Abstract
A fundamental property of choice functions is stability, which, loosely speaking, prescribes that choice sets are invariant under adding and removing unchosen alternatives. We provide several structural insights that improve our understanding of stable choice functions. In particular, (1) we show that every stable choice function is generated by a unique simple choice function, which never excludes more than one alternative, (2) we completely characterize which simple choice functions give rise to stable choice functions, and (3) we prove a strong relationship between stability and a new property of tournament solutions called local reversal symmetry. Based on these findings, we provide the first concrete tournament—consisting of 24 alternatives—in which the tournament equilibrium set fails to be stable. Furthermore, we prove that there is no more discriminating stable tournament solution than the bipartisan set and that the bipartisan set is the unique most discriminating tournament solution which satisfies standard properties proposed in the literature.
This is a preview of subscription content, access via your institution.





Notes
We refer to Monjardet (2008) for a more thorough discussion of the origins of this condition.
Together with Theorem 3, Theorem 2 also implies that, for any stable tournament solution S, \(\lceil {S}\rceil \) is a coarsest generator of S. When only considering generators that satisfy \({\widehat{\alpha }}_{_\subseteq }\), \(\lceil {S}\rceil \) is also the coarsest generator of S. In addition, since simple choice functions trivially satisfy \({\widehat{\alpha }}_{_\subseteq }\), the two theorems imply that \(\lceil {S}\rceil \) is the unique simple choice function generating S.
It can be checked that we obtain an equivalent condition even if we require that all outside alternatives have to be removed. When defining local \({\widehat{\alpha }}\) in this way, it can be interpreted as some form of transitivity of stability: Stable sets of minimally stable sets are also stable within the original feasible set (cf. Brandt 2011, Lem. 3).
It is open whether there are tournaments in which \( BP \) and \( TEQ \) are disjoint.
Brandt’s proof relies on a lemma that essentially showed that the generators he considers always satisfy local \({\widehat{\alpha }}\).
The name of this axiom is inspired by a social choice criterion called reversal symmetry. Reversal symmetry prescribes that a uniquely chosen alternative has to be unchosen when the preferences of all voters are reversed (Saari and Barney 2003). A stronger axiom, called ballot reversal symmetry, which demands that the choice set is inverted when all preferences are reversed, was recently introduced by Duddy et al. (2014).
To see that discriminative power is not captured by the axioms, observe that the trivial tournament solution satisfies stability, monotonicity, regularity, and composition-consistency.
However, these analytic results stand in sharp contrast to empirical observations that Condorcet winners are likely to exist in real-world settings, which implies that tournament solutions are much more discriminative than results for the uniform distribution suggest (Brandt and Seedig 2016).
Note that \((T_7)^g\) is the smallest tournament in which \({ BA}\) and \({ UC}\) differ (Brandt et al. 2015).
References
Aizerman, M., Aleskerov, F.: Theory of Choice. Studies in Mathematical and Managerial Economics, vol. 38. North-Holland, Amsterdam (1995)
Allesina, S., Levine, J.M.: A competitive network theory of species diversity. PNAS 108(14), 5638–5642 (2011)
Bordes, G.: Some more results on consistency, rationality and collective choice. In: Laffont, J.J. (ed.) Aggregation and Revelation of Preferences, Chapter 10, pp. 175–197. North-Holland, Amsterdam (1979)
Brandl, F., Brandt, F., Seedig, H.G.: Consistent probabilistic social choice. Econometrica 84(5), 1839–1880 (2016)
Brandt, F.: Minimal stable sets in tournaments. J. Econ. Theory 146(4), 1481–1499 (2011)
Brandt, F.: Set-monotonicity implies Kelly-strategyproofness. Soc. Choice Welf. 45(4), 793–804 (2015)
Brandt, F., Harrenstein, P.: Set-rationalizable choice and self-stability. J. Econ. Theory 146(4), 1721–1731 (2011)
Brandt, F., Seedig, H.G.: On the discriminative power of tournament solutions. In: Selected Papers of the International Conference on Operations Research, OR2014, Operations Research Proceedings, pp. 53–58. Springer (2016)
Brandt, F., Fischer, F., Harrenstein, P., Mair, M.: A computational analysis of the tournament equilibrium set. Soc. Choice Welf. 34(4), 597–609 (2010)
Brandt, F., Chudnovsky, M., Kim, I., Liu, G., Norin, S., Scott, A., Seymour, P., Thomassé, S.: A counterexample to a conjecture of Schwartz. Soc. Choice Welf. 40(3), 739–743 (2013)
Brandt, F., Brill, M., Fischer, F., Harrenstein, P.: Minimal retentive sets in tournaments. Soc. Choice Welf. 42(3), 551–574 (2014)
Brandt, F., Dau, A., Seedig, H.G.: Bounds on the disparity and separation of tournament solutions. Discrete Appl. Math. 187, 41–49 (2015)
Brandt, F., Brill, M., Harrenstein, P.: Tournament solutions. In: Brandt, F., Conitzer, V., Endriss, U., Lang, J., Procaccia, A.D. (eds.) Handbook of Computational Social Choice, chapter 3. Cambridge University Press, Cambridge (2016a)
Brandt, F., Harrenstein, P., Seedig, H.G.: Minimal extending sets in tournaments. Math. Soc. Sci. (2016b) (forthcoming)
Chernoff, H.: Rational selection of decision functions. Econometrica 22(4), 422–443 (1954)
Duddy, C., Houy, N., Lang, J., Piggins, A., Zwicker, W.S.: Social dichotomy functions. Working paper (2014)
Dutta, B.: Covering sets and a new Condorcet choice correspondence. J. Econ. Theory 44(1), 63–80 (1988)
Felsenthal, D.S., Machover, M.: After two centuries should Condorcet’s voting procedure be implemented? Behav. Sci. 37(4), 250–274 (1992)
Fey, M.: Choosing from a large tournament. Soc. Choice Welf. 31(2), 301–309 (2008)
Fishburn, P.C.: Probabilistic social choice based on simple voting comparisons. Rev. Econ. Stud. 51(4), 683–692 (1984)
Fisher, D.C., Reeves, R.B.: Optimal strategies for random tournament games. Linear Algebra Appl. 217, 83–85 (1995)
Fisher, D.C., Ryan, J.: Tournament games and positive tournaments. J. Graph Theory 19(2), 217–236 (1995)
Houy, N.: Still more on the tournament equilibrium set. Soc. Choice Welf. 32, 93–99 (2009a)
Houy, N.: A few new results on TEQ. Mimeo (2009b)
Hudry, O.: A survey on the complexity of tournament solutions. Math. Soc. Sci. 57(3), 292–303 (2009)
Kreweras, G.: Aggregation of preference orderings. In: Mathematics and Social Sciences I: Proceedings of the seminars of Menthon-Saint-Bernard, France (1–27 July 1960) and of Gösing, Austria (3–27 July 1962), pp. 73–79 (1965)
Laffond, G., Laslier, J.-F., Le Breton, M.: More on the tournament equilibrium set. Math. sci. hum. 31(123), 37–44 (1993a)
Laffond, G., Laslier, J.-F., Le Breton, M.: The bipartisan set of a tournament game. Games Econ. Behav. 5(1), 182–201 (1993b)
Laffond, G., Laslier, J.-F., Le Breton, M.: The Copeland measure of Condorcet choice functions. Discrete Appl. Math. 55(3), 273–279 (1994)
Landau, H.G.: On dominance relations and the structure of animal societies: I. Effect of inherent characteristics. Bull. Math. Biophys. 13(1), 1–19 (1951)
Laslier, J.-F.: Tournament Solutions and Majority Voting. Springer, Berlin (1997)
Masatlioglu, Y., Nakajima, D., Ozbay, E.Y.: Revealed attention. Am. Econ. Rev. 102(5), 2183–2205 (2012)
McGarvey, D.C.: A theorem on the construction of voting paradoxes. Econometrica 21(4), 608–610 (1953)
Mnich, M., Shrestha, Y.R., Yang, Y.: When does Schwartz conjecture hold? In: Proceedings of the 24th International Joint Conference on Artificial Intelligence (IJCAI), pp. 603–609. AAAI Press (2015)
Monjardet, B.: Statement of precedence and a comment on IIA terminology. Games Econ. Behav. 62, 736–738 (2008)
Moser, S.: Majority rule and tournament solutions. In: Heckelman, J.C., Miller, N.R. (eds.) Handbook of Social Choice and Voting, chapter 6, pp. 83–101. Edward Elgar, Cheltenham (2015)
Saari, D.G., Barney, S.: Consequences of reversing preferences. Math. Intell. 25, 17–31 (2003)
Schjelderup-Ebbe, T.: Beiträge zur Sozialpsychologie des Haushuhns. Z. Psychol. 88, 225–252 (1922)
Schwartz, T.: Cyclic tournaments and cooperative majority voting: a solution. Soc. Choice Welf. 7(1), 19–29 (1990)
Scott, A., Fey, M.: The minimal covering set in large tournaments. Soc. Choice Welf. 38(1), 1–9 (2012)
Sen, A.K.: Choice functions and revealed preference. Rev. Econ. Stud. 38(3), 307–317 (1971)
Slater, P.: Inconsistencies in a schedule of paired comparisons. Biometrika 48(3–4), 303–312 (1961)
Yang, Y.: A further step towards an understanding of the tournament equilibrium set. Technical report. arXiv:1611.03991. (2016)
Acknowledgements
This material is based on work supported by Deutsche Forschungsgemeinschaft under Grants BR 2312/7-1 and BR 2312/7-2, by a Feodor Lynen Research Fellowship of the Alexander von Humboldt Foundation, by ERC Starting Grant 639945, by a Stanford Graduate Fellowship, and by the MIT-Germany program. The authors thank Christian Geist for insightful computer experiments and Paul Harrenstein for helpful discussions and preparing Fig. 1.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1: Examples for Remark 2
Brandt and Harrenstein (2011, p. 1729) mention that \({\widehat{\alpha }}\) and \({\widehat{\gamma }}\) are independent from each other in the context of general choice functions. Here, we prove that the same holds even in the context of tournament solutions.
Proposition 1
There exists a tournament solution that satisfies \({{\widehat{\alpha }}}\), but not \({{\widehat{\gamma }}}\).
Proof
Let S be a stable tournament solution. As mentioned in Sect. 3, \(\lceil {S}\rceil \) satisfies \({\widehat{\alpha }}\). However, it is easily seen that, unless S is trivial, \(\lceil {S}\rceil \) violates \({\widehat{\gamma }}\). Hence, the statement follows from the existence of non-trivial stable tournament solutions (such as \( BP \)). \(\square \)
Proposition 2
There exists a tournament solution that satisfies \({{\widehat{\gamma }}}\), but not \({{\widehat{\alpha }}}\).
Proof
Let S be a stable tournament solution. Define the tournament solution \(S'\) such that for each tournament \(T=(A,\succ )\),
It can be shown that \(S'\) satisfies \({{\widehat{\gamma }}}\), but may violate \({{\widehat{\alpha }}}\). For the latter, let \(S= TC \) and consider a transitive tournament \((\{a,b,c\},\succ )\) such that \(a \succ b\), \(b \succ c\), and \(a \succ c\). By definition, \(S'(\{a,b,c\})=\{a\}\), but \(S'(\{a,b\})=\{a,b\}\). \(\square \)
Appendix 2: Examples for Remark 3
\(\textit{BA}\) satisfies local \({\widehat{\alpha }}\), but \({\widehat{\textit{BA}}}={ ME}\) violates \({\widehat{\alpha }}\) (Brandt et al. 2016b).
Similarly, there exists a tournament solution S for which \({\widehat{S}}\) is well-defined, but \({\widehat{S}}\) is not stable. For a stable tournament solution S, we have by definition that \(S={\widehat{S}}\) and hence that \({\widehat{S}}\) is also stable. The following proposition shows that \({\widehat{\alpha }}\) does not carry over from S to \({\widehat{S}}\) even if S is simple and \({\widehat{S}}\) is well-defined.
Proposition 3
There exists a simple tournament solution S satisfying \({\widehat{\alpha }}\) such that \({\widehat{S}}\) is well-defined but \({\widehat{S}}\) does not satisfy \({\widehat{\alpha }}\).
Proof
Let S be the tournament solution that always chooses all alternatives, with two exceptions:
-
If the tournament is of order 2, then S chooses only the Condorcet winner.
-
If the tournament is the tournament \(T_4\) given in Fig. 6, then S chooses alternatives a, b, and c.
Clearly, S is simple and satisfies \({\widehat{\alpha }}\). Since \({\widehat{S}}\) chooses alternatives a, b, and c from \(T_4\), but chooses only the Condorcet winner from the transitive tournament of order 3, it does not satisfy \({\widehat{\alpha }}\).
It remains to show that \({\widehat{S}}\) is well-defined. One can check that every tournament contains an S-stable set. Suppose for contradiction that some tournament T contains two distinct minimal S-stable sets, which we denote by B and C. Then B and C are also S-stable in \(B\cup C\). If B is a singleton, then B is the Condorcet winner in \(B\cup C\), which means C cannot be S-stable, a contradiction. Hence both B and C are transitive tournaments of order 3, and \(4\le |B\cup C|\le 6\). One can check all the possibilities of \(B\cup C\) to conclude that this case is also impossible. \(\square \)
For the tournament solution S defined in the proof of Proposition 3, we have that \(\widehat{{\widehat{S}}}\) is not well-defined. Even though no tournament contains two distinct minimal \({\widehat{S}}\)-stable sets, \(T_4\) does not contain any \({\widehat{S}}\)-stable set. This example also shows that for a tournament solution \(S'\), \(\widehat{S'}\) may fail to be well-defined not because it allows two distinct minimal \(S'\)-stable sets in a tournament but because some tournament contains no \(S'\)-stable set.
Appendix 3: Example for Remark 6
We show that there exists a tournament solution different from \( BP \) that satisfies \( LRS \), monotonicity, regularity, and Condorcet consistency.
To this end, we define a new tournament solution called \( POS \) which chooses all alternatives with positive relative degree. More precisely, an alternative is chosen by \( POS \) if it dominates strictly more than half of the remaining alternatives, is not chosen if it dominates strictly less than half of the remaining alternatives, and goes to a “tie-break” to determine whether it is chosen if it dominates exactly half of the remaining alternatives.
For tournaments of even size, \( POS \) chooses exactly the alternatives that dominate at least (or equivalently, more than) half of the remaining alternatives. Hence we do not need a tie-break for tournaments of even size. The tie-breaking rule for tournaments of odd size \(2n+1\) is as follows: For any (unlabeled) tournament T of order 2n and any partition of it into two sets B and C of size n, consider two tournaments \(T_1\) and \(T_2\) of order \(2n+1\). The tournament \(T_1\) contains T and another alternative a that dominates B but is dominated by C, while the tournament \(T_2\) contains T and another alternative a that dominates C but is dominated by B. If \(T_1\) or \(T_2\) is regular, \( POS \) chooses a in that tournament and not in the other one. Otherwise, \( POS \) arbitrarily chooses a in exactly one of \(T_1\) and \(T_2\).
Proposition 4
\( POS \) satisfies \( LRS \), monotonicity, and regularity.
Proof
We need to show that the tie-breaking rule in the definition of \( POS \) is well-defined. First, we show that if we perform a local reversal on alternative a, we do not get an isomorphic tournament with alternative a mapped to itself. Indeed, if a were mapped to itself, it would mean that no tournament solution satisfies \( LRS \), which we know is not true since \( BP \) satisfies \( LRS \). Secondly, the tournament obtained by performing a local reversal on an alternative in a regular tournament is not regular. Hence we do not obtain a conflict within the tie-breaking rule.
It follows directly from the definition that \( POS \) satisfies \( LRS \), monotonicity, and regularity. \(\square \)
The tournament \(T_4\) given in Fig. 6 shows that \( POS \) violates composition-consistency and \({\widehat{\alpha }}\) (and hence stability).
Interestingly, \( BP \) (and all of its coarsenings) always intersect with \( POS \) while there exists a tournament for which \( BA \) (and all of its refinements such as \( TEQ \) and \( ME \)) do not overlap with \( POS \). This follows from results on the Copeland value by Laffond et al. (1993b, 1994).
Appendix 4: Example for Remark 7
We construct a tournament solution that satisfies monotonicity and stability, but violates regularity and composition-consistency.
Every tournament solution has to be regular on tournaments of order 5 or less because of non-trivial automorphisms. Consider the tournament \(T_7\) shown in Fig. 7. \(T_7\) admits a unique non-trivial automorphism that maps each of the six alternatives in the two 3-cycles to the next alternative in its 3-cycle and maps alternative g to itself.Footnote 14
Now, define the simple tournament solution \(S_7\), which always returns all alternatives unless the tournament is \(T_7\) or it can be modified from \(T_7\) by weakening alternative g. In the latter case, \(S_7\) returns all alternatives except g.
We check that this definition is sound. First, we know that in \(T_7\), there is no automorphism that maps alternative g to another alternative. When we weaken g, it is the unique alternative with the smallest out-degree and hence cannot be mapped by an automorphism to another alternative. Now, the alternatives a, b, c form an orbit, and \(S_7\) excludes g whenever it is dominated by d, e, f (and has any dominance relationship to a, b, c). This yields four non-isomorphic tournaments for which \(S_7\) excludes g.
Proposition 5
\(\widehat{S_7}\) satisfies stability and monotonicity.
Proof
First, observe that \(S_7\) trivially satisfies local \({\widehat{\alpha }}\) because \(S_7\) only excludes an alternative in tournaments of order 7. By virtue of Theorem 3, it therefore suffices to show that \(\widehat{S_7}\) is well-defined.
One can check that every tournament contains an \(S_7\)-stable set. Let \(T_6\) denote the tournament obtained by removing alternative g from \(T_7\). Suppose for contradiction that there exists a tournament T that contains two distinct minimal \(S_7\)-stable sets, which we denote by B and C. Then B and C are also \(S_7\)-stable in \(B\cup C\). Moreover, \(T|_{B}\) must correspond to the tournament \(T_6\), and each alternative in \(C\backslash B\) either has the same dominance relation to B as the alternative g does to \(T_6\) or has a dominance relation that is a weakening of g. The same statement holds for C. We consider the following cases.
Case 1: \(10\le |B\cup C|\le 11\). The tournament \(T|_{B}\) has one of its alternatives corresponding to alternatives d, e, and f in Fig. 7 outside of \(B\cap C\), and this alternative must dominate all of the alternatives in C. Similarly, there exists an alternative in \(C\backslash B\) that dominates all of the alternatives in B. But this implies that some two alternatives dominate each other, a contradiction.
Case 2: \(7\le |B\cup C|\le 9\). At least one of the two tournaments \(T|_{B}\) and \(T|_{C}\) must have all of its alternatives corresponding to alternatives d, e, and f in Fig. 7 in the intersection \(B\cap C\), for otherwise we obtain a contradiction in the same way as in Case 1. Assume without loss of generality that \(T|_{B}\) has its alternatives corresponding to d, e, and f in the intersection. Hence three alternatives in \(B\cap C\) that form a cycle dominate the same alternative in C. But this does not occur in \(T_6\), a contradiction.
It follows from Theorem 4 that \(\widehat{S_7}\) satisfies monotonicity. \(\square \)
Clearly, \(\widehat{S_7}\) is not regular since it excludes an alternative from the regular tournament \(T_7\). Moreover, it is not a coarsening of \( BP \) since \( BP \) selects all of the alternatives in \(T_7\). Hence we have that stable and monotonic tournament solutions are not necessarily coarsenings of \( BP \).
Rights and permissions
About this article
Cite this article
Brandt, F., Brill, M., Seedig, H.G. et al. On the structure of stable tournament solutions. Econ Theory 65, 483–507 (2018). https://doi.org/10.1007/s00199-016-1024-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00199-016-1024-x