Operations Management

Professor Michael Pinedo

Session 22

Inventory Management
under Demand Uncertainty

Perishable Inventories and

Revenue (Yield) Management
1
 Demand Uncertainty
• Cisco’s 2.2 B $ inventory write-off in 2001.
• A must-read: http://www.cio.com/archive/080101/cisco.html
• L.L.Bean Case : estimated cost of lost sales and excess
inventory 20 M $
• Mark-downs: clothing, automobiles…
• Boeing: 1.6 B $ write-off in 1997
• On a typical afternoon in a typical U.S. supermarket, 8.2%
of the items are out of stock and this number is nearly
doubled for items that are advertised.
• In 34% of stock-outs, consumers refuse to buy an
alternative and often take their business to the competition.
• The costs of stock-outs in U.S. supermarkets alone are
estimated to be $7 to $12 billion of sales (Anderson
Consulting 1996).
• Sony PlayStation: Stock-outs in 2000
 Newsvendor Model
• Suppose we must decide a suitable stock level for a
perishable product which has historically random
demand. (Example: Newspapers)
• At the beginning of each day, the newsvendor
determines how many newspapers to buy. (One time
procurement)
• Assume: Once the newsvendor runs out of stock, it is
impossible to get more newspapers.
• How should we approach making this stocking decision?
 Define an objective
 Identify what data are required
 Measure consequences of decision
 Demand Time Series

65
60
55
50
Historical Demand

45
40
(units)

35
30
25
20
15
10
5
0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
Period

Source: Professors John Muckstadt, David Murray, Jim Rappold

 Demand Time Series

70 3000
Historical Demand
Cumulative Demand
60
2500

Cumulative Units Demanded

50
2000
Units Demanded

40
1500
30

1000
20

500
10

0 0
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97
Period
Source: Professors John Muckstadt, David Murray, Jim Rappold
 Cost Data

• What are the cost implications of our decision?

• Example: Purchase Cost = 1 $
• Example: Selling Price = 10 $
• Leftover newspapers can be scrapped (no cost/revenue)
 Example:Cost of having too much stock: O = $ 1 per
unit (overage cost)
 Example: Cost of having too little stock: U = $ 9 per
unit (underage cost)

Source: Professors John Muckstadt, David Murray, Jim Rappold

 Decision Structure
• Let D represent the customer demand that will occur
(random variable).

• Let R represent our stocking decision.

• What if we had stocked 30 units each period for the

previous 100 periods?
¾ What would the costs have been each period?
¾ What fraction of the time would we have had
enough stock to satisfy demand?
¾ What fraction of demand would we have satisfied?
¾ What’s the value of this information?

(See OilRigExampleTimeSeries.xls)

Source: Professors John Muckstadt, David Murray, Jim Rappold

 Operating Costs
Stock Level = 30 units
100 $300
Historical Demand
Total Accumulated Costs are $3,708
90 Cost Overage Costs are $542
Underage Costs are $3,166

Overage and Underage Costs

$250
80 Probability of No Stockout is 59%
Historical Demand (units)

Fill Rate is 87%

70
$200
60

($)
50 $150

40
$100
30

20
$50
10

0 $0
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97
Period
Source: Professors John Muckstadt, David Murray, Jim Rappold
 Operating Costs
Stock Level = 40 units
100 $300
Historical Demand
Total Accumulated Costs are $2,012
90 Cost Overage Costs are $1,272
Underage Costs are $740

Overage and Underage Costs

$250
80 Probability of No Stockout is 85%
Historical Demand (units)

Fill Rate is 97%

70
$200
60

($)
50 $150

40
$100
30

20
$50
10

0 $0
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97
Period
Source: Professors John Muckstadt, David Murray, Jim Rappold
 Operating Costs
Stock Level = 100 units
100 $300
Historical Demand
Total Accumulated Costs are $7,190
90 Cost Overage Costs are $7,190
Underage Costs are $0

Overage and Underage Costs

$250
80 Probability of No Stockout is 100%
Historical Demand (units)

Fill Rate is 100%

70
$200
60

($)
50 $150

40
$100
30

20
$50
10

0 $0
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97
Period
Source: Professors John Muckstadt, David Murray, Jim Rappold
 Method 1

• Use a spreadsheet to construct the costs in the last

100 periods (for example) for different stock levels.
• Choose the stock level with the lowest average cost.
• Let us now see two more mathematical models.
 Representing the Decision Problem

• What is the cost if the demand (D) and the supply

(R) were known?
 What if demand exceeds supply?
 What if supply exceeds demand ?

• What is the total cost per period?

 Holding (or overage) and shortage (or
underage) costs

Source: Professors John Muckstadt, David Murray, Jim Rappold

 Representing the Decision Problem
Mathematically
• If D were known, the cost function would be:
C(R,D)=Cost(R,D) = O·Max{0, R-D} + U·Max{0, D-R}
= O [R - D]+ + U [D - R]+

(had too little stock)

(had too much stock)
Total
-U O
Cost
(in dollars)

R-D = 0 units ( R - D ) units

• There exists a fundamental economic tradeoff.

Source: Professors John Muckstadt, David Murray, Jim Rappold

 Random Demand
• Demand may be represented as a probability distribution.
Demand Histogram
7% 7%
Frequency
Normal Probability Distribution
6% 6%

Probability of Occurrence
Frequency of Occurrence

5% 5%

4% 4%

3% 3%

2% 2%

1% 1%

0% 0%
0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54 57 60
Period Demand (units)

Source: Professors John Muckstadt, David Murray, Jim Rappold

 Building a Cost Model
• Recall that the cost of (R) units is:
O $ per unit when supply exceeds demand; and
U $ per unit when demand exceeds supply
• P(D) is the probability that demand is equal to D.
• The cost when demand equals D is given by the relationship:
O [R - D]+ + U [D - R]+
• The expected cost is a weighted average of all possible
demands and costs.

• Remember: D is a random variable. R is a decision variable

Source: Professors John Muckstadt, David Murray, Jim Rappold

 Expected Cost Function
Holding and shortage costs given demand D and stock R.
Cost when stock
equals R and
Expected Cost demand equals D

G ( R) = E[C ( R, D )] = å C ( R, D)×P{D }
All possible
demand values

Probability that
demand equals D

We can now evaluate G(R) for any value of R.

 Expected Cost Function
$300
D ~ N(30,102 )
$250
O = $1 per
$200 unit
U = $9 per unit
G(R)
$150

$100

$50

$0
0 20 40 60 80 100
Inventory Level R (in Units)
R* = 42.8 units

Safety Stock:  R *  mean  13 units

Source: Professors John Muckstadt, David Murray, Jim Rappold

 Question
Let us consider several products all with the same
mean demand.
Assume demands for all these products are normally
distributed.
What can you say about the optimal stocking
level for products with greater variability?
What can you say about the minimum attainable
expected cost for products with greater
variability?

See ExpectedCostCurves.xls (sigma = 10, underage cost = 9)

 Expected Cost Function
$300
$275 Expected Cost Curve for Mean = 30, StdDev = 10
$250 Expected Cost Curve for Mean = 30, StdDev = 20
Expected Cost Curve for Mean = 30, StdDev = 30
$225
Deterministic
Expected Cost ($)

$200
$175

$150
$125
$100
$75
$50
$25
$0
0 10 20 30 40 50 60 70 80 90 100
Stock Level (units)
Source: Professors John Muckstadt, David Murray, Jim Rappold
 Observations
• Select an operating policy based on minimizing the
long-run costs of making a decision.
 Period by period costs may vary substantially.
• The expected cost function is relatively flat about its
minimum value.
 Deviations from the optimum stock level do not
increase the expected cost dramatically.
• The greater the variation, the higher the required stock
level.
• The greater the variation, the higher the expected cost
(or lower the expected profit).

Source: Professors John Muckstadt, David Murray, Jim Rappold

 Method 2

• Construct a probability distribution of demand based

on historical data.
• Construct a spreadsheet to evaluate G(R), the
expected cost incurred by choosing a stock level R.
• Pick the value of R that attains the lowest expected
cost.
• Let us now see if there is a simpler way of
determining the optimal value of R.
 A Formula for Determining the Optimal
Stock Level
• We know (or estimate) the demand distribution.
• We know the overage and underage costs.
• Can you think of a simple way to find a formula for the
optimal stock level using economic arguments?
 Hint: Marginal analysis.
 Determining the Optimal Stock Level
• Minimize the expected holding plus shortage costs:
• Let the optimal stocking level be R*.
• We know that at the optimal level, the marginal benefit of
having an extra unit will be (approximately) the same as the
marginal cost of having that unit.
• Marginal benefit of having an extra unit = U if D > R*.
• Marginal cost of having an extra unit = O if D  R*.
• Expected marginal benefit = U P(D>R*)
• Expected marginal cost = O P(D  R*) Newsvendor
• So, U P(D>R*) = O P(D  R*). Formula
• If D is a continuous random variable, this is the same as
saying P(D  R*) = U/(U+O).
• If demand and inventory are integer valued, R* is the
smallest integer such that P(D  R*)  U/(U+O).
 Question

• Newsvendor Formula: P(D<R*) = U/(U+O).

• If we can salvage left-over inventory at a price
greater than the purchase price, what is the optimal
inventory level to stock?
 Question

• Newsvendor Formula: P(D<R*) = U/(U+O).

• If we can salvage left-over inventory at a price
greater than the purchase price, what is the optimal
inventory level to stock?
• To be meaningful, the newsvendor model requires
that the Overage Cost > 0.
 If not, there is an incentive for us to have infinite
stock!
 Example
• Example 2: An inventory manager at Gap, Inc. is trying to
determine how many units of a new tank top to stock for the
coming season. The demand for this product is forecasted to
be normally distributed with mean 10,000 units and standard
deviation 3000 units. The product costs $22 and sells for
$40. Excess units are sent to an outlet store where they can
be sold for $15 a piece. How many units of the tank top
should the inventory manager buy?

Source: Professor Vishal Gaur

 Example
• Example 2: An inventory manager at Gap, Inc. is trying to
determine how many units of a new tank top to stock for the
coming season. The demand for this product is forecasted to
be normally distributed with mean 10,000 units and standard
deviation 3000 units. The product costs $22 and sells for
$40. Excess units are sent to an outlet store where they can
be sold for $15 a piece. How many units of the tank top
should the inventory manager buy?
• We have overage cost = $(22 – 15) = $7 and underage cost
= $(40 – 22) = $18. Thus, the manager should choose Q
such that Probability [Demand < Q] = 18/(7 + 18) = 0.72.
• For a normal distribution, this corresponds to a Z-score of
0.58.
• Thus, the inventory manager should stock 10000 +
0.58*3000 = 11,740 units of the new tank top.

Source: Professor Vishal Gaur