You are on page 1of 12

Probabilistic Inventory

models
NEWSBOY/NEWS Vendor Problems

Inventory Control with Uncertain


Demand
The demand can be decomposed into two parts,

D = DDet + DRan
where

DDet = Deterministic component of demand


and
DRan

= Random component of demand.

Inventory Control with Uncertain


Demand
There are a number of circumstances under which

it would be appropriate to treat D


as being
D
deterministic even though
Ran is not zero. Some
of these are:
ZWhen the variance of the random component,
DRan is small relative to the magnitude of D .
ZWhen the predictable variation is more
important than the random variation.
ZWhen the problem structure is too complex to
include
an
explicit
representation
of
randomness in the model.
3

Inventory Control with Uncertain


Demand
However, for many items, the random component of
the demand is too significant to ignore.
As long as the expected demand per unit time is
relatively constant and the problem structure not
too complex, explicit treatment of demand
uncertainty is desirable.

Inventory Control with Uncertain


Demand
Example:
A newsstand purchases a number of copies of The
Computer Journal (weekly). The observed
demands during each of the last 52 weeks were:
15
14
8
6
8
13

19
11
9
7
11
12

9
6
5
12
11

12
11
4
15
18

9
9
4
15
15

22
18
17
19
17

4
10
18
9
19

7
0
14
10
14

8
14
15
9
14

11
12
8
16
17

Inventory Control with Uncertain


Demand
Example :
6
5
4
3
2
1
0

10

12

14

16

18

20

22

Inventory Control with Uncertain


Demand
Example :
Estimate the probability that the number of copies
of the Journal sold in any week.
The probability that demand is 10 is estimated to
be 2/52 = 0.0385, and the probability that the
demand is 15 is 5/52 = 0.0962.
Cumulative probabilities can also be estimated in
a similar way.
The probability that there are nine or fewer copies
of the Journal sold in any week is (1 + 0 + 0 + 0 + 3
+ 1 + 2 + 2 + 4 + 6) / 52 = 19 / 52 = 0.3654.
7

Inventory Control with Uncertain


Demand
We generally approximate the demand history
using a continuous distribution.
By far, the most popular distribution for inventory
applications is the normal.
A normal distribution is determined by two
parameters: the mean and the variance 2

Inventory Control with Uncertain


Demand
These can be estimated from a history of demand
D
by the sample mean
and the sample
variance s 2 .
D=

1 n
Di
n i =1

1 n
(Di D )2
s =

n 1 i =1
2

Inventory Control with Uncertain


Demand
The normal density function is given by the
formula
1 x 2
1
f (x ) =
exp
for < x < +

2
2

We substitute D as the estimator for and


estimator for .

s as the

10

Inventory Control with Uncertain


Demand

0.12
0.10
0.08
0.06
0.04
0.02
0
-4

-2

10

12

14

16

18

20

22

24

26
11

Optimization Criterion
In general, optimization in production problems
means finding a control rule that achieves
minimum cost.
However, when demand is random, the cost
incurred is itself random, and it is no longer
obvious what the optimization criterion should be.
Virtually almost all of the stochastic optimization
techniques applied to inventory control assume
that the goal is to minimize expected costs.
12

The Newsboy (girl) / newsvendor Model


(Continuous Demands)
The demand is approximately normally distributed
with mean 11.731 and standard deviation 4.74.
Each copy is purchased for 25 cents and sold for
75 cents, and he is paid 10 cents for each unsold
copy by his supplier.
One obvious solution is approximately 12 copies.
Suppose Mac purchases a copy that he doesn't
sell. His out-of-pocket expense is 25 cents 10
cents = 15 cents.
Suppose on the other hand, he is unable to meet
the demand of a customer. In that case, he loses
75 cents 25 cents = 50 cents profit.
13

The Newsboy Model


(Continuous Demands)
Notation:
co = Cost per unit of positive inventory remaining
at the end of the period (known as the overage
cost).
cu = Cost per unit of unsatisfied demand. This can
be thought of as a cost per unit of negative ending
inventory (known as the underage cost).
The demand D is a continuous nonnegative
random variable with density function f ( x ) and
cumulative distribution function F ( x ) .
The decision variable Q is the number of units to
14
be purchased at the beginning of the period.

The Newsboy Model


(Continuous Demands)
The cost function G(Q) is convex
Q

G (Q ) = co (Q x ) f ( x )dx + cu ( x Q ) f ( x )dx
Q

dG
= 0 + co f ( x )dx + 0 + cu [ f ( x )]dx
0
Q
dQ
Q
Q
= co f ( x )dx cu 1 f ( x )dx

0
0

= (co + cu ) f ( x )dx cu = (co + cu ) F (Q) cu


Q

d 2G
= (co + cu ) f (Q) 0
dQ 2

The optimal solution equation

( )

F Q* =

cu
co + cu

15

The Newsboy Model


(Continuous Demands)
Determining the optimal policy for convex
function

(Thousands)

G(Q)

12
10
8
6
4
2
0
-10

-6

-2 0
O

100
Q

200
Q*

300

400
16

The Newsboy Model


(Continuous Demands)
Example (continued):
Normally distributed with mean = 11.73 and standard
deviation = 4.74.
co = 25 10 = 15 cents.
cu = 75 25 = 50 cents.
The critical ratio is c (c + c ) = 0.50/0.65 = 0.77.
u

Purchase enough copies to satisfy all of the weekly


demand with probability 0.77. The optimal Q* is the
77th percentile of the demand distribution.
17

The Newsboy Model


(Continuous Demands)
Example (continued):
f(x)

Area = 0.77

11.73

Q*

x18

The Newsboy Model


(Continuous Demands)
Example (continued):
Using the data of the normal distribution we
obtain a standardized value of z
= 0.74. The
optimal Q is
Q * = z + = 4.74 0.74 + 11.73 = 15.24 15

Hence, he should purchase 15 copies every week.

19

The Newsboy Model (Discrete Demands)


Optimal policy for discrete demand:
The procedure for finding the optimal solution to
the newsboy problem when the demand is
assumed to be discrete is a natural generalization
of the continuous case.
The optimal solution procedure is to locate the
critical ratio between two values of F(Q) and
choose the
Q
corresponding to the higher
value. That is
Q*
c
p( x ) = F Q* c +u c
o
u
x =0

( )

20

The Newsboy Model (Discrete Demands)


Example:
Q

0
1
2
3
4
5
6
7
8
9
10
11

f (Q )

1/52
0
0
0
3/52
1/52
2/52
2/52
4/52
6/52
2/52
5/52

F (Q )

1/52(0.0192)
1/52(0.0192)
1/52(0.0192)
1/52(0.0192)
4/52(0.0769)
5/52(0.0962)
7/52 (0.1346)
9/52 (0.1731)
13/52(0.2500)
19/52 (0.3654)
21/52 (0.4038)
26/52 (0.5000)

f (Q )

F (Q )

12
13
14
15
16
17
18
19
20
21
22

4/52
1/52
5/52
5/52
1/52
3/52
3/52
3/52
0
0
1/52

30/52 (0.5769)
31/52 (0.5962)
36/52 (0.6923)
41/52 (0.7885)
42/52 (0.8077)
45/52 (0.8654)
48/52 (0.9231)
51/52 (0.9808)
51/52 (0.9808)
51/52 (0.9808)
52/52 (1.000)
21

The Newsboy Model (Discrete Demands)


Example :
The critical ratio for this problem was 0.77, which
corresponds to a value of F (Q ) between Q = 14
and Q = 15.
*
Since we round up, the optimal solution is Q = 15.
Notice that this is exactly the same order quantity
obtained using the normal approximation.

22

The Newsboy Model


XLRI is holding its annual convention in Dubai. Six months
before the convention begins XLRI must decide how many
rooms should be reserved in the convention hotel. At this time
XLRI can reserve rooms at a cost of $50 per person, but six
months before the convention XLRI doesn't know with certainty
how many people will attend the convention. XL believes that
the number of rooms required is normally distributed with a
mean of 5000 rooms and a standard deviation of 2000 rooms. If
the number of rooms required exceeds the number of rooms at
the convention hotel, extra rooms will have to be found at
neighboring hotels at at a cost of $80 per room. It is
inconvenient for convention participants to stay at neighboring
hotels. We measure this inconvenience by assessing an
additional cost of $10 for each room obtained at a neighboring
hotel. If the goal is minimize the expected cost to XLRI, how
many rooms should XLRI reserve at the convention hotel?
23

The Newsboy Model


Ticket Prices for Dubai-India Flight is $200 (?!). Each Plane
can hold upto 100 passengers. Usually some of the
passengers who have purchased tickets for a flight fail to
show up (no-shows). To protect against n-shows, the
airline will try to sell more than 100 tickets for each flight.
Rules stipulate that any ticketed customer who is unable to
board the flight is entitled to compensation, say, $100. Past
data indicate that the number of no-shows in these flight is
normally distributed with a mean of 20 and a standard
deviation of 5. In order to maximize expected revenues less
compensation costs , how many tickets should the airline
sell for each flight? Assume that anybody who doesnt use
a ticket receives a $200 refund.
24

You might also like