09. decision analysis

Upload: nicasavio2725

Post on 07-Aug-2018

213 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/21/2019 09. Decision Analysis

    1/25

    CHAPTER 9 DECISION ANALYSIS

    9.1 Introduction to Decision Analysis

    Eleents!

    "1#Decision maker "$%e'

    "(# Choices"decisions' strate)ies#'

    "*#Payoffs

    Selection problems. De+ine a ,inary -aria,leyj+or eac

    decision %it yj / 1' i+ %e a0e decision j' and

    oter%ise. A selection 2ro,le %it n 2ossi,le

    decisions %ill ten +eature te constrainty13y(3 43

    yn/ 1.

    5ulti2le criteria "MCDM/ multiple criteria decision

    making-s ulti2le states o+ nature 6 decision analysis

    /games against nature.

    Example:Consider a 2ro,le %it tree decisions d1'

    d(' and d*' and +our states o+ nature s1' s(' s*' and s7. Te

    2ayo++s are so%n ,elo%.

  • 8/21/2019 09. Decision Analysis

    2/25

    s1 s( s* s7

    d1 * ( 7 8

    d( ( 7 1

    d* ( *

    Tis )ae is

    "1# asyetric! te decision a0er is rational"loo0s at

    te 2ayo++s#' %ile nature is a rando 2layer

    "(# a siultaneous )ae "%e do not 0no% in ad-ance

    %at state o+ nature %ill ,e cosen#.

    Consider te continuu ,et%een

    certainty: 4 : risk: 4: uncertainty

    Certainty! %e 0no% e;actly %ic strate)y nature %ill

    2lay.

    Risk! %e 0no% te 2ro,a,ility distri,ution nature uses

    to 2lay er strate)ies "e.).' ,y %ay o+ 2ast

    o,ser-ations#

    ncertainty: %e do not 0no% e-en te 2ro,a,ility

    distri,ution o+ nature

  • 8/21/2019 09. Decision Analysis

    3/25

    Certainty is tri-ial! Since %e 0no% %at colun

    nature 2lays' all %e a-e to do is coose te ro% tat

    leads to te i)est 2ayo++.

    9.( =isuali>ations o+ Decision Pro,les

    5acro -ie%!!nfluence diagrams.

    5icro -ie%!Decision trees.

    !nfluence diagrams: Decision nodes' rando nodes'

    conse?uence nodes. So%s )eneral interrelations

    ,et%een decisions' cance e-ents' @ resultsoutcoes.

    Example:

    Decision Rando e-ent Conse?uenceAdd electronics

    de2artent

    Beneral econoic

    conditions

    Pro+it

    Relocate

    de2artent

    into a se2arate

    ,uildin)

    Local acce2tance

    o+ ser-ices

  • 8/21/2019 09. Decision Analysis

    4/25

    Te ,ro0en arcs! 2ossi,le in+luences "local acce2tanceo+ an electronics de2artent or store ay ,e

    in+luenced ,y te e;istence o+ an electronics

    de2artent in our de2artent store and our

    co2etitors< reaction to our introduction o+ te

    de2artent#.

    Decision tree +or te sae 2ro,le!

  • 8/21/2019 09. Decision Analysis

    5/25

    9.* Decision Rules nder ncertainty and

    Ris0

    Start %it uncertainty. "Little in2ut on our 2art' onlycrude in+oration %ill coe out#.

  • 8/21/2019 09. Decision Analysis

    6/25

    E;a2le!

    s1 s( s*

    d1 ( ( d( 1

    d* ( 1 1

    d7 ( * 7

    e+ore coencin)' cec0 +or dominances. One

    decision "ro%# doinates anoter' i+ its 2ayo++s are all)reater or e?ual tan tose o+ a sin)le oter ro%.

    Here' d1 doinates d7. Doinated decisions can ,e

    deleted. Colun doinances do not e;ist.

    Cec0in) +or doinances re?uires a total o+ Fm"m1#

    2airs o+ co2arisons. Beneral 2rocedure +or all

    decision rules! Deterine anticipated outcomes +or all

    decisions.

    Decision rules under uncertainty!

    "1# "ald#s rule "2essiist

  • 8/21/2019 09. Decision Analysis

    7/25

    In our e;a2le' te antici2ated outcoes are (' 1' 1'

    @ * "in case %e did not eliinate decision d7#' te

    a;iu is 1' %ic ,elon)s to d*. Tis is te cosen

    decision.

    "(# O2tiist

  • 8/21/2019 09. Decision Analysis

    8/25

    te. Tis a0es Jald

  • 8/21/2019 09. Decision Analysis

    9/25

    is M' %ile te o22ortunity cost +or ne%s2a2ers as

    ,een estiated to ,e 1M +or eac ne%s2a2er tat

    could a-e ,een sold ,ut %as not due to te lac0 o+

    su22ly. Su22ose tat te 2urcasin) strate)ies are 1'(' *' @ 7.

    Payo++ atri;

    s1 s( s* s7

    A /

    4

    3

    2

    1

    d

    d

    d

    d

    00.2850.1900.1150.2

    50.1900.2150.1200.400.1150.1200.1450.5

    50.200.450.500.7$

    Bi-en 2ro,a,ilities o+ 2 / K.8' .(' .1' .1' te decisions

    a-e e;2ected 2ayo++s o+ .9' .7' .9' and .7'

    ,uy * ne%s2a2ers @ e;2ect a daily 2ayo++ o+ .9.

    "# sin) target $alues *. Idea! Plot 2ayo++ -alues

    a)ainst te 2ro,a,ility tat te -alue can ,e acie-ed.

  • 8/21/2019 09. Decision Analysis

    10/25

    Te ori)inal 2ayo++ atri; %as

    s1 s( s*

    d1 ( ( d( 1

    d* ( 1 1

    d7 ( * 7

    2 / K.' .*' .(.

    d1! solid line

    d(! ,ro0en line

    d*! dotted line

  • 8/21/2019 09. Decision Analysis

    11/25

    Decision rules ",y %ay o+ u22er en-elo2e#!

    I+ * (' any decision %ill acie-e te tar)et.

    I+ *K(' 1' d(and d*are ,est. ot %ill reacte tar)et %it a 2ro,a,ility o+ 1.

    I+ *K1' 1' d*is ,est. It reaces te tar)et %it

    a 2ro,a,ility o+ 1.

    I+ *K1' (' d1is ,est. It reaces te tar)et %it

    a 2ro,a,ility o+ ..

    I+ *K(' ' d1and d(are ,est. ot acie-e tetar)et %it a 2ro,a,ility o+ .(.

    I+ *K' ' d(is ,est. It acie-es te tar)et %it

    a 2ro,a,ility o+ .(.

    I+ *Q ' none o+ te decisions %ill ,e a,le to reac

    te tar)et.

    9.7 Sensiti-ity Analyses

    $Jat i+& indi-idual 2ayo++s aijcan)e

    s1 s( s*

    d1 ( ( d( 1

    d* ( 1 1

    d7 ( * 7

  • 8/21/2019 09. Decision Analysis

    12/25

    2 . .* .(

    Su22ose tat %e are uncertain a,out a(*. Re%rite te

    2ayo++ as a(* / 3 %it an un0no%n K(' *'

    eanin) tat %e e;2ect te 2ayo++ to ,e ,et%een @1.

    E;2ected 2ayo++s!

    EM(/

    +

    9.

    5.1

    2.1.1

    4.1

    .

    Clearly' d1 @ d7 are doinated. Te 2ayo++s +or te

    reainin) strate)ies are so%n in te +i)ure ,elo%.

  • 8/21/2019 09. Decision Analysis

    13/25

    Tis leads to

    I+ Q ( "i.e.' a(*Q 9#' ten decision d(is ,est' @

    i+ ( "i.e.' a(*9#' decision d*is ,est.

    Anoter source o+ uncertainty relates to te a)nitude

    o+ te 2ro,a,ilities.

    Su22ose tat %e are unsure a,out p1. Siilar to te

    a,o-e' %e can use p1 3 %it soe un0no%n .Ho%e-er! Te su o+ 2ro,a,ilities ust e?ual 1' so i+

    p1increases' te oter 2ro,a,ilities ust decrease ,y

    . Assue tat te oter t%o 2ro,a,ilities decrease ,y

    te sae aounts' i.e.

    2 / K. 3 ' .* F' .( F.

    Bi-en te sae 2ayo++ atri;

    s1 s( s*

    d1 ( (

    d( 1

    d* ( 1 1d7 ( * 7

    2 . .* .(

    %e can co2ute te e;2ected -alues as

  • 8/21/2019 09. Decision Analysis

    14/25

    EM("# /

    +

    +

    +

    5.19.

    15.1

    31.1

    5.4.1

    .

    Su22ose %e estiate tat p1%ill ,e ,et%een .* @ .8.

    "Alternati-ely' startin) %it p1 /.' te can)e

    K .(' 3.1.

    Jitin tis ran)e' decision d*doinates d1@ d7' %ic

    can ,e deleted.

    Te e;2ected onetary -alues +or d(@ d*are so%n in

    te +i)ure ,elo%!

  • 8/21/2019 09. Decision Analysis

    15/25

    Decision rule!

    I+ .1 "i.e.'p1.7#' ten decision d(is ,est' @

    i+ Q .1 "i.e.'p1Q .7#' ten decision d*is ,est.

    Di++erent e;a2le! Sae 2ayo++ atri;' ,ut asp1 '

    p( @p* . Te e;2ected 2ayo++s are ten

    +

    +

    +

    3

    8

    3

    5

    3

    5

    9.

    5.1

    1.1

    4.1

    .

    Decision rule!

    I+

    .1 "i.e.'p1 .*#' ten decision d(is o2tial'i+ .1 "i.e.'p1.*#' ten decision d*is o2tial.

  • 8/21/2019 09. Decision Analysis

    16/25

    9. Decision Trees and te =alue o+

    In+oration

    =alue o+ in+oration. E;tree case +irst!

    Expected $alue of perfect information "E(P!#!

    Di++erence ,et%een te e;2ected 2ayo++ %it 2er+ect

    in+oration inus te e;2ected 2ayo++ %itout

    in+oration ",eyond te 2rior 2ro,a,ilitiesp#.

    Pre-ious e;a2le!

    s1 s( s*

    d1 (U (

    d( 1 U

    d* (U 1U 1

    d7 (U * 72 . .* .(

    2 / K.' .*' .(.

    Te ,est res2onses to nature

  • 8/21/2019 09. Decision Analysis

    17/25

    As te e;2ected 2ayo++ %itout 2er+ect in+oration

    "i.e.' te e;2ected onetary -alue o+ te ,est strate)y

    EM(,# %as 1. "acie-ed ,y usin) d*#' te expected

    $alue of perfect informationis ten

    E(P!/EPP! EM(U / (. 1. / 1.(.

    No% i2er+ect in+oration.

    Ia)e a +orecastin) institute tat uses indicators !1'!('

    4 to +orecast te states o+ nature s1' s(' 4 . Clearly' te

    indicators @ te states o+ nature sould ,e related.

    "ty2ical e;a2les are deand @ %olesaler

  • 8/21/2019 09. Decision Analysis

    18/25

    Decision tree!

    ro le+t to ri)t' te tree de2icts te se?uence o+

    e-ents.Decision modes"%e decide# are s?uares' e$ent

    nodes"nature a0es a rando coice# are circles' @

    terminal nodes "tere are no +urter u-es# as

    trian)les.

  • 8/21/2019 09. Decision Analysis

    19/25

    Note! te lo%er 2art o+ te tree e?uals %at %e a-e

    already done in te atri; %en %e deterined te

    EM(U strate)y.

    Nu,ers tat are needed!

    "1#Payoffs"at te terinal nodes#

    "(# !ndicator probabilitiesP"!# at nature

  • 8/21/2019 09. Decision Analysis

    20/25

    Here!

    or!1' %e co2uteP"!1# andP"sV!1#

    s P"s# P"!1Vs# P"!1Vs#P"s# P"sV!1#

    s1 . .8 .* .791

    s( .* .9 .( .77(8

    s* .( .( .7 .88

    P"!1# / .81

    or!(' %e co2uteP"!(# andP"sV!(#

    s P"s# P"!(Vs# P"!(Vs#P"s# P"sV!(#

    s1 . .7 .( .1(

    s( .* .1 .* .89

    s* .( . .18 .71*P"!(# /.*9

    Te co2lete decision tree is ten as +ollo%s!

  • 8/21/2019 09. Decision Analysis

    21/25

    Te nu,ers ne;t to te nodes are co2uted ,yback'ard recursion. Te recursion starts at te

    trian)ular terinal nodes @ %or0s ,ac0%ards to te

    root o+ te tree.

  • 8/21/2019 09. Decision Analysis

    22/25

    T%o rules a22ly!

    "1# ac0 into a decision node! coose te ,est strate)y'

    i.e.' te one %it te i)est "e;2ected# 2ayo++.

    "(# ac0 into an e-ent node! co2ute te e;2ected

    -alue o+ all successor nodes.

    Here' te result is tepayoff 'ith imperfect information

    EP!!/ (..

    No% co2are! %itout in+oration ",eyond 2rior

    2ro,a,ilities#' %e can )etEM(U / 1.. Jit additional

    in+oration' %e can )et EPII / (..

    Hence te expected $alue of perfect information

    E(S!/ (. 1. / ..

    A standardi>ed easure is e++iciencyE' %ic is

    E/ .1.( / .7*.

  • 8/21/2019 09. Decision Analysis

    23/25

    Te sae e;a2le %it di++erent nu,ers!

    s1 s( s*

    P"!-s#! !1 .9 .8 .(!( .1 .7 .

    Note! Tis atri; is te sae as ,e+ore %it soe

    coluns e;can)es. Ho%e-er' no% s1is stron)ly lin0ed

    to te indicators' @P"s1# / .' so te result sould ,e

    "at least soe%at# ,etter.

    Co2utation o+P"!1# andP"sV!1#!

    s P"s# P"!1Vs# P"!1Vs#P"s# P"sV!1#

    s1 . .9 .7 .818

    s( .* .8 .1 .(8

    s* .( .( .7 .9P"!1# / .8

    Co2utation o+P"!(# andP"sV!(#!

    s P"s# P"!(Vs# P"!(Vs#P"s# P"sV!(#

    s1 . .1 . .11

    s( .* .7 .1( .*8*8s* .( . .18 .77

    P"!(# /.**

    Je ten o,tainEP!!/ (.1(' so tat

  • 8/21/2019 09. Decision Analysis

    24/25

    E(S!/ (.1( 1. / .8( @E/ .8(1.( / .18.

    E;tree e;a2les! Rando indicators result in te2rior 2ro,a,ilities e?ualin) te 2osterior 2ro,a,ilities

    @ E=SI / .

    On te oter and' %ile a +orecast tat is al%ays

    correct %ill a-eE(S!/E(P!@E/ 1' a +orecast tat

    is al%ays %ron) as te sae +eatures it is te

    consistency o+ te indicators tat is i2ortant' not

    teir actual eanin).

    9.8. tility Teory

    E;2ected -alues are eanin)+ul' only i+ decisiona0ers are risk neutral. Tis eans' tey sould ,e

    indi++erent to eiter

    "1# recei-in) 1' in cas' no ?uestions as0ed' or

    "(# 2layin) te lottery %it a W o+ %innin) ('

    @ a W cance o+ %innin) notin).

    5ost 2eo2le %ould 2re+er "1# i.e.' tey are not ris0

    neutral.

  • 8/21/2019 09. Decision Analysis

    25/25

    Once te utilities a-e ,een deterined' tey can ,e

    used instead o+ 2ayo++s. All 2rocedures reain

    uncan)ed.