Decision Tree
Decision Tree
Let us assume a decision-making problem represented by the following conditional profit table :
Alternatives of production units
State of nature Probability
Al (50) A2 (100)
Si (High demand) 0.4 5000 10000
S2 (Low demand) 0.6 6000 — 4000
The decision tree can be drawn as follows to represent the above problem.
0.4 Rs. 5000
Sp P
rto
NV' 6°Il
Fig. 10.2
Steps in Decision Tree Analysis
1. List the decision points and the strategies (alternative courses of action) for each decision
point in a systematic manner.
2. Determine the probability and payoff associated with each alternative for each decision point.
3. Find out the expected payoff (EMV) of each course of action, starting from the extreme right
Courtesy : My Bibliotheca Page: 19 22/05/2021
working backward to the left.
DECISION THEORY 431
4. Select the course of action that gives the best payoff for each alternative.
5. Continue working backward to the next decision point on the left.
6. Continue with this process until the first decision point on the extreme left is reached.
7. Consider the situation on the whole and find its course of action to be adopted from the
beginning to the end under different possible outcomes.
Advantages of the Decision Tree Approach
1. It is systematic, orderly, logical and sequential approach.
2. It lists all possible outcomes and helps decision-nodes to examine each one of them.
3. It is easy to understand. Its graphical representation can be communicated to others with ease.
4. Since a decision now affects the decision-making in future, decision trees are particularly
useful in such situations.
5. This approach can be applied to different decision problems, such as introduction of new
product, investment decisions, etc.
Limitation of Decision Tree Approach
1. In real life situations, the decisions are made under a large number of variables. In such cases,
the diagram becomes extremely complicated.
2. It assumes utility of money is linear with time, which is not the case.
3. Decision Trees yields only and 'average' value solution as the problem is analyzed on the
basis of expected values.
4. The assignment of probabilities for different events is many a times not exact and only a
reasoned value.
Example 10.15. A farmer is not sure whether he should dig a tube well in his field. He is presently using
the canal water for irrigation of his fields for which he pays Rs. 5000 per year. The history of tube well
digging in the village has not been very encouraging, only 50 per cent of the wells dug up to 200 feet
yielded water. Some farmers drilled further up to 300 feet but only 25 per cent of them struck water at
300 feet. The cost of drilling is Rs. 100 per feet. The farmer has to make the following three decisions
(a) Shall he drill up to 200 feet ?
(b) If no water is struck at 200 feet should he drill up to 300 feet ?
(c) Should he continue to buy water from government for next 5 years, as the life of the tube well is only five
years ?
Solution. The decision tree diagram of the above problem is drawn below :
Fig. 10.3
Courtesy : My Bibliotheca Page: 20 22/05/2021
432 OPERATIONS RESEARCH
At decision node 1, the farmer has to take a decision before drilling up to 200 ft., or not drilling.
If he drills he pays Rs. 25000 @ Rs. 5000 per year for 5 years. If he drills up to 200 ft., there are two
probabilities 0.5 of water found and no water being struck. If the water is found, the cost he incurs is
Rs. 20000 as he digs 200 feet @ Rs. 100 per ft. If no water is found at 200 ft., he takes the decision
of drilling up to 300 feet or not drilling. If he does not drill 300 ft., he incurs expanses of Rs. 45000
because he has already spent Rs. 20000 for drilling up to 200 feet and he has to pay Rs. 25000
@ Rs. 5000 per year. If he drills up to 300 ft., there is an assured probability of 0.25 that water
will be found and of 0.75 that water will not be found. If water is found he spends Rs. 100 per
ft. for 300 ft. If it is not found he spend Rs. 55000 as he has already spent Rs. 30000 on digging
up to 300 ft., but he has also to spend Rs. 25000 @ Rs. 5000 per year for five years.
As explained earlier, in such problems we work backward
EMV of node B = 0.25 x 30000 + 0.75 x 55000
= Rs. (7500 + 41250) = Rs. 48750
EMV of node 2 = Rs. 45000 (Choosing the lesser of the two of Rs. 48750 and Rs. 45000)
EMV of node A = Rs. [0.5 x20000 + 0.5 x48750] = Rs. (10000 + 24375) = Rs. 34375
EMV of node 1 = 25000 (lesser of the two values Rs. 34375 and Rs. 25000)
Hence, it can be easily seen the best course of action for the farmer is not to drill and pay.
Rs. 25000/ for water from canal to the government for five years.
Example 10.16. A complex airborne navigating system incorporates a sub-assembly, which unrolls a map
of the flight plan synchronously with the movement of the aeroplane. The sub-assembly is bought on very
good terms from a sub-contractor, but is not always in perfect adjustment on delivery. The sub-assemblies
can be readjusted on delivery to guarantee accuracy at a cost of Rs. 50 per sub-assembly. It is not however,
possible to distinguish visually these sub-assemblies that need adjustments.
Alternatively, the sub-assemblies can each be tested electronically at a cost of Rs. 10 per sub-assembly
tested. Past experience shows that about 30 per cent of those supplied are defective; the probability of a
test indicating a bad adjustment of the sub-assembly is 0.8, while the probability that the test indicates
a good adjustment when the sub-assembly is properly adjusted is 0.7. If the adjustment is not made and
the sub-assembly is found to be faulty when the system has its final check, the cost of subsequent
rectifications will be Rs. 140.
Draw up an appropriate decision tree to show the alternatives open to the purchaser and use it to
determine his appropriate course of action.
Solution.
Rs. 50
Courtesy : My Bibliotheca Page:10.4
Fig. 21 22/05/2021
DECISION THEORY 433
The purchaser can either have the sub-assembly tested electrically and pay Rs. 10 or can·have_
it readjusted at a cost of Rs. 50. If he gets them tested, probability of defective sub-assembly is 0.3
and good is 0.7. Those which are defective out of these 80% (0.8) will be with bad adjustment and
need readjustment at the cost of Rs. 50. Only 20% (0.2) are OK as per test but if these are found to
be faulty in the final check, the purchaser has to spend Rs. 140. Out of 70% of sub-assemblies, which
are not defective, 70% are OK as per test and such assemblies will not cost anything to the purchaser
but 30%, which are bad as per test, will have to be readjusted at the cost of Rs. 50 per sub-assembly.
EMVof node B = Rs. (0.7 + 0.350) = Rs. 15
EMVof node C = Rs. (0.2 x 140 + 0.8 x 50) = Rs. 68
EMVof node A= Rs. (0.7x 15 + 0.3x 68) = Rs. 30.90.
EMVof node 1 = Rs. (10 + 30.90) = 40.90.
(When test is carried out)
EMVof node 1 = Rs. 50
(when no test is carried out)
Since the least cost is Rs. 40.90; the decision the purchaser has to take is to get sub-assemblies tested.
Example 10.17. XYZ Ltd., wants to update/change its existing manufacturing prices for product A. jt
wants to strengthen its R & D cell and conduct research for finding a better product of manufacturing
which can get them higher profits. At present the company is earning a profit of Rs. 20000 after paying
for material, labour and overheads. XYZ Ltd., has the following four alternatives :
(a) The company continues with the existing process:
(b) The company conducts research P, which costs Rs. 20000, has 75% probability of success and can get
the profit of Rs. 5000.
(c) The company conducts research Q, which costs Rs. 10000, has 50% probability of success and can get
the profit of Rs. 25000.
. (d) The company pays Rs. 10000 as royalty for a new product and can get profit of Rs. 20000.
The company can carryout only one out of the two types of research P. and Q because of certain
limitations. Draw a decision tree diagram and find the best strategy for XYZ Ltd.
Solution. The decision tree is drawn as shown below
Points 1, 2, 3, 4 and 5 are decision boxes. The four alternatives to the company are shown
coming out of decision box 1. Research P succeeds (probability 0.75) or it fails (Probability 0.25).
If it fails the company has three alternatives; conduct research Q, continue with the existing
process or pay royalty. If Q fails the company is left only with the option of paying royalty or
continuing with existing process. The payoffs are written at the end of the branches.
Let us now calculate the EMV starting from node 5 (working backward)
EMV at decision node 5 = Maximum out of
- (a) Rs. 20000
(b) Rs. 20000 — 10000 = 10000 = Rs. 20000
EMV at decision node 4 = Maximum out of B Rs. 20000 and Rs. (20000 — 10000) = Rs. 20000
EMV at decision node C = Rs. (0.5 x 20000 + 0.5 x 20000) = Rs. 22500
EMV at decision node D — Rs. (0.75 x 50000 + 0.25 x 20000) = Rs. 42500
EMV at decision node 3 = Maximum out of
(a) Rs. 20000
(b) Rs. 20000 — 10000 = Rs. 10000 ?
EMV at decision node 2 = As for decision node 3, i.e., Rs. 20000
EMV at decision node B = Rs. (0.5 x 20000 + 0.5 x 20000) = Rs. (22500)
EMV at decision node A = Rs. (0.75 x 50000 + 0.25 x 20000) = Rs. 42500
EMV at decision node 1 = Maximum out of four alternatives
(a) Rs. (42500 — 20000) = Rs. 22500
(b) Rs. (22500 — 10000) = 12500
(c) Rs. 20000
(d) Rs. (20000 — 10000) = Rs. 10000
Max. = Rs. 22500
The company should conduct research P to find a new process to earn a maximum profit
of Rs. 22500.
Example 10.18. ABC Lt., has invented a picture cell phone. It is faced with selecting one alternative out
of the following strategies :
(a) Manufacture the cell phone.
(b) Take royalty from another manufacturer.
(c) Sell the rights for the invention and take a lump sum amount.
Profit in thousands of rupees which can be incurred and the probability associated with such alternatives
are shown in the table below :
Event Probability Manufacturer Royalty Sell rights
High 0.25 200 60 50
Medium 0.4 50 40 50
Low 0.35 —10 20 50
Courtesy : My Bibliotheca Page: 23 22/05/2021
DECISION THEORY 435
Represent the company problem in the form of the decision tree and suggest what decision the company
should take to maximize profits.
Solution.
Fig. 10.6
EMV at node B = Rs. (0.25 x 60000 + 0.4 x 40000 + .35 x 20000)
= Rs. (15000 + 16000 + 7000) = Rs. 38000
EMV at node A = Rs. (0.25 x 200000 + 0.4 x 50000 + 0.35 x — 10000)
= Rs. (50000 + 20000 — 3500) = Rs. 66500
EMV at decision node 1 = Maximum out of Rs. 50000, 38000, 66000 = Rs. 66500
Thus, the best decision by ABC Ltd., is to manufacture the picture cell phone itself to get profit
of Rs. 66500.
Example 10.19. The investment staff of TNC Bank is considering four investment proposals for clients,
shares, bonds, real estate and saving certificates; these investments will be held for one year. The past data
regarding the four proposals is given below :
Shares : There is 25% chance that shares will decline by 10%, 30% chance that they will remain stable
and 45% chance that they will increase in value by 15%. Also the shares under consideration do not pay
any dividends.
Bonds : These bonds stand a 40% chance of increase in value by 5% and 60% chance of remaining
stable and they yield 12%.
Real Estate : This proposal has a 20% chance of increasing 30% in value, a 25% chance of increasing
20% in value a 40% chance of increasing 10% in value, a 10% chance of remaining stable and a 5%
chance of loosing 5% of its value.
Saving Certificates : These certificates will yields 8.5 with certainty.
Use a decision tree to structure the alternatives available to the investment staff, and using the expected
value criteria choose the alternative with the highest expected value.
Solution. Let us assume that we have Rs. 1000 to invest. The decision 're shown below.
Fig. 10.7
EMV at node 5 = Rs. 1085
EMV at node 4 = Rs. (0.2 x 1300 + 0.25 x 1200 + 0.04 x 1100 + 0.05 x 9950)
= Rs. (260 + 300 + 440 + 497.50) = Rs. 1497.50
EMV at node 3 = Rs. (0.4 x 1370 + 0.06 x 1120) = Rs. (548 + 672) = Rs. 1220
EMV at node 2 = Rs. (0.25 x 900 + 0.30 x 1000 + 0.045 x 1150)
= Rs. (225 + 300 + 517.50) = Rs. 1042.50
EMV at decision node 4 is maximum, i.e., 1497.50. So the decision should be to invest in real estate.
Example 10.20. Grow Fast company is evaluating four alternatives single-period investment opportuni-
ties where returns are based on the state of economy. The possible state of the economy and the associated
probability distribution is as follows :
State Fair Good Great
Probability 0.2 0.5 0.3
The return for each investment opportunity and each state of the economy are as follows :
Statement of economy
Alternative
Fair (Rs.) Good (Rs.) Great (Rs.)
W 1000 3000 6800
X 500 4500 6800
Y 0 5000 8000
Z — 4000 6000 8500
Using the decision tree approach, determine the expected return for each alternative. Which alternative
investment proposal would you recommend if the expected monitory value criteria is to be employed ?
Rs. 1300
Rs. 1500
Courtesy : My Bibliotheca Page: 26 22/05/2021
438 OPERATIONS RESEARCH
At point 2
1. Accept B
Outcome Probability Conditional Value . Expected Value
Success 0.4 24000 9600
Failure. 0.6 -10000 - 6000
3600
2. Stop
Expected value = 0
At point 1
1. Accept A
Outcome Probability. Conditional Value Expected Value
Success 0.6 20000 + 3600 14160
Failure 0.4 - 10000 - 4000
10160
2. Accept B
Outcome Probability Conditional Value Expected Value
Success 0.4 24000 + 8000 12800
Failure 0.6 -10000 - 6000
6800
3. Do Nothing
Expected value = 0
Hence, the best strategy is to accept a first and if A is successful then accet B.
9. (a) Write a note on HurWicz criterion. Define a the index of optimism connected with it. For
what values of a the criterion is too optimistic and too pessimistic.
(b) Define 'maximum criterion'. Illustrate with an example the 'Savage minimax regret criterion'.
10. What is EMV ? How is it computed to be used as a criterion of decision-making and which ?
11. Define EVPI. How is it calculated ?
12. Explain the difference between EOL and EVPI.
13. Bring out the significance of "utility" as a superior decision criterion as compared to expected
value criterion.
14. (a) What do you understand by Decision tree analysis ?
(b) What is node in a decision tree ?
15. A decision problem has been expressed in the following payoff table :
Outcome
Action I II III
A 10 20 26
B 30 30 60
C 40 30 20
What will be the marketing managing decision of (i) Maximum (ii) Minimix (iii) Maximax
(iv) Laplace and (v), Regret criterion is applidd.
18. The estimated sale of proposed types on perfumes are as under :
Estimated level of sales (units)
Types of Perfume Rs. 20000 Rs. 10000 Rs. 20000
A 25 15 10
B 40 20 5
C 60 25 3
(a) For each of the following decision. State the optical action and specify the value leading
to its selection
(i) Maximin (ii) Maximax (iii) Laplace (iv) Minmax Regret.
(b) What will be the optimal act if payoff entries represents the costs instead of sales ?
19. Consider the following payoff (profit) matrix :
Q1 Q2 Q3 Q4
al 5 10 18 25
a2 8 7 8 23
a3 21 18 12 21
a4 30 22 19 15
Solve the using Hurwicz criterion with a = 0.73.
20. A training programme is to be organized with a batch size of al, a2, a3 and a4, which meets the
expenses S1, S2, S3 and S4, indicates the level of attendance. The table below indicates the
additional cost or incurred do to the level of attendance in different batches.
SI S2 S3 S4
ai 6 11 19 26
a2 7 6 11 22
a3 20 17 11 20
a4 29 21 18 14
Find the optimum batch size using Laplace, Minimax and Maximin criterion.
[M.B.A. II Sem. GNDU, April. 2001]