DECISION TREES: THE ALTERNATIVES?

• Company Chocolatey (C) currently has assets of 150000

and wants to decide whether to market a new chocolate
(NC). The company has three alternatives:
– Test market (TM) NC locally.
Then utilize the results of the market study to
determine whether to market NC nationally.
– Market NC nationally without TM.

– Decide (without TM) not to market NC nationally.

 SUCCESS AND FAILURE NATIONAL MARKET

• Estimated: in absence of market study, NC has

– 55% chance of national success and 45% chance of

national failure.
– If former occurs, C’s asset position will  by 300000;
else,  by 100000.
– i.e., if former, C’s asset position=150000+300000,

– Otherwise, C’s asset position=150000-100000

 IF A MARKET STUDY IS PERFORMED…

• If C performs a market study (cost: 30000), there is

– 60% chance that study will indicate local success

– 40% chance that study will indicate local failure.

– If former, there is 85% chance that there will be

national success,
– Else, 10% chance of national success.

DM= decision maker

 BASIC IDEAS DRAWING DECISION TREES

• List various options as in accompanying diagram.

• Rectangle represents a decision fork (i.e. where a

decision needs to be made by the DM) and a circle
an event fork (i.e. where outside forces determine which
of several random events will happen).
• Figures accompanying any fork represent opt. value if
that fork is followed. Figures in larger rectangles
alongside connecting lines denote particular options.
 FOR EXAMPLE…

• E.g. Leftmost rectangle gives 2.7 (meaning 2, 70, 000), which

indicates that the best option will yield a final expected asset
position of 2.7.
• At any node, we determine best return at that node...
• By working backwards from the last stage returns and then
computing EV of different options, given the probs.
• Let TM=Test market, LS =local success, LF= local failure Similarly
NS/ NF
•
 EVALUATING EACH BRANCH of THE TREE

• Let us follow branch TM, LS leading out from leftmost

small
• Let us look at the small following the LS. This is a
decision fork, since DM has to decide between options of
TM, no TM.
• We know: if NS, final asset position will be 1.5+3-.3=4.2.

• If NF, final asset position will be 1.5-1-.3=0.2.

Decision Tree Example Maximization
1.5-.3=
1.2
Don’t
MN NS, .85

0. S ,
3.6

L
6
1.5-.3+3=
MN 4.2
2.64
3.6
NF,
.1 5
1.5-.3-
TM 1.2 Don’t MN 1=.2
L F, 1.5-.3
0.4 =1.2
2.7
MN 0.6 1.5-.3+3=
4.2
NS, .1

NF 1.5-..3
,. .9
-1=.2
2.7
NS, 55 1.5+3=
No

4.5
TM

MN

NF
, .4
5
1.5-1=.5
2.7
Do
n
MN ’t
1.5
 COMPARING THE ALTERNATIVES

• Pr. (NS) = .85, Pr. (NF) = .15

• Exp. value =.85*4.2 +.15*.2=3.6, if markets nationally.

• Else, it loses just the cost of conducting the market

study. Thus the expected value is: 1.5-.3=1.2.
• Hence, MN option would be chosen, and thus the value
of the LS option is 3.6.
• Similarly, expected values of all options can be obtained
• THE CONCLUSION…?
 VALUE OF INFORMATION

• Market Nationally without Test Marketing

• First suppose market study is costless. Then gain in TM

branch is 2.64+.3=2.94, which is > cost without TM =2.7.
• Hence we say Expected Value with Sample Information
(EVWSI) is 2.94 (i.e., if we had the information given by
study, we would get this amount).
• Largest amount of gain without this information is 2.7
(diag.), called Exp. Value with Original Information (EVWOI)
 VALUE OF SAMPLE INFORMATION

• Difference (2.94-2.7=.24) is called Expected Value of

Sample Information (EVSI)—reason should be obvious.
• One use of this information: decide whether conduct study
at all. Since cost of the market study (.3) exceeds EVSI,
Company C should not conduct study.
• Now suppose all uncertain events affecting C’s asset
position still occur with the given probabilities, but
• C finds out whether or not NC is a national success
before making decision to market NC nationally or not.
 VALUE OF PERFECT INFORMATION

• In other words, the DM has perfect information about

which state has occurred before making a decision.
• Then, we have Expected Value with Perfect Information
(EVWPI) = .55*4.5+.45*1.5=3.15, since 4.5 and 1.5 are
the final asset positions in case of NC and NF.
• Exp. Value of Perfect Info. (EVPI)= EVWPI- EVWOI.

• Amount up to which we would be willing to pay for perfect

information
 Decision Tree Example Minimization

• E-Education develops and markets MBA courses offered

over Internet. Company currently located in Chicago,
employs 150 people and needs additional office space.
Options:
• Lease additional space at current location in Chicago for
next two years, after that move to a new building (1).
• Move entire operation to small Midwest town immediately
$1 mil. (2).
• Lease new building (NB) in Chicago (C) immediately (3).
 ALTERNATIVES BEFORE THE COMPANY

• If Co. chooses option 1 and leases new space at current

location, it can at end of two years either lease New
Building in Chicago or move to small Midwest town.
• Company has 75% chance of surviving next two years.

• Leasing additional space for 2 years at current location in

Chicago would cost $750,000 per year.
• Leasing space in small Midwest town would cost only
$500,000 per year.
 ALTERNATIVES BEFORE THE COMPANY (2)

• Moving to a new building in Chicago would cost $200,000

• Leasing new building’s space would cost $650,000/ year.

• The company can cancel the lease at any time.

• The company will build its own building in five years, if it

survives.
• This is a minimization problem, again we compute EVs,
and in this case, decision will go for min. value.
• Computation, next slide.
Decision Tree Minimization: Values of alternatives

ALTERNATIVE CALCULATION VALUE

Stay in Chicago, lease new space (750.000) x 2 + 200,000 + $3,650,000

for two years, survive, lease (650,000) x 3 =
new building in Chicago
Stay in Chicago, lease new space (750,000) x 2 + 1,000,000 + $4,000,000
for two years, survive, move (500,000) x 3 =
to Midwest
Stay in Chicago, lease new space (750,000) x 2 = $1,500,000
for two years, fail
Stay in Chicago, lease new 200,000 + (650,000) x 5 = $3,450,000
building in Chicago, survive.
Stay in Chicago, lease new 200,000 + (650,000) x 2 = $1,500,000
building in Chicago, fail
Move to Midwest, survive 1,000,000 + (500,000) x 5 = $3,500,000
Move to Midwest, fail 1,000,000 + (500,000) x 2 = $2,000,000
Decision Tree For Minimization Problem
SE
LEA n C
i
ve NB 3650
rv i
Su 5 )
(. 7 3650
Mov
e 4000
M. W to
est
C.
in NS Fail
ay rs 3112.5
S t a se ye a (0 . 2 5
) 1500
Le r 2
fo 3450
Stay in C.
Lease new Survive
Building (.75)

Fail (0.25)
2962.5 1500

E-
education
Survive (.75)

3500
M idwe
ov
M

e s t to

3125
to wn

Fail (0.25)
2000
 WHICH ALTERNATIVE COSTS LESS…

• Consider first option: If survive, then two options are

open, and of these two, the lower-cost alternative is
Lease New Space in Chicago, so this is chosen.
• If fail, there is only one alternative. And its cost is
$1,500,000.
• EV of the first option of staying in Chicago and leasing
space for the first two years is .75 x 3,650,000 + 25 x
1,500,000 = $3,112,500.
 THE CONCLUSION IS…

• The second option, staying in Chicago and leasing a new

building now, has an expected value of .75 x 3,450,000 +
.25 x 1,500,000 = $2,962,500.
• Finally, the third option of moving to the Midwest
immediately has an expected value of .75 x 3,500,000
+ .25 x 2,000,000 = $3,125,000.
• From this, it looks like second opt is best alternative, to
stay in Chicago and lease a new building immediately.
SENSITIVITY ANALYSIS (P of NATIONAL SUCCESS)

• Let us consider the Max problem

• In the absence of a survey we have P(National

Success)=.55 and P(National Failure)=.45
• Now suppose this Probability is variable, call it p

• Then the expected asset position=p(4.5)+0.5(1-p)=4p+.5

• For this to be better than the “Test Market and proceed

accordingly” option, we must have
SENSITIVITY ANALYSIS (P, NATIONAL SUCCESS-2)

• 4p+.5≥2.64, i.e.,

• p≥.535

• Hence, as long as p≥.535, it will be better to “Do not Test

Market and proceed accordingly”
• Also, the expected asset position will be given by 4p+.5

• When p<.535, “Test Market and proceed accordingly”

will be better
 SENSITIVITY ANALYSIS (P of LOCAL SUCCESS)

• Local Success, Test Market

• We find, under MN option, expected asset position

=.85(4.2)+.15(.2)
• whereas under “do not MN” expected asset position=1.2

• Again, let P(National Success|local success) be variable.

• Call it p

• So expected asset position=p(4.2)+(1-p)(.2)=4p+.2

SENSITIVITY ANALYSIS (P, LOCAL SUCCESS-2)

• For “MN” to be better, 4p+.2≥1.2, so that p≥.25

• Hence as long as P(National Success | local

success)≥.25, we go for MN
• The expected asset position under MN will then be 4p+.2

• Similarly, we could also experiment with P(local failure)

• Or, even with two sets of probabilities; we would then

have simultaneous equations
 BAYESIAN ANALYSIS

• Sometimes, original estimates of Pr. can be improved by

• Using info. obtained via surveys etc.

• Called prior and posterior, respectively

• Suppose, launching a new product in a market

• Will market be favorable/unfavorable (FM/UM)?

• Suppose initially nothing known, or otherwise, say 50-50

• (It could equally well be any other values)

 BAYESIAN ANALYSIS

• From history etc., we know that:

• Bayes’ theorem: P(Ai|Bj)=P(Bj|Ai)*P(Ai)/ƩP(Bj|Ai)*P(Ai)

(summation being over all possible values of A)
• Law of Total Probability (get 1, odd)
RESULT OF SURVEY FAV. MKT. (FM) UNFAV. MKT. (UM)
Positive (predict P(survey P(survey
favourable mkt.) positive|FM)=.7 positive|UM)=.2
Negative (predict P(survey P(survey
unfavourable market) negative|FM)=.3 negative|UM)=.8
 BAYESIAN ANALYSIS

• P(FM|survey pos.)=

• P(survey pos.|FM)* P(FM)/ {P(survey pos.|FM)* P(FM)+

P(survey pos.|UM)* P(UM)}=
• (.7*.5)/( .7*.5+.2*.5)=.78

• P(UM|survey pos.)=

• P(survey pos.|UM)* P(UM)/{P(survey pos.|UM)* P(UM)+

P(survey pos.|FM)* P(FM)}=
• (.2*.5)/( .2*.5+.7*.5)=.22

•
 BAYESIAN ANALYSIS (CONTD.)

• P(FM|survey neg.)= P(survey neg.|FM)* P(FM)/

• P(survey neg.|FM)* P(FM)+ P(survey neg. |UM)* P(UM)

• =(.3*.5)/( .3*.5+.8*.5)=.27

• P(UM|survey neg.)= P(survey neg.|UM)* P(UM)/

P(survey neg.|UM)* P(UM)+P(survey neg. |FM)* P(FM)
• =(.8*.5)/( .8*.5+.3*.5)=.73