unit ordered. The holding cost is
mainly the opportunity cost of the money invested in inventory. But it should represent all
inventory carrying costs, including the cost of warehouse space, material handling, insurance
and obsolescence. The penalty cost is the cost of not having sufficient inventory to meet
all customer demands. These shortage costs can be interpreted as the loss of customers
goodwill and the subsequent reluctance to do business with the firm, the cost of delayed or
no revenue, and any possible extra administrative costs.
When demand is stochastic, shortages cannot be avoided. A service level is used in a
supply chain to measure the performance of such inventory systems. The most common
measures of service are (1) service level, (2) service level, and (3) service level. The
first type of service level is an event-oriented criterion. It measures the probability that all
customer demands are satisfied within a replenishment cycle. This definition is also called
the cycle service level, since it measures the fraction of cycles in which a stock out occurs.
The service level, or fill rate, is a quantity-oriented measure that represents the fraction
of the demand satisfied directly from stock on hand. An example to illustrate the difference
between the cycle service level and the fill rate is provided in Table 2. The total directly
satisfied demand from stock on hand is 12 units, while the total demand is for 15 units.
5

order cycle
1
2
3

inventory level
4
6
3

demand
5
8
2

satisfied demand
4
6
2

Table 2: An example to show the difference between cycle service level and fill rate.
Therefore, the fill rate equals 12/15 = 80%. Since the demand exceeds the inventory level
in two out of three order cycles, the cycle service level equals 33.3%. The service level
incorporates the waiting time of the demands backordered. This service performance measure
is the fraction of time during which there is no stock out. This service level definition is also
called the ready rate.

(R, Q) Policy with Continuous Reviews

When the inventory levels are monitored continuously, a fixed order size replenishment policy
is most often used when there are fixed order costs charged per order. Since the EOQ-formula
is robust in the sense that the cost function is rather flat around the optimal order size, the
size of the order is set according to this formula. Consequently, only the reorder level has
to be determined. First, consider a cycle service level to set this parameter. That is, the
probability that there is no out of stock occurence during a replenishment cycle. For an
(R, Q) replenishment policy with continuous reviews, an order will arrive L time periods
after the inventory position reaches the reorder level. Therefore, the cycle service level
is defined as the probability that demand during the lead time does not exceed R units;
P r(DL < R) . When demand follows a normal distribution this translates to


R AV G L

,
(1)

ST D L
where AV G and ST D represent the average and standard deviation of the demand per
time period, respectively, and () is the cumulative distribution function of a standard
normal distribution. This latter quantity can be found in the appropriate look-up table (see
Appendix A). Equivalently to Equation (refeq:R),

(2)
R = AV G L + z ST D L,
where z corresponds to the safety factor of a standard normal distribution with probability
. Table 3 illustrates several values of the safety factor z for different values of .
Example 1. A television manufacturing company produces its own speakers, which are
used in the production of its television sets. The television sets are assembled on a production line where one speaker is needed per set. The speakers are produced in batches because
they do not warrant setting up a continuous production line, and relatively large quantities
can be produced in a short time. Therefore, the speakers are placed into inventory until
they are needed for assembly into television sets on the production line. Another reason to
keep inventory has to do with the fact that sales of TV sets have been quite variable. Con6

0.75
0.68

0.8 0.85
0.85 1.04

0.9 0.92 0.94 0.95 0.96 0.98

1.29 1.41 1.56 1.65 1.76 2.06

Table 3: The value of the safety factor z for different values for the cycle service level when
demand follows a normal distribution (see also Appendix A).
sequently, the demand for the speakers is quite variable as well.
There is a lead time of 6 weeks between ordering a production run to produce speakers
and having speakers ready for assembly into television sets. The annual demand for speakers
is a random variable that has a normal distribution with a mean of 96,000 units and a
standard deviation of 6,930 units. To minimize the risk of disrupting the production line
producing the TV sets, management has decided that the safety stock for speakers should
be large enough to avoid a stockout during this lead time for 95 percent of the time. The
company is interested in determining when to produce a batch of speakers and how many
speakers to produce in each batch.
Several costs must be considered: Each time a batch is produced, a setup cost of $12,000
is incurred. This cost includes the cost of tooling up, administrative costs, record keeping,
and so forth. The production of speakers in large batches leads to a large inventory. The
estimated holding cost of keeping a speaker in stock is $0.30 per month. This cost includes
the cost of capital tied up in inventory. Since the money invested in inventory cannot be
used in other productive ways, this cost of capital consists of the lost return (referred to
as the opportunity cost) because alternative uses of the money must be forgone. Other
components of the holding cost include the cost of leasing the storage space, the cost of
insurance against loss of inventory by fire, theft, or vandalism, taxes based on the value of
the inventory, and the cost of personnel who oversee and protect the inventory.
The best replenishment policy is an (R, Q) policy with continuous reviews since productline orders can be placed at all times. As mentioned before,
p the EOQ-formula is rather
robust. Therefore, the best order quantity is given by Q = 2 AV G K/h. The setup
cost to produce the speakers is K = $12, 000, the unit holding cost is h = $0.30 per speaker
per month. Since the holding cost is per month and the demand per year, we need to make
sure that they are both expressed in the same time units. Consider
years as time units. The
p

annual holding cost is $3.60 per speaker. As a result, Q = 2 96, 000 12, 000/3.60 =
25, 298.22 speakers. After comparing the total costs for Q = 25, 298 (that is $91, 073.5966
per year) and for Q = 25, 299 (that is $91, 073.5967 per year), the production size should
be 25,298 speakers per batch.
The next step is to find the value of the reorder level R. The criterion used in this
example is a cycle service level of 95%. The formula to use is the following: R = AV G
L + z ST D. The value of z corresponds to the value found in the lookup-table for
the cumulative density function of a standard normal distribution. This means z = 1.65.
When we assume that there are 4 weeks in one month, then the lead time equals 1.5
months. Consequently, the demand distribution is converted to monthly demand. To find
the standard deviation of the demand per month, it is not allowed to divide the standard
deviation over 12 months (similar to the average). Any algebraic calculations are only
allowed based on the variance. The variance of the yearly demand is 6, 930 6, 930 =
48, 024, 900. The variance per month is 4, 002, 075, and consequently, the standard devia7

tion of the monthly demand is 2, 000.52 speakers. Furthermore, the average demand is
8, 000 units per month. As a result,

R = AV G L + z ST D L

= 8, 000 1.5 + 1.65 2, 000.52 1.5

= 16, 042.71.
Since the reorder level represents a number of speakers, it should be rounded up to the
nearest integer to guarantee that the service level is satisfied. In summary, the reoder level
should be 16,043 speakers.
In Equation (2) it is assumed that lead times are constant. Whenever the lead time has
a mean and standard deviation denoted by AV GL and ST DL, respectively, then

(3)
R = AV G AV GL + z AV GL ST D2 + AV G2 ST DL2 .
Note that this expression is the same as Equation (2) when the lead time is constant (that
is, when AV GL = L and ST DL = 0).
Example 1 (continued). If the lead time would be stochastic with the same average
as before (that is 6 weeks), but with a standard deviation of 2 weeks, then the order size
would remain the same (Q = 25, 298 speakers), but the reorder level becomes
p
(4)
R = 8, 000 1.5 + 1.65 1.5 2, 000.522 + 8, 0002 (2/4)2 = 19, 740.
This means that the reorder level increases by 3,697 speakers, since the safety stock needs
to hedge against more demand uncertainty during the lead time.
Besides the cycle service level, a commonly used definition for service is the fill rate,
which corresponds to the fraction of demand satisfied immediately by stock on hand:
expected demand satisfied immediately in a replenishment cycle
expected demand in a replenishment cycle
expected shortage in a replenishment cycle
= 1
.
expected demand in a replenishment cycle

fill rate =

(5)

When excess demand is assumed to be backordered and the inventory level is continuously
monitiored, a new order is placed whenever Q units have been demanded. This corresponds
to the expected demand in a replenishment cycle. On the other hand, when excess demand is
assumed to be lost this amount equals Q units plus the expected shortage in a replenishment
cycle (abbreviated as ESC). This means
ESC
Q
ESC
fill rate = 1
Q + ESC

backorder: fill rate = 1

lost sales:

For simplicity, only the backorder setting is discussed in the remainder of this document.
8

Next, ESC needs to be calculated and is dependent on the reorder level. Since ESC
represents the expected demand in excess of the reorder level in a replenishment cycle:
ESC = E[max{DL R, 0}]. When the demand follows a normal distribution, this simplifies
to

ESC = ST D L L(z),
(6)
where L(z) is known as the loss function and
z=

R AV G L
.
ST D L

(7)

This latter expression was already derived in Equation (2). The loss function is given by
L(z) = (z) z [1 (z)] and its values can be found in a table (see Appendix B).
For instance, when the safey factor z equals 1.65, this corresponds to L(z) = 0.0206. This
means that for a cycle service level of 95%, the safety factor equals z = 1.65 and the expected
demand to be exceeding the reorder level equals 0.0206 when demand would follow a standard
normal distribution during the lead time. Other values of L(z) are provided in Table 4.
z
L(z)

0.68
0.1478

0.85
0.1100

1.04
0.0772

1.29
0.0465

1.41
0.0359

1.56
0.0255

1.65
0.0206

1.76
0.0158

2.06
0.0072

Table 4: The value of the loss function L(z) for different values for the safety factor z when
demand follows a normal distribution (see also Appendix B).
Note that the lead time is assumed to be constant. When the lead time is stochastic with
mean AV GL and variance ST DL2 , then

(8)
ESC = AV GL ST D2 + AV G2 ST DL2 L(z)
and

R AV G AV GL
.
(9)
AV GL ST D2 + AV G2 ST DL2
This is exactly the same expression as Equation (7) when the lead time is constant (that is,
when AV GL = L and ST DL = 0).
z=

When the values of the reorder level R and order quantity Q are determined, the amount
of safety stock (SS) is given by
SS = R AV G AV GL,

(10)

and the average on-hand inventory level equals

inventory level =

Q
+ SS.
2

(11)

Example 1 (continued). Lets consider the impact of stochastic lead times. Recall that
with a constant lead time the reorder level shoud be 11, 291 units, whereas it should be
equal to 12, 654 units when the lead time is stochastic. To compute the fill rate, the z-value
needs to be determined first. However, for both scenarios this is already determined,namely
9

z = 1.65. As a result, the loss function can be found in the standard normal loss table:
L(z) = 0.0206. Lets first consider the fill rate when the reorder level equals 16, 043 units.

ESC = 2, 000.52 1.5 0.0206 = 50.47

50.47
= 0.9980.
fill rate = 1
25, 298
However, when the lead time is stochastic,
p
ESC =
1.5 2, 000.522 + 8, 0002 (2/4)2 0.0206 = 96.63
96.63
= 0.9962.
fill rate = 1
25, 298
This example illustrates two things: (1) when the fill rate is rather high, the lead time
variance is not of great influence, and (2) the fill rate is higher than the cycle service level.
This latter observation is always true and is of importance for many practical settings.
In most text books the cycle service level is used to define service level, whereas in many
real-world applications the fill rate is used as service performance measure. When the text
book formulas are used, the reorder level is structurally set too high. Consequently, the
safety stock is higher as well as the average on-hand inventory level. This will result in
higher inventory costs, whereas the fill rate constraint would still be satisfied with a lower
reorder level.

(S 1, S) Policy with Periodic Reviews

A base-stock policy is often used when no fixed costs are charged for each order. In order
to keep the inventory level as low as possible, an order is placed at each review where
enough items are orderd to meet demand until the next order delivery. For an (S 1, S)
replenishment policy with period reviews, the order that is placed at a review moment is
such that the inventory position equals S units after ordering. This order arrives after the
lead time of L time periods. Since an order is placed at each review moment, the next order
will arrive after r time periods. This means that the order-up-to level S should cover the
demand during L + r time periods. The question to be answered is how to set the value of
S.
First, lets use the cycle service level to determine the vaule of S. In an (S 1, S) policy
with periodic reviews, this service level definition corresponds to the probability that the
demand during L + r time periods does not exceed S units; P r(DL+r < S) . When
demand follows a normal distribution this translates to


S AV G (L + r)

,
(12)

ST D L + r
and
S = AV G (L + r) + z ST D

L + r,

(13)

where z corresponds to the safety factor of a standard normal distribution with probability
. When these equations are compared to Equation (1) and (2), it becomes clear that they
10

are similar except for the time period during which the inventory level needs to hedge against
demand uncertainty.
Equation (13) assumes that lead times are constant. Similar to the previous section, a
stochastic lead time could be included where the average and standard deviation of the lead
time are denoted by AV GL and ST DL, respectively. Consequently,
p
S = AV G (AV GL + r) + z (AV GL + r) ST D2 + AV G2 ST DL2 .
(14)
This is exactly the same expression as Equation (13) when the lead time is constant (that
is, when AV GL = L and ST DL = 0).
Example 2. The central warehouse for a pharmacy orders antibiotics every two weeks
(that is, every 14 days). The daily demand has an average of 2,000 boxes with a standard
deviation equal to 800 boxes. An order will arrive after exactly 5 days. Since the warehouse
should always have enough antibiotics on inventory, they want to provide their customers
with a service level of 99%. The manager of the warehouse would like to know how many
boxes to order with a 99% cycle service level.
Since no order costs are mentioned, it is best to order at each review moment to keep
inventory holding costs as low possible. Consequently, the correct inventory replenishment
policy to consider for the warehouse is a base-stock policy. A 99% cycle service level
corresponds to a z value equal to 2.33. The lead time is constant. As a result, Equation
(13) should be used to set the base-stock level:

(15)
S = 2, 000 (5 + 14) + 2.33 800 5 + 14 = 46, 124.99.
To satisfy the service constraint it is best to round this number up to 46,125 boxes. This
means that the warehouse manager should order such an amount that the inventory position
(that is, on-hand inventory minus backorders plus inventory on order) after ordering is equal
to the base-stock level of 46,125 units.
Besides the cycle service level, the fill rate could also be used to set the base-stock level
S. See Equation (5) for a definition. Since an order is placed at each review moment, the
replenishment cycle equals r time units. Consequently, the expected demand in a replenishment cycle equals AV G r, whereas the expected shortage in a cycle (or ESC) is given by
E[max{DL+r S, 0}]. Similar to the previous section,

(16)
ESC = ST D L + r L(z)
and

S AV G (L + r)

,
(17)
ST D L + r
for the particular inventory policy under study in this section. Whenever the lead time is
stochastic, with a mean and standard deviation of AVGL and STDL, respectively, then
p
ESC = (AV GL + r) ST D2 + AV G2 ST DL2 L(z)
(18)
z=

and
z=p

S AV G (AV GL + r)
(AV GL + r) ST D2 + AV G2 ST DL2
11

(19)

This is exactly the same expression as in Equation (17) when the lead time is constant (that
is, when AV GL = L and ST DL = 0).
When the value of order-up-to level S is determined, the amount of safety stock (SS) is
given by
SS = S AV G (AV GL + r),
(20)
and the average on-hand inventory level equals
inventory level =

r AV G
+ SS.
2

(21)

Example 2 (continued). As it turned out, the supplier of the antibiotics was not really
reliable. Consequently, the warehouse manager wants to include the standard deviation of
the lead time, which is 2 days. The average lead time turned out to be 6 days. Furthermore,
the warehouse manager has read that cycle service level is not the only performance measure
for service towards customers, and decided to use a 99% fill rate instead.
In order to find the correct base-stock level based on this new information, the expected
shortage needs to be computed first. Since the fill rate constraint is 99%, the following
expression must hold:
fill rate = 1

ESC
AV G r

0.99 = 1

ESC
.
2, 000 14

(22)

This means that the expected shortage in a cycle of 14 days should be at most 280 boxes.
ESC is expressed as
p
ESC = (AV GL + r) ST D2 + AV G2 ST DL2 L(z).
(23)
Consequently,
280 =

(6 + 14) 8002 + 2, 0002 22 L(z),

(24)

such that L(z) = 0.0522. Since L(z) represents the loss function, we need to find the smallest value of z in the standard normal loss function that is smaller than 0.0522. Otherwise,
the fill rate constraint is violated. This corresponds to a z value of 1.24. Finally, Equation
(19) is used to find the value of the correct base-stock level:
S 2, 000 (6 + 14)
1.24 = p
(6 + 14) 8002 + 2, 0002 22

(25)

As a result,
S = 1.24

p
(6 + 14) 8002 + 2, 0002 22 + 2, 000 (6 + 14) = 46, 654.54.

The new base-stock level becomes 46,655 boxes.

12

(26)

(s, S) Policy with Periodic Reviews

When there are fixed cost involved with placing an order, it is no longer efficient anymore
to order at each review. As a result, a base-stock policy (as in the previous section) is not
sufficient. A reorder level should be introduced to trigger the placement of an order (similar
to the (R, Q) policy for continuous reviews). However, since orders can only be placed at a
review moment, the inventory position can be lower than the reorder level at the moment
an order is placed. The difference between the reorder level and the inventory position at
order placement is called the undershoot. Consequently, the order size equals the difference
between order-up-to level S and reorder level s plus the undershoot. The average undershoot
is given by
ST D2 + r AV G2
(27)
AV GU =
2 AV G
and the variance of the undershoot by
ST DU 2 =

3 r ST D2 + r2 AV G2
AV GU 2 .
3

(28)

Since the EOQ-formule is robust, it is common to set S s + AV GU equal to the EOQ.

This means that once the reorder level is specified, the order-up-to level can be found easily
as well. Therefore, the focus in the remainder of this document is on how to set the reorder
level.
Lets first consider the cycle service level to set the reorder level. Recall that the cycle
service level equals the probability of no stockout occurrence during a replenishment cycle.
This means that the reorder level should cover the demand during the undershoot period
plus the lead time, since it takes another L time units before an order is delivered. This
time period is also called the risk period. Consequently, P r(Drisk < R) . The average
demand during the risk period equals
AV GR = AV GU + AV G L,

(29)

and the variance of the demand during this time period equals
ST DR2 = ST DU 2 + ST D2 L.

(30)

Consequently, the same formulas as for the (R, Q) policy with continuous reviews could be
used, but with a different time period during which the reorder level should hedge against
demand uncertainty. Here, this time period is the risk period consisting of the lead time
plus the period of undershoot ratherthan just the lead time. As a result, AV G L should
be replaced by AV GR and ST D L by ST DR. This means that,
s = AV GR + z ST DR.

(31)

When the lead time is not constant, AV GR and ST DR need to be updated to

AV GR = AV GU + AV G AV GL,
ST DR2 = ST DU 2 + AV GL ST D2 + AV G2 ST DL2 .
13

(32)
(33)

Besides the cycle service level, the fill rate could also be used to set the reorder level s.
Recall that this service level definition corresponds to the fraction of demand satisfied immediately by stock on hand. Similar to the (R, Q) policy with continuous reviews, the expected
demand in a replenishment cycle is equal to the EOQ-formula (as explained above), whereas
the expected shortage in a cycle (or ESC) is given by E[max{Drisk s, 0}]. Consequently,
ESC = ST DR L(z)

(34)

and

s AV GR
.
(35)
ST DR
When the values of the reorder level s and order-up-to level S are determined, the amount
of safety stock (SS) is given by
SS = s AV GR,
(36)
z=

and the average on-hand inventory level equals

inventory level =

S s + AV GU
+ SS.
2

(37)

Example 3. The new head of the automotive section of Nichols Department Store has the
responsibility to ensure that reorder quantities for the various items have been correctly
established. He decides to test one item and chooses Michelin tires, XW size 185 x 14
BSW. The placement of an order will cost $20, whereas the cost for each tire is $35. The
holding cost is 20% of the tire cost per year. Orders can only be placed once a week and
the delivery lead time is on average 5 days with a standard deviation of 1 day. The annual
demand follows a normal distribution with an average of 1,000 tires, whereas the standard
deviation of the daily demand is only 2 tires. Because customers generally wait in case of
a stockout and do not go elsewhere, the head of the section decided on a medium level of
service, so he wants to ensure an 85% probability of not stocking out on this specific brand
of tires. Assume the demand occurs 7 days per week, and one month equals 4.3 weeks.
Because there are fixed order cost charged with each order, these costs need to be
balanced with the holding cost. This means that the appropriate replenishment policy
should include a reorder level. Since orders can only be placed at fixed time intervals, it
is a periodic replenishment policy. Therefore, the (s, S) policy with periodic reviews is the
correct replenishment policy for this example. As mentioned before, the EOQ-formula is
robust in the sense that this amount gives a reasonable answer to the number of units to
order when an order is placed. In this example, lets use weeks as time unit.
The average
demand per week is 1, 000/(12 4.3) = 19.38 with a standard deviation of 22 7 = 5.29
tires. The weekly holding cost is 0.2 35/(12 4.3) = 0.136. Consequently,
r
2 19.38 20

= 75.593.
(38)
Q =
0.136
Since this is the average order size, this value does not have to be rounded to an integer
value. The order size consists of the difference between the reorder level s and the order-

14

up-to level S plus the average undershoot (that is, the amount of units below the reorder
level when an order is placed). Since the demand follows a normal distribution, the average
undershoot equals
5.292 + 1 19.382
= 10.41 tires,
(39)
AV GU =
2 19.38
whereas the standard deviation of the undershoot is given by
r
3 1 5.292 + 12 19.382
ST DU =
10.412 = 6.69 tires.
(40)
3
To compute the cycle service level, the demand during the risk period has to be considered.
This equals the undershoot plus the demand during the lead time. The average demand
during the risk period is
AV GR = 10.41 + 19.38 (5/7) = 24.255,
and the standard deviation of the demand during the risk period equals
p
ST DR = 6.692 + (5/7) 5.292 + 19.382 (1/7)2 = 8.51.

(41)

(42)

Since the restriction on the cycle service level is 85%, this corresponds to a z-value of 1.04.
Finally, the reorder level can be computed by
s = 24.255 + 1.04 8.51 = 33.11.

(43)

That is, the reorder level equals 34 tires. This vaue also specifies what the order-up-to
level should be, since the result of the EOQ-formula (Q = 75.593) shoud be equal to
S s + AV GU . As a result,
S = Q + s AV GU = 75.593 + 34 10.41 = 99.18.

(44)

This means that the order-up-to level should be set to 99 tires (the value is rounded to the
nearest integer).
For these specific values of the inventory replenishment policy (that is, s = 34 and S = 99),
the fill rate is given by
ST DR L(z)
,
(45)
fill rate = 1
S s + AV GU
where z = (34 24.255)/8.51 = 1.145. Note that this z value is not the same as for the
cycle service level since the reorder level s is rounded in between. This z value corresponds
to L(z) = 0.0627 and the fill rate equals
fill rate = 1

8.51 0.0627
= 0.9929.
99 34 + 10.41

(46)

Even though the cycle service level is only 85%, the fill rate is rather high with 99.3%.

15

References
F. Harris. How many parts to make at once. Factory, The Magazine of Management, 10:
135136, 1913.
A.C. Hax and D. Candea. Production and Inventory Management. Prentice- Hall, Englewood
Cliffs, 1984.
E.A. Silver and H.C. Meal. A heuristic for selecting lot size quantities for the case of a
deterministic time-varying demand rate and discrete opportunities for replenishments.
Production and Inventory Management Journal, 2nd Quarter:6474, 1973.
H. Wagner and T.M. Whitin. Dynamic version of the economic lot size model. Management
Science, 5:8996, 1958.
P.H. Zipkin. Foundations of inventory management. McGraw-Hill, New York, 2000.

16