Decision Making Under Uncertainty and risk

2.4 Posterior probabilities and Bayesian analysis Compiled by Dr. A.K. Chaudhary

New information is frequently discovered through the search and evaluation of decision alternatives. The test of
alternatives is revised and expanded as new information is concerned that affects the identification of alternatives.
Consequences are restated when it concerns the effects. The states of nature itself are either reconsidered or their
probabilities are revised when the uncontrolled factors are involved. The value of new information is seen in its
impact on the expected payoff. The expected value and the cost of the new information are compared to determine
whether it is worthwhile.

A prior probability is a probability of an uncertain quantity calculated to express one's beliefs about this quantity
before some evidence is taken into account. In other words, an initial probability statement used to evaluate expected
payoff is called a prior probability distribution. The one which has been revised in the light of information which has
come to hand is called a posterior probability distribution. It will be evident that what is a posterior to one sequence
of state of nature becomes the prior to others which are yet to happen. A further analysis of problems using posterior
probabilities with respect to new expected payoffs with additional information is called prior-posterior analysis.

Posterior Analysis Technique /Bayesian Decision Rule

So far, we have used the prior probability distribution of the various states of nature to evaluate EMV or EOL for
different decision alternatives/ acts and accordingly choose the optimal decision alternative/act. Bayesian decision
rule / posterior analysis is an extension of this criterion. EMV and EOL techniques of decision making discussed so
far come under the Prior Analysis technique in as much as both are based on the prior probabilities assigned by the
decision maker on the basis of his/her perfect information, experience and subjective judgment. So, prior probability
distribution is not always the best predictor. To optimize his/her action, the decision maker might be interested to
look in for the best (exact or most accurate) predictor. For this, the posterior analysis technique of decision making
is more appropriate that is based on the posterior probabilities which are determined on the basis of Bayes theorem
in the light of the additional information obtained by sampling techniques, market survey, experiments, and test
research and through some other agencies. We have already discussed the concept of EVPI, which provides the
upper limit to the amount of money, the decision maker would be willing to spend to gather the additional
information about various states of nature.

In posterior analysis, the prior probabilities are revised to obtain posterior probabilities on the basis of the additional
information gathered. These posterior probabilities are then used to obtain the expected payoffs for different
combinations of decision alternatives/acts and events. Finally, the decision alternative/act with the maximum EMV
is chosen as the best decision on alternative act. This process is likely to result in better decision.

This technique is applied only when the EVPI or the EOL (i.e. the cost of transferring a risk into certainty) obtained
under any prior analysis technique appears to be very high on account of imperfect information. In such cases, the
prior probabilities are converted into the posterior probabilities and then the revised EVPI (EOL) is computed again
on the basis of such revised or posterior probabilities. According to Bayes' theorem, the posterior probabilities are
obtained by the following model:

Prior probability of an event×Conditional probability of the event

Posterior probability of an event = Sum of products of the prior and conditional probability of all the events

Joint probability of the event

=
Revised Marginal probabilty of a strategy
In posterior analysis, certain factors related to the technique to compute and use of posterior probabilities in decision
making may be noted as under:

1. Average Minimum Expected Opportunity Loss (AMEOL)

2. Expected Value of Sample Information (EVSI)
3. Expected Net Gain from Sampling (ENGS)
1. Average Minimum Expected Opportunity Loss (AMEOL)
This represents the expected value of the market research which is obtained on the basis of the posterior
probabilities. In other words, this represents the revised EVPI which is obtained by the following formula:
AMEOL = ∑ Posterior EOL × Revised marginal probability
2. Expected Value of the Sample Information (EVSI)
The information, on the basis of which the prior probabilities are revised to posterior probabilities is generally called
sample information. The expected value of the sample information represents the difference between the EVPI and
the AMEOL. It is denoted by EVSI. Thus,
EVSI = EVPI –AMEOL
Alternately,
EVSI= Expected payoff with sample information − Expected payoff without sample information
= ∑[(Probability of forecast ( predicted) outcome) × ( Maximum posterior payoff over different acts)]
− Maximum EMV based on prior probabilities,
the summation being taken over different states of nature.
The expected value of sample information (EVSI) measures the “expected reduction in expected loss” from a given
research study. In decision theory, the expected value of sample information (EVSI) is the expected increase in
utility that a decision-maker could obtain from gaining access to a sample of additional observations before making
a decision. The additional information obtained from the sample may allow them to make a more informed, and thus
better decision, thus resulting in an increase in expected utility. EVSI attempts to estimate what this improvement
would be before seeing actual sample data; hence, EVSI is a form of what is known as pre-posterior analysis. The
gain in expected return obtained from knowing with certainty the future state of nature is called Expected Value of
Perfect Information (EVPI).
The Efficiency of EVSI
The efficiency of EVSI is determined by the efficiency index (EI), given by
EVSI
EI = EVPI × 100 = x % ( say)
This shows that the sample information is x % as efficient as perfect information. The EI enables the decision maker to
decide if he/she should go in for additional information. If x < 1, it is not worth going for the additional information.
3. Expected Net Gain from Sampling (ENGS)
This is obtained by subtracting the expected cost of sampling or market study from the EVSI. Thus,
ENGS = EVSI – ECS,
ECS = Cost of Sampling
By cost of sampling, we mean the expenditure incurred on obtaining the additional information about different states of
nature.
If ENGS is negative, then it is not worth to collect the additional information.
Steps for Posterior Analysis
For application of the posterior analysis technique, the following steps are to be taken up in turn:
(i) Find the EVPI according to a prior analysis technique viz. EMV, EOL etc.
(ii) Verify that the EVPI thus obtained appears to be high and that there is additional information available
relating to the conditional probability of the different states of nature.
(iii) Find the posterior probability of each of the events applying the Bayes’ technique as per the following
table:
Compiled by Dr. A.K. Chaudhary
 Bayes’ Posterior Probability Distribution
Event Prior prob.(i) Conditional prob.(ii) Joint prob.(i×ii) Posterior prob.=
(J.P. ÷Revised Marginal prob.)

Total 1 Revised Marginal prob. 1

(iv) Construct the expected opportunity loss table again on the basis of the posterior probabilities and find the
posterior EOL for each of the acts thereby.
(v) Determine the value of the net revised EOL of the market research i.e. AMEOL by the following table.
Event Optimal act Posterior EOL Revised marginal Expected posterior
prob. of the event EOL

Total ∑EOL
(vi) Determine the EVSI by the model:
EVSI = EVPI – AMEOL
(vii) Determine the ENGS by the model:
ENGS = EVSI – ECS
(viii) If there is ENGS, state that the expected cost of sampling is desirable otherwise not.
Example1. There are three varieties of machines viz. A, B and C of which any one is to be purchased by a firm. The
profits expected from the different machines under different markets conditions are tabulated as under:
Profit matrix (in ‘000 Rs.)
Market Machine
A B C
Poor 0.50 0.00 −1.50
Fair 1.00 1.50 0.50
Good 1.50 2.50 3.50
The probabilities of the market being poor, fair, and good were estimated by the firm to be 0.3, 0.5 and 0.2
respectively. But a research group opined that these probabilities can not to be relied upon to be accurate and they
assessed the following chances after a careful survey of the market conditions.
Actual state Indicated state
Poor Fair Good
Poor 0.7 0.2 0.1
Fair 0.2 0.7 0.1
Good 0.0 0.2 0.8
From the above particulars
(i) Construct the expected opportunity loss table basing on the market assessment made by the firm and
determine the optimal act using EOL criterion. Also, determine the EVPI therefrom.
(ii) Construct the EOL table basing upon the information furnished by the research group and determine the net
revised EOL (i.e. AMEOL) thereby.
(iii) State the maximum amount that should be spent after the market research.
(iv) Also, find the expected net gain from the sampling (ENGS), if the decision maker is prepared to obtain
additional information for Rs.110.
Solution:
Prior Analysis
The payoff table/profit matrix is given as follows:
 Payoff table [Profit matrix (in ‘000 Rs.)]
Act A B C
Event
Poor 0.50 0.00 −1.50
Fair 1.00 1.50 0.50
Good 1.50 2.50 3.50
The regret table (opportunity loss table) can be obtained by subtracting payoff value of each event in the row from
maximum payoff value of corresponding event of that row.
Regret table [OL table]
Act A B C
Event
Poor 0.50−0.50=0 0.5−0= 0.5 0.5−(−1.50) =2
Fair 1.50−1.00=0.5 1.50−1.50=0 1.50−0.50=1
Good 3.50−1.50=2 3.50−2.50=1 3.50−3.50=0

The expected opportunity loss (EOL) value of each act by using formula [expected loss = Prob. ×OL] can be
obtained as follows:
Expected opportunity table [EOL table]
Act A B C
Event Prob.
Poor 0.3 0.3×0=0 0.3×0.5=0.15 0.3×2=0.6
Fair 0.5 0.5×0.5=0.25 0.5×0=0 0.5×1=0.5
Good 0.2 0.2×2=0.4 0.2×1=0.2 0.2×0=0
EOL Total 0.65 0.35 1.10

From the above EOL table constructed on the basis of the prior probabilities, we have found that the EOL for act B
is least value i.e. 0.35 or Rs.350. Thus, best act by prior analysis is machine B. We know that EVPI means minimum
EOL. Thus,
EVPI = Minimum EOL = Rs.350 which is the cost of transferring the risk to uncertainty. This represents the
maximum amount of money that decision maker has to pay to get additional information about various states of
nature before decision has to be made.
Posterior Analysis
Since the research group advice not to rely on the prior probabilities thus estimated and to revise such probabilities
in the light of the additional information i.e., the given conditional probabilities, the posterior analysis of the
problem is to be made as under:
Event Prior Conditional Probability Joint Probability Posterior Probability
Prob. Poor Fair Good Poor Fair Good Poor Fair Good
Poor 0.3 0.7 0.2 0.1 0.21 0.06 0.03 0.68 0.13 0.12
Fair 0.5 0.2 0.7 0.1 0.10 0.35 0.05 0.32 0.78 0.21
Good 0.2 0 0.2 0.8 0 0.04 0.16 0 0.09 0.67
Total 1 0.31 0.45 0.24 1 1 1

Revised Marginal Probability

The posterior probability for each of the states have been calculated by using following formula:
 The particular joint probability
Posterior probability = Corresponding Marginal Probability
0.21
Thus, posterior probability for the (Poor, Poor) = = 0.68
0.31
0.10
Posterior probability for the (Fair, Poor) = = 0.32
0.31
0
Posterior probability for the (Good, Poor) = = 0 and so on
0.31
Now, the revised EOL table will be constructed as under:
Revised EOL table
Act A B C

Posterior OL EOL Posterior OL EOL Posterior OL EOL

Event Probability Probability Probability
Poor 0.68 0 0 0.13 0.5 0.065 0.12 2 0.24
Fair 0.32 0.5 0.16 0.78 0 0 0.21 1 0.21
Good 0 2 0 0.09 1 0.09 0.67 0 0
Total Posterior EOL 0.160 Total Posterior EOL 0.155 Total Posterior EOL 0.450

Determination of the Net Revised EOL

Event Optimal Act Posterior EOL (i) Revised marginal Probability (ii)Expected Posterior EOL (in
‘000’Rs) (i × ii)
Poor A 0.160 0.31 0.05
Fair B 0.155 0.45 0.07
Good C 0.450 0.24 0.11
Total AMEOL 0.23
From the above table it must be seen that the EOL could be reduced to 0.23 (‘000’Rs) from 0.35 (‘000’Rs) by the
market research. Thus, the net gain from the market research is
EVSI = EVPI –AMEOL =Rs.350 – Rs.230 = Rs.120
(iii). Hence, the maximum amount that should be spent after the market research is Rs.120.
To sum up, the values of the various factors are as under:
(i) EVPI as per prior analysis = Rs.350 [i.e. 0.35 in ‘000’Rs.]
(ii) AMEOL as per posterior analysis = Rs.230 [i.e. 0.23 in ‘000’Rs.]
(iii) EVSI = EVPI –AMEOL =Rs.350 – Rs.230 = Rs.120
(iv) ENGS = EVSI – ECS (expected cost of sampling) = Rs.120 – Rs.110 = 0

Example2. A manager is faced the problem of choosing one of the three products (acts): product A (act A1), product
B (act A2) and product C (act A3), for production. The potential demand (states of nature) for each product may turn
out to be poor (N1), moderate (N2) and good (N3). The payoff table (in Rs.) for different combinations of acts and
events is given in following Table.
Acts A1 A2 A3
States of nature
N1 −20 −50 200
N2 200 −100 −50
N3 400 600 300
The probabilities of the states of nature are 0.3, 0.4 and 0.3 respectively.
(i) Determine the optimal act using the EMV criterion.
 (ii) Also, find the expected value of perfect information (EVPI).
(iii) If the decision maker is prepared to obtain additional information for Rs.110, should the information be
collected?
(iv) If he/she collects the additional information and gives the following conditional probabilities for the states
of nature, obtain the posterior probabilities and the posterior payoff table.
Actual state of nature Predicted State of nature (Demand)
(Demand) X1(Low) X2(Moderate) X3(Good)
N1 (Low) 0.5 0.3 0.2
N2(Moderate) 0.3 0.6 0.1
N3(Good) 0.2 0.1 0.7
(v) What is the optimal decision based on posterior analysis?
(vi) Also, find the expected value of sample information (EVSI) and the expected net gain from sampling
(ENGS).
Solution:
(i) The payoff table in Rs. and the probabilities of the states of nature are given below:
States of nature Probability Payoff (in Rs.)
Act A1 Act A2 Act A3
N1 0.3 −20 −50 200
N2 0.4 200 −100 −50
N3 0.3 400 600 300
The EMV for each act can be calculated as follows:
EMV for A1 = ∑Payoff ×Probability
= −20×0.3 + 200×0.4 + 400×0.3= −6 + 80 + 120= Rs.194
EMV for A2 = ∑Payoff ×Probability
= −50×0.3 + (−100×0.4) + 600×0.3= −15 − 40 + 180= Rs.125
EMV for A3 = ∑Payoff ×Probability
= 200×0.3 + (−50×0.4) + 300×0.3= 60 − 20 + 90= Rs.130
Since EMV for act A1 has maximum value by EMV criterion, act A1 is the optimal act.
(ii) The expected profit under perfect information (EPPI) is given by
EPPI = ∑Maximum payoff over different acts ×Probability
= 200×0.3 + 200×0.4 + 600×0.3= 60 + 80 + 180= Rs.320
EVPI = EPPI – Maximum EMV under uncertainty= Rs.320 − Rs.194 =Rs.126
(iii) EVPI represents the maximum amount of money that the decision maker has to pay to get additional
information before decision has to be made.
Here, the cost of additional information = Rs.110 and EVPI = Rs.126
Net gain = Rs.126 – Rs.110 = Rs.16
Since EVPI = Rs.126 > Rs.110, the decision maker should decide to go in for the additional information to
get net gain Rs.16.
(iv) Posterior Analysis

Compiled by Dr. A.K. Chaudhary

 Joint and Marginal Probabilities
State Prior Conditional Joint Probability: P(N∩X)
of Prob. Probability
nature X1 X2 X3 X1 X2 X3
N1 0.3 0.5 0.3 0.2 0.15 0.09 0.06
N2 0.4 0.3 0.6 0.1 0.12 0.24 0.04
N3 0.3 0.2 0.1 0.7 0.06 0.03 0.21
Total 1 P(X1)= P(X2)= P(X3)
0.33 0.36 =0.31

Revised Marginal Probability

The revised conditional probabilities i.e. the posterior probabilities for various states of nature are obtained
on dividing the column probabilities in the above table by the corresponding column totals and are given in
the following table.
Posterior Probabilities
States of nature Posterior probabilities: P(Ni/Xj)
P(Ni/X1) P(Ni/X2) P(Ni/X3)
N1 0.15/0.33=0.45 0.09/0.36=0.25 0.06/0.31=0.19
N2 0.12/0.33=0.36 0.24/0.36=0.67 0.04/0.31=0.13
N3 0.06/0.33=0.19 0.03/0.36=0.08 0.21/0.31=0.68
Total 1 1 1
Next step is to compute the expected posterior payoffs (EMV’s) i.e. revised payoffs for various acts, given
that the forecast (Xj), assuming that the different states of nature follow the posterior probability
distribution as given in the following table.
Calculations of expected posterior payoffs
States of Posterior Conditional Payoff (Rs) Expected Payoff (Rs.)
nature Prob. Act A1 Act A2 Act A3 Act A1 Act A2 Act A3
P(N/X)
N1 0.45 −20 −50 200 −9 −22.5 90
N2 0.36 200 −100 −50 72 −36 −18
N3 0.19 400 600 300 76 114 57
Total Total Posterior EMV 139 55.5 129
N1 0.25 −20 −50 200 −5 −12.5 50
N2 0.67 200 −100 −50 134 −67 33.5
N3 0.08 400 600 300 32 48 24
Total Total Posterior EMV 161 −31.5 40.5
N1 0.19 −20 −50 200 −3.8 −9.5 38
N2 0.13 200 −100 −50 26 −13 −6.5
N3 0.68 400 600 300 272 408 204
Total Total Posterior EMV 294.2 385.5 235.5
Posterior EMV /Expected Posterior Payoff (in Rs.) table
Outcome Posterior EMV Maximum Marginal p×M
Act A1 Act A2 Act A3 payoff in Probability
row (M) (p)
X1 139 55.5 129 139 0.33 45.87
X2 161 −31.5 40.5 161 0.36 57.96
X3 294.2 385.5 235.5 385.5 0.31 119.5
 Total ∑pM= 223.33
(v) The expected value of sample information is
EVSI =∑[(probability of outcome Xi) × (Maximum posterior payoff over different acts)
−Maximum EMV based on prior probabilities
= ∑p×M –Rs.194 = Rs.(223.33 −194) = Rs.29.33
The expected net gain from sampling is
ENGS = EVSI – ECS, where ECS is expected cost of sampling (market survey)
= Rs.( 29.33 −110) = −Rs.80.67
Since ENGS is negative, it is not worth for the management to go in for the additional information. Hence, the
optimum decision is the act A1 (i.e., to produce the product A) as in part (i).
2.5 Decision Making with Utilities

"Utility" is the level of satisfaction or benefits that someone gains from consuming a given amount of goods or
services. In economics, the unit of utility is known as "Utils".

So far problems were analyzed with probabilities of states of nature where selection of an optimal course of action
was based on the criterion of expected profit (or loss) expressed in monetary terms. However, in many situations,
such criteria involving expected monetary payoff may not be appropriate. This is because of the fact that different
individuals attach different utility to money under different conditions. The term utility is the measure of preference
that individuals have for various alternatives available to them. The utility of a given alternative is unique to
individual decision-makers and, unlike a simple monetary amount, can incorporate intangible factors or subjective
standards from their own value systems.

2.5.1Utility Functions

Utility functions typically describe the relative preference value that individuals have for a given amount of the
criterion such as money, goods, etc. They are often determined by proposing a situation to the subjects whereby they
must choose between receiving a given amount, say Rs.20, 000 for certain thing versus a 50-50 chance of gaining a
larger amount or nothing say Rs.60, 000 or zero. The gamble amount Rs.60, 000 is then adjusted upward or
downward until the individual is indifferent to whether decision-maker receives the certain amount of Rs.20, 000 or
the gamble. This indifference point then establishes one experimentally determined value on the individual's utility
curve and other points are similarly determined.

Utility Function is a mathematical function that transforms monetary values into utility values. In other words, the
utility function is the mathematical representation of a consumer’s preferences. Once derived, a utility function can
be used to convert a decision criteria value into utils so that a decision can be made on the basis of maximizing the
expected utility value (EUV) rather than, say EMV. Three general types of utility functions are (1) Risk-Averse (2)
Risk-Neutral and (3) Risk-Seeking (or loving) .

2.5.2 Expected Utility

Expected utility is a theory in economics that estimates the utility of an action when the outcome is uncertain. It advises
choosing the action or event with the maximum expected utility. At any point in time, the expected utility will be the
weighted average of all the probable utility levels that an entity is expected to reach under specific circumstances.
Under this alternative criterion, it is assumed that a rational decision-maker will choose that alternative which optimizes
the expected utility rather than expected monetary one. Once we know the individual's utility function along with the
probability assigned to outcomes in a particular situation, then total expected utility for each course of action can be
obtained by multiplying utility values with their probabilities. Thus,
Expected Utility = ∑Utility value ×Probability =PAU(A) + PB U(B) + PcU(C)
The strategy which corresponds to optimum utility function is called the equal strategy. Expected utility is used as a tool
for decision-making under circumstances where the outcomes of decisions are not known.
The expected utility theory considers it a logical choice to choose the event with the maximum expected utility. However,
in case of risky outcomes, decision-makers may not choose the action with a higher expected utility. The decision to
choose an action will also depend on the entity’s risk aversion and other entities’ utility. While some entities choose the
option with the riskier highest expected utility, some highly risk-averse entities prefer the low-risk option even if it shows
a lower expected value.

Expected utility theory also helps to explain the reason for people taking out insurance policies. It is a situation where the
payback is not immediate; however, insurance policies cover individuals for several risks. Insurance policyholders receive
tax benefits and a certain income at the expiry of a predetermined period. Hence, when one compares the expected utility
to be received from paying insurance premiums with the expected utility of investing the amount on other products,
insurance appears to be a better choice.

The concepts of marginal utility and expected utility are related. The expected utility of wealth or a reward reduces when
the entity possesses sufficient wealth. Such entities may go for the safer alternative instead of the riskier ones.

The addition of $1,000 to the income may not impact the marginal utility of two different entities in the same way. For
example, if the annual income of a low-earning family is increased from $1,250 to $2,250, it will improve their quality of
life as well as the marginal utility. On the contrary, if the income of a high-earning family increases from $120,000 to
$121,000 in a year, there is a very small utility improvement.
Example3. Here is the information given below for playing two gambles 1 and 2.
Gamble1 Gamble 2
20% chance of winning Rs.6000 25% chance of winning Rs.4000
80% chance of winning Rs.0 75% chance of winning Rs.0
Suppose consumer’s utility of money is U(X) =X.
(i) Find expected utility of each gamble.
(ii) Select best option using EUV criterion.
Solution:
The expected utility of gamble1 =EUV1= p1U1+p2U2 = 0.20 ×Rs.6000 +0.80×Rs.0 = Rs.1200
The expected utility of gamble2 =EUV2= p1U1+p2U2 = 0.25 ×Rs.4000 +0.75×Rs.0 = Rs.1000
Since EUV1 > EUV2, consumer will choose gamble 1 that maximizes the expected utility.

2.5.3 Utility Curve/Indifference Curve

A utility curve that relates utility values to rupees values is constructed. Such a curve usually is obtained by placing
the decision-maker in various hypothetical decision situations and plotting the decision-maker’s pattern of choices in
terms of risk and utilities. The following figure shows several utility curves and the risk preference associated with
each.
The utility curve exhibits whether a consumer is risk aversion (avoider), risk indifference (neutrality) or risk affinity
(seeker or loving). It determines whether a given utility function is consistent with a consumer being risk averse, risk
neutrality or risk loving.

Risk Preferences and Shape of the utility function

Case1. Risk neutrality: Risk neutral consumers always choose the option with the highest expected value.
Maximizing expected utility  Maximizing expected value
For risk neutral consumer, the utility of getting the expected value of a gamble for certain equals the expected utility
of the gamble. Thus, E (U) = U
Risk neutral = Linear utility function
Case2. Risk Averse
Risk Averse = Concave utility function
For risk averse consumer, the utility of getting the expected value of a gamble for certain is greater than the expected
utility of the gamble. Thus, U > E (U)
Case3. Risk Loving (Seeker)
Risk Loving = Convex utility function
For risk loving consumer, the utility of getting the expected value of a gamble for certain is less than the expected
utility of the gamble. Thus, U < E (U).
2.6 Decision making under ignorance

Ignorance is an extreme form of uncertainty. In most narrow and technical sense, it means inability to assign a
meaningful probability to the phenomena of interest. Decisions making under ignorance are decisions made when
one cannot assign a subjective probability to the possible outcomes of any act i.e., under ignorance, decision maker
knows all possible states of nature but does not know the probability of occurrence. For decisions making under
ignorance there are no probabilities available to the agent. So, the standard principle for decision making, that is, the
principle of maximizing expected utility cannot be used. Under the decision making under ignorance, the following
three criteria are used to select the best decision alternative (best act).

(a) Maximax criterion

(b) Maximin criterion
 (c) Laplace-Bayes criterion
(a) Maximax Criterion
Under maximax criterion, decision maker first selects the “best of the best” strategy with the optimistic point
of view. He/she evaluates each decision by the maximum possible return associated with that
decision ( if cost data is used, the minimum return is “best”).Then, decision maker would select
those decision alternative (act) which gives the maximum of these maximum returns (maximax).
Under maximax criterion, for “risk takers”, decision maker doesn’t consider the “down side” risk
And ignores the possible losses from the selected alternative.
(b) Maximin Criterion
Under maximin criterion, decision maker selects the “best of the worst” strategy with the
pessimistic point of view. He/she evaluates each decision by the minimum possible return
associated with the decision. Then, the decision maker would select those decision alternative (act
or strategy) which gives the maximum value of the minimum returns (maximin) i.e. would select
the strategy with the smallest possible loss. For “risk averse” decision makers, select a “protect”
strategy and focus on the worst case scenario.
(c) Laplace-Bayes Criterion
Under Laplace-Bayes criterion, decision maker assumes that all states of nature are equally likely to occur. He
/she finds average payoff of each decision alternative. Then, he/she would select those decision alternative
which gives best average payoff.
Example4. Kelly Construction wants to get in on the boom of student condominium construction. The company
must decide whether to purchase enough land to build a 50-, 100-, or 150-unit condominium complex. Many other
complexes are currently under construction, so Kelly is unsure how strong demand for its complex will be. If the
company is conservative and builds only a few units, it loses potential profits if the demand turns out to be high. On
the other hand, many unsold units would also be costly to Kelly. The following table has been prepared, based on
three levels of demand.
Kelly Construction payoff table (Rs.)
Alternative actions States of nature (Demand)
Low (50 units) Medium (100 units) High (100 units)
Build 50 400000 400000 400000
Build 100 100000 800000 800000
Build 150 200000 500000 1200000

Under decision making with ignorance, select best alternative action based on (i) maximax criterion (ii) maximin
criterion and (iii) Laplace-Bayes criterion.
Solution:
(i) Here,
Kelly Construction payoff table (Rs.)
Alternative actions States of nature (Demand) Maximan criterion
Low (50 units) Medium (100 units) High (100 units) Maximum payoff
Build 50 400000 400000 400000 400000
Build 100 100000 800000 800000 800000
Build 150 200000 500000 1200000 1200000 maximax
Under decision making with ignorance based on maximax criterion, decision maker should build 150 which gives
maximum payoff Rs.1200000 among all selected maximum payoff values Rs.400000, Rs.800000 and Rs.1200000.
(ii) Here,
Kelly Construction payoff table (Rs.)
Alternative actions States of nature (Demand) Maximin criterion
Low (50 units) Medium (100 units) High (100 units) Minimum payoff
Build 50 400000 400000 400000 400000Maximin
Build 100 100000 800000 800000 100000
Build 150 200000 500000 1200000 200000
Under decision making with ignorance based on maximin criterion, decision maker should build 50 which gives
maximum payoff Rs.400000 among all selected minimum payoff values Rs.400000, Rs.100000 and Rs.200000.
(iii) Here,
Kelly Construction payoff table (Rs.)
Alternative actions States of nature (Demand) Laplace-Bayes criterion
Low (50 units) Medium (100 units) High (100 units) Average payoff Rs
Build 50 400000 400000 400000 400000
Build 100 100000 800000 800000 566667
Build 150 200000 500000 1200000 633334
Under decision making with ignorance based on Laplace-Bayes criterion, decision maker should build 150 which
gives maximum average payoff Rs.633334 among all selected minimum payoff values Rs.400000, Rs.566667 and
Rs.633334.
2.7 Risk analysis and Sensitivity analysis.
Risk analysis helps the decision maker recognize the difference between: the expected value of a decision
alternative, and the payoff that might actually occur. The risk profile for a decision alternative shows the possible
payoffs for the decision alternative along with their associated probabilities.
Sensitivity analysis can be used to determine how changes to the following inputs affect the recommended decision
alternative:
• probabilities for the states of nature
• values of the payoffs
If a small change in the value of one of the inputs causes a change in the recommended decision alternative, extra
effort and care should be taken in estimating the input value.