Newsvendor Problem

Copyright 2011 Clamshell Beach Press, www.ClamshellBeachPress.
com
CBP WP 57-23

THE NEWSVENDOR PROBLEM

Arthur V. Hill

The John and Nancy Lindahl Professor, Curtis L. Carlson School of Management, University of
Minnesota, Operations & Management Science Department, 321 19-th Avenue South,
Minneapolis, MN 55455-0413, USA. Voice 612-624-4015. Email [email protected].

Copyright 2011 Clamshell Beach Press
Revised July 7, 2011

1. I ntroduction

Early each morning, the owner of a corner newspaper stand needs to order newspapers for that
day. If the owner orders too many newspapers, some papers will have to be thrown away or sold
as scrap paper at the end of the day. If the owner does not order enough newspapers, some
customers will be disappointed and sales and profit will be lost. The newsvendor problem is to
find the best (optimal) number of newspapers to buy that will maximize the expected (average)
profit given that the demand distribution and cost parameters are known.

The newsvendor problem is a one-time business decision that occurs in many different business
contexts such as:

- Buying seasonal goods for a retailer Retailers have to buy seasonal goods (sometimes
called style goods) once per season. (A season can be a day, week, year, etc.) For
example, most swimsuits can only be purchased seasonally. If a buyer orders too few
swimsuits, the retailer will have lost sales and dissatisfied customers. If the buyer orders too
many swimsuits, the retailer will have to sell them at a clearance price or even throw some
away. Gupta, Hill, and Bouzdine-Chameeva (2006) extend the newsvendor model to handle
multiple seasons (periods), each with a different price elasticity of demand.
- Making the last buy or last production run decision Manufacturers have to make a last
buy (or last production run) for a product (or component) that is near the end of its life cycle.
If the order size is too small, the firm will have stockouts and disappointed customers. If the
order size is too large, the firm will only be able to sell the items for their salvage value.
Hill, Giard, and Mabert (1989) considered a similar problem within the context of selecting a
keep quantity for an aging service parts inventory.
- Setting safety stock levels A distributor has to set the safety stock level for an item. If the
safety stock is too low, stockouts will occur. If safety stock is too high, the firm has too
much carrying cost. Nearly all safety stock models are newsvendor problems with the selling
season being one order cycle or one review period.
Clamshell
Beach
Press
The Newsvendor Problem

Copyright 2011 Clamshell Beach Press, www.ClamshellBeachPress.com
Page 2
- Setting target inventory levels A salesperson carries inventory in the trunk of a vehicle.
The inventory is controlled by a target inventory level. If the target is too low, stockouts will
occur. If the target is too high, the salesperson will have too much carrying cost.
- Selecting the right capacity for a facility or machine If the capacity of a factory or a
machine over the planning horizon is set too low, stockouts will occur. If capacity is set too
high, the capital costs will be too high.
- Overbooking customers If an airline overbooks too many passengers, it incurs the cost of
giving away free tickets to inconvenienced passengers. If the airline does not overbook
enough seats, it incurs an opportunity cost of lost revenue from flying with empty seats.

All of these newsvendor problem contexts share a common mathematical structure with the
following four elements:

- A decision variable (Q) The newsvendor problem is to find the optimal Q for a one-time
decision, where Q is the decision quantity (order quantity, safety stock level, overbooking
level, etc.). Q* denotes the optimal (best) value for Q.
- Uncertain demand (D) Demand is a random variable defined by the demand distribution
(e.g., normal distribution, Poisson distribution, etc.) and estimates of the distribution
parameters (e.g., mean, standard deviation). Demand may be either discrete (integer) or
continuous. This paper develops the newsvendor models for both cases.
- Unit overage cost (c
o
) This is the cost of buying one unit more than the demand during the
one-period selling season. In the standard retail context, the overage cost is the unit cost (c)
less the unit salvage value (s), i.e., c
o
= c s. The salvage value is the salvage revenue less
the salvage cost required to dispose of the unsold product.
- Unit underage cost (c
u
) This is the cost of buying one unit less than the demand during the
one-period selling season. This is also known as the stockout (or shortage) cost. In the retail
context, the underage cost is computed as the lost contribution to profit, which is the unit
price (p) less the unit cost (c), i.e., c
u
= p c. The lost customer goodwill (g) associated with
a lost sale can also be included (i.e., c
u
= p c + g). However, it is difficult to estimate the g
parameter because it is the net present value of all future lost profit from this customer and
all other customers affected by this customers negative reports (negative word of mouth).

Since c
o
and c
u
are both cost parameters, taxes should be considered for both or neither. Given
that the newsvendor problem is in a single period, cash flows do not need to be discounted.

This paper is intended to give readers both a mathematical and intuitive understanding of the
newsvendor model that is used to solve the newsvendor problem. This model is one of the most
celebrated models in all of operations management and operations research and has been in the
literature for over 100 years (Edgeworth 1888; Arrow, Harris, & Marschak 1951).

This paper presents the newsvendor problem in the standard retail context. The reader is
encouraged to explore the Excel workbook Newsvendor Model.xls from Clamshell Beach Press.
The remainder of this paper is organized as follows. Sections 2 and 3 present the newsvendor
problem with discrete (integer) demand and continuous (non-integer) demand. Section 4

Page 3
presents a simple example with graphs for the continuous demand case using both the triangular
and normal distribution approaches. Section 5 presents a simple way to estimate the critical ratio
that is needed for these two models. Section 6 then discusses behavioral issues related to the
newsvendor problem. Section 7 concludes the paper with a summary of the main concepts.
Appendices 1 and 2 derive the newsvendor model with discrete and continuous demand.
Appendix 3 derives the expected values. Appendix 4 presents the VBA code for the inverse
Poisson CDF and Appendix 5 presents the VBA code for the inverse triangular CDF.

2. Newsvendor problem with discrete demand

The model

When demand only takes on integer (whole number) values, it is said to be discrete.
1
With
order quantity Q and specific demand D, the cost for the one-period selling season is:
2

( ) if
( , )
( ) if
o
u
c Q D D Q
Cost Q D
c D Q D Q
<
=

>
(1)

For discrete demand, the demand distribution is defined by the probability mass function
3
p(D).
The equation for the expected cost, therefore, is given by:

1
0 0
( ) ( ) ( , ) ( )( ) ( )( )
Q
o u
D D D Q
ECost Q p D Cost Q D c p D Q D c p D D Q

= = =
= = +

(2)

The first term in equation (2) is the expected overage (scrap) cost and the second term is the
expected underage (shortage) cost. As shown in Appendix 1, the optimal order quantity Q* can
be found at the Q value where the expected cost function is flat. This is where the expected costs
for Q and 1 Q+ units are approximately equal (i.e., ( ) ( 1) ECost Q ECost Q ~ + ). Therefore, Q* is
the smallest value of Q such that the following relationship holds true:

*
0
( *) ( )
Q
u
D
u o
c
P Q p D
c c
=
= >
+
(3)
Appendix 1 derives equation (3). The value / ( )
u u o
R c c c = + is called the critical ratio or
critical fractile and is always between zero and one. The optimal Q is denoted as Q* and can
be found with a simple search procedure starting at Q = 1 and increasing Q until the above

1
A discrete random variable only takes on integer (whole number) values. This could be based on the Poisson
distribution, another theoretical discrete distribution, or an empirical discrete distribution (using historical data).
2
A mathematically concise expression is ( , ) ( ) ( )
o u
C Q D c Q D c D Q
+ +
= + , where ( ) max( , 0) x x
+
= .
3
The probability mass function p(D) is the probability that demand is exactly the integer D. The cumulative
distribution function P(D) is the probability that demand is less than or equal to D.

Page 4
relationship is satisfied. When c
u
= c
o
, the critical ratio is R = 0.5, which is consistent with the
intuition that suggests that Q* should be equal to the median demand when the costs are equal.

Newsvendor example with the Poisson distribution

For example, a buyer for a manufacturer must decide how much to make with the last
manufacturing run before a product is discontinued. The firm currently has zero in stock and the
forecast for the lifetime demand is 4 = units. (The forecast is the mean of the distribution.)
The demand over the lifetime of the product is assumed to be a Poisson distributed random
variable (a reasonable assumption). The underage cost (c
u
) and overage cost (c
o
) are estimated to
be $1000 and $100 per unit, respectively. The c
u
parameter is large because a stockout will
disappoint customers and because the product will not be manufactured again. The critical ratio
is / ( )
u u o
R c c c = + = 1000/1100 = 0.909. Hill, Giard, and Mabert (1989) developed a decision
support system to help managers solve the newsvendor problem in this business context.

Figure 1 shows the Poisson probabilities p(D) and the cumulative Poisson probabilities P(D).
The optimal (maximum expected profit) value of Q can be found by finding the smallest value of
Q such that ( ) 0.909 P Q > . The optimal value of Q for this problem, therefore, is Q* = 7.

I mplementing the model in Excel

The cumulative Poisson distribution can be implemented in Excel with the function
POISSON(Q, , TRUE). While Excel does not provide a function for the inverse of the
cumulative Poisson, it is easy to find the Q that satisfies equation (3) with a simple search.
Appendix 4 implements a simple VBA function for Excel for the Poisson inverse Cumulative
Distribution Function (CDF) where Q* = poisson_inverse(R, ).

Figure 1. Poisson probabilities with mean =4

D

p(D) P(D)
0 0.018 0.018
1 0.073 0.092
2 0.147 0.238
3 0.195 0.433
4 0.195 0.629
5 0.156 0.785
6 0.104 0.889
7 0.060 0.949 Q*
8 0.030 0.979
9 0.013 0.992
10 0.005 0.997
11 0.002 0.999

0.000
0.050
0.100
0.150
0.200
0.250
0 1 2 3 4 5 6 7 8 9 10 11
Demand (D )
P
r
o
b
a
b
i
l
i
t
y

p
(
D
)

Page 5
Figure 2 shows the expected profit for this example, again showing the optimal value at Q* = 7
units. Notice that the expected profit does not change very much with small deviations from Q*
and that it is better to err on the high side than on the low side for this example. As derived in
Appendix 3, the expected profit is ( )( ( 1) ( )) ( ) p s g P Q QP Q p c g Q g + + + , which is
$3,607 for this example.

Figure 2. Expected profit versus order quantity for the example with Q* =7

3. The newsvendor model with continuous demand

The model

As with the discrete demand case, the cost for order quantity Q and specific demand D is:

( ) if
( , )
( ) if
o
u
c Q D D Q
Cost Q D
c D Q D Q
<
=

>
(4)

We assume that demand (D) is a continuous random variable
4
with density function ( ) f D and
cumulative distribution function ( ) F D . The expected cost function is given by:

0 0
( ) ( , ) ( ) ( ) ( ) ( ) ( )
Q
o u
D D D Q
ECost Q Cost D Q f D dD c Q D f D dD c D Q f D dD

= = =
= = +
} } }
(5)

4
A continuous random variable can take on any real value, including fractional values.
$-
$500
$1,000
$1,500
$2,000
$2,500
$3,000
$3,500
$4,000
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Order Quantity (Q)
E
x
p
e
c
t
e
d

P
r
o
f
i
t

(
$
)

Page 6

This equation is analogous to equation (2) for the discrete demand problem. In order to find the
optimal Q, we take the derivative of the expected cost function and set it to zero to find:

1
( )
( ) (1 ( )) 0
( )
*
o u
u
u o
u
u o
dECost Q
c F Q c F Q
dQ
c
F Q
c c
c
Q F
c c
= =
=
+
| |
=
|
+
\ .
(6)

Testing the second derivative proves that Q* is a global optimum.

As mentioned before, / ( )
u u o
R c c c = + is critical ratio and is always between zero and one. In
order to find Q*, the optimal value of Q, it is necessary to find the Q associated with the
cumulative probability distribution so that ( *) / ( )
u u o
F Q c c c = + . Mathematicians write this as
1
* ( / ( ))
u u o
Q F c c c
= + , where
1
(.) F
is the inverse of the cumulative distribution function (also

called the inverse distribution function).
5
Appendix 2 presents the derivation for equation (6).
Appendix 3 derives expressions for the expectations for the number of units sold, lost sales, units
salvaged, cost, and profit. This appendix also shows the relationship between the expected profit
and expected cost.
I mplementing the continuous demand model with the normal distribution

Microsoft Excel includes the inverses for several cumulative distributions, including the normal,
lognormal, and gamma distributions. For the normal distribution, the Excel function for the
optimal Q
*
is NORMINV(R, , ). For example, a newsvendor problem has costs c
o
= $100 and
c
u
= $1000 and a critical ratio of 0.909 R ~ . The demand is normally distributed with 4 = and
1 o = units. The optimal order quantity is then Q* NORMINV(0.909, 4, 1) 5.34 units. The
Encyclopedia of Operations Management (Hill, 2012) includes Excel functions for the inverse
cumulative distributions for all commonly used continuous probability distributions.

I mplementing the continuous demand model with the triangular distribution

When little or no historical information about demand is available and/or the demand
distribution is not symmetrical, the triangular distribution is a practical approach. An
experienced person (or team) estimates three parameters the minimum demand (D
min
), most
likely demand (D
ml
), and maximum demand (D
max
). It is best to start with D
min
and D
max
so that
people do not anchor on the mode. Excel does not include the triangular distribution or its

5
Mathematicians also write this as * arg min( ( ))
Q
Q ECost Q = .

Page 7
inverse, but the inverse for the triangular distribution is easy to derive and implement (Hill &
Sawaya 2004). With the triangular distribution, the optimal order quantity for critical ratio
/ ( )
u u o
R c c c = + is:

min
min max min ml min ml min max min
1
max max min max ml min max min
max
for 0
( )( ) for 0 ( ) / ( )
* ( )
(1 )( )( ) for ( ) / ( ) 1
for 1
ml
D R
D R D D D D R D D D D
Q F R
D R D D D D D D D D R
D R
+ < s
= =

< <
(7)

Appendix 5 presents the VBA code for this function.

4. Newsvendor example with continuous demand

The problem

A retailing firm buys swimsuits for the summer season. The firm buys its swimsuits from a low
cost provider in Asia, but is only able to make a single purchase per year. The estimated demand
is 5000 units, with a minimum of 2000 and a maximum of 8000 units. The selling price is p =
$20 per unit. The firm pays c = $5.00 per unit. The firm can sell excess inventory outside North
America for a salvage value of s = $2.00 per unit. The management believes that no significant
goodwill is lost with a lost sale (i.e., g = 0). Therefore, the underage cost (cost of a lost sale) is
u
c p c g = + = $15 and the overage cost (cost of one unit of extra inventory) is
o
c c s = = $3.
The critical ratio is then / ( ) 15 / 18 83.3%
u u o
R c c c = + = ~ .

The solution with the triangular demand distribution approach

With the triangular distribution, we have (
min ml max
, , D D D ) = (2000, 5000, 8000). Using the
inverse cumulative distribution function for the triangular distribution (equation (7)), the optimal
order quantity is
1
* (0.833) 6, 268 Q F
~ ~ units and the optimal expected profit is approximately

$69,464. Figures 3 and 4 show the demand and the expected profit graphs with the triangular
distribution for demand.


Page 8
Figure 3. Demand distribution for the example problem with the triangular distribution

Figure 4. Expected profit for the example problem with the triangular distribution

The solution with the normally distributed demand approach

With the triangular distribution, the estimated standard deviation is
max min
( ) / 6 D D o = = (8000
2000)/6 = 1000 units. Using the inverse cumulative normal distribution, the optimal order
quantity is
1
* (0.833) 5967 Q F
~ ~ units and the optimal expected profit is about $70,503. In

Excel Q* can be found with NORMINV(0.833, 5000, 1000). Figures 5 and 6 show the graphs.
Demand Distribution
0.000000
0.000050
0.000100
0.000150
0.000200
0.000250
0.000300
0.000350

2
,
0
0
0

2
,
3
7
5

2
,
7
5
0

3
,
1
2
5

3
,
5
0
0

3
,
8
7
5

4
,
2
5
0

4
,
6
2
5

5
,
0
0
0

5
,
3
7
5

5
,
7
5
0

6
,
1
2
5

6
,
5
0
0

6
,
8
7
5

7
,
2
5
0

7
,
6
2
5

8
,
0
0
0

Demand (D)
D
e
n
s
i
t
y

f
u
n
c
t
i
o
n

f
(
D
)
Expected Profit versus Order Quantity
$-
$10,000
$20,000
$30,000
$40,000
$50,000
$60,000
$70,000
$80,000

2
,
0
0
0

2
,
3
7
5

2
,
7
5
0

3
,
1
2
5

3
,
5
0
0

3
,
8
7
5

4
,
2
5
0

4
,
6
2
5

5
,
0
0
0

5
,
3
7
5

5
,
7
5
0

6
,
1
2
5

6
,
5
0
0

6
,
8
7
5

7
,
2
5
0

7
,
6
2
5

8
,
0
0
0

Order Quantity (Q)
E
x
p
e
c
t
e
d

P
r
o
f
i
t

(
$
)

Page 9

Figure 5. Demand distribution for the example problem with the normal distribution

Figure 6. Expected profit for the example problem with the normal distribution

In this example, the triangular and normal distribution approaches have practically the same
optimal order quantity and optimal expected profit. However, when the most likely value is not
close to the midpoint of the minimum and maximum values (i.e., the distribution is skewed), the
optimal solutions may be far apart. When this is true, the triangular distribution approach will
likely be a better method than the normal distribution approach.

Demand Distribution
0.000
0.010
0.020
0.030
0.040
0.050
0.060
2
,
0
0
0
2
,
4
8
0
2
,
9
6
0
3
,
4
4
0
3
,
9
2
0
4
,
4
0
0
4
,
8
8
0
5
,
3
6
0
5
,
8
4
0
6
,
3
2
0
6
,
8
0
0
7
,
2
8
0
7
,
7
6
0
8
,
2
4
0
8
,
7
2
0
Demand (D)
D
e
n
s
i
t
y

f
u
n
c
t
i
o
n

f
(
D
)
Expected Profit versus Order Quantity
$-
$10,000
$20,000
$30,000
$40,000
$50,000
$60,000
$70,000
$80,000
2
,
0
0
0
2
,
4
8
0
2
,
9
6
0
3
,
4
4
0
3
,
9
2
0
4
,
4
0
0
4
,
8
8
0
5
,
3
6
0
5
,
8
4
0
6
,
3
2
0
6
,
8
0
0
7
,
2
8
0
7
,
7
6
0
8
,
2
4
0
8
,
7
2
0
Order Quantity (Q)
E
x
p
e
c
t
e
d

P
r
o
f
i
t

(
$
)

Page 10
5. Estimating the critical ratio

The critical ratio is normally calculated using the equation / ( )
u u o
R c c c = + , which requires
estimates of both the underage and overage costs. As mentioned in the introduction, in the retail
context, the underage cost is price minus unit cost plus lost goodwill (i.e.,
u
c p c g = + ) and the
overage cost is unit cost less salvage value (i.e.,
o
c c s = ). While price, cost, and salvage are
usually fairly easy to estimate, the lost goodwill (g) is often difficult to estimate, which means
that the underage cost and the critical ratio are also hard to estimate.

Goodwill is hard to estimate because it is difficult to predict how customers react to a stockout
situation. Some disappointed customers might be satisfied with an alternative product from the
retailer or might return at a later date, which means that g is close to zero. However, some
customers might leave the store disappointed and never come back, which means that the retailer
would lose the net present value of the lifetime stream of profit from those customers. Even
worse, some customers might give a bad report to many others, which might further damage the
retailers brand, sales, and profits. Goodwill is also hard to estimate in other newsvendor
problem contexts such as the overbooking context where it is difficult to estimate the lost
goodwill associated with turning away a passenger at the boarding gate for a departing flight.

Another approach for calculating the critical ratio estimates the ratio of the cost parameters
/
u o
r c c = without estimates of either the underage or overage cost. It is easy to show that once r
is known, the critical ratio can be calculated as / ( 1) R r r = + .
6
When / 1
u o
r c c = = , the critical
ratio for the newsvendor problem is R = 1/(1+1) = 0.50, which is the median of the demand
distribution. When r is much greater than one, the critical ratio R is close to 1, which leads to an
optimal Q on the far right tail of the demand distribution. When r is close to zero, the critical
ratio R is close to zero, which leads to an optimal Q on the far left tail of the demand distribution.

For example, it might be difficult to estimate goodwill parameter g and the associated underage
cost c
u
, but the decision makers might be able to estimate that the cost of a lost sale is ten times
higher than the cost of having too much inventory at the end of the period. For this situation, the
ratio /
u o
r c c = is 10 and the critical ratio is / ( 1) R r r = + = 10/(10+1) = 10/11 0.909.

6. Behavioral issues with the newsvendor problem

Reward systems

In this authors experience, decision makers (buyers, analysts, managers) who solve the
newsvendor problem as a part of their job often make bad decisions. In this authors opinion, the

6
Proof:
(1/ ) /
(1/ ) / 1 1
u u o u o
u o u o o u o
c c c c c r
R
c c c c c c c r
= = = =
+ + + +
, where /
u o
r c c = .

Page 11
main issue is that reward systems are not aligned with the economics. In other words, the
decision makers are instructed to optimize expected profit, but are then also told to reduce lost
sales or reduce excess inventory. These mixed messages lead the decision makers to respond to
the voice that is yelling the loudest at the moment and to ignore (or at least discount) the hard
to measure economics.

For example, in a project this author did with a large music retailer, the cost of excess inventory
for new releases was low because the retailer could return CDs to the manufacturer for a small
restocking fee (about 15% of the cost). The cost of a stockout for the retailer was high due to
high margins (about $10 per CD). Applying the newsvendor model, the retailers buyers should
have been aggressively overbuying on a consistent basis. However, excess inventory was easy
to measure and lost sales were hard to measure, which often led buyers to give more weight to
excess inventory and less weight to lost sales in their buying decisions. In other words, it
appeared that buyers were being driven by their reward system to under-buy even though the
economics should have led them to overbuy. This meant that the critical ratio for the buyers was
different from the critical ratio for the retailer, which resulted in a misalignment between the
goals for the buyers and the retail firm. The buyers overage cost was
o
c c s b = + , where b is
the buyers reputational cost per unit remaining at the end of the selling season. The b and g
parameters are both difficult to estimate, but can sometimes be imputed (inferred) from historical
buyer behavior (Olivares, Terwiesch, & Cassorla, 2008).
7

Low margin (critical ratio) conditions

Several research papers have used human experiments to study the behavioral issues with the
newsvendor problem (e.g., Schweitzer & Cachon 2000; Bolton and Katok 2008). Nearly all of
this research has found that subjects tend to under-buy in high critical ratio (high margin)
situations and overbuy in low critical ratio (low margin) situations. This pattern cannot be
explained by risk aversion, risk-seeking preferences, loss avoidance, waste aversion, or
understanding opportunity costs. Moritz and Hill (2010) found that subjects have cognitive
dissonance in low margin situations because they are faced with the dilemma of meeting
customer demand (which suggests a large order quantity) and optimizing expected profit (which
suggests a small order quantity). They conclude that subjects tend to estimate a positive goodwill
parameter (e.g., g > 0), even when told that goodwill is not lost with a shortage (e.g., g = 0).

High margin (critical ratio) conditions

Moritz, Hill, and Donohue (2010) found that decision makers in a high margin condition tended
to anchor on the previous period demand when making newsvendor decisions over several
periods. They also found that the simple three-question cognitive reflection test (CRT)
developed by Frederick (2005) predicted the degree to which decision makers anchored on the
previous period demand, where high CRT people had significantly less anchoring. CRT

7
Do not confuse the buyers reputational cost parameter (b), which is for overbuying, and the retailers lost
goodwill parameter (g), which is for underbuying.

Page 12
measures the degree to which people allow their system 1 (automatic, impulsive) thinking to be
moderated by their system 2 (analytical) thinking. In other words, the CRT measures the degree
to which individuals are more patient and less impulsive when presented with a judgmental task.

7. Conclusions

The newsvendor logic is fundamental to solving many operations problems. The newsvendor
model provides both useful intuition and a useful tool. Explicitly defining the overbuying and
under-buying costs, and calculating the critical ratio can often lead to better economic decisions
than those made only on the basis of experience, intuition, myopic reward systems, politics,
power, or personalities.

If managers are willing and able to make some assumptions about the form of the demand
distribution and estimate the demand distribution and cost parameters, they can imbed the
newsvendor model in a decision support system to help buyers make better economic decisions.
This type of decision support system is particularly valuable for retail style goods where many
decisions have to be made routinely every day and where these decisions have a significant
financial impact on the firm.

The inputs to newsvendor model include (a) the form of the demand distribution (e.g., normal,),
(b) the parameters of the demand distribution (e.g., mean), and (c) estimates of the overage and
underage cost parameters. The goodwill parameter, a component of the underage cost, is often
difficult to estimate. This paper introduced the concept of the buyers reputational cost per
unit remaining at the end of the selling season. This parameter is also difficult to estimate.

Decision makers (buyers, analysts, managers) who solve the newsvendor problem as a part of
their job often make bad decisions. The reward systems should be designed to align buyer
behavior with the firms economic objectives. Also, it appears that in low margin situations,
buyers tend to be biased towards inferring a higher goodwill parameter and in high margin
situations, buyers with low CRT tend to put too much weight on last seasons actual demand.

The newsvendor economic logic appears in many different business contexts such as buying for
a one-time selling season, making a final production run, setting safety stocks, setting target
inventory levels, and making capacity decisions. These contexts all have a single decision
variable, random demand, and known overage and underage costs. The newsvendor model
provides a useful tool for solving these problems and practical insights into how to think about
these problems.


Page 13
8. References

Arrow, K., T. Harris, and J. Marschak (1951). Optimal Inventory Policy, Econometrica, 19
(3) 250-272.
Bolton, G.E. and E. Katok (2008). Learning by Doing in the Newsvendor Problem: A
Laboratory Investigation of the Role of Experience and Feedback, Manufacturing and
Service Operations Management, 10(3), 519-538.
Edgeworth, F.Y. (1888). The Mathematical Theory of Banking, Journal of the Royal
Statistical Society, 53, 113-127.
Frederick, S. (2005). Cognitive Reflection and Decision Making, Journal of Economic
Perspectives, 19(4), 25-42.
Gupta, D., A.V. Hill, and T. Bouzdine-Chameeva (2006). A pricing model for clearing end of
season retail inventory, European Journal of Operational Research, 170 (2), 518-540.
Hadley, G., and T.M. Whitin (1963). Analysis of Inventory Systems, Prentice-Hall, Englewood
Cliffs, NJ.
Hill, A.V. (2012). The Encyclopedia of Operations Management, Financial Times Press, New
York, New York.
Hill, A.V., V. Giard, and V.A. Mabert (1989). A Decision Support System for Determining
Optimal Retention Stocks for Service Parts Inventories, IIE Transactions, 21 (3), 221-
229.
Hill, A.V., and W.J. Sawaya III (2004). Production Planning for Medical Devices with an
Uncertain Approval Date, IIE Transactions, 36 (4), 307-317.
Landsman, Z., and A.E. Valdez (2005). Tail conditional expectations for exponential
dispersion models, ASTIN Bulletin, 35 (1), 189-209.
Moritz, B.B., and A.V. Hill (2010). Asymmetric Ordering Behavior in Newsvendor Inventory
Decisions: Customer Service and Cognitive Dissonance, University of Minnesota
working paper.
Moritz, B.B., A.V. Hill, and K.L. Donohue, Individual Differences in the Newsvendor Problem:
Behavior and Cognitive Reflection, under second review by Management Science
(October 5, 2010).
Olivares, M., C. Terwiesch, and L. Cassorla (2008). Structural Estimation of the Newsvendor
Model: An Application to Reserving Operating Room Time, Management Science, 54
(1), 41-55.
Schweitzer, M.E. and G.P. Cachon (2000). Decision bias in the newsvendor problem with
known demand distribution: experimental evidence, Management Science, 46, 404-420.
Silver, E.A., D.F. Pyke, and R. Peterson (1998). Inventory management and production
planning and scheduling, Third edition, John Wiley & Sons, New York.
Winkler, R.L, G.M. Roodman, and R.R. Britney (1972). The Determination of Partial
Moments, Management Science, 19 (3), 290-296.


Page 14
Appendix 1: Derivation of the newsvendor model with discrete demand
8

The optimal expected cost will be where the expected costs for ordering Q units is approximately
the same as the expected cost for ordering Q + 1 units. This is the point at the bottom of the total
expected cost curve where the curve is flat, which is the global minimum expected cost. Setting
( ) ( 1) ECost Q ECost Q = + and applying equation (2), we find:

1
0 0 1
( )( ) ( )( ) ( )( 1 ) ( )( 1)
Q Q
o u o u
D D Q D D Q
c p D Q D c p D D Q c p D Q D c p D D Q

= = = = +
+ = + +

(8)

However, the left side can be rewritten as
0 1
( )( ) ( )( )
Q
o u
D D Q
c p D Q D c p D D Q
= = +
+

because
0 D Q = when Q D = . Combining terms and defining the cumulative distribution function as
0
( ) ( )
Q
D
F Q p D
=
=
yields:

0 1
0
( ) ( ) 0
( ) (1 ( )) 0
( ) ( )
Q
o u
D D Q
o u
Q
u
D
u o
c p D c p D
c F Q c F Q
c
F Q p D
c c
= = +
=
=
=
= =
+

(9)

Therefore, the optimal order quantity for the discrete demand newsvendor problem can be found
by finding the smallest Q such that
0
( ) ( ) / ( )
Q
u u o
D
F Q p D c c c
=
= > +
, where the quantity

/ ( )
u u o
R c c c = + is the critical ratio (fractile). Note that this result requires no assumptions about
the demand distribution other than it must be a discrete distribution. The next section proves that
expression (9) also holds for continuous demand distributions.

Appendix 2: Derivation of the newsvendor model for continuous demand

For order quantity Q and specific demand D, the cost is:

( ) for
( , )
( ) for
o
u
c Q D D Q
Cost Q D
c D Q D Q
s
=

>
(10)

For continuous demand with density ( ) f D , the expected cost is:

8
The derivations in the appendices show many more intermediate mathematical steps than are normally shown in a
research paper. This is done to help all readers understand the details of the derivations.

Page 15

0
0
0 0
( ) ( , ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ( )) (1 ( ))
( ) ( ) ( ) ( )
D
Q
o u
D D Q
Q Q
o o u u
D D D Q D Q
o o u u
o u o u u
ECost Q Cost D Q f D dD
c Q D f D dD c D Q f D dD
c Q f D dD c Df D dD c Df D dD c Q f D dD
c QF Q c H Q c H Q c Q F Q
c QF Q c QF Q c H Q c H Q c
= =

= = = =
=
= +
= +
= +
= + +
}
} }
} } } }
( )( ( ) ( )) ( )
u
u o u
c Q
c c QF Q H Q c Q = + +
(11)

where ( ) F Q is the demand distribution function evaluated at Q and ( ) ( )
D Q
Df D dD H Q
=
=
}
.
Note: In equation (11), the mean demand is
0
( )
D
Df D dD

=
=
}
and is called the complete
expectation because the range of integration is (0, ) . The partial expectation of demand is
0
( ) ( )
Q
D
H Q Df D dD
=
=
}
with a range of integration (0, ) Q .
9
Given that ( ) H = , it is clear that
( ) ( )
D Q
Df D dD H Q
=
=
}
.

By the fundamental law of calculus, '( ) ( ) F Q f Q = and according to Leibnizs rule
10

'( ) ( ) H Q Qf Q = . Therefore, the first derivative of ECost(Q) with respect to Q is:

( )
( )( '( ) ( ) '( ))
( )( ( ) ( ) ( ))
( ) ( )
u o u
u o u
u o u
dECost Q
c c QF Q F Q H Q c
dQ
c c Qf Q F Q Qf Q c
c c F Q c
= + +
= + +
= +
(12)

9
Winkler, Roodman, and Britney (1972) use the term partial moment rather than partial expectation. We assume
that demand is always non-negative (i.e., D 0). Winkler et al. use the notation
0
( )
Q
E D for the partial expectation
of the random variable D in the range (0,Q). This paper will use the simpler notation ( ) H Q .
10
Leibnizs rule states that
( ) ( )
( ) ( )
( , ) ( ) ( )
( , ) ( ( ), ) ( ( ), )
x h y x h y
x g y x g y
d r x y dh y dg y
r x y dx dx r h y y r g y y
dy y dy dy
= =
= =
c
= +
c
} }
. In this
situation,
0
'( ) ( ) / ( ) ( )
Q
D
d
H Q dH Q dQ Df D dD Qf Q
dQ
=
= = =
}
, where y Q = , x D = , ( ) h Q Q = , ( ) 0 g Q = , and
( , ) ( , ) ( ) r x y r D Q Df D = = .

Page 16
Setting this derivative to zero leads to:

( )
u
u o
c
F Q
c c
=
+
(13)

where the quantity / ( )
u u o
R c c c = + is the critical ratio (fractile). The second derivative of
( ) ECost Q is
2 2
( ) / ( ) ( )
o u
d ECost Q dQ c c f Q = + , which is non-negative for all values of Q.
Therefore, ( ) ECost Q is a convex function and
1
* ( / ( ))
u u o
Q F c c c
= + is the globally optimal

order quantity. Note that this result does not require any assumptions about the demand
distribution and therefore is true for all continuous probability distributions.

Appendix 3: Derivations of expected values

The derivations in this section are developed for continuous demand. Expectations for discrete
demand are identical when integrals are replaced by summations, ( ) f D is replaced by ( ) p D .
These derivations show the details to make the mathematical reasoning as accessible as possible
for the non-mathematical reader.

Expected number of units sold

For order quantity Q and specific demand D, the number of units sold is:

for
( , )
for
D D Q
Sold Q D
Q D Q
s
=

>
(14)

An alternative expression for the number of units sold is min( , ) D Q . For continuous demand
with density ( ) f D , the expected number of units sold is then given by:

0
0
( ) ( , ) ( )
( ) ( )
( ) (1 ( ))
D
Q
D D Q
ESold Q Sold Q D f D dD
Df D dD Q f D dD
H Q Q F Q
= =
=
= +
= +
}
} }
(15)

Page 17

where ( ) H Q is the partial expectation of demand, which is defined as
0
( ) ( )
Q
D
H Q Df D dD
=
=
}
,
and ( ) 1 ( )
D Q
f D dD F Q
=
=
}
. Winkler, Roodman, and Britney (1972) prove that the partial
expectation (partial first moment) for a normally distributed random variable is:

0
( ) ( ) ( ) ( )
Q
u u
D
H Q Df D dD F z f z o
=
= =
}
(16)

where and o are the mean and standard deviation of demand, ( )
u
F z and ( )
u
f z are the
cumulative and probability density functions for the standard normal distribution, and
( ) / z Q o = . Note that ( ) ( )
u
F Q F z = , but that ( ) ( )
u
f Q f z = . Therefore, for normally
distributed demand, the expected number of units sold is:

( ) ( ) (1 ( ))
( ) ( ) (1 ( ))
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
u u
u u
u
u
ESold Q H Q Q F Q
F z f z Q F Q
F z f z Q QF Q
F Q f z Q QF Q
Q Q F Q f z
o
o
o
o
= +
= +
= +
= +
= +
(17)

Expected number of units of lost sales

For order quantity Q and specific demand D, the number of units of lost sales is:

0 for
( , )
for
D Q
Lost Q D
D Q D Q
s
=

>
(18)

Alternative expressions for lost sales include max( , 0) D Q and ( , 0) D Q
+
. For continuous
demand with density function ( ) f D , the expected units of lost sales is:


Page 18

0
( ) ( , ) ( )
( ) ( )
( ) ( )
( ) (1 ( ))
D
D Q
D Q D Q
ELost Q Lost Q D f D dD
D Q f D dD
Df D dD Q f D dD
H Q Q F Q
=

= =
=
=
=
=
}
}
} }
(19)

where ( ) ( )
Q
D D Q
Df D dD Df D dD

= =
= +
} }
and ( ) 1 ( )
D Q
f D dD F Q
=
=
}
.

For normally distributed demand, ( ) H Q can be replaced with equation (16) to find the expected
number of units of lost sales:

( ) ( ) (1 ( ))
( ) ( ) (1 ( ))
(1 ( )) (1 ( )) ( )
( )(1 ( )) ( )
u u
u
u
ELost Q H Q Q F Q
F z f z Q F Q
F Q Q F Q f z
Q F Q f z
o
o
o
=
= +
= +
= +
(20)

Expected number of units salvaged

For order quantity Q and specific demand D, the number of units salvaged is:

for
( , )
0 for
Q D D Q
Salvage Q D
D Q
s
=

>
(21)

Alternative expressions for the number of units salvaged include max( , 0) Q D and ( , 0) Q D
+
.
For continuous demand with density function ( ) f D , the expected number of units salvaged is:

0
0
0 0
( ) ( , ) ( )
( ) ( )
( ) ( )
( ) ( )
D
Q
D
Q Q
D D
ESalvage Q Salvage Q D f D dD
Q D f D dD
Q f D dD Df D dD
QF Q H Q
=
=
= =
=
=
=
=
}
}
} }
(22)

Page 19

For normally distributed demand, ( ) E Q can be replaced with equation (16) to find the expected
number of units salvaged:

( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
u u
u
u
ESalvage Q QF Q H Q
QF Q F z f z
QF Q F Q f z
Q F Q f z
o
o
o
=
= +
= +
= +
(23)

Expected cost

As developed in equation (11), the expected cost for continuous demand is:

( ) ( )( ( ) ( )) ( )
u o u
ECost Q c c QF Q H Q c Q = + + (24)

For normally distributed demand, ( ) E Q can be replaced with equation (16) to find the expected
cost:

( ) ( )( ( ) ( ) ( )) ( )
( )(( ) ( ) ( )) ( )
u o u u u
u o u u
ECost Q c c QF Q F z f z c Q
c c Q F Q f z c Q
o
o
= + + +
= + + +
(25)

where ( ) / z Q o = and ( )
u
F z and ( )
u
f z are the CDF and PDF for the standard normal.
(Note that ( ) ( )
u
F z F Q = and ( ) ( )
u
f z f Q = .)

Expected profit

For order quantity Q and specific demand D, the profit is:

( ) if
( , )
( ) if
pD s Q D cQ D Q
Profit Q D
pQ g D Q cQ Q D
+ s
=

<
(26)

The expected profit function, therefore, is:


Page 20

0
0
0 0 0
( ) ( , ) ( )
[ ( )] ( ) [ ( )] ( )
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( ) (1 ( ))
D
Q
D D Q
Q Q Q
D D D
D Q D Q D Q
EProfit Q Profit Q D f D dD
pD s Q D f D dD pQ g D Q f D dD cQ
p Df D dD sQ f D dD s Df D dD
pQ f D dD g Df D dD gQ f D dD cQ
pH Q sQF Q sH Q pQ F Q
= =
= = =

= = =
=
= + +
= +
+ +
= + +
}
} }
} } }
} } }
( ( )) (1 ( ))
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
( )( ( ) ( )) ( )
g H Q gQ F Q cQ
pH Q sQF Q sH Q pQ pQF Q g gH Q gQ gQF Q cQ
pH Q sH Q gH Q pQF Q sQF Q gQF Q pQ cQ gQ g
p s g H Q p s g QF Q p c g Q g
p s g H Q QF Q p c g Q g
+
= + + + +
= + + + +
= + + + +
= + + +
(27)

For normally distributed demand, ( ) H Q can be replaced with equation (16) to find the expected
profit:

( ) ( )( ( ) ( )) ( )
( )( ( ) ( ) ( )) ( )
( )(( ) ( ) ( )) ( )
u
u
EProfit Q p s g H Q QF Q p c g Q g
p s g F Q f z QF Q p c g Q g
p s g Q F Q f z p c g Q g
o
o
= + + +
= + + +
= + + +
(28)

Relationships between expected values

All units ordered at the beginning of the period must be either sold or salvaged at the end of the
period:

( , ) ( , ) Q Sold D Q Salvage D Q = + (29)

Taking expectations of both sides, leads to the relationship:

( ) ( ) Q ESold Q ESalvage Q = + (30)

The demand in a period must be converted into either a sale or a lost sale. In other words, the
demand is always the sum of the number of units sold and the number of units of lost sales. For
any order quantity and demand realization:

( , ) ( , ) D Sold D Q Lost D Q = + (31)


Page 21
Taking expectations of both sides, it is clear that the average demand is the sum of the expected
number of units sold and the expected number of units of lost sales:

( ) ( ) ESold Q ELost Q = + (32)

The expected cost in equation (24) can be related to expected profit as follows:

( ) ( )( ( ) ( )) ( )
( )( ( ) ( )) ( )( )
( )( ( ) ( )) ( ) ( )
( ) ( )( ( ) ( )) ( )
( ) ( )( ( ) ( )) ( ) )
(
u o u
ECost Q c c QF Q H Q c Q
p c g c s QF Q H Q p c g Q
p s g QF Q H Q p c g p c g Q
p c g p s g H Q QF Q p c g Q
p c g p s g H Q QF Q p c g Q g g
p c

= + +
= + + + +
= + + + +
= + + +
= + + + +
= ) [( )( ( ) ( )) ( ) ]
( ) ( ) ( )
( ) ( ) ( )
g p s g H Q QF Q p c g Q g g
ECost Q p c EProfit Q
EProfit Q p c ECost Q

+ + + +
=
=
(33)

In other words, for any order quantity Q, the expected profit is the profit for selling the average
demand, ( ) p c , less the expected cost. The expected cost, therefore, can be interpreted as the
cost of demand variability and the organization should be willing to pay that amount to reduce
uncertainty to zero. In other words, when the demand has no variability the expected profit is
( ) p c and the expected cost is zero. Given that the quantity ( ) p c is a constant and is
independent of Q, maximizing expected profit is equivalent to minimizing expected cost.

Summary of the expected value equations

Tables 1 and 2 summarize the expectations for all continuous and discrete demand distributions.
Table 3 summarizes the relationships between expected values for the newsvendor problem for
any demand distribution. Tables 4 and 5 summarize the expectations for normally distributed
demand (a continuous demand distribution) and Poisson distributed demand (a discrete demand
distribution).


Page 22
Table 1. Summary of expected values for all continuous demand distributions
11

Expected units sold
( ) ( ) (1 ( )) ESold Q H Q Q F Q = +
Expected salvage
( ) ( ) ( ) ESalvage Q QF Q H Q =
Expected lost sales
( ) ( ) (1 ( )) ELost Q H Q Q F Q =
Expected cost
12

( ) ( )( ( ) ( )) ( )
u o u
ECost Q c c QF Q H Q c Q = + +
Expected profit
( ) ( )( ( ) ( )) ( )
( ) ( ) ( )
p ESold Q s ESalvage Q g ELost Q cQ
= + + +
= +

Table 2. Summary of expected values for all discrete demand distributions
13

Expected units sold
( ) ( ) (1 ( )) ESold Q H Q Q F Q = +
Expected salvage
( ) ( ) ( ) ESalvage Q QF Q H Q =
Expected lost sales
( ) ( ) (1 ( )) ELost Q H Q Q F Q =
Expected cost
14

( ) ( )( ( ) ( )) ( )
u o u
ECost Q c c QF Q H Q c Q = + +
Expected profit
( ) ( )( ( ) ( )) ( )
( ) ( ) ( )
p ESold Q s ESalvage Q g ELost Q cQ
= + + +
= +

Table 3. Relationships between expectations for all demand distributions
All units ordered will always be either sold or salvaged.
( ) ( ) Q ESold Q ESalvage Q = +
The realized demand will always be equal to the actual
number of units sold plus the lost sales.
( , ) ( , )
min( , ) max( , 0)
D Sold Q D Lost Q D
D Q D Q
= +
= +

Average demand is always the expected units sold plus the
expected units of lost sales.
( ) ( ) ESold Q ELost Q = +
Expected profit is always the profit for the average
demand less the expected cost.
( ) ( ) ( ) EProfit Q p c ECost Q =

Table 4. Summary of expected values for normally distributed demand
15

Expected units sold
( ) ( ) ( ) ( )
u
ESold Q Q Q F Q f z o = +
Expected salvage
( ) ( ) ( ) ( )
u
ESalvage Q Q F Q f z o = +
Expected lost sales
( ) ( )(1 ( )) ( )
u
ELost Q Q F Q f z o = +

11
The partial expectation for continuous demand is defined as
0
( ) ( )
Q
D
H Q Df D dD
=
=
}
.
12
The expected cost is the sum of the expected underage cost plus the expected overage cost.
13
The partial expectation for discrete demand is defined as
0
( ) ( )
Q
D
H Q Dp D
=
=
.
14
The expected cost is the sum of the expected underage and overage costs.
15
The expected values for the normal distribution are derived by replacing ( ) H Q with ( ) ( )
u u
F z f z o , where
( ) / z Q o = . The proof for this relationship can be found in Winkler, Roodman, and Britney (1972).

Page 23
Expected cost
( ) ( )(( ) ( ) ( )) ( )
u o u u
ECost Q c c Q F Q f z c Q o = + + +
Expected profit
( ) ( )(( ) ( ) ( )) ( )
u
EProfit Q p s g Q F Q f z p c g Q g o = + + +

Table 5. Summary of expected values for Poisson
16
distributed demand with mean
Expected units sold
( ) ( 1) (1 ( )) ESold Q F Q Q F Q = +
Expected salvage
( ) ( ) ( 1) ESalvage Q QF Q F Q =
Expected lost sales
( ) ( 1) (1 ( )) ELost Q F Q Q F Q =
Expected cost
( ) ( )( ( ) ( 1)) ( )
u o u
ECost Q c c QF Q F Q c Q = + +
Expected profit
( ) ( )( ( 1) ( )) ( ) EProfit Q p s g F Q QF Q p c g Q g = + + +

The expected values for other distributions can be found using the partial expectation functions
presented in Table 6. These equations were derived by the author from the tail conditional
expectation functions presented in Landsman and Valdez (2005).

Table 6. Partial expectation functions
Continuous distributions
Normal
( ) ( ) ( )
u u
H x F z f z o = , where ( ) / z x o =
Lognormal
2
/ 2 2
( ) (1 (( ln( )) / ))
Normal
H x e F x
o
o o
+
= +
Exponential
( ) ( ) ( )
Exp
H x x F x x = +
Gamma
( ) ( | 1, )
Gamma
H x F x o | = +
Discrete distributions
Poisson
( ) ( 1)
Poisson
H x F x =
Binomial
( ) ( 1| , 1)
Binomial
H x F x p n =
Negative binomial
( ) ( 1| , 1)
NB
H x F x p o = +

Appendix 4: VBA code for the inverse of the triangular CDF

Function triangular_inverse(p, a, b, c) As Double
'
' Compute the inverse of the triangular distribution at probability p
' given triangularly distributed demand with parameters
' (minimum, most likely, maximum) = (a, b, c).
'
If p <= (b - a) / (c - a) Then
triangular_inverse = a + Sqr(p * (c - a) * (b - a))
Else
triangular_inverse = c - Sqr((1 - p) * (c - a) * (c - b))

16
Hadley and Whitin (1963) prove that the partial expectation for the Poisson distribution is ( ) ( 1) H Q P Q = .

Page 24
End If
End Function

Appendix 5: VBA code for the inverse of the Poisson CDF

Function poisson_inverse(p, lambda)
' p =cumulative probability and lambda = mean of the Poisson distribution.
' This routine truncates the result at xmax = 60.
Dim x As Integer
Const xmax = 60
For x = 1 To xmax
poisson_inverse = x
If Application.WorksheetFunction.Poisson(x, lambda, True) >= p Then Exit Function
Next x
MsgBox poisson_inverse( & Format(p, 0.00%) & ) was truncated at _
& Val(xmax) & ., vbExclamation
End Function

Related resources from Clamshell Beach Press: The companion Excel workbook
newsvendor model.xls is available for download from www.ClamshellBeachPress.com. Other
related Excel workbooks from Clamshell Beach Press include slowmove.xls and safety
stock.xls. The Seasonal Buying paper applies the newsvendor logic to the retail buying
context. The Triangular Distribution paper presents the details of the triangular distribution.

Acknowledgements: The author thanks Jonathan Hill for his helpful edits on the mathematics
in earlier versions of this paper. The author also thanks Sheryl Holt (Writing Studies
Department, University of Minnesota) and Lindsay Conner (Word Out Communications,
http://word-out.com) for their helpful edits of earlier versions of this paper.

Copyright 2011 Clamshell Beach Press. All rights reserved. Copying or distributing any part of this document in
any form without prior written permission from Clamshell Beach Press is illegal. Written permission may be
obtained by sending an email to [email protected].

Newsvendor Problem

Uploaded by

Copyright:

Available Formats

Newsvendor Problem

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Newsvendor Problem

Uploaded by

Copyright:

Available Formats

What is the newsvendor problem?

What are some contexts where the newsvendor problem applies?

Copyright 2011 Clamshell Beach Press, www.ClamshellBeachPress.

is the inverse of the cumulative distribution function (also

~ ~ units and the optimal expected profit is approximately

~ ~ units and the optimal expected profit is about $70,503. In

, where the quantity

= + is the globally optimal

You might also like