0% found this document useful (0 votes)
230 views20 pages

Inventory Management Under Demand Uncertainty

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
230 views20 pages

Inventory Management Under Demand Uncertainty

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 20

Inventory Management under Demand Uncertainty

Özalp Özer
Date: 02/2004; Rev’d: 06/2011

0. Overview
In this unit, you will learn the essentials of inventory management under demand
uncertainty.

In Section 1, we introduce the concept of demand uncertainty. We define what is meant


by probabilistic forecast. We also provide some intuition about how demand
uncertainty affects inventory-stocking decisions.

In Section 2, we introduce the most basic but fundamental inventory management


problem under demand uncertainty. We present the classical newsvendor problem and
an inventory stocking tool for a manager who faces uncertain demand during a single
selling season. We use marginal cost-benefit analysis to decide on an optimal stocking
level prior to the selling season. This stocking level minimizes the cost of overage and
underage.

In Section 3, we provide several inventory management policies to manage inventory


over multiple selling seasons and stocking decisions. We introduce the concept of
leadtime demand uncertainty. We discuss reasons for holding safety stock when
demand is uncertain. We provide tools both for continuous and periodic review
inventory systems.

In Section 4, we introduce the concept of risk pooling over multiple stocking locations.
We illustrate the benefit and cost of consolidating two warehouses into a single
warehouse.

In Section 5, we provide a tool, an exchange curve, to trade off service level and dollar
investment in safety stocks.

In Section 6, we conclude with our final remarks on global inventory management


issues

By the end of this unit, you should be able to due the following.

• Communicate informative forecast statements.


• Understand the effect of demand uncertainty on stocking decisions.
Copyright © 2004 by Özalp Özer of Massachusettes Institute of Technolog and the University of Texas at Dallas. All
rights reserved. To request permission to reproduce materials, e-mail Dr. Özalp Özer at [email protected] or
[email protected]. No part of this publication may be reproduced, stored in a retrieval system, used in a
spreadsheet, or transmitted in any form or by any means –– electronic, mechanical, photocopying, recording, or
otherwise –– without the permission of the author. This lecture note was first written while Dr. Özalp Özer was a
faculty member at Stanford University.
• Identify a newsvendor setting and apply marginal cost/benefit analysis to decide on
an optimal stocking level.
• Choose an appropriate inventory review system for your products.
• Decide on an inventory management policy and determine its impact on inventory
holding cost.
• Quantify the effect on inventory of consolidating customer demand.

1. Demand Uncertainty
In the previous unit, we treated demand as a known quantity and provided models
with trades-offs between set-up (or fixed ordering) cost and holding cost. In reality,
however, demand for products displays variation over time. Several factors such as
changing consumer preferences and economic trends create uncertainties in demand.
Firms try to estimate future demand patterns using forecasting methods, which are
based on historical data and expertise about the market. Even though firms are aware of
the concept of demand uncertainty, they often come up with a single point estimate of
future demand. The following statement from a marketing manager is quite common.
“We expect demand for our product to be 10,000 units during the next selling season.”

Most of us may have also heard that forecasts are always wrong. Does this mean
that firms should not even try to come up with a forecast? No. Firms do need to
estimate demand for planning purposes. However, enhancing forecast statements by
providing scenario analysis improves the quality of the forecast. Consider following
two statements.

Firm A: “We expect demand for our product to be on average 10,000 units.
However, if the weather is sunny during our sales season we expect to sell 20,000
units; whereas, if it is rainy we expect to sell nothing.”

Firm B: “We expect demand for our product to be on average 10,000 units.
However, if weather is sunny we expect to sell 11,000; whereas, if it is rainy we
expect to sell 9,000 units.”

Both of these statements suggest that the average demand will be 10,000 units.
However, the first statement states that not selling even a single product is a possibility.
It also states that Firm A has a potential to sell almost twice more than Firm B. Either
scenario may be preferable, depending on risk preferences. However, the decision of
how much inventory to stock for Company A is more challenging than that for
Company B.

Probability of rain affects each company differently. Firm B would be less


concerned about not knowing the possibly of rain then Firm A. The impact of
uncertainty is less for Firm B. However, the story is different for Firm A. If Firm A
stocks 20,000 units, they would be quite happy to know that probability of a sunny day

Inventory Management under Demand Uncertainty 2


is higher than 95%. To capture the effect of demand uncertainty, we report information
about the possible variation in demand. Therefore, a good forecast should report at least
two data points. The first is the average demand. The second is information about the
possible variation from average demand. Reporting the standard deviation can convey
this information.

If no variability exists, hence the demand has no uncertainty, then the manager can
order exactly the average demand and avoid completely the possibility of stocking out.
However, when the demand is uncertain, the manager may run out of stock even if she
orders the mean demand. As discussed in the previous unit, stockout results in shortage
cost due to lost sales and loss of goodwill. To cope with uncertainty in demand, firms
may need to produce more than the average demand. The aim of this unit is to provide
you with some inventory management tools to counter the adverse effects of demand
uncertainty.

To do so, we first illustrate what we mean by a probabilistic forecast. A marketing


department can use historical data, current economic conditions and other factors to
estimate demand and come up with a probabilistic forecast. A good starting point is to
plot a histogram, which reports the frequency of observed demand during a selling
season. We provide an example below.

25%

20%

15%
Probability

10%

5%

0%
40 50 70 90 100 110 130 150 160
Unit Sales

FIGURE 1: Histogram For the Probabilistic Forecast

d 40 50 70 90 100 110 130 150 160


Probability 3% 6% 13% 18% 20% 18% 13% 6% 3%
F(d) 3% 9% 22% 40% 60% 78% 91% 97% 100%

TABLE 1: Tabulation of Probabilistic Forecast

For this example, the first row in Table 1 reports possible realized demand at the
end of a selling season. The second row reports the probability of observing d units of

Copyright © 2004 by Özalp Özer 3


demand ( Prob{Demand=d units} ). One interpretation for second row is the frequency
of observed demand during a sales season given the sales history. For example, in
Figure 1, the histogram suggests that out of 100 sales periods during 18 periods,
demand was 90 units. Another interpretation is that, the manager uses her judgment and
predicts that the company will sell 100 units with 20% chance. The last row in the
Table reports the cumulative probability; that is, probability that the demand during a
given selling period is less than or equal to d units. We use the notation F(d) =
Prob{Demand ≤ d units} to refer to the cumulative probability.

Often histograms may resemble a bell shaped curve. For those instances, a Normal
distribution is a good approximation to model demand uncertainty and to provide a
probabilistic forecast.

Likelihood

Prob{Demand is at most d units}


One STD

AVG d

FIGURE 2: Normal Distribution with Mean AVG and Standard Deviation STD.

Recall from your managerial statistics course that a Normal distribution is


determined by two parameters; the mean and the standard deviation. The shaded area
represents the cumulative probability of demand being less than or equal to d units. The
value of this probability can be obtained from the standard Normal tables.
___________________________________________________________________

Example: Suppose that demand follows a Normal distribution with mean 100 units
and standard deviation 20 units. What is the probability of observing demand less
than or equal to 133 units?

Answer: Recall that F(133)= Prob{Demand ≤ 133 units}= Prob{ Demand-


mean/STD ≤ 133-100/20}=Φ(1.65). From the standard Normal table Φ(1.65) =
95%.
__________________________________________________________________

The above concepts suffice to study the effect of demand uncertainty on stocking
decisions and to show why and how firms should protect themselves against demand
uncertainty. The detail of forecasting is the job of a statistician or a forecasting
specialist. During the rest of this unit, we assume that the probabilistic forecast of

Inventory Management under Demand Uncertainty 4


demand for the product is given. Next, we introduce the most basic but fundamental
inventory management problem under demand uncertainty.

2. Newsvendor’s Inventory Management Problem (Aka


Newsvendor Problem)
Consider a manager who faces a single selling season. Due to obsolescence of an item
or the nature of the business, the manager does not have the ability to store inventory
from one selling season to the next. Furthermore, the single selling season is not long
enough for the manager to place a second order; that is, she has only one opportunity to
stock inventory. In other words, the production or replenishment leadtime is longer than
the selling season itself. Hence, the manager has to order and stock inventory prior to
the selling season. Alas! Demand is also uncertain. How many items should the
manager order before the selling season starts?

Several managers from various industries face such notoriously difficult decisions.
A newsvendor located at the corner of your street has to decide how many copies of a
particular issue of a magazine she has to stock. A garment retailer who sells fashion
items with a short selling season (Zara’s winter collection and Zekitriko’s swim suit
collection to name a couple) has to decide how many items she has to stock in her retail
store. A toy manufacturer has to decide how many items to manufacture in Asia and
ship to North American retailers who sell during the Christmas season. A grocery store
manager who has to sell some of his products on a given day is another example.

All these managers face two fundamental trade-offs. At the end of the selling
season, if the realized demand turns out to be less than the available inventory, then the
manager needs to salvage leftover inventory at a discounted price. For certain products,
she may have to pay a disposal fee for leftover inventory. On the other hand, if realized
demand exceeds the available inventory, then the manager loses customers and hence
opportunity to make more money. Subsequently she may build a reputation of not
having enough products on hand to satisfy fickle consumer demand. The following
shopping season, consumers may choose to visit other retailers.

Historically, researchers motivated this single-period inventory management


problem under demand uncertainty with a newsvendor narrative. We will follow the
tradition here. Consider a newsvendor who purchases local newspapers for 20 cents
each in the morning and sells for 50 cents each throughout the day. He can return
unsold copies to his supplier for 11.5 cents. Demand for newspapers is uncertain.
However, the newsvendor estimates that during any given day 100 people on average
buy the newspaper.

The newsvendor has to decide how many copies to buy from the regional distributor
early in the morning. An obvious solution is to buy 100 copies, the average demand.
Note, however, that not having a copy when a customer wants one costs the
newsvendor

Copyright © 2004 by Özalp Özer 5


cu = 50 − 20 = 30 cents.

We refer to this cost as the cost of underage. On the other hand, having an unsold copy
at the end of the day costs the newsvendor

co = 20 − 11.5 = 8.5 cents.

We refer to this cost as the cost of overage. Intuitively, the vendor should buy more
than the average demand because the cost of underage is much higher than the cost of
overage. This observation suggests that optimal number of papers to stock should
depend on the costs of overage and underage.

Question: Which of the two costs are reflected in your company’s profit and
loss statement? Do you think the way these costs are reported in the P&L
statement affect inventory related decisions?

The cost of underage represents the marginal benefit of having one more unit of
inventory in stock when needed. Assume that the newsvendor is already committed to
stocking Q = 100 units in the morning. During the day, if she has one more unit of
paper in stock, she may earn cu = 30 cents more. From the historical data, the
newsvendor estimates that demand during a given day exceeds Q = 100 units with
probability 1 − F (Q) = Prob(Demand > Q units) = 50%. She can sell this additional unit
only if demand exceeds Q units. Hence, the expected marginal benefit of stocking one
more unit is

cu * (1 − F (Q)) ,

which is equal to 30 * 50% = 15 cents. On the other hand, the cost of underage represents
the marginal cost of having one more unit of inventory in stock than is not needed. The
expected marginal cost of having an additional unit is

co * F (Q) ,

which is 8.5*50% = 4.25 cents (Why?). The optimal stocking level should strike a
balance between the marginal benefit and cost of having an additional unit. The
newsvendor can increase his stocking level one unit at a time until the marginal benefit
of doing so is less than the marginal cost. In this particular example, increasing the
stock level from 100 to 101 is worthwhile. The optimal order size Q* should satisfy the
following

cu * (1 − F (Q)) = co * F (Q) .

Rearranging the terms we arrive at the famous critical ratio.

Inventory Management under Demand Uncertainty 6


cu
F (Q * ) = .
cu + c o

This ratio is the probability of satisfying all demand during the day if the
newsvendor stocks Q * units at the beginning of the day. For our example, this ratio is
30/ (8.5+30) = 0.78. Thus, the newsvendor should stock enough newspaper such that
the probability of no stock out is 0.78 * 100% = 78%.

Next the newsvendor must convert this desired probability of stockout into stocking
quantity. This conversion depends on the distribution of demand or the probabilistic
forecast. Based on the historical data and her judgment, the newsvendor estimates that
demand for the newspaper follows the probabilistic forecast provided in Table 1, from
which the optimal stock quantity is Q * = 110 units. Given this quantity, her expected
cost of underage and overage is given by

EC (Q * ) = 8.25 * {(110 − 40) * 3% + (110 − 50) * 6% + ... + (110 − 100) * 20%}


+ 30 * {(160 − 110) * 3% + (150 − 110) * 6% + ... + (130 − 110) *13%}
= 335.25

Notice that instead of a marginal cost benefit analysis, the newsvendor can also
calculate her expected cost under all possible stocking quantity scenarios that are
Q = 20,...,160 units. She can choose the quantity that minimizes the expected cost. This
solution gives us the same answer as the marginal cost/benefit analysis.

Assume that demand for the newspaper follows a Normal distribution with mean
100 and standard deviation 20. We use the Normal table to convert this critical ratio to
an inventory level. For 0.78, z=0.7721. Thus, Q* should lie 0.7721 standard deviations
from the mean demand 100, which is 100 + 0.7721*20 = 115. Note that, given a
demand distribution, we convert a desired probability of not stocking out (78%) to an
inventory level (115 units).

We make the following observations:

1. The newsvendor orders more than the average demand. The average demand is
100 units, but the optimal inventory decision is 110 units. This outcome is due to
two reasons. First, the cost of underage is larger than cost of overage. The
opportunity cost of not having one more unit is larger than the potential cost of
creating this opportunity. Second, when the probabilistic forecast is symmetric
around the average, the chance of observing a demand smaller than the average is
less than 50% (Why?). Intuitively, the newsvendor has a higher chance of making
more money than losing if she orders more than the average demand.

2. The marginal analysis suggests that allowing for the possibility of a stockout
could be a better choice than promising to meet 100% of demand. In other words,
stocking inventory to meet possibly less than 100% of realized demand, and hence

Copyright © 2004 by Özalp Özer 7


losing a few customers, may not be a bad idea. However, note that the above
underage cost does not include loss of good will or shortage cost. The newsvendor
can add an ill will cost/ shortage cost to modify the cost of underage. A larger ratio
of cu / c o implies that the newsvendor should carry more inventory. If the sales
price increases, hence the cost of underage increases, then the newsvendor should
stock more newspapers. On the other hand, if the purchase price increases, the
newsvendor should stock fewer newspapers.

________________________________________________________________

Exercise: If your local grocery store manager runs out of stock very rarely, can
you say that you are paying for a big markup? Why?

Exercise: How does the standard deviation of probabilistic forecast affect the
stocking decision of the newsvendor? If the standard deviation increases from
20 to 40, how much inventory should the newsvendor stock?
________________________________________________________________

3. Consider the above newsvendor example with the probabilistic forecast given in
Table 1. What should the newsvendor do if the critical ratio turns out to be 95%?
For this scenario, we do not have any Q that corresponds to this critical ratio. In this
case, the newsvendor should pick a quantity that gets her closer to this ratio and
results in lower expected cost. In this example, she can calculate expected cost both
for when Q=150 and for when Q=130. Next, she can choose the stocking level that
results in the smallest expected cost.

3. Multiple Order Opportunities


In the previous section, we presented an inventory model for a manager who faces
uncertain demand during a single selling season. We used marginal cost-benefit
analysis to solve the newsvendor problem in which the newsvendor can make only a
single stocking decision for the entire selling season. We mentioned several practical
examples in which the manager can use the newsvendor model to decide on a stocking
level. However, in several other practical situations the manager may need to order
products repeatedly at any time during the year. Next, we relax the single period
assumption and assume that the manager faces an ongoing inventory control problem.

Consider, for example, the distributor of Coke. The distributor faces random
demand for the product and receives the supply from the manufacturer. Delivery often
received after a fixed leadtime. For example, when the distributor places an order, he
receives the orders two weeks later. Demand is uncertain. Hence, the distributor may
need to hold inventory even if no fixed cost is charged for ordering the product. Three
main reasons for this are as follows.

Inventory Management under Demand Uncertainty 8


1. In order to meet demand during the leadtime, the distributor has to carry
inventory. Without any inventory at the distributor, a customer would have to
wait as long as it takes to receive an order from the manufacturer.
2. Since demand is uncertain, the distributor may want to protect the system
against larger-than-average demand during the replenishment leadtime.
3. The distributor may want to balance fixed ordering cost and inventory holding
cost as in the case of EOQ model.

Next we provide inventory models that help the distributor to better manage
inventory. She has to decide (i) how often the inventory status should be determined (ii)
when to place a replenishment order and (iii) how many cases of Coke to order.

3.1 Lost Sales versus Backlogging

For an ongoing inventory problem, the manager carries unsold inventory over to the
next time unit and incurs a holding cost. On the other hand, if demand for a product
exceeds available stock in the store, then the manager faces a stockout situation and
incurs a shortage cost.

The severity of the shortage cost depends on how the unmet demand is treated. Two
extreme alternatives are the following. All unmet demand can be backlogged; that is,
this demand will be satisfied through orders arriving in the future. Alternatively, all
unmet demand can be lost; that is, this demand disappears. Reality often lies
somewhere between these two extremes; that is, part of unmet demand is backlogged
while the rest is lost. The inventory policies discussed in this unit can safely be applied
to the full backlogging case. Analysis of a lost sale case is beyond the scope of this
unit. For a high service system, the model discussed here can be used as an
approximation for the lost sales problem. By service, we mean the percentage of
demand satisfied through on-hand inventory. Next, we define four important inventory
related concepts.

On-hand inventory is the physical inventory available in a store that the store can
use to satisfy demand directly. This quantity can never be negative.

Inventory on-order is the inventory that has been ordered but not yet received.

Inventory position= On-hand inventory + inventory on order - backorders

If on-hand inventory were positive, no demand would be backordered. Hence on-hand


inventory and backorders cannot both be positive at the same time. Inventory position
is an important consideration in deciding when to replenish on-hand inventory.
Ignoring the inventory on order in replenishment decisions is a very common error.

Safety stock is the inventory held in addition to average demand to hedge against
costly stockouts due to demand uncertainty. In other words, safety stock is the average

Copyright © 2004 by Özalp Özer 9


stock held just before receiving a replenishment order. The amount of safety stock
depends on demand uncertainty and service goals set by the management.

Question: Does the manager need to carry safety stock when demand is known
with certainty? Why?

3.2 Leadtime Demand Uncertainty

Consider a product for which the replenishment leadtime is zero and the fixed cost of
ordering is negligible. Before placing an order, the manager can wait until on-hand
inventory is depleted. By doing so, she avoids carrying unnecessary inventory while
satisfying all demand given that replenishment orders arrive immediately. Hence, zero
leadtime with zero fixed ordering cost make holding safety stock unnecessary.

Question: Can you think of a product for which the leadtime is zero?
Answer: Water supply.

For an ongoing inventory problem, the key issue is to protect the system against
uncertainty in demand over the replenishment leadtime. Let

AVG the average demand per time unit,


STD the standard deviation of demand per time unit,
L the replenishment leadtime.

The average demand during the leadtime is the product of the average demand per
period and the leadtime.

DLT = L * AVG

The standard deviation of leadtime demand is the product of the standard deviation of
demand per time unit and the square root of the leadtime.

σ LT = L * STD .

If the average demand is 40 cases per week, the standard deviation is 10 cases per week
and replenishment takes 4 weeks, then the average leadtime demand is 160 cases and
the standard deviation of leadtime demand is 20 units. (Remember the rule: do not add
standard deviations! Add the variances and take the square root to obtain the total
standard deviation.) The standard deviation of leadtime demand captures the demand
uncertainty over the leadtime.

As in the newsvendor problem, if the demand per unit time follows a Normal
distribution with AVG and STD then the leadtime demand is normally distributed with
mean L*AVG and standard deviation L * STD . In fact, researchers have shown that for
long replenishment leadtime problems, using a Normal distribution to model demand
uncertainty works very well. In other words, if the manager faces a long replenishment

Inventory Management under Demand Uncertainty 10


leadtime, she can use historical data to estimate the mean and the standard deviation of
demand and safely assume that demand follows a Normal distribution, that is, a bell-
shaped curve.

As we will see, anything management can do to reduce demand uncertainty over the
leadtime improves inventory performance. For example, helping suppliers to improve
operations may reduce demand variability during the replenishment leadtime. Such
improvements lead to better inventory performance as long as management is already
using a sound inventory management method.

The above discussion assumes that the leadtime is known with certainty. However,
leadtimes may be uncertain as well. For example, transportation time may depend on
weather conditions. If successive replenishment orders do not cross each other before
reaching to the store, and successive leadtimes do not depend on each other, then we
use the following formula to calculate the standard deviation of leadtime demand.

AVGL * STD 2 + AVG 2 * STD 2L ,

where AVGL is the average lead time and STD L is the standard deviation of the
leadtime.

3.3 Continuous Review versus Periodic Review

Recall from our discussion in the previous unit that when demand is known,
determining the order size is equivalent to determining the order interval. When
demand is uncertain, however, this equality does not hold. The manager has to decide
how often to review the inventory status and place replenishment orders. We either fix
the order size or fix the order interval. This distinction gives rise to the two main
inventory review systems. The first is the continuous review (fixed-order-quantity)
system and the second is the periodic review (fixed-time-period) system.

In a continuous review system, the manager knows the inventory status at all times.
Demand is recorded as it occurs. Due to the advances in information technology,
continuous review systems are easy to implement. The computerized checkout systems
in supermarkets, for example, instantly record all transmissions and update the
inventory level. Note, however, that knowing the inventory level at all times does not
prevent a stockout situation. (Why?) Continuous review is also sometimes known as
transaction reporting because all demand is recorded as it occurs.

In a periodic review system, the manager reviews inventory only at discrete points
in time. The manager counts inventory on-hand and places replenishment orders at
specific time intervals, such as at the end of each day, week or month. When necessary,
during these review periods, management places an order to bring the inventory
position back up to a desired level. A familiar example of a periodic review system is a
vending machine. The vendor company periodically checks and fills the vending

Copyright © 2004 by Özalp Özer 11


machine, perhaps at the end of each day. If a machine runs out of Coke in the morning,
it remains empty until the end of the day.

A periodic review system requires less bookkeeping then a continuous review


system but it also provides less direct control. To hedge against stockouts in between
inventory reviews, the manager stocks more inventory compared to a continuous
review system. A periodic review system may also provide economies of scale in fixed
ordering cost when a group of items are reviewed at the same points in time. Often
continuous review system is recommended for a slow moving but “critical” item.
Criticality of an item could be due to the dollar value invested. Alternatively, the item’s
availability could be critical for other operations. Firms often use continuous review
systems for spare parts inventory.

Question: Which review system would your supplier wish that your company is
using?
Answer: The supplier may prefer determined order intervals of the periodic
review system. With a continuous review system, he needs to be ready to ship
anytime.

Question: Consider a supermarket that uses a computerized checkout system


with inventory replenishment decisions made once each day after the store
closes. Is this a continuous review system?
Answer: No. The review system is still periodic. To be a continuous review
system, the manager should be able to place a replenishment order at any point
in time. Knowing the inventory status is not sufficient, unless the manager can
react to it. The above system is a periodic review system with periods of 1 day.

Question: If you were the manager of a mega supermarket, which of the


following products would you consider reviewing continuously Crest
(toothpaste), Gillette (razor), IBM Thinkpad (notebook computer).

3.4 Continuous Review: Reorder-point, Order-quantity (R,Q) Policy- Two


Bin Policy

Here we introduce the reorder-point order-quantity policy, which is an extension of the


EOQ model. Under this policy, the manager orders Q units whenever inventory
position falls below a pre-determined reorder point R. Orders arrive after L units of
time. For example, she orders 50 units whenever the inventory position drops to 10
units. This policy is also known as the two-bin policy. The manager stores R units in
the first bin. Whenever demand depletes the inventory in the second bin, the manager
places an order of size Q. During the replenishment leadtime, she satisfies demand from
the first bin. When the order arrives, he first fills the first bin and uses the remaining
order to fill the second bin.

Next we state our assumptions. We consider a single item. Demand is uncertain


over time; however, we assume that average demand is constant overtime. Demand

Inventory Management under Demand Uncertainty 12


arrives one unit at a time; that is, each customer demands one unit. The replenishment
leadtime L is constant. Inventory holding cost h is charged per item per unit time. The
manager incurs the fixed cost K per order. Management imposes a service level α,
which is defined as the probability of not stocking out. This probability can be
interpreted as the proportion of cycles in which no stockouts occur. This service level
is particularly relevant when a shortage has the same consequence independent of its
magnitude. Our objective is to find the reorder point R and an order quantity Q that
satisfies the required service level while keeping the inventory related costs to a
minimum.

In Figure 2, we plot inventory position and inventory on hand as a function of time


when the manager employs an (R,Q) policy to manage a single product.

Inventory Level
Inventory Position

R+Q

Inventory On-hand

Time

FIGURE 3: Inventory position and inventory on-hand over time under an (R,Q) policy

We need to determine the optimal values for two parameters, which are R and Q.
The EOQ formula provides a reasonable approximation to set the order quantity. We
set the reorder point to

R = DLT + ss

where ss stands for safety stock. Hence, the reorder point is the sum of two parts. The
first is the average demand during the replenishment leadtime, which equals L*AVG.
This part ensures that enough inventory is available to satisfy average demand until the
next order arrives. The second is the safety stock, which is necessary to protect the
system against demand uncertainty during the replenishment leadtime, that is, to protect
the system against larger-than-average demand during the replenishment leadtime. The
safety stock level is set to ensure that demand during the leadtime is less than or equal
to R during at least α of the order cycles. For example, if customer demand were
satisfied during 90 order cycles out of 100, then α is equal to 90%. In other words, R
satisfies the following equality

Copyright © 2004 by Özalp Özer 13


F ( DLT + ss) = α .

Similar to the newsvendor problem, the manager needs to know the leadtime
demand distribution to decide on the safety stock. Assume that leadtime demand
follows the probabilistic forecast provided in Table 1. If the service level is 97% then
safety stock is given by

F ( DLT + ss) = α ⇒ ss = 150 − 100 = 50 units.

If leadtime demand follows a Normal distribution with mean D LT and standard


deviation σ LT = L * STD then the safety stock is given by

ss = z * σ LT ,

where z is the constant associated with the given service level.

For a given R and Q, we calculate the average total annual cost as:

D Q
TC ( R, Q ) = ( ) K + ( + ss ) h .
Q 2
The first is the average set up cost and the second is the average holding cost. Note that
the average holding cost includes the cost of holding safety stock. Larger safety stock
results in higher holding costs. Cost of holding inventory is due to cycle stock, which is
defined in Unit 7, as well as safety stock. Recall that when demand is known, inventory
holding cost is due to the cycle stock only.

________________________________________________________________

Example: Consider a warehouse located in Oakland, CA. The warehouse


manager satisfies demand from grocery stores in the greater bay area. The
manager, John, is a good friend of yours. He asked you to help him establish an
(R,Q) policy for ColaTurka, a soft drink produced in Turkey. He mentioned that
an order from Istanbul arrives in Oakland after 2 weeks. From historical data,
demand for ColaTurka seems to follow a Normal distribution with mean 60,000
cases per year and standard deviation 1224 cases per year. When he places an
order, he incurs a fixed cost of $250 per case, and the holding cost is $1 per case
per unit time. John wants to set a high service level to make sure that customers
recognize John as a reliable source for ColaTurka. John wishes to have 97.5%
as his target service level. You would like to help him calculate the safety stock,
reorder point, order quantity and total average inventory cost under the (R,Q)
policy.

Answer: Since demand per unit time is Normal, leadtime demand also follows a
Normal distribution with

Inventory Management under Demand Uncertainty 14


2 2
D LT = * 60,000 = 2500 and σ LT = * 1,224 = 250 .
48 48

To calculate the desired level of safety stock, we obtain from the standard
Normal table the constant z that corresponds to 97.5%. It is 1.96. Hence,

ss = 1.96 * 250 = 490 units

Therefore, the reorder point is R=2500+490=2990 units. Using the EOQ


formula, the economic order quantity is Q * = 2 * 250 * 60,000 = 5477 cases. The
average inventory cost is
TC(5477) = (60,000 / 5477) * 250 + (5477 / 2 + 490) *1 = 5967 $/year. Note that this
cost excludes the variable purchasing cost, which has no impact on the
parameters of the inventory policy. (Why?)
________________________________________________________________

Question: Can a manager ever guarantee that probability of stockout is zero


when demand follows a Normal distribution?

Answer: The Normal distribution extends to infinity, a positive (but extremely


small) possibility of large demand always exists. However, high the service
level is, the manager cannot wipe out stockouts completely.

Exercise: Suppose that leadtime demand for ColaTurka follows the


probabilistic forecast in Table 1. How much safety stock should John hold to
have a 100% service level, that is, a stockout probability is zero. What is the
corresponding inventory holding cost if the fixed cost of ordering is $250.

3.5 Continuous Review: Order-Point, Order-Up-to-Level (s,S) Policy

The order-point, order-up-to policy is similar to an (R,Q) policy. We use an (s,S) policy
when the quantity demanded can be larger than one unit. For example, an arriving
customer may demand 15 units. For such systems, we set s=R and S=s+Q. The
manager places a replenishment order whenever the inventory position falls below the
order point s. Note that the order size under an (s,S) policy is variable. Due to a larger
demand, the inventory position may fall significantly below s. In this case, the order
quantity will be larger than Q units.

3.6 Continuous Review: Base Stock Policy (Order-up-to S Policy)

When the fixed cost of ordering is relatively unimportant or when it is negligible, we


use what is called a base stock policy. Under this policy the manager places a unit order
whenever the inventory position falls below S units. Note that this policy is a special
case of the (R,Q) policy where S=R and order quantity Q=1. Consider, for example, a
system with S=5 units. Assume also that the current inventory position is 5 units. If a

Copyright © 2004 by Özalp Özer 15


unit is demanded, the inventory position drops to 4 units. Next the manager orders one
unit to bring the inventory position back up to 5 units. Under a base stock policy, which
is also known as an order-up-to policy, the inventory holding cost is low. Because of
frequent replenishment, the manager does not need to keep large safety stock. However,
the number of orders issued is high. This policy should be used when orders are
infrequent and items are expensive.

Question: Can you think of an example from your company for which base
stock policy could be a good candidate?

3.7 Periodic-review, Order-up-to (T,S) Policy

Under a (T,S) policy, the manager reviews the inventory every T units of time. If the
inventory position is less than S, she orders enough units to bring the inventory position
up to S. In many cases, the review period T is given. For example, a firm may already
have a fixed review cycle due to physical constraints or a regular delivery schedule.
Alternatively, the manager can set T to balance holding cost and ordering cost using the
EOQ formula; that is, T=Q*/AVG where Q* is the solution to the EOQ formula and
AVG is the average demand. Note that, under this policy, the system replenishes
inventory every T+L periods of time, where L is the replenishment leadtime. Similar to
the logic we described in Section 3.4, the manager has to protect the system against
demand uncertainty during the leadtime plus a review period. The average demand for
T+L periods is

DTL = (T + L) * AVG

and the standard deviation is

σ TL = T + L * STD .

Now we can calculate the safety stock

ss = zσ TL ,

where z is the constant corresponding to the desired service level. We set the order-up-
to level to

S = DTL + ss .
______________________________________________________________

Exercise: The owner of the vending machines at your company visits and
replenishes the machines every day even if vending machines are full. He
estimates that demand for Pepsi during any given day follows a Normal
distribution with average demand 200 units per week and standard deviation 24
units per week. He also wishes to have a 97.5% service rate. To achieve this

Inventory Management under Demand Uncertainty 16


service level, what should be the corresponding order-up-to level? What is the
review period?
______________________________________________________________

4. Multiple Locations and Risk Pooling


Consider the distribution problem faced by Hewlett Packard (HP), who distributes
printing equipment to two customer regions in Europe. Customers located in France
receive items directly from the first warehouse in Paris. Customers located in Italy
receive printers directly from the second warehouse in Milan. Each warehouse carries
over 1000 SKUs.

The VP of the Supply Chain, Emily Chung, is considering consolidating the two
warehouses into one warehouse, which will satisfy demand from both France and Italy.
She reasons that such a consolidation would result in reduced overhead and better
coordination of operations with manufacturing partners located in China. Under the
current system it takes 4 weeks to ship products from China to the warehouses in
Europe. She estimates that this replenishment leadtime will remain the same under the
new system. She also claims that a higher-than-average demand at one location may be
offset by a lower-than-average demand at the other location. As a result, she expects to
carry less safety stock. She is also aware of the current distribution system’s advantage
over the single warehouse system. Under the current system, warehouses are located
closer to the end customer. This proximity enables HP to satisfy demand faster.
Nevertheless, she would like the new senior operations manager, Otis, to calculate
inventory related gains from such a consolidation. She states that under the current
system service level is 95%; that is, each warehouse carries enough safety stock for A
level items so that the probability of stocking out is 5%. Emily’s only requirement is
that under the new warehouse system the service level remain the same.

Otis starts his analysis by considering the inkjet printers. In France, annual demand
for inkjet printers follows a Normal distribution with mean 15,000 units and standard
deviation is 1,000 units. On the other hand, In Italy, annual demand for HP printers
follows a Normal distribution with mean 10,000 and standard deviation 900 units. Otis
calculates the constant z from the standard Normal distribution Table. For a service
level of 95%, z should be set equal to 1.65. Hence, the current safety stock at each
warehouse should be

ss F = 1.65 * 4 / 48 *1000 = 476 units


and
ss I = 1.65 * 4 / 48 * 900 = 429 units.

Total safety stock for inkjet printers held in Europe is 905 units. Under the new
warehouse system, demand will be satisfied through a single warehouse. Total demand
follows a Normal distribution with mean 15,000+9,000 units. Six years after

Copyright © 2004 by Özalp Özer 17


graduation, Otis still remembers that his professor used to say, “Add variances not
standard deviations!” Therefore, total demand variance is 1000 2 + 900 2 = 1,810,000.
Total safety stock for the consolidated warehouse is

sscon = 1.65 * 4 / 48 * 1,810,000 = 641 units.

Otis also finds out that the cost of holding inventory is $30 per inkjet printer per
unit time for both warehouses in Italy and France. Percentage savings in safety stock
30 * (905 − 641)
dollars for the inkjet printer is = 29% . He knows that HP’s total
30 * 905
inventory investment in France and Italy is approximately $6Million. If HP were to
achieve similar inventory reduction across all product categories, he concludes that the
consolidation has the potential to slash inventory related costs by about $1.74 Million.
________________________________________________________________

Exercise: After Otis runs all calculations for a sample of 100 products, he
reports final results to Emily. She expresses her appreciation about such a
thorough analysis. However, Emily points out that under this consolidation, HP
should be saving also from fixed ordering costs. She asks Otis to investigate
whether the order quantity for the inkjet printer remains the same after the
consolidation. She tells Otis that the fixed cost of ordering inkjet printers is
$500 per order, independent of the distribution strategy. Do you agree with
Emily? Why/Why not?
________________________________________________________________

The above analysis assumes that demand for HP products in Italy and in France are
independent. In other words, observing a higher-than-average demand at one location is
independent of observing a lower-than-average demand in the other location. However,
if demand at each location were positively correlated, then demand realizations would
be similar. Observing a higher than average demand at one location would signal a
similar pattern in the other location. Thus, the benefits from risk pooling diminish if the
demands in separate locations are positively correlated.
________________________________________________________________

Exercise: Consolidating warehouses results in consolidation of customer


demand. This consolidation reduces uncertainty in demand, hence reducing
inventory related investments. Similarly, we can better forecast demand for a
product category (HP inkjet printers) than a specific model (HP inkjet printer
1200s). In other words, aggregating demand across product categories provides
more accurate forecast information. Given this discussion can you think of
means to reduce work-in-process inventory for HP inkjet printers?
________________________________________________________________

Inventory Management under Demand Uncertainty 18


5. Exchange Curves
An exchange curve shows the relationship between the service level and the safety
stock for a given distribution of uncertain demand (or probabilistic forecast). It provides
a means for determining the preferred level of service for a given dollar investment in
inventory. Recall the relationship between safety stock and service level. Setting a high
service level requires a manager to carry substantial safety stock. The holding cost of
safety stock is

HC ( z ) = h * z * σ LT .

To delineate an exchange curve, we increase the service level and calculate the holding
cost using the above equation. Figure 4 is an example of an “exchange curve”.
Cost of Holding Safety Stock

… 90% 95% 100%


Service Level
FIGURE 4: Exchange Curve

Using exchange curves, management can judge whether current service levels are
supported by sound stocking decisions. Consider a firm whose inventory investment
and service level for the same product corresponds to point A in Figure 4. What can be
said about this firm? The firm can improve its inventory management practices and
reduce inventory investments with the same level of service.

The above analysis is for a single item. A similar approach, however, can be used to
construct exchange curves for aggregate values. Such aggregate measures can be used
to compare a firm’s performance to the performance of its competitors.

An exchange curve also enables a firm to quantify the effect of strategies to


improve inventory performance. A firm, for example, might be considering an
investment in Electronic Data Interchange. EDI works by providing a collection of
standard message formats for businesses to exchange data via any electronic messaging
service. First, notice that EDI can reduce the replenishment leadtime, hence reducing
demand uncertainty over the leadtime. This improvement reduces the safety stock. An
investment in EDI shifts the exchange curve downwards across all desired service

Copyright © 2004 by Özalp Özer 19


levels. The manager can carry out a scenario analysis for possible reductions in
replenishment leadtime and uncover the inventory related savings by investing in an
EDI initiative.

Question: Can you think of other IT investments that may shift the exchange
curve downwards?

6. Conclusion
In this unit, we introduce you to some of the most important inventory management
tools for uncertain demand environments. One important use of inventory models and
tools is the insight we obtain from our analyses. The tools enable a better understanding
of the role of safety stock, the relationships between different parameters, and the
sensitivity of policy parameters to inaccuracies in data.

Now, we know how to set appropriate inventory levels, how to monitor them and
when to place replenishment orders. We can quantify and estimate the impact of not
having inventory available when needed. To avoid costly stock outs, we can choose the
best safety stock levels. Note, however, that all of our discussion so far addresses a
single firm with a manager who has a single objective maximizing firm’s stakeholders’
value by better managing inventories.

To stay competitive, firms need to improve the processes within the firm as well as
across the global supply chain. Often supply chains are not vertically integrated.
Different firms own different parts of the supply chain and these firms often have
selfish objectives. Coordination and synchronization between supply chain members is
often difficult, leading to problems such as the “bullwhip effect”. The following unit
will touch upon some of these issues.

Suggested Reading: Warren Hausman, 2003. Supply Chain Performance Metrics.


Kluwer Academic Publishers. Chapter 3 in “The Practice of Supply Chain
Management”, edited by Corey Billington, Terry Harrison, Hau Lee and John Neale.

Inventory Management under Demand Uncertainty 20

You might also like