ch13 the art of modelling with spreadsheet

Upload: rajkzo

Post on 07-Aug-2018

226 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    1/44

    MANAGEMENT

    SCIENCEThe Art of Modeling with Spreadsheets

    STEPHEN G. POWELL

    ENNETH !. "AE!

    Co#pati$le with Anal%ti& Sol'er Platfor#(O)!TH E*ITION

    *ECISION ANAL+SIS

    CHAPTE! ,-

    POWE!POINT

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    2/44

    INTRODUCTION

    • Many business problems contain n&ertainele#ents that are impossible to ignorewithout losing the essence of the situation.

    In this chapter, we introuce some basicmethos for analy!ing ecisions a"ecte byuncertainty.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    0

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    3/44

    UNC#RT$IN %$R$M#T#R&

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    -

    • Now, we broaen our 'iewpoint to inclue n&ertaininpts(parameter 'alues sub)ect to uncertainty.

    • Uncertain parameters become *nown only after a ecisionis mae.

    +hen a parameter is uncertain, we treat it as if it coul ta*eon two or more 'alues, epening on inuences beyon ourcontrol.

    •  These inuences are calle states of natre, or moresimply, states.

    • In many instances, we can list the possible states, an foreach one, the corresponing 'alue of the parameter.

    • -inally, we can assign probabilities to each of the states sothat the parameter outcomes form a probability istribution.

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    4/44

    %$O-- T$/0#& $ND D#CI&ION CRIT#RI$

    • -or each action1state combination, the entryin the table is a measure of the economicresult.

     Typically, the payo"s are measure inmonetary terms, but they nee not be pro2t2gures.

    •  They coul be costs or re'enues in other

    applications, so we use the more generalterm payof .

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    5

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    5/44

    /#NC3M$R4 CRIT#RI$

    •  The Ma6i#a6 pa%o7 &riterion  see*s thelargest of the ma5imum payo"s among theactions.

     The #a6i#in pa%o7 &riterion see*s thelargest of the minimum payo"s among theactions.

    •  The #ini#a6 regret &riterion see*s the

    smallest of the ma5imum regrets among theactions.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    8

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    6/44

    INCOR%OR$TIN6 %RO/$/I0ITI#&

    • +e can immeiately translate this informationinto probability istributions for the payo"scorresponing to each of the potential

    actions.• +e use the notation EP to represent an

    e5pecte payo" 7e.g., an e5pecte pro2t8.

    • Note that the e5pecte payo" calculation

    ignores no information9 all outcomes anprobabilities are incorporate into the result.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    9

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    7/44

    U&IN6 TR##& TO MOD#0 D#CI&ION&

    • $ pro$a$ilit% tree epicts one or moreranom factors

    •  The noe from which the branches emanate iscalle a &han&e node, an each branch

    represents one of the possible states thatcoul occur.

    • #ach state, therefore, is a possible resolutionof the uncertainty represente by the chancenoe.

    • #'entually, we:ll specify probabilities for eachof the states an create a probabilityistribution to escribe uncertainty at thechance noe.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    :

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    8/44

    &IM%0# %RO/$/I0IT TR##

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    ;

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    9/44

     T3R## C3$NC# NOD#& IN T#0#6R$%3IC-ORM

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    <

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    10/44

    D#CI&ION TR##&

    • Decision1tree moels o"er a 'isual tool thatcan represent the *ey elements in a moel forecision ma*ing uner uncertainty an helporgani!e those elements by istinguishing

    between ecisions 7controllable 'ariables8 anranom e'ents 7uncontrollable 'ariables8.

    • In a de&ision tree, we escribe the choicesan uncertainties facing a single ecision1

    ma*ing agent.•  This usually means a single ecision ma*er,

    but it coul also mean a ecision1ma*inggroup or a company.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    ,1

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    11/44

    R#%R#NTIN6 D#CI&ION&

    • In a ecision tree, we represent ecisions ass;uare noes 7bo5es8, an for each ecision,the alternati'e choices are represente asbranches emanating from the ecision noe.

    •  These are potential actions that are a'ailableto the ecision ma*er.

    • In aition, for each uncertain e'ent, thepossible alternati'e states are represente asbranches emanating from a chance noe,labele with their respecti'e probabilities.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    ,,

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    12/44

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    13/44

    D#CI&ION TR##&9 RI&4 %RO-I0#&

    •  The istribution associate with a particular actionis calle its ris= pro>le.

    •  The ris* pro2le shows all the possible economicoutcomes an pro'ies the probability of each9 It is

    a probability istribution for the principal output ofthe moel.

    •  This form reinforces the notion that, when some ofthe input parameters are escribe in probabilisticterms, we shoul e5amine the outputs in

    probabilistic terms.• $fter we etermine the optimal ecision, we can

    use a probability moel to escribe the pro2toutcome.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    ,-

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    14/44

    D#CI&ION TR##& -OR $ RI#& O- D#CI&ION&

    • Decision trees are especially useful in situations wherethere are multiple sources of uncertainty an a se;uenceof ecisions to ma*e.

    • -or e5ample, suppose that we are introucing a newprouct an that the 2rst ecision etermines which

    channel to use uring test1mar*eting.• +hen this ecision is implemente, an we ma*e an

    initial commitment to a mar*eting channel, we can beginto e'elop estimates of eman base on our test.

    • $t the en of the test perio, we might reconsier ourchannel choice, an we may ecie to switch to another

    channel.•  Then, in the full1scale introuction, we attain a le'el of

    pro2t that epens, at least in part, on the channel wechose initially.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    ,5

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    15/44

    #=$M%0#

    • In the following e5ample, we ha'e epicte7in telegraphic form8 a situation in which wechoose our channel initially, obser'e the test

    mar*et, reconsier our choice of a channel,an 2nally obser'e the eman uring full1scale introuction.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    ,8

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    16/44

    D#CI&ION TR## +IT3 >U#NTI$0D#CI&ION&

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    ,9

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    17/44

    %RINCI%0#& -OR /UI0DIN6$ND $N$0

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    18/44

    RO00/$C4 %ROC#DUR# -OR $N$0

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    19/44

     T3# CO&T O- UNC#RT$INT

    • $n action must be chosen beore learninghow an uncertain e'ent will unfol.

    •  The situation woul be much more

    manageable if we coul learn about theuncertain e'ent 2rst an then choose anaction.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    ,<

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    20/44

    IM%#R-#CT F&. %#R-#CT IN-ORM$TION

    • +hen we ha'e to ma*e a ecision beforeuncertainty is resol'e, we are operating withimperfect information 7uncertain *nowlege8about the state of nature.

    •+hen we can ma*e a ecision afteruncertainty is resol'e, we can respon toperfect information about the state of nature.

    •  Our probability assessments of e'entoutcomes remain unchange, an we are still

    ealing with e5pecte 'alues.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    01

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    21/44

    #=%#CT#D F$0U# O- %#R-#CT IN-ORM$TION7EVPI)

    •  The e5pecte payo" with perfect informationmust always be at least as goo as thee5pecte payo" from following the optimalpolicy in the original problem, an it will

    usually be better.•  The EVPI measures the i"erence, or the gain

    ue to perfect information.•  The calculation of EVPI can also be

    represente with a tree structure, where we

    re'erse the se;uence of ecision an chancee'ent in the tree iagram, )ust as we i inthe calculations.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    0,

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    22/44

    D#CI&ION TR## -OR T3# #F%IC$0CU0$TION

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    00

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    23/44

    U&IN6 D#CI&ION TR## &O-T+$R#

    • It is often iGcult to create a layout for thecalculations that is tailore to the features ofa particular e5ample.

    • -or that reason, it ma*es sense to ta*ea'antage of software that has beenesigne e5pressly for representing ecisiontrees in #5cel.

    • *e&ision Tree is a tool containe in $nalytic

    &ol'er %latform for constructing an analy!ingecision tree.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    0-

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    24/44

    D#CI&ION TR## M#NU

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    05

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    25/44

    D#-$U0T INITI$0 TR## %RODUC#D / D#CI&ION TR##

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    08

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    26/44

    D#T$I0& -OR T3# -IR&T D#CI&ION NOD#

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    09

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    27/44

    #=%$ND#D INITI$0 TR## DI$6R$M

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    0:

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    28/44

    NOD# +INDO+ -OR T3# -IR&T #F#NTNOD#

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    0;

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    29/44

    -IR&T #F#NT NOD# %RODUC#D /D#CI&ION TR##

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    0<

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    30/44

    #=%$ND#D DI$6R$M +IT3 COND #F#NTNOD# CO%I#D

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    -1

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    31/44

    -U00 DI$6R$M

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    -,

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    32/44

    N&ITIFIT $N$0&I& +IT3 TR##%0$N

    • $ ecision1tree analysis retains the propertiesof a spreasheet.

    •  The wor*sheet prouce by Decision Tree

    contains inputs, formulas, an outputs, )ust asin any well1esigne moel.

    •  Thus, we can perform sensiti'ity analyses inthe usual ways.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    -0

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    33/44

    N&ITIFIT $N$0&I& -OR T3# #=$M%0#MOD#0

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    --

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    34/44

    MINIMI

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    35/44

    0OC$TION O- T3# M$=IMI

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    36/44

    ?M$=IMI

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    37/44

    #=%ON#NTI$0 UTI0IT -UNCTION

    • $lthough there are many ways of con'ertingollars to utils, one straightforwar methouses an e6ponential tilit% fn&tion9U  a J b e5p 7JD/R8

    where D is the 'alue of the outcome inollarsK U is the utility  'alue, or the 'alue ofan outcome in utilsK an a, b, an R represent parameters of the utility function.

    %arameters a an b are essentially scalingparametersK R inuences the shape of thecur'e an is *nown as the ris= toleran&e.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    -:

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    38/44

    $N$0&I& +IT3 UTI0ITI#&

    •  To carry out the analysis, we use this functionto con'ert each monetary outcome fromollars to utils, an then we etermine theaction that achie'es the ma5imum e5pecte

    utility.• $lthough Decision Tree allows the e5ibility of

    setting three i"erent parameters, we usuallya'ise setting a  b  ?.

    •  This choice ensures that the function passes

    through the origin, so that our remaining tas*is 2ning a 'alue of R that captures theecision ma*er:s preferences.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    -;

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    39/44

    6R$%3 O- UTI0IT -UNCTION -OR T3##=$M%0#

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    -<

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    40/44

    U&IN6 TR## %0$N +IT3 #=%ON#NTI$0 UTI0IT-UNCTION

    • In Decision Tree, it is necessary to specify the threeparameters in the e5ponential utility function.

    •  These three 'alues must be entere in the tas*

    pane on the Moel tab, along with esignating the

    'alue for Certainty #;ui'alents to be the#5ponential Utility -unction.

    • $fter the user esignates the use of #5ponentialUtility -unction, Decision Tree isplays aitional

    calculations in columns /, -, an . Immeiatelybelow the monetary payo"s the isplay shows thesame 2gures con'erte to utils.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    51

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    41/44

    MODI-IC$TION O- T3# #=$M%0# MOD#0 -OR #=%ON#NTI$0UTI0ITI#&

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&.

    5,

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    42/44

    • $ ecision tree is a speciali!e moel for recogni!ing the role ofuncertainties in a ecision1ma*ing situation.

    •  Trees help us istinguish between ecisions an ranom e'ents,an more importantly, they help us sort out the se;uence inwhich they occur.

    • %robability trees pro'ie us with an opportunity to consier the

    possible states in a ranom en'ironment when there are se'eralsources of uncertainty, an they become components of ecisiontrees.

    •  The *ey elements of ecision trees are ecisions an chancee'ents. $ ecision is the selection of a particular action from agi'en list of possibilities.

    • $ chance e'ent gi'es rise to a set of possible states, an eachaction1state pair results in an economic payo".

    • In the simplest cases, these relationships can be isplaye in apayo" table, but in comple5 situations, a ecision tree tens tobe a more e5ible way to represent the relationships anconse;uences of ecisions mae uner uncertainty.

    &UMM$R

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&. 50

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    43/44

    &UMM$R 7CONT:D8

    •  The choice of a criterion is a critical step in sol'ing a ecisionproblem when uncertainty is in'ol'e.

    •  There are benchmar* criteria for optimistic an pessimisticecision ma*ing, but these are somewhat e5treme criteria. Theyignore some a'ailable information, incluing probabilities, inorer to simplify the tas* of choosing a ecision.

    • The more common approach is to use probability assessmentsan then to ta*e the criterion to be ma5imi!ing the e5pectepayo", which in the business conte5t translates into ma5imi!inge5pecte pro2t or minimi!ing e5pecte cost.

    • Using the rollbac* proceure, we can ientify those ecisionsthat optimi!e the e5pecte 'alue of our criterion. -urthermore,we can prouce information in the form of a probability

    istribution to help assess the ris* associate with any ecisionin the tree.

    • Decision Tree is a straightforwar spreasheet program thatassists in the structuring of ecision trees an in the calculationsre;uire for a ;uantitati'e analysis.

    Chapter ,- Cop%right / 01,- 2ohn Wile% 3Sons4 In&. 5-

  • 8/20/2019 Ch13 The Art of Modelling with Spreadsheet

    44/44

      All rights reserved. Reproduction or translation ofthis work beyond that permitted in section 117 of the 1976

    United tates !opyright Act without e"press permission of

    the copyright owner is unlawful. Re#uest for further

    information should be addressed to the $ermissions

    %epartment& 'ohn (iley ) ons& *nc. +he purchaser may

    make back,up copies for his-her own use only and not for

    distribution or resale. +he $ublisher assumes no

    responsibility for errors& omissions& or damages caused by

    the use of these programs or from the use of the information

    herein.

    !$/R*0+ 2 3415 ' (*8/ ) & *!.