Anda di halaman 1dari 7

Preference Relations and Revealed Preference

BINARY RELATIONS
For a given set X ,let X x X denote the usud Cartesian p d a c t of all ordered pairs ( z ,y ) , where both z and y are from X . A binary relation B on the set X is formdly defined as a aubat of X x X - write B E X x X,and ( ty) , E B if the o r d d pair ( z ,y ) is in the relation B . Another, quicker way to m-te (z,y) E B is zBy ,which can be read a s ' z Bees y " or " z stands in the relation B to y." If (z,y) 4 B , I'll write "not zByn or tBy.

(8) Let X = {1,2,3) and B = {(1,1), (1921,(1,319 (2,319 (3,111. (b) Let X = all people in the world and I d B be the relation "&am at least one given name with.* (c) Let X = R (the red line, ranember) and let B be the nldian "greater or equal ton; that is, B =?. (d) Let X = R and let B be the relation: zBy if j z yl> 1. (e) L e t X = R and let B be the relation zBy if t y is an integer multiple of 2.

There is a long list of properties that a given b i i d a t h might or might not have. The properties that will be importsat in this book are the following. A binary relation B on a set X is:

if z ~ f z ord z EX; symmetric if zBy implies yBz; asymmetric if zBy implies y j z ; antisymmetric if zBy and yBz imply z = y ; tmnsitiue i f zBy and yBa imply zBz;

rt@xitx if zBz for all z E X;

Refnence Rehations and Rrornlai P @ m a

orns are never exclusive in this book unless s p d c d y mentioned); ureaklyumncctedifforall s,y E X , z = y or zBy or yBz; acyclic if zl Bz2, zzBzs,. z,-~Bz, imply zl # z, .

ncgaiidy imnsitm if zBY and yBz imply rib; complrtc or canccted if for all z, y E X, zBy or yBz (or both;

..

Example (a) (above) is weakly connected, but nothing else. Exrunple (b) is reflexive and symmetric. Example (c) is d e x i v e , antisymmetric, transitive, negatively transitive, complete, and w d y connected. Example (d) is i d e x i v e , symmetric. Example (e) is reflexive, symmetric, transitive.

[{not(zBy or yBz for all y E X)) implies {not zBz)J which is [{there exists y E Z with ~8~ and y ~ t implies ) {zi)z)], whicb is negative transitivity.
Now bad; to F Is negative transitivity reasonable? b it r e d t o say that if z is strictly preferred to z, then for dl y either t irr h'ctly preferred to y or y is strictly preferred to z? As anormati* propaty, I think it is (barely) masonable. But as a descriptive property, 1 don't think it is reasonable. Suppose X = (0,oo) x (0, oo), whac t = (zl, t 2 ) E X is interpreted as the commodity b d e z l bottles of beer and 2 2 bottles of wine. Tot- (if his tastes are like my own) would certainly say that (10,lO) + (9,9). But -dm (15,6). Totrep might not be willing to say either that (10,lO) + (15,6) ap (15,6) + (9,9) he might plead that comparisons d l e d for an too difficult for him to make. Despite these difficulties with negative transitivity, it is standard to proceed assuming that + is asymmetric and negatively transitive.

Pnwf. (Very pedantic.) The statement [M implies N] is the mne aa the statement [N or not M], thus [M implies N] is the uane as [not N implies not M j . (The second equivalence is called contrapmition.) Thus [(zBz) implies (+By or yBz for aII y E X)] is the same as

PREFERENCE RELATIONS
In this section, we take up the following simple story. There is a set of items X , and Totrep ia willing to express his preferences among these items by making paired comparisons of the fonn: "I strictly prefer z to yn which is written z >- y. "Strict preferencen is a binary relation on X . Consider the following properties that this binary relation might possess:
(a) Asymndty if z is strictly preferred to y, then y is not strictly preferred to z. (What do you think of this? Reasonable normatively? HOWabout descriptively? Think of these questions for each of the following.) (b) ~mifmrfmrfy if z is strictly p r e f d to y and y is strictly preferred to z , thexi z is strictly preferred to z .
(c) r+ti&

Definition (2.2). A binary relation + on a set X is called a pnfnnrce relafion if it is asymmetric and negatively transitive.
Proposifia (23). If + is a preference relation, then >- is idexive, transitive, md acyclic.

- no z

i s strictly prefenwi to itself.

(d) Nqatioe ttnnsitivity if z is not strictly preferred to y and y is not strictly preferred to z, then z is not strictly preferred to z. Negative transitivity ia a hard property to deal with intuitively o let me develop an dtemative statement that is in the form given, s completely equivalent.

(2.1)), z + y implies that either z >- y or t >- z . But z y is impossible because y F z is assumed and + is asymmetric Thus z z, which is transitivity; (c) If 21 >- t 2 , z ~ + 25,. ,zn-1 + 2,. then by t-tivity + x,. Since >- is imflexive, this implies z l # z,. Thus is acyclic.

Proof. (a) Asymmetry directly implies irdexivity. (b) Suppose z >- y and y + z. By negative transitivity (and L a ~ l s

..

When we are given a binary relation that arprescs strict pnterence, we use it to define two other binary relations:

only if) zBz implies that, for all y E X , zBy or yBz.

(2.1). A b

i relation B is negstively transitive iff (if and

Prtfemuc Relations and Retmlui Pqkencc


where we are using as shorthsnd for >= or for %ot + The relation is d e d racllk pnfmne~, although i t redly expresses the absence of is called intiifereme it exprstrict ptefetence The relation the absence of strict pteference in either direction, which is perhaps not quite the same thing as active indifference.

1 1

." -

2 . This is a plot that is followed in many developments of &dm theory, a d in the standard treatment it leads to the same mathematical results:

W f M l (2.4). If + is a preference relation, then: (a) For d 2 and y, exactly one of z y,y z or z (b) 2 is complete and t d t i v e . (c) i s reflexive, symmetric, and transitive. (d) w + z , z - y , y + z i m p l y w + y and r + z . (e) 2 ) - y i f T z + y o r z - y . (f) z k y and y t z i m p l y z - y .

+
y holds.

Proposition 05). Given a binary relation 2 ' on a set X , d& new binary relations hJ and -' from 2' by

taro

Then if k' is complete and transitive, +' will be a preference relation. Moreover, if we start with a binary relation k ' , define and -' aa above from k',and then define k and from +I by

+'

Proof. (a) foII0ws h m the definition of and the fact that + is asymmetric. (b) By the asymmetry of +, either z 3 y or y 3 z (or both) for dl z and y, thus 2 is complete. For transitivity of t,note that this follows immediately from the negative transitivity of (c) N i s reflexive because + is irreilexive. is symmetric because the is symmetric. For transitivity, suppafe t y z . definition of Then z y 3 z and z 3 y 3 t. By negative transitivity of +, z ) ( z + z , o r Z-z. (d) If w z y, then by part (a) one of w + y or y w or y + w . But y >- w is impossible, since then trsnsitivity of + would imply y + 2. And y w is impossible, since then transitivity of would imply z w, cantradicting w + t. Thus w + y must hold. The other part is pimilarIy done. (e) z 2 y iff y 3 z iff z y or z gt (by part (a)). (f) This is'immediate from the definitions of & and

then 2' and

will agree, as will

and

-.

- -

+ -.

+.

--

T h e proof is I& as an exercise. So it doesn't matter whetha we ntart with a strict preference relation that is asymmetxic and negatively transitive or with a weak preference relation that is complete and transitive - we end up in the same place. For reawns of intapretation I prefer to take strict preference as being basic But it is a matter'of personal taste, and most authors do it the other way.

REVEALED PREFERENCE THEORY

Note well the plot: Totrep expresses strict preferences, from which we define weak preferences and indifference. It is strict preference that is basic Totrep is not being called upon to express any judgments concerning weak preference or indifference, and he might disagree with our use of those terms to describe the negation of strict prdennce. Another possible plot would be to ask Totrep to e x p m weak preferencar or preferepce or indifference. That is, the basic relation is

In the previous section, t6e story was that Totrep waa m a U q paired comparisons between items in X . But especially ftom a dt scriptive point of view, we would like to start with an even more basic arpnsssd. c~ concept that of choices made rather than p r e f ~ ~ l l c That is, from a descn'ptin point of view what we stx is an individual's choice behavior - we have to connect that behavior as best we can with his preferences which are neva directly e x p d . The individual's choice behavior reveals his preferences, hence the name of this subject: revealed preference thwry. This subject also hs, som normative justifications- taking preferences as given, how should choices be made? But this subject is of greatest interest from the d d p t i v e viewpoint.

To keep matters simple, throughout this section I'll assume that the choice set X is finite. Eepeddy if the application you are thinking of is demand for consumption bundles or for any item that is infinitely divisible, this is not a very nice simplification. For nonempty subsets of X , I'll use notation such as A, B, etc. The set of all nonempty subsets of X will be denoted P(X).

DeFnition (2.6). A choice fundia for a (finite) set X is a function c :P(X) + P ( X ) such that for all A E X , c(A) E A.
The interpretation is: If Totrep is offered his choice of anything in the set A, he says that any member of c(A) will do just he. If Totrep's preferenas are given by the b i i relation + (and by the corresponding >- and -), it is natural to suppose that he &cms according to the d e that from a set A, anything that is undominufad will be okay. In symbols, define a function c(-, +) :P(X) by

is nonempty. Suppose it was empty then for esch z E A then exids a y E A such that y z. Pick zl E A (A is nonempty), and let z l be zl 's "y Let z3 be zl's y ", and so on. In 0th- words, z l ,23, zj, is a sequence of elemeats of A where ...2 , 2 , 1 z1 +z1. Because A is a finite set, there must exist some rn and n such that z, = z , and m > n. But thie muld be a cyde, and + ia assum& to be acyclic. The neassary contradiction is establishad.

...

".

+ ...+

Note the following instant oxolhy: If is a prdercn~b Athen <-,+) is a choice function. Also, we can strengthen (27) M folhs.

i f T

Propwitia (2.8). For a binary &ion

+ is acyclic.

+, c(., >-)is a choice function

It is dear that for any + , c(A, +) E A, but it isn't dear whether c(A, +) # 0. Thus it isn't clear that c(-, +) is a choice function. That will be something to be investigated T h e o t h a questions to be looked at are: (a) Fiom the normative point of view: Ginn a relation >- (not necessarily a preference relation), when is o(., +) a choice function? If + is a preference relation, what propeaties does c(-, +) have? (b) Rom the descriptive point o f view: Givm a choice function c, when is there, a b i i relation + such that c(-) = c(-, +)? W h e n is this b i i relation a pnference nlation? (NB., this last question is the &tit$ one, aa we're going to be building models where individuals are ausumed to be maximizing their prdermces according to some preference relation.)

Proving this is left ss an exercise. Next we survey some properties of choice functions. Th dsb sic axiomatic property of choice is Houthakker's &om of naacd pnfereLlce.

Houfhakkds &om (29). If z and y are both in A and B and if z E c(A) and y E d B ) , then z E c(B).

In words, if z is sometimes chosen (&om A) when y is availab15 then whenever y is chosen and z is available, z is a l s o chosen. Houthskker's axiom is broken into two pieces by Sen:

Sen3 prop@

a (2.10). If z E B G A and t E c(A) ,then z E c(B).

Sen's paraphrase of this is: If the world champion in some game C a Pakistani, then he must slso be the champion of P a k i s t a n .

PIopwitia (2.7). choice function.

It a binary relation

+ is acyclic, then c ( . , +)

is a

z E c(B).

Sen's propnty @ (23U.If z, y E c(A), A G B and

E c(B), then

Proof. We need to show that for A E P(X), the set

Sen's paraphrase: If the world champion in some game is s Pakidmi, then all champions (in this game) of Pakistan are also world champi-

ons.

F
D-

- 8 .

. ,

R Y

VJ' u *
rn-

CL

5'

90
Y
rn

9
V

ix 8

Pnfcnurc Relntions and Rmtnlcd Pnfamu


A binary nIntion + is acydic if and only if thncdtsachuicefundion c=c(.,+) for which implies that, but is not implied by e satisfics Sm's a, and none of the previous three imply or are implied by c safisfit5 Sm's p.

17

(3) Prove Proposition (2.5).

(4) Show that the properties (for a b i i relation) of asymmetry a d negative transitivity are independent.

(5) Prow Proposition (2.8): For a binary relation (on a finite set X), c(., +) is a choice function if and only if is acyclic. In what sense is it important here that X is finite?

This is just an introduction to revealed p r e f m e e theory there is a very large literature on the subject. Two important questions, both related to the applications of these ideas to classical demand theory in microemnomica, that we haven't discussed are: What if X is an infinite e t ? (Where did wt use the finiteness of X in the development above?) For one approach to an answer to this question, see problem 8 of the next chapter. Suppose we can't observe Totrep's choices from d subsets of X That is, suppcwe c is defined only for a subset of P ( X ) . You can s e t ? how this would provide problems, especially if sets of the form (2, y) are not in the domain of c. And you can see why this is a natural question What

(6) Prove that for any binary relation +, 4,+) dc&d as in the display on page 12 satisfies Sen's a. (Is it important here that X is finite? In what sense is it important that + is acyclic, even though you weren't told to assuxne acycliaty?)

(7) Give an example of a finiteset X and an acyclic binary nlation where c(-, +) does not satisfy Sen's property 8.

can be said in such c a w ?

- espccidly if we have in mind a descriptive theory of choice.

(8) In and around Propositions (2.13) and (2.14), I seem to get very confused about whetha negative transitivity is dl I need to prow Houthskker's axiom, and thus asymmetry. Unw& me. Have I implicitly used asymmetry in the proof of (2.13). and if so, when? Deal with the proof. I know (and so do you, if you did problem (4)) that negative transitivity dasn't imply asymmetry, and I don't want an example of that - I want to know what is going on in the pmof.

PROBLEMS
(1) For eseh of the five examplea on page 7, show that the b ' i relation has precisely the list of properties that are &bed to it on page 8 (froal the l i s t of properties on pages 7 and 8).
(2) A binary &tion E that b refiexive, symmetric and transitive is called an q u i d relation. (For example, if >- is a preference relation, then ia an equivalence relation; d Pmposition (2.4)(c).) H e n is an essy proof that if E is symmetric and transitive, then it is automaticdly d e x i v e (thus reflexive could be deleted from the list of properties): Fix E X and take some y such that zEy. Then y E t by symmetry, and hence xEz by transitivity. Unfortuuately, this easy proof is &m. Why?

+ such that c(.,

(9) Give an example of a finite set X and a choice function q on P(X) that satisfies Sen's a but such that there is no binary &ion +) = c.

Anda mungkin juga menyukai