Skip to main content
Loading Explore guides…
Dynamic Choice Economics · Acedra
Home
Explore
Economics
Dynamic Choice Economics
Sign in to save
Chat
If YouTube asks you to sign in, open the video on YouTube.
Watch on YouTube
Reload player
Transcript
106 segments
Copy
Hide video
0:00
[Music] I'm an economic theorist uh I'm mostly interested not in particular topics in economics but in how economists do their business which is what theorists do so this is may seem a little dry but you know this is what one end of the business school does when they want to talk about how the discipline does what it does in particular uh I'm going to give you a look at the research program I've been on it's a little short on details it's going to be long on slides and I have 55
0:32
minutes to get through this what is it about well my research program is about Dynamic Choice uh economists have a way I'll describe that way in a bit as to how to model deal with Dynamic Choice when they go out and try and understand Dynamic choices that people make uh I'm interested in providing economists with more realistic more consistent with observed Behavior models of dynamic Choice that's the research program now to understand what that means and why
1:01
there's an issue here uh we got to take a step back and the step back is to talk about how economics as a discipline works so let's do that economists do two things they talk about people sometimes firms making choices not any choices but consistent choices so a lot of the subject is about what's called Choice Theory how does this person choose how does that person choose what's the basis on which they choose always with consistent choices and then they talk
1:30
about how those choices are equilibrated in various institutional settings normally markets but increasingly other institutional settings uh Mike Harrison how many people here know Mike Harrison right okay so Mike Harrison uh who's just recently retired another institution in the school or an institution in the school describes these two pieces as the first is the theory of cats and the second is the theory of cats in a bag so you should think of that now economists are somewhat um what the
2:00
word they they like to extend their influence wherever they can and so they sometimes instead of calling them consistent choices they say it's rational choices and I think it's a bad you know the the implied pejorative is that choices that don't conform to the economic model are irrational that's bad language I'd like to avoid that language I'm going to use the word consistent choices so there's choices and then there's the equilibration I'm all about the choices part of the subject
2:30
so what is consistent choice to an economist uh it's everything but the following so man walks into a Roadside Diner and says I want a piece of pie and a cup of coffee please waitress says we have apple and peach pie today guy says in that case I'd like a piece of apple pie she says I just remembered we have one slice of banana cream left as well and he says you also have banana cream well in that case give me Peach okay right that's the one thing that economists don't want to deal with okay
3:10
consistent choice is everything but what that guy just did now it's more formal formally if an economic agent chooses an object X and not y when both of them are available then she never chooses y when X is available having shown that she likes X more than y she should always like X more than y that's consistent choice to an economist and as long as that property holds we can model her choices it's a mathematical theorem we can maximize her choices as if she's
3:41
maximizing a numerical function that Maps every option into a number and she's going to take whatever option has the biggest number now there are some technical conditions that you have to meet for that really to follow from just that property there but they're not very stiff uh conditions in particular if the universe of things she can chooses finite the only thing you have to add to that condition the formally condition is that you give her a set of objects to
4:06
choose from and she says okay I'll have this one or if you prefer this one or this one she makes a choice she doesn't stand there and say oh I can't choose that plus the first thing that means you can model her choices if she's maximizing utility and that's what economists do they say consumers are utility maximizers now in special circumstances especially when there's uncertainty in the world we might make more structural assump asss on the nature of her choice
4:31
uh say that she maximizes expected utility or subjective expected utility but by and large the essence of economic choice is that level of consistency nothing else now empirically it doesn't work and a good example comes from the marketing literature designer of maale odor catalog uh says you know he's he's trying to design a web page or a catalog page to get you to pay $200 to buy a particular printer he wants you to choose the printer over keeping $200 in
5:04
your pocket so how does he do that well he might do one or two things he'll advertise two printers on the page the one he wants you to buy for 200 and a second one which has a few more features at a much higher price and you're supposed to conclude wow that's $200 printer is a great deal you know I only get a little bit more for $350 I'll buy the $200 printer or going the other way he advertises the printer he wants to buy and a second much worse printer for
5:33
$190 so you say I just have to spend $10 more I get all these extra features that $200 printer is a great deal now the guy who's designing the the web page here is exactly playing on the frame that you put the choice into which is the one thing economists don't like they don't want the frame to influence your choice so you know this is just like the availability of banana cream piie seem to frame the choice between Apple and Peach in a way that got the guy to change what he's doing that's what
6:01
you're doing and people do that nonetheless economists say no they don't right economic models by and large have kept to the consistent Choice model of individual Behavior because they say once we admit the frames change what you're doing there's not much we can say about your choices so economists ignore that kind of stuff which the marketing people love because that way they make fun of us but uh economists believe in consistent Choice as a good way to attack problems of economic choice in
6:31
the world now what's Dynamic choice right that's the subject I'm interested in I if I'm choosing a kind of pie to eat at a Railside Restaurant or whether to buy a particular printer that's a static choice right at the moment I have a choice between different things I choose to spend my money on the printer I choose to have apple pie or peach pie or whatever it happens it's over if we're talking about what career you're going to follow or where to go to school or how much you're going to save
6:59
that's more of a Dynamic Choice first of all the choice you make today is going to affect the opportunities you have tomorrow for future choices and today's Choice needs to be informed about the value that you attach to tomorrow's opportunities or lack of opportunities so it's that kind of choice the choice of career how much to save where to go to school those are Dynamic choices because they have implications going forward in the future those are the kind of choices that I'm interested in now
7:25
how do economists model Dynamic Choice well they use the statically optimized strategically and then just do it or sat swoosh approach uh what is this approach well it starts off with modeling the problem the choice problem the individual has so you have a decision maker who faces a dynamic decision problem there are choices to be made today but also tomorrow next month next year we have a bunch of dates at which the decision maker has to make choices that's the there are connections between the
8:02
choices what you choose today will have an influence on what options you're going to have to choose tomorrow and the day after that and the day after that so your choices today not only impact what you get today but what choices you'll have in the future while you're making these choices information may arrive that you can use in use in making later choices so you think about what information am I going to get later on and your choices today can affect what information do or do not
8:30
get so it's a very complex structure that Economist will start by modeling that's the dynamic decision problem the person has and for every Dynamic decision problem there's a range of strategies that the person can do things that say well today I'm going to choose this option and tomorrow if I get this information between now and tomorrow I'll make this choice if I get this other information I'll make that choice I'll full what I'm going to do for the rest of my life in all the situations
8:58
that I'm going to have to make choices that's a strategy and we first formulate all the strategies that we have in this decision problem and then every strategy we have Maps into an outcome how much I get to eat today tomorrow the next day possibly the outcome is random there's uncertainty in the world in which case we think of the distribution over our consumption over many days as the outcome the the outcome is the distribution of possible consumption streams we have
9:29
and what we do as in this satoo approach is we say okay here's a strategy let me write a little bit on the board so here's my decision
9:44
problem with a whole bunch of structural stuff and that gives us a set of
9:56
strategies now each strategy gives us an outcome what we're going to get so that's outcome from
10:10
strategy number one this is the outcome from strategy number oh 7,321 th000 I mean lots of strategies uh typical decision problem in the financial literature goes on forever you have an infinite number of strategies and what you do is for every strategy you have over here you figure out what will be the outcome for me and then you say I've got preferences or utility over the outcomes so that I look at every strategy I could do I look at what outcome it produces I look at how
10:53
good an outcome that is for me I figure out that this is the outcome or yeah this is the outcome out of all the possible outcomes I could have from all the possible strategies that's the outcome that maximizes my utility which means this is the best strategy to play that's what I'm going to do and then mindlessly I just Implement that strategy through time so you choose at the outset right we I'm sitting at this decision problem I'm about to choose where to go to school that has an impact
11:24
on my career I think through the entire scheme of my life what's going to happen to me for the next you know 60 years what are the strategies I could have for 60 years what's the outcome from every strategy which of those outcomes do I like the best that means I should go to Dartmouth where I went once I have that strategy in mind I just do it that's the just that's the swoosh part of this that's the way economists model Dynamic choice at the outset pick a strategy
11:55
pick an optimal strategy based on the outcomes and just implement uh this is viewed by economists as obvious this is how you would make the right decision uh and it's so obvious you know I've written a couple of Economics textbooks in every economics textbooks I've ever written I discuss the problem of dynamic choice and what is the standard approach I don't call it Sato in the books but that's essentially what it is and it's the only textbooks that ever discusses everybody else says
12:27
well it's obvious that's what you're going to do uh it's just unworthy to talk about the assumptions that go in to this particular way of viewing Dynamic choice so uh oh before I get to that now nobody really believes that people sit down trying to pick a college thinking through their entire lifetime it's an as if model and the idea there is that it's not how people think about things but the way they behave is as if that's what they did a good example here is you go back to uh
13:02
utility maximization the basic model of Economics I mean if you read an economics textbook it says when you go over to the grocery store you got money in your pocket you got a list of things you might buy you consult your utility function you consult your budget constraint you optimize subject to a budget constraint now nobody does that right you go down the aisle with your cart you look at something you say that looks good I'll put that in the cart you know occasionally what happens is you go
13:26
down another aisle you say oh that looks even better and you go back to the aisle and put something back you make your choices sort of as you go along well economists say okay that's how you make choices but it's as if when you're in Safeway you're maximizing your utility subject to a budget constraint it's not how you think about it but it's the net result of what you do and there's a famous analogy which is due to Milton Friedman that says that well let me give
13:52
you a good example of an as if model trees in the spring decide where to put their leaves and a good as if model of how trees set their Leaf canopy is as if they're maximizing their exposure to Sun over the course of a day now trees don't really do the calculation right I mean trees have trouble with simple sums but if you model the setting of a leaf canopy as if a tree was maximizing it it turns out to be a pretty good and useful model of how the tree will set up its Leaf canopy so
14:27
this is viewed as an as if model it's not how people make choices people don't sit down and think about things this way but their choices are as if they do this and so maybe that works its safe way uh there's empirical evidence on both sides uh evidence that says yeah people roughly conform as if they were maximizing utility subject to a budget constraint the marketing guys come in and tell us about the way they design mail Auto cataloges to make this model
14:52
not work but in particular the question I I I've asked for you know many years now is is is a good as if model of dynamic choices there's something about the Dynamics of the choice that says that the satoo approach is not a good as if model so that's what I'm interested in is not so much what's the answer to that question I say the answer to that question is no it's really what are the reasons it's not a good model and then what are you going to do about it so
15:22
I've got three reasons that I think it's not a good as if model of dynamic Choice first of all people's tastes change okay you're making a decision here where to go to school stuff like that well you know three years down the line you may decide you know I thought that's where the path I wanted to follow but it's not I'm going to follow a different path and I'll get back to this in a minute it has two and to some extent contradictory consequences for the nature of decisions
15:50
I make as I go along the fact that I recognize that my taste will change there are things that you can't anticipate sitting here thinking about what you're going to do for the rest of your life when you're choosing where to go to school you know something unanticipated is going to happen to you you know in your sophomore year let alone five years out on the job and the actions you take today take account of the fact that you know there are things you can't anticipate so that's reason
16:18
number two and third you know this problem is just too hard to think about it's too complex you can't really do the optimization you know so this is strategy 7 million 21,000 Etc I mean there are an infinite number of strategies and nobody in most realistic problems can figure out which is the best According to some numerical measure so you couldn't do this if you wanted to if you had all the data instead what people do is they follow juristic decision rules and those
16:48
juristic decision rules can't always be justified using this approach if you use a urtic decision rule it's not as if you are solving an optimization problem like this so in each of those three what I'm going to do now is talk about each of those three what there is and a little bit about what I've worked on in that particular regard so first of all changing tastes now you can't doubt people's tastes change as time passes uh you plan you're going to do something in the
17:21
middle you say you know I don't want to do that anymore I'm going to go do something else uh I do that a lot now the question is if you know your tastes are going to change is that going to affect what you do today and there are three options at least in economic modeling that you have to say three answers to the question what are you going to do about it today knowing that your taste may change one option is nothing what you say is well my taste might change I don't care today I'm
17:53
going to figure out what I think is best for me given my taste today I'll keep going until my taste change and then I'll change my plan a little bit more sophisticated is you can strain your later choices so you get what you want today and uh this goes the the original story here uh comes from the um Odyssey uh when odys is going past the aisle of the sirens he knows that once he hears the siren song his tastes are going to change he's going to throw himself off so he
18:23
constrains himself he has them tie him he has his Sailors tie him to the mass so that he can't have the option of throwing himself off the ship later on more modern examples that people talk about and as one deals with me is I try to make sure that there isn't any sweet food in the refrigerator late at night because I know if it's there late at night my will will get weaker and I'll start eating that stuff uh one of the original papers on this uh was about Christmas clubs why do
18:54
people you do people even know what a Christmas Club is anymore how do people know what a Christmas Club is okay so for those of you who don't used to be a big thing uh at least when I was being raised in in New Jersey a long time ago is that the banks would very happily take some money from you every week or every month and not let you have it until two weeks before Christmas the idea idea being is that you know if you start putting money away in a regular
19:21
bank account you can always withdraw it and then the money won't be there for Christmas but if you put the money in a Christmas Club you can't touch until it's time to go buy the Christmas presents this is a way to force yourself to save and not spend the money and the banks used to love this because they paid no interest I mean you know this the the banks are holding your money for you not letting you have access to it well that's exactly this idea that you
19:46
know you'd know at some point you might want to spend the money that you're saving for Christmas presents so we'll make sure you can't get at it well you constrained Yourself by going there now a very recent but pretty large literature about this this is about hyperbolic discounting uh hyperbolic discounting is the following phenomenon I give you a choice between so this is a choice between $100 uh today or let's see what's a good number 120 next week so I give you that choice a lot of
20:32
people I don't know if I quite got the right numbers here but a lot of people say well I prefer $100 today but if I give you the choice of $100 in 51 weeks or 120 in 52 weeks then people say well you know 51 weeks 52 weeks what difference does it make $20 is good so they go that way and the problem is is that in 51 weeks if I offered you the choice of $100 today or 120 next week youd say oh I made a mistake I want the $100 today so that's hyperbog discounting it it's as time
21:17
passes you put a lot of weight on immediate consumption versus next week the difference between 51 weeks and 52 weeks isn't as great and that going to shift as you get closer and closer to 51 and 52 weeks out now what are the consequences of this well it affects how much money you save the form you save it in in particular today if you think that you know your preferences for 51 versus 52 weeks or right you want to lock in money that you save so that you don't
21:49
pull the money out with a big penalty stuff like that subject of some okay so you could do nothing you can constrain your later ches on the other hand with sort of the exact opposite you say look I don't know what my future tastes are going to be but what's important to me today is to give myself the option to do what's best for myself what I think will be best for myself later on so this goes with the following sort of story you're making reservations at a restaurant you're not going to go to
22:23
this restaurant for 3 weeks so you're making reservations way in advance how big menu of choices so today you're picking the menu of the restaurant and Suppose there was menu one and menu two and you look at menu one you say yeah you know there's some good stuff there menu two there's some good stuff there on balance I prefer menu one to menu two so this one is better but if there was a super restaurant that had menu one and two all in the same menu why not go with
23:00
this in case something on that menu 3 weeks from now is going to have a lot of appeal to me so that's you know I had
23:10
prefer for flexibility I want to leave myself the flexibility to make a choice later on because I don't know what my tastes are going to be I don't know what my preference going to be later on but I don't want to constrain myself today I would allow my later self to have what it wants so that's complet completely different from number two number two you're saying what I want today is what I want to force myself later on to have now you're saying well what I'm going to
23:36
want later on is what I should allow myself to have two different very different ways of dealing with the idea that tastes are changing so my interest in this the constrain your later self the thing that goes with hyperbolic discounting and Christmas clubs it's big important topic lots of work has been done in economics not me this is not something I've ever worked on instead I work on the two other things things the two other directions the leave yourself
24:02
flexibility Direction well paper I wrote back in 1979 a long time ago says okay if you think your tastes are going to change exactly what is the form that this preference for flexibility will take so trying to say mathematically how do I put together a formal model which is what economists do how do I put together a formal model of this kind of preference for flexibility where it's being driven by the fact I don't know what my future taste will be and I want
24:30
to give my later self the chance to make a choice so that's one thing I've worked on the other thing is something I'm working on right now which I find very fascinating this is comes out of my 10 years as senior associate Dean uh I'm very interested in motivation and how do you motivate people now you know you guys deal with the faculty so it won't surprise you to learn that they're a hard group to manage right I no it won't surprise you in particular you know as senior asso you got the job
25:01
of motivating them to do what's good for the school and not what they want to do necessarily and so I've spent a lot of time thinking about motivation especially in the context of the faculty now economists if you go to an economist and you say okay we're talking about
25:25
motivation on the job motivating faculty members of the GSB to spend time thinking about their teaching actually put some effort into teaching the NBAs not spending all their time doing research or whatever else it is they do uh you say to an economist I've got a question about motivation on the job an economist the standard Economist says I know the answer to your question the answer to your question is incentives right economists have one answer to motivation which is pay them
25:57
more money if do good stuff that's it so this is the economist
26:09
answer uh promise more money if the employee does good
26:25
stuff now the problem with this the the of incentives uh it's a theory that is very important in economics um it's one of the things I think we're now up to like three Nobel prizes given for various aspects of the theory of incentives the big issue in incentive theory is when good stuff happens do you know that it's because of what the employee did if there's uncertainty between what the person tries to do and what the outcome actually is promising them more money if they do good stuff
26:59
well what it should say if the employee seems to have done good stuff and that introduces uncertainty into the employee about their compensation if I tell my fellow employ my fellow instructors or fellow professors I will pay you more money if you I'll give you bonus if your teacher ratings are very high they can try to get their teacher ratings very high but it might not work and they don't like the uncertainty in their compensation worse than that another
27:32
problem with this is that if I tell them you know I'll give you a lot more money if your ratings get high what are they going to do they're going to start pandering to the students right I don't want them to Pander to the students he says that I can't tell whether the high ratings are due to pandering or to the fact that they're actually teaching a better course so there's an economic theory of incentives which for the kind of job at least that faculty members have and a lot more jobs bes sides
27:58
including you know the kind of jobs you have it doesn't really work very well it works a little bit but not very well that's what economists do now go across the street to my friends in OB these particular the social
28:15
psychologists they have very different theories about motivation I have theories that say talk about self-determination and self-perception and in particular I'm give you an illustration of this which is a hospital in Boston called Beth Israel uh anybody hear from Boston no okay so Boston has a lot of hospitals in particular there's a there's a little District of town known as the hospital District which has Massachusetts
28:47
General and brigam and
28:53
woman's and Beth Israel and there's one more of the four is hospital called deonis all of these are the Harvard Medical School teaching hospitals they are the hospitals that you know people who are at Harvard Medical School go and staff they take rounds there stuff like that residents at these hospitals are connected to the Harvard Medical School stuff like that now it's very clear in the 1970s when the Story begins that there was a clear order amongst these hospitals and it was Mass General was
29:31
the best briam and wom's number two Beth Israel three deaconness number four and everybody agreed you know the prestige positions were in Mass General new guy comes to Beth Israel a CEO guy named
29:48
rabkin this is around 1972 I think and he wants to move up in the pecking order so he's trying to figure out what can I do to make Beth Israel stand out and maybe be as thought of as good as Mass General and he hit on a system of we're going to become so his theory or his strategy is become the hospital patients
30:22
prefer so he's going to emphasize service to the patients and in particular the way he did this is he went to the senior RN the senior registered nurses and he said we're going to do nursing differently here we are going to have what's called primary nursing if a patient comes and is going to be a patient at Beth Israel that person is assigned a primary nurse it's like you meet your doctor you meet your primary nurse you sit down and your primary nurse and I'll use she because
30:52
most of them were she's at the time your primary nurse is dealing with you and maybe two possibly three other patients at any one time when she's on duty for eight her eight hour shift she's doing everything for you she's managing your care she's cleaning up your bed stuff like that but she's also the center of your care the doctors talk to her about what needs to be done next for you uh when she's off duty the the the 16 hours a day she's off duty she's still on call if there's an issue about
31:24
you you know primary nurses would get phone calls at 3: in the morning saying the family of one of your patients has a concern talk to them okay now did they give the nurses more money for doing this no and the net result of this is that the nurses loved it they thought this was great the cue to become a primary nurth at Best Israel for people who are nurses at Mass General briam woman's just got bigger and bigger and bigger what they did is they enriched the job
31:51
content being a nurse you were doing something you're doing something important you were getting psychological rewards for doing that something important and the nurses responded really really well to that now it's a great story except it has a very bad outcome a bad ending which is primary nursing is a little more expensive and as times change subsequent CEOs said well we've got to get rid of primary nursing it's too expensive given the way we're compensated and as you can imagine
32:20
it was a disaster they made a psychological count contract with the nurses which they then broke and bringing a psych iCal contract is much worth well okay so this all fits in with the psychological theory of self-determination which says people will give you better output employees will give you better output if they have more self-determination if they're doing something as important they perceive as important if they're connected to other people doing the same sort of thing and
32:51
also with the theory of self-perception which is people people when they do stuff they try and figure out after the fact why did I behave in this particular way if they did it because they got a Bonus they say I'm doing it for the money and then they will continue to do things for the money whatever it takes to get the bonus is what they'll do if on the other hand they say they do it because my patients are important to me or because I love the work they will increasingly love the
33:22
work they will think that their patients they'll internalize the welfare of their patients more well that from economic terms say social psychologists don't think of it this way but from a point of view of Economics what is that that's what you do and how you do it and what reward you get for it changes your taste changes your utility function so one of the projects I'm working on and not meeting with greatest of success because my colleagues in economics really don't
33:47
like to talk about changing tastes that much but what I'm trying to do is convince them that things like this are real they're important and you know I won't go into all the ways I Tred to manipulate my colleages when I was associate Dean but I was trying uh you know this is more important motivational Force than the stuff that economists say is important but if you want to talk about it in economic terms you have to talk about changing tastes and the way people's tastes change what changes them
34:17
how do they change and that takes you from economics into psychology you got to understand social psychology cognitive psychology and try to import that to economics so one of the things that I'm doing right now one of the things that I think is very important for economics is to try and do better on how tastes change and then what's the impact on that for dynamic Choice okay so that's number one changing tastes unanticipated contingencies um in the saso model every
34:47
contingency that might ever be realized is anticipated and each one is given its probability this has probability one 10 million this one has probability one in a thousand you do a grand optimization over all those things now formally you could have things that you don't anticipate it's just that you make no allowance for them in the decision you make today because they have zero probability you thought now there's a s branch of Economics called transaction cost
35:12
economics uh one of the great pioneers of this is is a guy up at Berkeley named olly Williamson um and point out olly got the Nobel Prize for for all the work he did on transaction cost economics which says you know the standard economic model the SAT model applied to relationships between two firms or relationships between an employee and employer doesn't make any sense because the world is not driven by the things you can anticipate it's driven by the things you can't anticipate in a
35:40
long-term relationship and so what's important and understanding when a relationship will be efficient or not is not what do you plan is going to happen but how do you plan to meet the things that you didn't anticipate that's what transaction cost economics is all about it's about the governance structure which is who has which decision rights when decisions later on have to be made that you didn't anticipate so you know that's what unanticipated that's why uned
36:07
unanticipated contingencies is important and my contribution as I go back to this paper I talked about before about preference for flexibility viewed One Way say I want to model the idea that my taste will change what are the consequences for the choices I make over menus this goes the other way it says if I see the choices that you over menus and you have preference for flexibility what are the consequences for how I model your choices what sort of expected utility model model do I use and so
36:37
there's paper 1979 follow-up papers in ' 86 and ' 92 I won't go into a lot of details about this because this is old business for me but trying to come to grips with the idea that there are unanticipated contingencies that you prepare for very important in economics and in long-term relationships very important to Dynamic choice and uh not something that economists have been very good at with the exception of the people who follow OE and then finally there's complexity
37:06
in your ristics what's the issue here well most dynamic decision problems are amazingly complex in fact they're far too complex for any Economist to solve uh so when you have papers in economics that talk about Dynamic decision problems you have one of two choices you either assume that the consumers in your model your model of economy are solving problems that you know you could never solve and there's a whole literature literature on General equilibrium that assumes people sitting
37:39
here today A lot of finances this way too people sitting here are making portfolio decisions making projections of future prices of things nobody can make those projections but we assume our agents can and we derive results based on the assumption that they can okay or when you want to be a little more structural you only deal with the kinds of problems that you can solve which is tremendously limiting in the the range of dynamic decision problems the economists will actually talk
38:09
about so uh let me give you an example of a problem that's pretty simple to describe and pretty difficult to solve contemplate hiring staff for a professional service firm that takes one projects so you are in charge of hiring the staff for this Professional Services firm as you look at the PO IAL employees you're unsure how valuable different candidates will be so you got a bunch of
38:35
candidates there they are and they all are carrying their CVS and they all have letters of recommendation that's not going to tell you exactly how good they're going to be and it's certainly not going to tell you how well they'll work together well this guy with this CV and this these letters of recommendation will these people work well together or not so how do you find out who's good who's not well you hire them you set them to work and you see how it goes so
39:03
the question is you look at the CVS you look at what information you have you know you don't have complete information and then you got to decide who do I hire okay very standard problem and it's not just hiring people it's what tools do you use what assets do you provide for yourself uh questions at the outset is how many people do you hire on your team Do you want to build in some redundancy and in particular suppose you need you know there's a particular
39:33
specialty so these people are part of a specialty you need one person on your team with that specialty you've got let's say three candidates you don't know which one of those three candidates is going to be the best so should you hire two of them and find out should you just hire one and hope that person works out understanding that if you hire one person and they work out kind of okay you got no information about how those two guys are going to do if you hire two
40:00
of them at least you'll have information about how good each of them is and then you can make a choice very simple problem to describe impossible to solve you cannot solve if if I formally describ this with a model with uncertainty stuff like that this is an
40:20
example of what's called a multi-arm bandit problem
40:33
okay and just see anybody heard the term before one two okay so what is a multiarm bandit problem it's called multi Arm Bandit problem because it's about a slot machine so here is a slot machine in Las Vegas which has the following weird characteristic for a slot machine there's an arm on this side and there's an arm on this side and here's a slot where you put your money you put your coin you put your coin in and that enables you to pull one arm or the other but you can't pull both now
41:07
you don't know what is the return from this arm of the Bandit you don't know what the return is from that arm of the Bandit and the only way you can get information is by pulling the arm so suppose let's take a very simple case suppose you knew that this arm wins oh 10% of the time and you think this arm wins 5% of the time or 14% of the time and each of those is equally likely well then if you're only going to pull one arm once this arm is better on average but if if you got a little bit
41:58
of time maybe you want to try this arm a few times collect statistics on how often it wins and see you know am I getting the 14% machine because then I know that's the machine I want to play that's the arm I want to play that's a multi-arm bandit problem is you got a course of different actions you could take you're not sure what the returns from each course of action is and the only way to find out is to try it well this is a multi-arm banded problem now the solution how do you optimally play
42:27
the multiarm Bandit this was a very difficult problem for years and years when I started in 1975 sort of at the business school when I was in graduate school 1972 this was an unsolved problem in any sense nobody knew how to solve this problem and then a guy came along named gens who said well in the very special case where there's no information about this arm in what goes on there and vice versa where the arm arms the uncertainty about them is independent you don't
43:00
learn anything about one arm by pulling another arm then here's how you solve the problem and it's a beautiful solution I mean it's a brilliant piece of work the problem is this is a multi-arm banded problem where the arms are the different teams you could assemble and those are certainly not independent if I hire this guy and this guy and this guy I learn a hell of a lot about the performance of a team that just consists of those two so they're not independent nobody has a clue how to
43:29
solve this problem and you know there's good reason to believe computer scientists o guys have been working on this problem good good reason to believe that no one will ever be able to solve these until we get computers which are much faster much faster than any computer we have today so what are you going to do about this um well people will use urtic on how to solve this problem what's an example of a urtic uh the best examp I give you a ristic comes from the book and then the
43:59
movie Moneyball people know about the the book and movie Moneyball so this is about the Oakland general manager Billy Bean and the euristic that people used to use in baseball in assembling a B team of baseball players is they would send old baseball players out to watch them play and sort of says you know I think this guy is better than that guy based on how he looks and the big innovation that Moneyball talk about is Billy Bean said ah you know Screw how
44:30
people look give me their statistics and in particular both in the book and the movie he talks about how do you assemble a lineup in a baseball team you add up their on base averages that's a simple Uris yeah is that the right thing to do obviously not but it's a rule that get you a decision you get a team out on the field which has maximized the sum of their on base averages you got a team out there and you know you hope it's a good euristic so what I'm doing this is
44:58
in a a project that I'm doing with a former student now at bone named Alejandro frantic some of you may know Alejandro what we're doing is we are looking at different Uris sixs for solving this problem over here and seeing how well they perform based on simulations now we have no idea how they perform relative to the optimal Uris because or the optimal strategy because nobody knows what the optimal strategy is but at least we can say this Uris is better than that and to give you an
45:33
example uh a simple euristic which you can find in textbooks on uh on managerial accounting is the aristic does the person pay for himself or herself I hire this person this person this person say this is the team that I assemble I let them work together for a few months and then I say okay how much value has this guy generated how much has he cost me in terms of salary has he paid for himself if he does I'll keep him if he hasn't I fire him okay there's a ristic terrible
46:11
I mean you can quickly figure out even theoretically why but that urtic just it is bad urtic on the other hand there's a closely related euristic called the incremental value euristic which actually does pretty well against all other tics but one there's one euristic that that we find in the paper that does great the only problem is it's a really hard urtic to implement I mean it takes a lot of computer power just to get the urtic to work but so what's the point of
46:37
this what we're trying to do is we're saying look we can't solve this problem but this is an important problem in Dynamic Choice people facing this sort of problem uh you know they're going to use euristic what urtic what can we discover by simulating this situation which eristics seem to do well why do they do well when do they do well okay now uh yes there's been in economics literature uh this is something that keeps coming up like every 10 or 15 years uh there was a period of time in
47:10
the uh late 70s through the late 80s where people were interested in juristic for learning what information can be gained from market prices and I've got one paper there with uh Margaret Bray although uh Tom Sergeant who used to be at Hoover and the economics department has now gone to NYU uh Tom is the real hero of this particular literature in particular if you have a taste for books in economics Tom has This brilliant book called uh the
47:42
conquest of American
47:49
inflation which describes what happened in the end of the Carter Administration in the early days of the Reagan why is it that all of a sudden we got rid of stagflation we got rid of the inflation part of stagflation his story is basically that the way economists thought about modeling the economy was with a euristic called ordinar Le squared regression so they were regressing they didn't knew the model that they were using the statistical model they were using was not right but
48:18
you know it was close and so we'll pretend like it's right we'll do what it says and uh basically what he said is when you run ordinarily squar regression on the data they were looking at they got lucky right the policy prescriptions that came out which just you know by luck they got a series of numbers that told them to do what was right they never would have figured it out on their own very nice book uh then you know a few years later there was a literature on how to play games against random
48:47
opponents and this is something i' you know I wrote a bunch of papers with a guy named Drew fudenberg on and now as I said I'm looking at precisely this kind of problem precisely this problem over here with uh Alejandro okay so I got five minutes left and I was hoping to leave a little bit of time for questions but here's my bottom line Dynamic decisions made by economic actors they're the most important decisions economic decisions that people make and they're the ones
49:11
that we really want to understand the best unfortunately the way that economists go about trying to understand those decisions they use the saso model they use it over and over again and it's inadequate even as an as if model model it just doesn't really capture what goes on uh you know this has been a 30 40e Battle mainstream economists resist moving away from saso because they they think well you know you're not being rational anymore you know R we know what rationality is
49:44
and this is saying people behave irrationally and if we do that you know we'll become sociologist and we'll get paid like sociologists don't want to do that at all uh but you know the business of Economics that should say the business of Economics ought to be in modeling how real people make economic decisions and not how some ideal PE people from another planet do so and if that gets us sociologist salaries that's too bad so uh you know I've spent 40 years nibbling around the
50:13
edges of dynamic choice I hope the discipline will begin to take some bigger bites out of this subject because it's really really important and they don't do a good job of it at all
Here to help you learn
Sign in to keep a private conversation with your course tutor.
Ask me anything about the material…
Sign in to chat