Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

Master Course in

Business Administration

Today(we(will(talk(about(trees(

Decision(Theory(
(Decision(Trees(
Quan0ta0ve(Methods(for(
Managerial(Decision8Making(
Bura;o(Alessandra(

(Decision Trees(

Decision(Theory(
What(do(decision(trees(need(for?(

Fundamentals:(
(
(ALTERNATIVES((set(of((strategies:(mutually(exclusive).(
(
EVENTS((states(of(nature(uncontrollable,(mutually(exclusive((and(
comprehensive)(!(in(chronological(order.(
(PROBABILITIES(for(each(event.(
(PAYOFF(associated(to(every(couple(alterna0ve/event((MONEY)(

(to(make(decisions!(

(DECISION(RULE((criterium)(to(value(the(alterna0ves.(
(

((((((The(decision(maker(chooses(one(of(the(alterna0ves(

Best(decision:(
with(respect(to(which(criterium?(
Decision(rules:(
MAXIMIN:(best(choice(among(the(worst((pessimism(and(risk(
aversion).( ((
((
MAXIMAX:(best(choice(among(the(best((op0mism(and(risk(
proneness().(
EXPECTED(VALUE:(the(greater(expected(payo(((weighted(
expecta0on(of(payos(with(probabili0es((as(weights().(

Valua0on(of(probabili0es(
Importance(of(INFORMATION(availability,(to(understand(
with(which(probability(the(event(will(occur.(
(
Frequen0st(Deni0on((available(informa0on):((

((
(p =

#. positive outcomes of an experiment


#. total trials

((Subjec0ve(Deni0on((without(informa0on):((
((((trus_ulness(in(the(events(realiza0on.(

Lets(be(P(E)(the(probability(that(event(E(takes(place:(its(a(
number(
(
0 P( E ) 1
(
In(case(of(n(mutually(exclusive(realiza0ons(E1,,En(:(
(
((((((((((((((((((((((((((((((((((((((((P(E1)(+((+(P(En)(=(1(
(
Example:(toss(of(a(dice((
((((
((((E1(=(even(number(
P(E1)(=(0.5(
((((E2(=(odd(number(
((((((((P(E2)(=(0.5(
(

Example(1(

Events

Alternatives

A1
A2

P(E1)(+(P(E2)(=(0.5(+(0.5(=(1(

Example(1:(decision(criteria(
!

Events

Alternatives

A1
A2

E1

E2

(
200(

(
270(

(
160

(
800(

(
(((((Maximin:((A1(((((Min({200,270}(=(200(
((((((((((((((((((((((((((A2(((((Min({160,800}(=(160(
(((((Maximax:(A1((((Max({200,270}(=(270(
((((((((((((((((((((((((((A2((((Max({160,800}(=(800(
((

E1

E2

(
200(

(
270(

(
160

(
800(

Payo(matrix(

(
Example(1:((
Expected(Monetary(Value(criterium(
(
P(E1)(=(p((((((((P(E2)(=(18p(
EMV((Expected(Monetary(Value)(=(Weighted(mean(of(payos(
with(probabili0es(as(weights.(
Events
Alternatives

A1

E1

E2

(p)!

(1&p)!

(
200(

(
270(

(EMV)

((Max({200,160}=200(
((I(choose(A1(
((((((((Max({270,800}=800(
(((((((((I(choose(A2(

(((EMV(A1)(=((200*(p(+(270*(18p))(=( (

Example 1:
Expected Monetary Value criterium(
P(E1)(=(0.4((40%)(((((((P(E2)(=(0.6((60%)(

Example 1:
Expected Monetary Value criterium(
P(E1)(=(0.4((40%)(((((((P(E2)(=(0.6((60%)(
(

Events

Alternatives
Events
Alternatives

A1

E1

E2

(EMV)

(0.4)!
(
200(

(0.6)!
(
270(

242(

EMV(A1)(=((200*0.4+270*0.6)(=(

242(

A1

A2

!
!EMV(A1)(=(242(
(EMV(A2)(=(544!!!!

E1

E2

(EMV)

(0.4)!
(
200(

(0.6)!
(
270(

(
(200*0.4+270*0.6)=242(

(
160

(
800(

(
(160*0.4+800*0.6)=544(

Max({242,544}(=(544, (

(I(choose(A2(

Example 1:
Expected Monetary Value criterium(

P(E1)(=(0.95((95%)(((((((P(E2)(=(0.05((5%)(
Events
Alternatives

A1

A2

!
!EMV(A1)(=(203.5(
(EMV(A2)(=(192(((

E1

E2

(0.95)! (0.05)!
(
(
200(
270(
(
160

(
800(

(EMV)

(
(200*0.95+270*0.05)=203.5(
(
(160*0.95+800*0.05)=192(

Max({203.5,192}(=(203.5,(

((((I(choose(A1(

The(op0mal(decisions(depend(on(
probabili0es(

In(simple(cases(its(sucient(to(write(
the(payo(matrix(
(
(

(in(some(other(cases(its((
(useful(the(representa0on(through(
Decision(Trees.(

Decision(Tree(
Useful(scheme(to(solve(decision(problems(with(sequen0al(
decisions.(
(The(decision(taken(at(a(given(instant((inuences(future(decisions.((
Structure:(
DECISIONES(NODES((squares)(from(which(BRANCHES(of(
alterna0ves(exit.(
PROBABILITY(NODES((bullet)(from(which(event((((((((((((((
BRANCHES(exit. (
(
(
((
((probabili0es(of(event(branches(exi0ng(from(the(((((((((((((((((((
same(node(sum(to(1).(
PAYOFFS(are(associated(to(nal(leaves(only.(
20( 84(
200(

Example(2(

Example(1:(Decision(Tree(
!
Ev.

E1
(0.4)!

E2
(0.6)!

EMV

A1

(
200(

(
270(

(
242(

A2

(
160

Alt.

242( E1(
544

(
800(

(
544(

200(

Events

E2(

270(

A2(

E1(

160(

544( E2(

800(

E2

E3

(
1(

(
1(

(
1(

(
810

(
20(

(
40(

(
30(

(
70(

Alternatives

A1
A1(

E1

A2

A3

830(
(

Max({242,544}(=(544(
Optimal decision: A2 with expected payoff 544

Payoff matrix

20(

Example(2:(decision((tree(
Ev.
Alt.

A1
A2
A3

E1

E2

E3

((0.4)((

((0.4)(

((0.2)(

(
1(

(
1(

(
1(

(
1(

(
810

(
20(

(
40(

(
12(

(
30(

(
70(

(
14(

830(
(

Bidding(for(a(government(contract(

1(

EMV
A1(

12(

14(

E1(

A2(

810(
E2(

40(

E3(
A3(

14(

20(

E1(

830(
E2(

E3(

30(
70(

Max({1,12,14}(=(14(
Optimal decision: A3 with expected payoff 14

Scitools(wins((
if(nobody(else(bids(
If(the(compe0tors(bid(a(higher(sum.(

On( the( basis( of( past( contracts,( it( will( cost( 5.000( ( to(
prepare( a( bid,( and( in( case( of( victory,( 95.000( ( to(
supply(the(instruments.(

Scitools(believes(there(is(a(30%(chance(that(there(will(
be(NO#compe0ng(bids.(

Scitools,( a( company( specialized( in( scien0c(


instruments,( has( been( invited( to( submit( a( bid( on( a(
government(contract.((
( The( contract( calls( for( a( specic( number( of( these(
instruments(to(be(delivered(during(the(coming(year.(
The( bids( must( be( sealed( (so( that( no( company( knows(
what( the( others( are( bidding)( and( the( low( bid( wins( the(
contract.((
(((((Thanks(to(Scitools(good(reputa0on,((
in(case(of(equal(bids,(Scitools(wins.(

Is(it(convenient(for(Scitools(to(bid?(
((
If(yes,(how(much(to(bid
(
(
(
(((to(maximize(its(EMV)?(

Costs(and(gains(
COSTS:(
Bid(cost:( (5.000((
supply:(
(95.000((
PAYOFFS:(
NO(bid(case:((0((
bid(case:(amount(R((gain)(and(wins(:(
(((((gains8(costs(=(R(8(5.000(((95.000((=(R(8(100.000((
(
bid(case:(amount(R((gain)(and(DOES(NOT(win(
(
(((gains8(costs(=(0((8(5.000((=(85.000((

Bids(typology(
Fixed(bids:((

(((((((((
(((((((115.000((,(120.000((,(125.000((

Free(bids(

N.B.(Scitools wins if either nobody else bids or the competitors bid


a higher sum.(

Compe0tors(Bid(Probabili0es(
(xed(bids)(
On the bases of past contracts of this type,
Scitoools believes that, in case of bidding
competitors, the competitor behavior will be the
following

bid
115.000(

Probability
0.3

120.000(

0.4

125.000(

0.3

Probability(of(NO(compe0ng(bids(is(0.3(
(
Probability(of(compe0ng(bids((is((1((0.3(=(0.7(
(
X(0.7(

bid
115.000(
120.000(
125,000(

Probability
0.3
0.4
0.3
0.3+0.4+0.3=1

Probability
0.21
0.28
0.21
0.21+0.28+0.21=0.7

Solving(approach(

Payo(matrix(
Compe0tors(bids(

Develope(a(decision(tree(which(contains(for(each(feasible(
strategy(both(of(the(decision(maker(and(the(compe0tors:(
The(payo(of(Scitools(
The(probabili0es(to(obtain((each(payo.(

No(bid(
(
Scitools(bid(

Doesnt(bid(
Bids(115((
Bids(120((
Bids(125((

Calcula0on(of(EMV(for(each(node(of(the(tree.(

Probabilities

Decision(Tree(
Doesnt(bid(
bids115((
bids120((
bids125(( 0

0(
15((
20((
25((

bid(
115(
(0.21)(
0(
15(
85(
85(

0
NB

125

115

120

bid(
120(
(0.28)(

0,79(

20

0,21(
0,51(

-5
25

0,49(

-5

0(
15(
85(
85(

Bid(
(125(
0(
15(
20(
85(

0(
15(
20(
25(

0.3

0.21

0.28

0.21

EMV(Calcula0on(
bid(
(125(
(0.21)(

0(
15(
20(
85(

0(
15(
20(
25(

Euro(1000s(

0.3((+(
0.28+(
0.21=(
888888888(
0.79(

EMV(NP)(=(0(

0
NB

115

15

125

EMV(115)(=(15(

15

14.75

Euro(1000s(

15

0(
15((
20((
25((

Bid(
120(

Euro(1000s(

The(problem(is(solved(by(folding(back(the(decision(tree,(
and(choosing(the(strategy(associated(to(the(higher(EMV.(

NO(bid(
(0.3)(

Bid(
115(

120
10.3

0.79(

20

0.21(
0.51(

-5
25

0.49(

-5

Max({0,15,14.75,10.3}(=(15((

(
(

EMV(120)=(
((0.79*20+0.21*(85)=14.75(
EMV(125)=(
((0.51*25+0.49*(85)=10.3(
Scitools(should(bid(
115.000(((

Compe0tors(Bid(Probabili0es(
(free(bids)(
bid
Less than 115.000

Probability
0.2

Between 115.000
and 120.000
Between 120.000
and 125,000
More than 125.000

0.4
0.3
0.1

(
Free(bids:(0((and(115.000((
(

Probability of NO competing bids is 0.3


(
Probability of competing bids is 1((0.3(=(0.7(

X(0.7(

Low bid

Probability

Probability

Less than115.000

0,2

0.14

Between 115.000
and 120.000
Between 120.000
and 125,000
More than125.000

0,4

0.28

0,3

0.21

0,1

0.07

0.2+0.4+0.3+0.1=1(

0.14+0.28+0.21=0.7(

(
Free(bids:(0((and(115.000((
(

Competitors bids

No(bid(
Doesnt(bid(
0(
Probability(
0.3(

Bid(
<115(
0(
0.14(

Bid(
1158120(
0(
0.28(

Bid(
1208125(
0(
0.21(

Competitors bids
Bid(
125(
0(
0.07(
Euro(1000s(

Doesnt(bid(
bids115((
Probability(

No(
(bid(
0(
15((
0.3(

Bid(
<115(
0(
85(
0.14(

Bid(
Bid(
1158120( 1208125(
0(
0(
15(
15(
0.28(
0.21(

Bid(
125(
0(
15(
0.07(

Euro(1000s(

(
Free(bid(:(117.000((
(
Competitors bids

(
Free(bid:(114.000((
(
Competitors
bids
No(bid(
(
bids(114((
14((
Probability( 0.3(

Bid(
<115(
?(
0.14(

Bid(
Bid(
1158120( 1208125(
14(
14(
0.28(
0.21(

Bid(
125(
14(
0.07(

Non(
Puntano( Puntano( Puntano( Puntano(
puntano( <115(
1158120( 1208125(
125(
Bids(117((
17((
85(
?(
17(
17(
Probability( 0.3(
0.14(
0.28(
0.21(
0.07(

Euro(1000s(

Euro(1000s(

We(do(not(know(the(winning(probability(with(such(a(free(bid,(
therefore(we(cannot(take(a(decision!!!(
(
We(only(know(that(with(14%(probability((the(compe0tors(bid(less(
than(115.000$.(

We(do(not(know(the(winning(probability(with(such(a(free(bid,(
therefore(we(cannot(take(a(decision!!!(
(
We(only(know(that(with(28%(probability((the(compe0tors(bid(
between(115.000$(and(120.000$.(
.(

Decision(Tree(

Payo(matrix(
Competitors bids

Scitools(bids(

No(bid(
(
Doesnt(bid(
Bids(115((
Bids(120((
Bids(125((

0(
15((
20((
25((

Bid(
<115(
0(
85(
85(
85(

Bid(
Bid(
Bid(
1158120( 1208125( 125(
0(
0(
0(
15(
15(
15(
85(
20(
20(
85(
85(
25(

Doesnt((bid(
Bids(115((
Bids(120((
Bids(125((

Euro(1000s(
0.3

0.14

0.28

0.21

0(
15((
20((
25((

0(
85(
85(
85(

0
NB

Probability

Bid(
<115(
(0.14)(

No(bid(
(0.3)(

125

15

0.86(

0.14( -5

115
0.58(

0.07

Bid(
Bid(
Bid(
1158120( 1208125( 125(
(0.21)(
(0.28)(
(0.07)(
0(
0(
0(
15(
15(
15(
85(
20(
20(
85(
85(
25(

120

0.42(

0.37(

20
-5
25

0.63(

-5

0.3((+(
0.21+(
0.28+(
0.07=(
888888888(
0.86(

EMV(Calcula0on(

New(product(introduc0on(
Euro(1000s(

12.2

NB

125

EMV(NB)(=(0(

0
12.2

0.86(

15
0.14( -5

9.5

0.58(

20

115

120

6.1

0.42(

0.37(

-5

EMV(115)(=((
((((((0.86*15+0.14*(85)=12.2(
EMV(120) =
((((((0.58*20+0.42*(85)=9.5(

25
0.63(

-5

EMV(125)(=((
(((((0.37*25+0.63*(85)=6.1(

A( rm( has( to( decide( whether( to( introduce( a( new(


product(into(the(market(or(not.(
Such( launch( requires( costs( and( it( may( have( dierent(
results(
( ((Great((Fair(8(Awful)(
(((((((((
Cost(of(launch:(7(milions((Euros(
Unit(margin:(((((((((((((((18(Euro(
(dierence(between(price(and(unit(cost)
((((

Max({0,12.2,9.5,6.1}(=(12.2((

(
(

Scitools(should(bid(
115.000(((

(x(18)(8(7.000.000(

(
(
(

(( Launch!result!
Great((LG)(
(
(((( (
Fair((LF)(

Awful((LA)(

Sold!quan<<es!

((

Prot!from!launch!

600.000((

((((3.800.000(

300.000(

(((81.600.000(

90.000(

(((85.380.000(
(

((((Prior(probabili0es(of(na0onal(market(results(
Launch!result!

Probability!

Great((LG)(

0.45(

Fair((LF)(

0.35(

Awful((LA)(

0.20(

(
(
(
(

((((
(
(
(

(
((
((
((((

((
(
(
(

(
(
(

((
((
((

((

Market(Test(
The( rm( wonder( if( conduc0ng( a( pre8launch( test( (with(
cost( 100.000( ),( that( is( introducing( the( product( in( a(
given( region,( before( launching( the( product( to( the(
na0onal(market.(
(
If(we(invest(in(the(test,(we(have(to(wait(for(its(results,(
which( can( be( GREAT,( FAIR,( AWFUL,( according( to( the(
expected(number(of(sales.(

(x(18)(8(7.000.000(8(100.000(

Possible!quan<<es!
Test!result!
sold!(1000s!of!units)!
in!na<onal!market!

Test!
(revenue!!test!cost)!

600.000((

Great((TG)(

(((((3.700.000(

300.000(

Fair((TF)(

((((81.700.000(

90.000(

Pessimo((TP)(

((((85.480.000(

Historical(Data(

Historical(Data(
(((Among(products(whose(launch(was(GREAT,(the(
64%(was(the(same(in(the(test,(the(26%(was(
evaluated(Fair(and(the(10%(was(awful.(
Prob((TG(|(LG)=0.64(
Prob((TF(|(LG)=0.26(
Prob((TA(|(LG)=0.10(

Historical(Data(

(Among(products(whose(launch(wasFAIR,(the(18%(
was(Great(in(the(test,(the(57%(was(Fair(and(the(
25%(was(Awful.(

(Among(products(whose(launch(was(AWFUL,(the(
9%(was(Great(in(the(test,(the(48%(was(Fair(and(the(
43%(was(Awful.(

Prob((TG(|(LF)=0.18(
Prob((TF(|(LF)=0.57(
Prob((TA(|(LF)=0.25(

Prob((TG(|(LA)=0.09(
Prob((TF(|(LA)=0.48(
Prob((TA(|(LA)=0.43(

Decision(Tree:(Payos(

Firm(strategy(

No Launch

Deciding(if(either(making(the(test(or(not.((
(
In(any(case,(deciding(if(introducing(the(new(
product(or(not.(

Euro(1000s(
0
LG
LF

Launch

No
Test

-1.600

LA

-5.380

No Launch

TG

Developing(a(decision(tree.(

3.800

LG

-100

LG

TF

-100

No Launch

Decision(Tree:(Probabili0es(
No Launch

No
Test

LF

No
Test

-5.380

No Launch

TG

-100

LG

-100

LG

LA

Launch
No Launch

TF

TA

Launch

LF
LA

Launch
No Launch

LF

-100

LG
LF
LA

-5.480

0
LG

-1.600

LA

LF
LA

Launch

Euro(1000s(

3.800

Prob(((LO()(=(0.45(

Test

No Launch

Launch

LG

3.700
-1.700
-5.480
3.700
-1.700
-5.480
3.700
-1.700

Decision(Tree:(Probabili0es(

Euro(1000s(
LG

LF
LA

Launch

TA

LF
LA

Launch
No Launch

Test

-100

3.700
-1.700
-5.480
3.700
-1.700
-5.480
3.700
-1.700
-5.480

LF

Launch

3.800
-1.600

LA

-5.380

Prob(((LB()(=(0.35(
No Launch

TG
Test

LG

-100

LG

TF
Launch
No Launch

-100

LG

Launch

LF
LA

Launch
No Launch

TA

-100

LF
LA

LF
LA

3.700
-1.700
-5.480
3.700
-1.700
-5.480
3.700
-1.700
-5.480

Decision(Tree:(Probabili0es(
Euro(1000s(

No Launch

LF

Launch

-100

LG

-100

LG

LF
LA

Launch
No Launch

TF

LF
LA

Launch

-100

No Launch

LG
LF
LA

Launch

3.700
-1.700
-5.480
3.700
-1.700
-5.480
3.700
-1.700

Euro(1000s(
0
LG

No
Test

LF

Launch

3.800
-1.600

LA

-5.380

TG
Test

-100

LG

-100

LG

No Launch

TF

TA

Launch

LF
LA

Launch
No Launch

LF
LA

Launch

-100

LG
LF
LA

3.700
-1.700
-5.480
3.700
-1.700
-5.480
3.700
-1.700
-5.480

-5.380

No Launch

TG
Test

TF
Launch
No Launch

TA

-100

LG

-100

LG

-100

LG

LF
LA

Launch
No Launch

LF
LA

Launch

LF
LA

3.700
-1.700
-5.480
3.700
-1.700
-5.480
3.700
-1.700
-5.480

Bayess(rule(
(condizional(probabili0es)(
((((((Prob((TG(|(LG)=0.64(!((Prob((LG(|(TG)(=(?(
prob( B | A) =

Prob(((LO(|(TO()(=(?(
No Launch

Prob(((TG()(=(?(

3.800
-1.600

LA

-5.480

Decision(Tree:(Probabili0es(
No Launch

LF

Launch

No
Test

-5.380

No Launch

TA

0
LG

Prob(((LP()(=(0.20(

TG

Euro(1000s(

3.800
-1.600

LA

Test

No Launch

0
LG

No
Test

Decision(Tree:(Probabili0es(

prob( B A)

prob( A)

prob( B A) = prob( B | A) * prob( A)

prob ( A B )
=
prob ( B )
prob ( B A)
prob ( B | A) * prob ( A)
=
=
prob ( B )
prob ( B )
prob ( A | B ) =

prob( LG | TG ) =

prob(TG | LG ) * prob( LG )
prob(TG )

prob( B) = prob( B ) =
= prob( B | E1 ) * prob( E1 ) + prob( B | E2 ) * prob( E2 ) + ! + prob( B | En ) * prob( En )
prob(TG ) =
= prob(TG | LG ) * prob( LG ) + prob(TG | LF ) * prob( LF ) + prob(TG | LA) * prob( LA)

Prob (TG )= 0.3690


Prob (TF) = 0.4125
Prob (TA) = 0.2185
prob(TG | LG ) * prob( LG )
=
prob(TG )
prob(TG | LG ) * prob( LG )
=
prob(TG | LG ) * prob( LG ) + prob(TG | LF ) * prob( LF ) + prob(TG | LA) * prob( LA)
prob( LG | TG ) =

Prob((TG(|(LG)=0.64(!(Prob((LG(|(TG)(=(0.7805(
Prob((LG(|(TF)(=(0.2836(
Prob((LG(|(TA)(=(0.2059(
Prob((LF(|(TG)(=(0.1707(
Prob((LF(|(TF)(=(0.4836(
Prob((LF(|(TA)(=(0.2327(
Prob((LA(|(TG)(=(0.0488(
Prob((LA(|(TF)(=(0.2327(
Prob((LA(|(TA)(=(0.3936(

Decision(tree:(expected(values(
74(

No Launch
Launch

Euro(1000s(
LG

EMV=74(

LF

3.800

-100

TG
TF

8100(
8100(

No Launch

LG
LF
LA

Launch
No Launch

Launch

TA

2330.24(
-100

81048.07(
-100

LG
LF
LA
LG
LF
LA

Launch

82075.65(

Euro(1000s(
0
LG

74(

LF

No
Test

-5.380

2330.24( No Launch

No Launch
Launch

-1.600

LA

796.76(

Test

74(

No
Test

EMV(Test)=796.76(

Decision(tree:(Solu0on(

3.700
-1.700
-5.480
3.700
-1.700
-5.480
3.700
-1.700
-5.480

3.800
-1.600

LA

-5.380
-100

796.76(

2330.24( No Launch

Test

EMV=796.76(

TG
TF

8100(
8100(

No Launch

Launch

LG
LF
LA

Launch
No Launch

Launch

TA

2330.24(
-100

LG

-100

LG

LF
LA

LF
LA

3.700
-1.700
-5.480
3.700
-1.700
-5.480
3.700
-1.700
-5.480

Decision(tree:(Solu0on(

Example:(Oil(extrac0on(

Euro(1000s(

It turn out convenient to make the Test, and

if the result is GREAT to launch the product,


in the other two cases (FAIR, AWFUL) not to lunch it.

-100

796.76(

2330.24( No Launch

Test

TG

EMV=796.76(

TF

LG
LF
LA

Launch

8100(

No Launch

Launch

8100(
TA

2330.24(

No Launch

-100

LG

-100

LG

LF
LA

LF
LA

Launch

Example:(Oil(extrac0on(
Alterna<ves!

!Oil!

No!Oil!

Drilling(

700(

8100(

Selling(the(ground(

90(

A(priori(Probabili0es(

0.25(

90(
0.75(

3.700
-1.700
-5.480
3.700
-1.700
-5.480
3.700
-1.700

A(decision(maker(possesses(a(plot(of(ground(where(he(
supposes(there(is(oil.(
A(geological(study(stated(that(there(is(1(possibility(over(
4(to(nd(oil.(
Another( oil( society( oered( to( buy( that( ground( for(
$90.000(and(to(drill(for(oil(by(himself.(
The(drilling(cost(is($100.000(
If(someone(get(the(oil,(the(revenue(is($800.000,(if(no(oil(
appears,(then(there(is(no(revenue.(

-5.480

Example:(Oil(extrac0on(without(test(
Alterna<ves!

!Oil!

No!Oil!

Drilling(

700(

Selling(the(ground(

90(

A(priori(Probabili0es(

0.25(

8100(
90(
0.75(

MAXMIN: max{min{700,-100}, min{90,90}}=


max{-100, 90}=90 " Its preferable to sell the ground
EMV(Drilling)=0.25*700+0.75*(-100)=100
EMV(sale)=0.25*90+0.75*(90)=90
! Its preferable to drill

Example:(Oil(extrac0on(
What( is( the( minimum( probability( of( nding( oil,( under( which( its(
be;er(to(sell(the(ground?(
Alterna<ves!

!Oil!

Example:(Oil(extrac0on(
EMV(Drilling)=p*700+(1-p)*(-100)=800*p-100
EMV(sale)=p*90+(1-p)*(90)=90
It is convenient to sell #! 90> 800*p-100 #! p<190/800=0.2375

No!Oil!

Drilling(

700(

Selling(the(ground(

90(

A(priori(Probabili0es(

p(

EMV(Drilling)=800*p-100

8100(
90(
18p(

EMV(Trivellazione)=p*700+(1-p)*(-100)=800*p-100
EMV(sale)=p*90+(1-p)*(90)=90
It is convenient to sell #! 90> 800*p-100 #!
p<190/800=0.2375

p
0.2375
-100

Oil(example(with(test(
Before taking the decision, we could conduct a deep analysis of
the ground in order to obtain a higher information about the
possibility of finding the oil.
The cost of such inspection is $30.000.
Possible Risults:
TU: Test unfavourable: low probability to find oil
TF: Test favourable: high probability to find oil
In the past:
Prob(TU|O)= 0.4
Prob (TU|NO)=0.8

Prob (TF|O)=1-Prob (TU|O)=1-0.4=0.6


Prob (TF|NO)=1-Prob (TU|NO)=1-0.8=0.2

EMV(sale)

90

Oil(example(with(test(
Bayess rule
Pr(TU | O)Pr(O)
0.4 * 0.25
=
=
Pr(TU)
0.4 * 0.25 + 0.8 * 0.75
= 1 / 7 0.142857
Pr(TU | NO)Pr(NO)
0.8 * 0.75
Pr(NO | TU) =
=
=
Pr(TU)
0.4 * 0.25 + 0.8 * 0.75
= 6 / 7 0.857143
Pr(O | TU) =

Pr(TF | O)Pr(O)
0.6 * 0.25
=
= 1/ 2
Pr(TF)
0.6 * 0.25 + 0.2 * 0.75
Pr(TF | NO)Pr(NO)
0.2 * 0.75
Pr(NO | TF) =
=
= 1/ 2
Pr(TF)
0.6 * 0.25 + 0.2 * 0.75
Pr(O | TF) =

Oil(example(with(test(

Informa0on(for(Oil(problem(

Tree

EXPECTED VALUE OF PERFECT INFORMATION (EVPI) = MAXIMUM


THRESHOLD FOR TEST COST
Espected Value of best payoffs - Expected Value without perfect inf.
Oil
700
25%

No Oil
90
75%

EMV(IP)

EMV(NOTEST)

242,5

- 100

EVPI

142,5

EXPECTED VALUE OF SAMPLE INFORMATION (EVSI)

prob(TU)
prob(TF)

70%
30%

Decision+30

EMV(TEST)+ TEST COST

90
300

123 + 30 = 153

EVSI = (EMV(TEST)+ TEST COST) - EMV(NOTEST)


SAMPLE INFORMATION

PERFECT(INFORMATION(marke0ng(

EXPECTED(VALUE(OF(SAMPLE(INFORMATION((EVSI)

((

Other(examples(

EMV(PI)=3800*0,45+0*0,35+0*0,20=1710
MAX value = 1710-74=1636

Financing(a(project(

Expected value of sample information


EVSI = (EMV(TEST)+ TEST COST ) - EMV(NOTEST)
EMV(TEST)+ TEST COST =

796,76 + 100 = 896,76

EVSI = (EMV(TEST)+ TEST COST ) - EMV(NOTEST) =


896,76-74= 822,76

Environmental(decisions(

= 53

>30.

Thanks
For your attention
I hope to have put
some light upon trees

AND widen your Horizons

You might also like