Home Final Topology for Preference Spaces
Article
Licensed
Unlicensed Requires Authentication

Final Topology for Preference Spaces

  • Pablo Schenone ORCID logo EMAIL logo
Published/Copyright: February 5, 2025

Abstract

We say a model is continuous in utilities (resp., preferences) if small perturbations of utility functions (resp., preferences) generate small changes in the model’s outcomes. While similar, these two questions are different. They are only equivalent when the following two sets are isomorphic: the set of continuous mappings from preferences to the model’s outcomes, and the set of continuous mappings from utilities to the model’s outcomes. In this paper, we study the topology for preference spaces defined by such an isomorphism. This study is practically significant, as continuity analysis is predominantly conducted through utility functions, rather than the underlying preference space. Our findings enable researchers to extrapolate continuity in utility as indicative of continuity in underlying preferences.

JEL Classification: B4; D

Corresponding author: Pablo Schenone, Fordham University, Rose Hill Campus, 441 E. Fordham Road, Bronx, NY 10458, USA, E-mail: 

Acknowledgments

I wish to thank Kim Border, Laura Doval, Federico Echenique, Mallesh Pai, Omer Tamuz, and participants of the LA theory fest for insightful discussions that helped shape this paper. All remaining errors are, of course, my own.

Appendix A: Proofs

We start with a well known result. Under the final topology, U is open in P if and only if F −1(U) is open in U . Consequently, F is continuous. Proofs are included for completeness.

Lemma 1.

U is open in P if and only if F −1(U) is open in U . As a consequence, F is continuous.

Proof.

(⇐) Let U P be open. Let g : P P be the identity function, so g is continuous by definition. By definition, continuity of g implies continuity of (Fg). Because (Fg) = F, then F is continuous. Thus, F −1(U) is open.

(⇒) Let U P and assume F −1(U) is open. Let Z = {0, 1} and let τ Z = {Z, ∅, {1}} be the open sets in Z. Define g : P Z as g(≻) = 1 if and only if ≻ ∈ U. Then Fg is continuous, so that g must be continuous. Therefore, U = g −1(1) is open.□

We now start characterizing the open sets in P . We introduce two pieces of notation. First, for any preference, ̂ , and any set, SX, we let ̂ | S denote the restriction of ̂ to S. Formally, ̂ | S = { ( x , y ) S × S : x ̂ y } . Second, let P and let AX such that for each pair x, yA either xy or yx. Then, we let B ( , A ) = { ̂ : ̂ | A = | A } .

Lemma 2.

Let P and let x, yX be such that xy. Then, the set B(≻, {x, y}) is open in P .

Proof.

Pick any P and x, yA as in the statement of the lemma. It suffices to show that F −1(B(≻, {x, y})) is open in U . Notice that F 1 ( B ( , { x , y } ) ) = r R A ( r ) where A ( r ) = { u U : u ( x ) > r > u ( y ) } . Because A(r) is open in U for each r R , then r R A ( r ) is open in U . Indeed, A(r) ≡ Π zX U z where U z = R when z ∉ {x, y}, U x = (r, ∞) and U y = (−∞, r), which is open in the topology of pointwise convergence. Therefore, F −1 B(≻, {x, y}) is open in U and so B(≻, {x, y}) is open in P .□

We now use Lemma 2 to show that F is an open map. To do this, we show F maps the basis of U into open sets in P .

Lemma 3.

Let U be an element in the basis of U . Then, F(U) is open in P .

Proof.

Let U be an element in the basis of U . Then, there exists a finite set AX such that U = Π xX I(x) where, for each xX, I ( x ) R is open, I ( x ) = R if xA, and I ( x ) R if xA. Furthermore, without loss of generality, I(x) is an interval for each xA. Therefore, for each xA there are numbers a x , b x R such that I x = (a x , b x ).

Case 1: xA I(x) ≠ ∅. We will show that F ( U ) = P , which is open and hence completes the proof. Take any P . Let ( a , b ) R be such that (a, b) ⊂ ∩ xA I(x). Such an interval exists because ∩ xA I(x) is non empty and open. Let uF −1(≻). Then there exists a scaling f : R R such that f(u)(x) ∈ (a, b) for all xA and F(f(u)) = F(u) = ≻. Because f(u) ∈ U by construction and F(f(u)) = ≻ then ≻ ∈ F(U). Because P was arbitrarily chosen this shows P F ( U ) , and this completes the proof.

Case 2: xA I(x) = ∅. Because ∩ xA I(x) = ∅ there exists x 0, y 0A such that I(x 0) ∩ I(y 0) = ∅. Let A = { { x , y } : I ( x ) I ( y ) = } . Because { x 0 , y 0 } A then A . Thus, ( u , v U ) ( { x , y } A ) , u(x) ≠ u(y), v(x) ≠ v(y) and u(x) > u(y) ⇔ v(x) > v(y). This follows from u(x) ∈ I(x), u(y) ∈ I(y) and I(x) ∩ I(y) = ∅. Let P be any preferences such that, for each { x , y } A , xy if and only if I(x) > I(y) in the strong set order.[8] In particular, for all { x , y } A , xy iff u(x) > u(y) for all uF(U). Let B ( , A ) = { x , y } A B ( , { x , y } ) . B ( , A ) is open because each B(⊳, {x, y}) is open and the intersection is finite. We now show that F ( U ) = B ( , A ) which proves that F(U) is open.

F ( U ) B ( , A ) : Let ≻ ∈ F(U). Then there is uU such that F(u) = ≻. Because uU, then F(u)|{x, y} = ⊳|{x, y} for each { x , y } A . Thus, ≻|{x, y} = ⊳|{x, y} and thus B ( , A ) .

B ( , A ) F ( U ) : Pick B ( , A ) . Thus ≻|{x, y} = ⊳|{x, y} for each pair { x , y } A . Thus, for any { x , y } A and any uF −1(≻), u(x) > u(y) ⇔ xy. Up to a rescaling, there exists vF −1(≻) such that v(x) ∈ I(x) for each xA. Thus, there exists vUF −1(≻) so that ≻ = F(v) ∈ F(U).□

As a consequence of the previous Lemma, F is an open map.[9] Consequently, if D is a basis for the topology of U then F ( D ) is a basis for the final topology on P . The previous lemma effectively shows F ( D ) = { B ( , A ) : P , A α ( ) } . This last observation proves Theorem 1. For completeness, we include an explicit proof of Theorem 1 below.

Part of Lemma 3’s proof involves showing that although sets of the form B(≻, A) generate the basis of T P , they do not form a basis. Example 6 below shows this in a simple example.

Example 6.

Suppose X = {1, 2, 3}. The following set is an element of the basis of R X : U = I 1 × I 2 × I 3 where I 1 = (0, 1), I 2 = (2, 5), I 3 = (2, 5). Pick any uU. Then u represents a preference such that 2 F(u) 1 and 3 F(u) 1. However, we could have 3 F(u) 2, 2 F(u) 3, or neither. Thus, pushing forward U according to F recover all preferences ≻ such that 2 ≻ 1 and 3 ≻ 1 but we cannot conclude anything about the ranking of 2 and 3. Hence F(U) = B(F(u), {1, 2}) ∩ B(F(u), {1, 3}) where uU. However, B(F(u), {1, 2, 3}) = B(F(u), {1, 2})∩B(F(u), {1, 3})∩B(F(u), {2, 3}) ≠ B(F(u), {1, 2})∩B(F(u), {1, 3}) = F(U) because U could include functions v such that u and v disagree on how to rank elements 2 and 3. For instance u(1) = 0.5, u(2) = 3, u(3) = 4, and v(1) = 0.5, v(2) = 4, v(3) = 3.

A.1 Proof of Theorem 1

Proof.

Let D be the basis for U generated by products of open intervals. That is, each D D is of the form D = Π x I x where each I x is an open interval and I x R for a finite subset of elements xX. We show B = F ( D ) . Because F is open and D is a basis for U , then B is a basis for P .

B F ( D ) : Let B ( , A ) B , where P and A α ( ) . Let u be any representation of ≻. Let A A , so that A is ≻-strictly ranked. Let ɛ = min{|u(x) − u(y)|: x, yA, xy} > 0, where positivity is guaranteed because A is finite and ≻-strictly ranked. For each xA, let I x = ( u ( x ) ε 2 , u ( x ) + ε 2 ) R . For each xA, let I x = R . Notice that I x I y = ∅ for all x, yA, xy. Let D = Π x I x D . Then, B ( , A ) = F ( D ) F ( D ) for each A A . Therefore B ( , A ) = A A B ( , A ) F ( D ) so B F ( D ) .[10]

B F ( D ) : Pick any F(D) where D D . Let A = { x : I x R } , which is finite by definition. First, assume ∩ xA I x ≠ ∅. Then, ∩ xA I x = (a, b) for some a , b R . Thus, F ( D ) = P B . Second, assume ∩ xA I x = ∅. Then there exists A 0A such that x A 0 I x = . Let A = { A 0 A : x A 0 I x = } Therefore, if v, uD and x, yA 0 then u(x) ≠ u(y), v(x) ≠ v(y) and u(x) > u(y) ⇔ v(x) > v(y). That is, any utility function in D ranks the alternatives in each A 0 A the same way and no two alternatives receive equal utility. Then F ( D ) = A 0 A B ( , A 0 ) for all ≻ ∈ F(D). Therefore, F ( D ) B for all D D and so F ( D ) B .□

A.2 Proof of Theorem 2 and Proposition 1

Theorem 2.

The space P s is Hausdorff and path-disconnected.

Proof.

We first show P s is Hausdorff. Let 0 , 1 P s such that ≻1 ≠ ≻0. Then there exists x, yX such that x0 y and y1 x. Consider the sets B 0 = B(≻0, {x, y}) and B 1 = B(≻1, {x, y}). By Theorem 1, B 0 and B 1 are neighbordoods of ≻0 and ≻1 respectively. Furthermore, B 0B 1 = ∅. This shows P s is Hausdorff.

We now argue that P s is totally path-disconnected. It suffices to show that U s is totally path-disconnected. Take u , v U s to be distinct functions. Assume there is a continuous path t joining u and v. That is, t : [ 0,1 ] U s with t(0) = u, t(1) = v and t continuous. Since uv and u , v U s , there are points x, yX such that u(x) > u(y) and v(y) > v(x). Let Δ(s) = t(s)(x) − t(s)(y) for s ∈ [0, 1]. Then, Δ(0) > 0 > Δ(1). Thus, there is a point s* such that Δ(s*) = 0. Then, t(s*)(x) = t(s*)(y), contradicting that t ( s * ) U s . Thus, no two distinct functions are connected via a continuous path.□

Proposition 1.

Let P 0 be Hausdorff and P s P 0 . Then, P 0 = P s .

Proof.

We proceed in two steps. Recall that ≃ denotes the total indifference preference.

Step 1: if P 0 then P 0 is not Hausdorff. By Theorem 1, the only open neighborhood of ≃ is P 0 . Therefore, ≃ is topologically indistinguishable from any other preference, and thus P 0 is not Hausdorff.

We may alternatively proceed using limit arguments. Let 0 denote constant function 0. Clearly F(0) = ≃. Let P s be any preference and let u be any representation of ≻. Then u n = 1 n u represents ≻. Since u n 0, F(u n ) = ≻ for all n, F(0) = ≃, then { } ̄ , so P 0 cannot be Hausdorff.

Step 2: if P 0 then P 0 is not Hausdorff. Take P 0 \ P s . For each xX, let < x > be the indifference class of x according to ≻. For each < x > , let f < x > : < x > R be any function such that f(a) ≠ f(b) for all a, b ∈ < x >.[11] Construct 0 P s as follows: for any pair x, yX, if xy then x0 y; if xy then x0 y if and only if f <x>(x) > f <x>(y). Notice that 0 P s and that ≻0|A = ≻|A for any finite set A that is ≻-strictly ranked. Let G be any open set such that ≻ ∈ G. Then there is a finite collection A and a preference 1 P 0 such that B ( 1 , A ) . Without loss of generality B ( 1 , A ) = B ( , A ) .[12] By construction, 0 B ( , A ) . Thus ≻ and ≻0 are topologically indistinguishable, so P 0 is not Hausdorff.□

A.3 Proof of Proposition 2 and Theorem 3

For our next lemma, we start by recalling notation we introduced in the main text. For all xX, < x > represents the indifference class of x. Then, define P c i = : ( x X ) i f A < x > A is not open}.

Lemma 4.

Let P c i . Then, ≻ is locally strict in ( X , T X ( ) ) .

Proof.

Let (x, y) ∈ X × X be such that xy. If xy, then the locally strict property holds vacuously. Assume then that xy. Then, let N be a neighborhood of (x, y). Then, there is an open set G = G 1 × G 2X × X such that (x, y) ∈ GN. Assume that for all (z, z′) ∈ G 1 × G 2 we had zz′. Because xG 1, for any z′ ∈ G 2 we get xz′. Thus, G 2 ⊂ < x >. By analogous reasoning, G 1 ⊂ < y >. Then, G 1 ⊂ < y > and G 2 ⊂ < x >, contradicting that G is open. Hence, there must exist (z, z′) ∈ G such that zz′, completing the proof.□

Proposition 2.

If X is finite, then P c i = . If X is infinite, then P c i is not Hausdorff in the final topology.

Proof.

Part 1: If X is finite, then P c i = .

Let ≻ be a preference that is locally strict. We start with two observations about ≻: first, ≻ ≠ ≃ and, second ≻ must contain an infinite number of distinct indifference classes. Indeed, assume that ≻ only contains finitely many distinct indifference classes (say, N indifference classes). Then we can pick a representative from each indifference class and label them in an increasing way: x 1x 2 ≻ (…) ≻ x N . Then < x 1 > is open as it is the upper contour set of x 2, thus ≻ is not locally strict. We conclude that if ≻ is locally strict, then it has infinitely many (distinct) indifference classes.[13] Consequently, if ≻ is locally strict, then X must have infinitely many members. Alternatively, if X is finite, then P c i = .

Part 2: If X is infinite, then P c i is not Hausdorff in the final topology. We start with an observation about the topology T X ( ) . The basis for such a topology is given by sets of the form I(z, y) = {x: yxz}, where y, zX ∪ {∞, − ∞}, and we use the convention I(−∞, y) ≡ Lower(y) and I(z, ∞) ≡ Upper(z).[14] Furthermore, we say a point x is isolated if the following is true: there are points a and b in X such that (1) axb, (2) I(x, a) = I(b, x) = ∅. Next, we make two observations about locally strict preferences. First, a locally strict preference cannot have isolated points. If x was an isolated point then < x > = I ( a , b ) for the points a and b that make x isolated. This is a contradiction because it would imply < x > is open. Second, if ≻ is locally strict, ≻ must have infinitely many distinct indifference classes. This is a consequence of the arguments in Part 1. Consequently, there is a point xX such that x is neither maximal nor minimal (i.e., there exist a, bX such that axb) because otherwise every point would be either maximal or minimal and there would only be two indifference classes, contradicting that ≻ is locally strict. Putting these observations together we conclude that any locally strict preference must have a point that is neither maximal nor minimal, and this point cannot be isolated. We now use these observations to prove that P c i is not Hausdorff. Pick any triplet x, y, zX and choose any preference P c i that satisfies zxy. Such a preference exists because every preference must have an x that is neither maximal nor minimal. Because x cannot be isolated then at least one of the following must hold: either I(x, z′) ≠ ∅ for all z′ ∈ Upper(x), or I(y′, x) ≠ ∅ for all y′ ∈ Lower(x). We will now construct a preference ≻1 such that ≻1 ≠ ≻ and ≻ belongs to every neighborhood of ≻1, thus showing P c i is not Haursdorff. Without loss of generality, assume I(x, z′) ≠ ∅ for all z′ ∈ Upper(x); we can carry out an analogous construction for the other case. For each pair a, bX define ≻1 as follows: if aUpper(x) and bUpper(x) then ≻|{a, b} = ≻1|{a, b}, aUpper(x) and bUpper(x) then a1 b, and if aUpper(x), bUpper(x) then a1 b. We make three observations about ≻1. First, ≻1 ≠ ≻ because x1 y and xy. Second, ≻1 is locally strict. To see this it suffices to show that no subset of the indifference class of x under ≻1 is open in T X ( 1 ) . Let G ⊂ {wX: w1 x} be an open set. Because G is open there must be a non-empty set I(a, b) such that I(a, b) ⊂ G.[15] Because every point in G is minimal by construction, a = −∞. Then, there must be an alternative bX such that I(−∞, b) = Lower(b) ⊂ G. Because I(x, z′) ≠ ∅ for all z′ ∈ Upper(x), such a bX cannot exist: if b ∈ {w: w1 x} then Lower(b) = ∅, and if b1 x then there exists b′ ∈ Lower(b) ∩ Upper(x) so Lower(b) ⊄ G. Thus, G cannot be open. Lastly, ≻1 satisfies the following property: a1 bab for all a, bX. This last observation implies that if N is a neighborhood of ≻1 then ≻ ∈ N. Therefore, P c i is not Hausdorff.□

Theorem 3.

The space P clsg is Hausdorff.

Proof.

Let , ̂ P clsg , ̂ . We will show that there are neighborhoods of ≻ and ̂ that are disjoint. Because ̂ , there must be x, yX such that xy and x ̂ y .

Case 1: y ̂ x . Then B(≻, {x, y}) and B ( ̂ , { x , y } ) are disjoint neighborhoods of ≻ and ̂ respectively, and this concludes the proof.

Case 2: y ̂ x . Because ≻ has no gaps, then {w: xwy} ≠ ∅. Furthermore, {w: xwy} is open in T X by continuity of ≻. First, observe that there must be z ∈ {w: xwy} such that z ̂ x , otherwise {w: xwy} would be an open set such that z ̂ z for all z, z′ ∈ {w: xwy}, contradicting that ̂ is locally strict. Take z ∈ {w: xwy} such that z ̂ x . If x ̂ z then we get y ̂ x ̂ z . Thus, y ̂ z and zy, implying B(≻, {z, y}) and B ( ̂ , { z , y } ) are disjoint neighborhoods of ≻ and ̂ thereby concluding the proof. If z ̂ x then we get z ̂ x and xz, implying B(≻, {z, x}) and B ( ̂ , { z , x } ) are disjoint neighborhoods of ≻ and ̂ thereby concluding the proof.

Therefore, we can always find neighborhoods of ≻ and ̂ that are disjoint, and this concludes the proof.□

References

Border, K. C., and U. Segal. 1994. “Dynamic Consistency Implies Approximately Expected Utility Preferences.” Journal of Economic Theory 63 (2): 170–88. https://doi.org/10.1006/jeth.1994.1038.Search in Google Scholar

Borkovsky, R. N., U. Doraszelski, and Y. Kryukov. 2010. “A User’s Guide to Solving Dynamic Stochastic Games Using the Homotopy Method.” Operations Research 58 (4-Part-2): 1116–32. https://doi.org/10.1287/opre.1100.0843.Search in Google Scholar

Chambers, C. P., F. Echenique, and N. S. Lambert. 2021. “Recovering Preferences from Finite Data.” Econometrica 89 (4): 1633–64. https://doi.org/10.3982/ecta17845.Search in Google Scholar

Debreu, G. 1967. “Neighboring Economic Agents.” In La décision: Agrégation et dynamique des ordres de préférence, Vol. 2, 85. Paris, France: Centre National de la Recherche Scientifique.Search in Google Scholar

Debreu, G. 1954. “Representation of a Preference Ordering by a Numerical Function.” Decision Processes 3: 159–65.Search in Google Scholar

Dubra, J., and F. Echenique. 2000. “A Full Characterization of Representable Preferences.” In Documento de Trabajo/FCS-DE; 12/00.Search in Google Scholar

Eaves, B. C., and K. Schmedders. 1999. “General Equilibrium Models and Homotopy Methods.” Journal of Economic Dynamics and Control 23 (9–10): 1249–79. https://doi.org/10.1016/s0165-1889(98)00073-6.Search in Google Scholar

Grodal, B. 1974. “A Note on the Space of Preference Relations.” Journal of Mathematical Economics 1 (3): 279–94. https://doi.org/10.1016/0304-4068(74)90017-2.Search in Google Scholar

Hildenbrand, W. 1970. “On Economies with Many Agents.” Journal of Economic Theory 2 (2): 161–88. https://doi.org/10.1016/0022-0531(70)90003-7.Search in Google Scholar

Kannai, Y. 1970. “Continuity Properties of the Core of a Market.” Econometrica: Journal of the Econometric Society: 791–815. https://doi.org/10.2307/1909693.Search in Google Scholar

Mas-Colell, A. 1974. “Continuous and Smooth Consumers: Approximation Theorems.” Journal of Economic Theory 8 (3): 305–36. https://doi.org/10.1016/0022-0531(74)90089-1.Search in Google Scholar

Nishimura, H., and E. A. Ok. 2023. “A Class of Dissimilarity Semimetrics for Preference Relations.” Mathematics of Operations Research 2249–70, https://doi.org/10.1287/moor.2022.0351.Search in Google Scholar

Rader, T. 1963. “The Existence of a Utility Function to Represent Preferences.” The Review of Economic Studies 30 (3): 229–32. https://doi.org/10.2307/2296323.Search in Google Scholar

Shiomura, T. 1998. “On the Hicksian Laws of Comparative Statics for the Hicksian Case: The Path-Following Approach Using an Alternative Homotopy.” Computational Economics 12 (1): 25–33 https://doi.org/10.1023/a:1008602109014.10.1023/A:1008602109014Search in Google Scholar

Received: 2024-03-12
Accepted: 2025-01-10
Published Online: 2025-02-05

© 2025 Walter de Gruyter GmbH, Berlin/Boston

Downloaded on 1.12.2025 from https://www.degruyterbrill.com/document/doi/10.1515/bejte-2024-0037/html?lang=en
Scroll to top button