ࡱ> U@ /bjbj /jz%}vvvv. . .  [[[8[<\ Z\(\^@]@]@]+++َێێێێێێ$Rb. ++++vv@]@]) +.v@]. @]َ +َ B M2 . M@]\ N[YMՍ*0ZM S MB d @vvvv . M++ +++++ ,:  : HYPERLINK "file:///C:\\Documents%20and%20Settings\\macleodk\\My%20Documents\\My%20Webs\\DSCI%206613.htm" Back Expected Value of Perfect Information Figure 1 shows the decision tree we developed in the previous lecture. Having completed the drawing, we are now ready to make our decision, when the phone rings.          Figure 1: Decision Tree It is our boss, telling us that he has someone on the other line who wants to sell us some market research data. He wants to know if you are interested Q: How much does it cost? A: $5,000. D: Thats dirt cheap for market research data. We would typically expect to pay $15,000 to $30,000 for even a simple survey. Q: Why is it so cheap? A: The data comes from a survey conducted for another company, who paid full price for the work. The market research company is trying to make a little extra money by re-selling the data. Q: What does the data tell us? A: It will only tell us whether the market is favorable or unfavorable. Q: Can we use that type of information? A: Yes. D: If you look at the tree in Figure 1, then both high growth and moderate growth would be considered favorable. Low growth would be considered unfavorable. Q: Is it worth $5,000? A: We can use the Expected Value of Perfect Information (EVPI) to come up with a quick answer to that (remember your boss is on the phone with you) Q: What does perfect information mean? A: It means that whatever it tells you is correct. Q: Can you actually buy perfect information? A: No, not legally, but let me set up a hypothetical situation: D: The following is not intended to insult anyones religious beliefs. It is supposed to be a little bit funny and to make a point. You are sitting in your office, feet up on the desk and thinking about the decision tree, trying to make up your mind, when a celestial being appears in your office. It is angel in full regalia: wings, robes, halo and harp. The angel announces that s/he is raising money for a vacation to Disney World by selling information about the future. The information was gleaned by eavesdropping around the throne of God, up in heaven. At this point, the angel pulls out a stone tablet, which is kept turned away from you. Q: Why does the angel keep the tablet turned away from us? A: S/he doesnt want you to see the information before you buy it. D: Thats a funny thing about buying information. Think of buying a car: do you send a letter to a dealership, listing the features you want on your new car, have them select one for you, and they drop it off when you send them a check? Probably not, most of us go look at the car before we buy it. If you rent an apartment, do you just sign the contract or do you go look at the apartment first (actually, there have been times in my life where I have been desperate enough to just signed the contract, but not by choice)? If you buy even a loaf of bread, do you turn it over to make sure it isnt all green on the bottom? Of course you do. Whenever we buy something, we look at it first. You dont do that with information. Q: Why do we buy information sight-unseen? A: If they let you see the information first, you might not pay them (though Im not sure Id recommend doing this to a celestial being). D: Remember that the next time you hear someone railing away at our Intelligence agencies. The CIA and friends spend their entire lives dealing with a product they must purchase before seeing (imagine running the purchasing department for your company like that). They naturally spend a lot of their time trying to verify information they have already purchased, and thats before they can begin to analyze it. It must be frustrating, to say the least. Q: So what is the angels information worth to us? A: That depends on what the information says. D: If we could see the information, then we would know what to do and we could judge what it is worth by how much we gain (or how big a loss we avoid). Unfortunately, we wont know what it says until after we hand over a check. Q: How do we know how much to pay? A: Thats a different question, and one we can answer via the EVPI calculation. Q: How do you calculate EVPI? A: Calculate a weighted average of the best payoffs for each state-of-nature. D: We dont know exactly what the stone tablet will say, but we can narrow down the things it might say, by considering what possible futures exist for our company. We can get a list of possible futures by looking at the decision tree. Q: Is that all the futures we have to consider? A: Let me answer that question with another question: Q: What is the assumption of exhaustiveness as it applies to states-of-nature? A: The assumption of exhaustiveness is that we have a complete list of the states-of-nature. D: That means the tree shows all possible futures, which further means that the tree shows all the possibilities of what might be written on the stone tablet. Q: Are the assumptions for a decision tree the same as the assumptions for a payoff table? A: Yes. D: You might have guessed that when I said (in an earlier lecture) that any payoff table can be drawn as a tree. The only difference is that the assumptions apply to the nodes (circles and squares), not the tree as a whole. Thus, each node is mutually exclusive, exhaustive, and deterministic. Q: Could you give an example of calculating the EVPI? A: Sure. D: Start off by listing the possible futures, as shown in Figure 2: HighModerateLowFigure 2: Possible Futures for EVPI Calculation Then, for each state-of-nature, look at the tree (Figure 1 is repeated as Figure 3, below) and select the best payoff.          Figure 3: Decision Tree (repeated) To help, I have shown all the High state-of-nature payoffs in green, all the Moderate payoffs in blue and all the Low payoffs in red. I have also put the winner for each state-of-nature in bold. Copying the winning (highest) payoff for each state-of-nature into our table, we get Figure 4: HighModerateLow25001200-50Figure 4: Possible Futures for EVPI Calculation The instructions say to calculate a weighted average of these best payoffs, so we need some weights. Q: What is our current best estimate that the stone tablet will read High Growth? A: 40% D: This came from the decision tree, our probability of High Growth occurring. That is likely to be our best estimate because it is our only estimate. Taking the rest of the probabilities from the tree, we get the calculation in Figure 5: HighModerateLow0.40*(2500)+0.40*(1200)+0.20*(-50)=1470Figure 5: EVPI Calculation Now we have calculated EVPI, but we still dont know how much to pay the angel. Q: Would we pay the EVPI amount, $1,470K (remember that the payoffs are in 000s)? A: No. D: We have to stop and ask ourselves how well we could do without the angels help. Looking up at Figure 3 again (I refuse to copy down that tree a third time), our best expected value is $1,380K. Considering that, the angels information, perfect though it is, is worth only the difference between those two, as shown in Figure 6: EVPI $1470K - EVw/oI - 1380K Purchase Price 90K Figure 6: Purchase Price for Perfect Information Q: We would pay $90K for the perfect information? A: Yes. D: It might seem that we are cheating the angel (perhaps not all that bright an idea, given his/her connections) because this information will be worth a lot more than merely $90K. Remember, though, that we can work without the information and make some decision, and that we still dont know what the information will say. Since we dont know, we have to consider all the possibilities, and that drives down the price. Q: Does all this have anything to do with our problem? A: Of course it does. D: To get back to the problem at hand, our boss is still on the phone, wanting to know if we are willing to pay $5K for some decidedly non-perfect information. Q: Will our boss really wait while we calculate EVPI? A: Yes. D: If you drop out my silly story, know how the calculation works and have the data in front of you, then this calculation should take less than a minute. Q: How does EVPI tell us what to pay for imperfect information? A: It doesnt, directly. D: What you do with EVPI is find out the cost of the information (we know that, $5K) and a quick estimate of the accuracy of the information. So you ask your boss if the market research company is any good. S/he says, Yes, weve worked with them before and theyve been pretty good. The only question, then, is: Q: Are you willing to spend $5K for pretty good information when you know you would pay $90K for perfect information? A: Well, I would. D: The EVPI acts as an upper limit for us. You will be amazed at how quickly that value ($90K) drops off as the accuracy decreases (we will calculate a version of that in the next lecture). The only virtue of the EVPI is that it is quick. Once your boss asks, you can perform the calculation and have a basis for making a decision in less than 60 seconds. It is an extremely rough estimate (quick ones usually are), but at least it is something to base your decision on. Q: Is there any other calculation, one that perhaps takes longer but is more accurate? A: Funny you should ask. See the next set of lecture notes, on Expected Value of Sample Information.  HYPERLINK "file:///C:\\Documents%20and%20Settings\\macleodk\\My%20Documents\\My%20Webs\\DSCI%206613.htm" Back  FILENAME Expected Value of Perfect Information, Page  PAGE 6 of  NUMPAGES 6 Page  PAGE 1 of  NUMPAGES 7 -50 Low (.2) Mod (.4) High (.4) 800 1500 910 -500 Low (.2) Mod (.4) High (.4) 1200 2500 1380 -100 Low (.2) Mod (.4) High (.4) 500 1200 660 Subcont Build Expand 1380 -50 Low (.2) Mod (.4) High (.4) 800 1500 910 -500 Low (.2) Mod (.4) High (.4) 1200 2500 1380 -100 Low (.2) Mod (.4) High (.4) 500 1200 660 Subcont Build Expand 1380 klmqrt> ? A B C E G H J M N P R S U V W X   * , 6 8   & ( t v ACDþþþþþþþþþþþþúúhWhU h25h2 hdw5jhaUmHnHuhhdw h$J0Jjh$JUh$Jjh$JUGst> @ A C F G I J N Q R T U W Y Z [ s t u  !!!!!!!!!!!!!!!!!!!!!!!!!$a$gddw$a$gddwgd$J$h^h`a$gd$Jz--/  ) * 5 6   % & s t !!!!!!!!!!!!!!!!!!!!!!!!!!$a$gd2$h^h`a$gddw$a$gddw @A QR12]^!!!!!!!!!!!!!!! !!!!!!!!!!!!$h^h`a$gddwRT245\]^`!pr46GIQST˿˻˻ha hfhf hf5hf hTq5 hr< 6hTqhWhr< hr< 5 h35h3 h,5hU h,Gop34!!!!!!!!!!!!!!!!!!!!!!!!!!!$h^h`a$gddwFGPQ!!!!!!!! $$Ifa$gdf$h^h`a$gdf$h^h`a$gddw TVWY\m]!M!M!M!M!M!M!$h^h`a$gda$h^h`a$gdakd$$IfTlF8$  t06    44 laTTUWXY[]^`cdfhiklmn Z^| ~ !!""""""##.$3$;$<$D$E$K$L$M$N$O$]$$$$$ h855 h,>*h,h,>*h85h85>*h85 h,6 h,5h, ha5haha5hahaB*phhahaB*phhahaB* phhajhaUmHnHu:\]_`dghjkmopq!!!!!!!!!!!!!!!! $$Ifa$gdIn$h^h`a$gda$h^h`a$gdamaaa $$Ifa$gdInkd$$IfTlF8$  t06    44 laT  | m]!M!M!M!M!M!M!$h^h`a$gda$h^h`a$gdakd;$$IfTlF8$  t06    44 laT !!!!!!!!!!!g2 $$Ifa$gdIn$h^h`a$gda !!!! $$Ifa$gdInkd$$IfTl֞`8wOYl?  t0644 laT!!!" " ""g2 $$Ifa$gdIn""!kd$$IfTl֞`8wOYl?  t0644 laT"."/"""""""+$,$:$M$b$$$$$$!!!!!!!!!!!!!!!!!!$ Hh^h`a$gd,$ Hh^h`a$gd,$ \ dh^h`a$gd85$h^h`a$gd,$h^h`a$gd,$$x&y&&&&&h'i'''''D(E((((())V*W*i*j*E,F,!!!!!!!!!!!!!!!!!!!!!!!!!!!$ Hh^h`a$gd,$$$v&x&y&{&&&&&i'k'''''E(G((((((u))))))))))U*V*W*Y*j*l*!,*,F,H,,,----q-r-s-w-x-y-z-{--ԺŴԭhJhJCJaJjhJhJCJUaJ hhdw h$J0Jjh$JUjh$JU hh$Jh$J h6 h5 h{5hh{h Ykh, h855h85:F,,,---y-z----------.....!!!!!!!!j!!jv:cv:cv:cv:gddw$a$gdJgd{$a$gdJ$a$gd$h^h`a$gd$J$ Hh^h`a$gd,-------------------------------------.......".#.(.).-...3.4.>.?.I.J.U.V.[.\.a.b.g.ǯ haCJhdw hdwCJhh{CJaJmHnHu h{hJhJh$JCJaJmHnHuhJhJCJaJhJCJaJmHnHujhJhJCJUaJ>..".#.(.).-...3.4.>.?.I.J.U.V.[.\.a.b.g.h.m.n.x.y....v:v:v:v:cv:cv:cv:v:v:v:v:cv:cv:cv:gddwg.h.m.n.x.y........................../// / //////)/*/5/6/;/*B*ph',27=HS_ekqw (3?EKQWbmy~'3210/.(&%$#"!    ! " # $ % &'()*+,-./0}|{zyxrponh2',27=HS_ekqw (3?EKQWbmy~    !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJ'j z z z z z z^ h'hst>@ACFGIJNQRTUWYZ[stu)*56%&st@A  Q R 1 2 ] ^ op34FGPQTVWY\]_`dghjkmopq|  ./+,:MbxyhiD E !!V"W"i"j"E$F$$$%%%y%z%%%%%%%%%%&&&&&&"&#&(&)&-&.&3&4&>&?&I&J&U&V&[&\&a&b&g&h&m&n&x&y&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&''' ' '''''')'*'5'6';'<'A'B'G'H'M'N'X'Y'c'd'o'p't'u'z'{''''''''''''000x000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000 0 0 0 000000000000000000000000 0 0 0 0 0 0 0 0000000000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00000000000000000000000000000000000 00 0 0 00 0 0 0 0 0 0 0 0x0x0x00@00@00000000000000000000000000000000000000000000000000000@00000000000000000000000000000000000000000000000000000000000000000000000000000s>@ACFGIJNQRTUWYZ[stu)*56%&st@A  Q R 1 2 ] ^ op34FGPQTVWY\]_`dghjkmopq|  ./+,:M%y%z%%%%&&&&&&"&#&(&)&-&.&3&4&>&?&I&J&U&V&[&\&a&b&g&h&m&n&x&y&&&&&&&&&&&&&&&&&&&&&&'' ''')'5';'A'G'M'X'c'o't'z'''''''@00000000000000000000000000 0 0 00 00 0 00 00(0 0`0`00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000@0@0@0@0@0000000000000000000000@000 00 @0G900T@0G900T@0G900T@0G900T,@0G900TH@0G900Td@0G900T@0G900T@0G90 0T@0G90 0T@0G90 0T@0G90 0T @0G90 0T(@0G900TD@0G900T`@0G900T|@0G900T@0G900T@0G900T@0G900T@0G900T@0G900T$@0G900T@@0G900T\@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@0@000 WWWY|T$-g.d//!+-/3 \ !!""$F,.../// "#$%&'()*,.0124/lq%r%w%'XX 19@BGRT^eglwy!!8@)c  )(  B  3 g f  s *g  ZB  s *Dg  TB  c $Dg TB  c $Dg ZB  s *Dg ZB   s *Dg  f   s *pG  f   s *pG  f   s *pG   h - 13  # #" r2  s *pGg T/>1 ZB  S Dg >..0ZB  S Dg >0.z2`B  c $Dg ...`B  c $Dg .z2z2`B  c $Dg >00f  S +. / `  C 0 1 l  c $pG.-. l  c $pG.T/0 l  c $pG.*1z2 `  C 1 13 h - 13 # #" r2  s *pGg T/>1 ZB  S Dg >..0ZB  S Dg >0.z2`B  c $Dg ...`B  c $Dg .z2z2`B   c $Dg >00f ! S  !+. /  ` " C  0 1  l # c $ pG#.-.  l $ c $ pG$.T/0  l % c $ pG%.*1z2  ` & C 1 13 h - 13 '# #"  r2 ( s *pG(g T/>1 ZB ) S Dg >..0ZB * S Dg >0.z2`B + c $Dg ...`B , c $Dg .z2z2`B - c $Dg >00f . S .+. / ` / C 0 1 l 0 c $pG0.-. l 1 c $pG1.T/0 l 2 c $pG2.*1z2 ` 3 C 1 13 B g 3 g f h s *Khg  KZB i s *Dg TB j c $Dg TB k c $Dg ZB l s *Dg ZB m s *Dg f n s *JpGn Jf o s *IpGo If p s *HpGp Hh - 13 q# #"  r2 r s *GpGrg T/>1 GZB s S Dg >..0ZB t S Dg >0.z2`B u c $Dg ...`B v c $Dg .z2z2`B w c $Dg >00f x S Fx+. / F` y C E0 1 El z c $DpGz.-. Dl { c $CpG{.T/0 Cl | c $BpG|.*1z2 B` } C A1 13 Ah - 13 ~# #" r2  s *@pGg T/>1 @ZB  S Dg >..0ZB  S Dg >0.z2`B  c $Dg ...`B  c $Dg .z2z2`B  c $Dg >00f  S ?+. / ?`  C >0 1 >l  c $=pG.-. =l  c $<pG.T/0 <l  c $;pG.*1z2 ;`  C :1 13 :h - 13 # #" r2  s *9pGg T/>1 9ZB  S Dg >..0ZB  S Dg >0.z2`B  c $Dg ...`B  c $Dg .z2z2`B  c $Dg >00f  S 8+. / 8`  C 70 1 7l  c $6pG.-. 6l  c $5pG.T/0 5l  c $4pG.*1z2 4`  C 31 13 3B S  ?>ACDGJKLNORUWTWYZ]`abdehkm' %t 2 TOt t  9t%dt  t @ th t  t t'%7t  3t  ootq%mtn2 7Otl tj  t~%Hto }th@ otgL tk  ti tq%tp gtm SStz%%&&'''z%%%&& &&&4&:&?&E&J&Q&n&t&y&&&&&&&&&&'''%'*'1'N'T'Y'_'d'k'':::::::::::::::::::t%%%y%z%%%%%%%%%%%%%%&&&&&&"&#&(&)&-&.&3&4&>&?&I&J&U&V&[&\&a&b&g&h&m&n&x&y&&&&&&&&&&&&&&&&&&&''z%%'MACLEODKSchool of Businesslh U r< ,@]W2F,85$JJPUEY YklInTqSq3f{dwac  '@y%y%cy%y%'`@UnknownGz Times New Roman5Symbol3& z Arial"qhǣrǣrDD24g%g%3QH)?Sq%Expected Value of Perfect InformationMACLEODKSchool of BusinessOh+'0 ,8 T ` lx&Expected Value of Perfect Informationofxpe MACLEODKValACLACLNormalKSchool of Businesse2hoMicrosoft Word 10.0@F#@ʁ@ʁ՜.+,D՜.+,h$ hp  East Carolina UniversitysDg%{ &Expected Value of Perfect Information Title 8@ _PID_HLINKSAT ^9http://www.business.ecu.edu/users/macleodk/DSCI 6613.htm^9http://www.business.ecu.edu/users/macleodk/DSCI 6613.htm  !"#$%&'()*+,-./012345789:;<=?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~Root Entry F Data 61Table> WordDocument/jSummaryInformation(DocumentSummaryInformation8CompObjj  FMicrosoft Word Document MSWordDocWord.Document.89q