MA30118 - Solution Sheet One

Simon Shaw
[email protected]

2005/06 Semester II

1. Lincoln City are considering the number of programmes to print for Sat-
urday’s big match. They believe that there is a 35% chance that there will
be a heavy turnout (S1 ), a 50% chance for a normal turnout (S2 ), and a
15% chance for a low turnout (S3 ). They must decide whether to print 500
copies (A1 ), 750 copies (A2 ), 1000 copies (A3 ), or 1250 copies (A4 ). The
payoff table, Table 1, is given below. Unsold programmes would result in a
loss.
S1 S2 S3
A1 100 100 100
A2 150 140 110
A3 200 160 75
A4 250 120 -50

Table 1: Payoffs for programme production

(a) Are all the actions admissible?

The payoff for A2 is always greater than that for A1 whatever the state of nature.
Thus, A2 dominates A1 and A1 is inadmissible. We may remove A1 from any
further consideration.
(b) Find the minimax regret decision.

We find the best payoffs under each state of nature. Thus, π ∗ (S1 ) = 250, π ∗ (S2 ) =
160 and π ∗ (S3 ) = 110. The corresponding opportunity regrets are shown in Table
2. From Table 2 we find the largest opportunity regret for each action. Thus,

S1 S2 S3
A2 250 - 150 = 100 160 - 140 = 20 110 - 110 = 0
A3 250 - 200 = 50 160 - 160 = 0 110 - 75 = 35
A4 250 - 250 = 0 160 - 120 = 40 110 - (-50) = 160

Table 2: Opportunity regrets for programme production

Action Maximum opportunity regret

A2 100
A3 50
A4 160

1
 The minimax regret decision is the one that minimises these. In this case, the
minimax regret decision is A3 , we print 1000 copies.
(c) Find the decision which maximises expected payoff.

We calculate the expected monetary value for each three of the admissible actions,
where, as there are three states of nature,
3
X
EM V (Ai ) = π(Ai , Sj )P (Sj ).
j=1

Thus,

EM V (A2 ) = 150(0.35) + 140(0.50) + 110(0.15) = 139,

EM V (A3 ) = 200(0.35) + 160(0.50) + 75(0.15) = 161.25,
EM V (A4 ) = 250(0.35) + 120(0.50) + (−50)(0.15) = 140.

Hence, EM V = max{139, 161.25, 140} = 161.25. The decision which maximises

EM V is A3 , we print 1000 copies.

2. A manager is considering his staffing needs and has the following strate-
gies: Lay off two staff employees (A1 ); maintain staff at current levels (A2 );
increase the staff size by one employee (A3 ); increase the staff size by two
employees (A4 ). There are four possible states of nature: Business will
decrease (S1 ); business will stay the same (S2 ); business will increase mod-
erately (S3 ); and business will increase rapidly (S4 ). The prior probabilities
are P (S1 ) = 0.15, P (S2 ) = 0.25, P (S3 ) = 0.25 and P (S4 ) = 0.35. The possible
payoffs, in pounds as profits, are given in Table 3 overleaf. Find the decision

S1 S2 S3 S4
A1 1451 1840 2050 2300
A2 -1091 1685 2430 2900
A3 -2015 1100 3060 3561
A4 -3460 -1350 3340 4300

Table 3: Payoffs for the choice of staffing levels.

which maximises expected payoff.

We calculate the EMV for each action.

EM V (A1 ) = 1451(0.15) + 1840(0.25) + 2050(0.25) + 2300(0.35) = 1995.15

EM V (A2 ) = −1091(0.15) + 1685(0.25) + 2430(0.25) + 2900(0.35) = 1880.1
EM V (A3 ) = −2015(0.15) + 1100(0.25) + 3060(0.25) + 3561(0.35) = 1984.1
EM V (A4 ) = −3460(0.15) − 1350(0.25) + 3340(0.25) + 4300(0.35) = 1483.5.

Thus, EM V = maxi EM V (Ai ) = 1995.1 under action A1 : we lay off two staff employ-
ees.

2
3. A firm decides to build a new factory which will last 10 years. Let h1 denote
the event that the demand is high for the first two years, l1 the event that
the demand is low for the first two years, h2 that the demand is high for
the last eight years and l2 that the demand is low for the last eight years.
These are the only possible outcomes. Marketing information reveals that
P (h1 , h2 ) = P (h1 ∩ h2 ) = 0.6, P (h1 , l2 ) = P (h1 ∩ l2 ) = 0.1, P (l1 , l2 ) = P (l1 ∩ l2 ) = 0.3.
Find P (h1 ), P (h2 |h1 ) and P (l2 |h1 ).

As demand for the final eight years can either be high (h2 ) or low (l2 ), then

P (h1 ) = P (h1 ∩ h2 ) + P (h1 ∩ l2 ) = 0.6 + 0.1 = 0.7.

Note that since demand for the first two years is either high (h1 ) or low (l1 ), it follows
immediately that

P (l1 ) = 1 − P (h1 ) = 0.3,

which, since we can’t follow initial low demand with high demand, agrees with P (l1 ∩l2 ).
Using the definition of conditional probability,

P (h1 ∩ h2 ) 0.6 6
P (h2 |h1 ) = = = ,
P (h1 ) 0.7 7
P (h1 ∩ l2 ) 0.1 1
P (l2 |h1 ) = = = .
P (h1 ) 0.7 7

Notice that, given h1 , demand for the last eight years is either h2 or l2 so that
P (h2 |h1 ) + P (l2 |h1 ) = 1.

4. The firm in question 3. may either build a big factory or a small factory
which may then be expanded. If it builds a big factory and demand is
high for the full ten years then the profit is £14million. If it builds big
and demand is high for the first two years and low for the remaining eight
then the profit is −£0.4million. A big factory with low demand for the
entire ten years gives a profit of −£4million. If the firm builds a small
factory and demand is high for the first two years then it may expand the
factory. Expansion followed by eight years of high demand yields a profit of
£6million, while expansion followed by eight years of low demand results in
a profit of -£4.4million. If the factory is not expanded and demand remains
high, then the profit is £3.2million, while if the demand falls to low for the
remaining eight years then the profit is £5.6million. A small plant with low
demand for the whole ten years returns a profit of £5.4million.

(a) Construct the decision tree for the firm.

This is given in Figure 1.

(b) Find the expected value of perfect information (EVPI) for the firm.

We first find the expected monetary value under certainty, EM V U C. For the
whole ten years, there are three possible outcomes, h1 ∩ h2 , h1 ∩ l2 and l1 ∩ l2 .

3
 Figure 1: The decision tree for the factory building question

Now,

π ∗ (h1 ∩ h2 ) = max{14, 6, 3.2} = 14, (1)

π ∗ (h1 ∩ l2 ) = max{−0.4, −4.4, 5.6} = 5.6, (2)
π ∗ (l1 ∩ l2 ) = max{−4, 5.4} = 5.4. (3)

Thus,

EM V U C = π ∗ (h1 ∩ h2 )P (h1 ∩ h2 ) + π ∗ (h1 ∩ l2 )P (h1 ∩ l2 ) + π ∗ (l1 ∩ l2 )P (l1 ∩ l2 )

= 14(0.6) + 5.6(0.1) + 5.4(0.3) = 10.58.

From Figure 1, EM V = 7.16, so that

EV P I = EM V U C − EM V = 10.58 − 7.16 = 3.42.

(c) Which decision maximises the expected payoff. Comment on this

choice of action.

We should choose to build a large plant. Notice that having taken this decision, we
will get a payoff of £14million with probability 0.6, −£0.4million with probability
0.1 and −£4million with probability 0.3. There is a (large) probability of 0.4 that
we will lose money. The EM V of this decision is dominated by the large payoff of
£14million. The EM V takes no account of the firms’ ability to absorb any loss,
and thus its’ attitude to risk.