SUMMARY OF STABLE ALLOCATION AND

THE PRACTICE OF MARKET DESIGN

By:
Pradhita Audi
1. Introduction

Economics has been studied on the ways to allocate resources from societies. The fundamental theory to
analyze resource allocation laid in 1962 from Gale and Shapley with a model of two sets of agents. The pairing
of two sets of agents defined to be stable if in the trade happened, there are no unexploited gains. However,
Gale and Shapley (1962) discovered that deferred-acceptance algorithm produces a stable matching. This idea
is supported with theory from Alvin Roth (1944) who founded that US market for new doctors was suffered by
several series of market failures when a centralized cleaning house improved their situation by implementing
a procedure that equals to Gale and Shapley's deferred-acceptance process. There are other theories in several
years ahead that rewarded greatly such as tools from non-cooperative and cooperative game theory. The non-
cooperative theory focused on a strategic problem faced by individual decision-makers. The development from
cooperative game theory tools notably known from Lloyd Shapley. These approaches (cooperative and non-
cooperative) complement and integrated as a whole. The two important factors for market, stability and
incentive compatibility, derive from these approaches. Stability is obtained from cooperative game theory
while incentive compatibility comes from mechanism design, a branch theory of non-cooperative game theory.
Then the scientific background firstly introduced the idea of stability in the perspective of cooperative game
theory .

2. Theory I : Stability

2.1 Coalitional games with Transferable utility

Coalitions are a group of individuals that cooperate with each other. Some coalitions can produce the worth
of economic surplus, which consist of a sum of money, with its own resources. The transferrable utility is a
utility where some members of coalitions split the surplus according to the suitable scheme and each
member’s utility equals to their surplus share. Cooperative game studies the incentives of the individual to
form coalitions, given that any potential conflicts of interest within a coalition. This situation can be solved by
binding agreements that induce the members of coalitions to proceed and action that maximize the surplus
of the coalition (Huang and Sjostrom, 2003, Koczy, 2007). In games of transferable utility, it is assumed that
players can freely transfer utility by making side-payments. However, side-payments most likely are
constrained and the utility is not transferable. Despite this, cooperative game studies that are well-developed
for non-transferable utility games could handle such situations.

2.2 Stability and the core

If the member of some coalition S can use their own resources to make themselves better off, then it is could
be said that this coalition can improve upon or block a certain period. When utility is transferable, coalition S
can be improved upon that period, thus the allocation from this period is unstable. If an allocation cannot be
improved upon by any coalition, that is by using its own resources can bring about an outcome which all
members preferred, it is said to be stable. The idea of stability in cooperative game theory corresponds with
the Nash equilibrium in non-cooperative game theory, which a situation that no individual can contradict and
make themselves better off. When individuals have unlimited ability and time to strike deals with each other
than the outcome must be stable. Otherwise, some coalition would have a motive to form and make its
members better off.

2.3 Do stable allocations always exist?

D.B. Gillies (1953a, 1959) and Shapley (1953, 1955) were the first to consider the core as an independent.
Through the given example, it shows the core (i.e. the set of stable allocations) is sometimes quite large. On

1
the other hand, in other examples, the core could be empty. In general, if there is no sufficient surplus then
it is impossible to have a stable allocation.

2.4 Core and competitive equilibrium

Edgeworth (1881) firstly argued that if some traders are not satisfied with their transactions, they could re-
contract. However, the contract curve is the set of results that could not be destabilized by re-contract. Then,
Edgeworth conjectured that within markets with enough traders, the contract curve would approximately
equal with competitive equilibrium. Debreu and Scarf (1963) verified Edgeworth’s argument under more
general takings: if the economy is replicated so the number of traders becomes greater, then the core
approximately equals the set of competitive equilibria (also see Shubik, 1959). Therefore, without having to
specify the exact rules that govern trade, the core provides a key foundation for competitive equilibrium.

3. Theory II : Matching Markets

Various goods are indivisible and heterogeneous while the market for the goods become very thin. In such
markets, the participants must be properly matched in order to do the transactions.

3.1 Two-sided matching

Gale and Shapley (1962) studied two-sides matching markets which consist of a market with two disjoint agents
(e.g. buyers and sellers) that must be matched to carry out a transaction. Things such as wages are ruled out
and not subject to the negotiation. In general, a stable matching only occurs if there is no coalition can improve
into it. The stable matching must have two conditions : (1) no agents find the matching unacceptable and (2)
no department-student would prefer to be matched rather staying with their current matches. Condition (1) is
individual rationality while condition (2) is pairwise stability. These conditions imply that neither any singleton
coalition can improve the matching. Then, the Gale-Shapley introduced a deferred-acceptance algorithm to
find a stable matching. A critical aspect of this algorithm is that desirable offers are not immediately accepted,
rather held on until deferred acceptance. In this process, each department starts by making its first offer to its
top-ranked applicant. If the offer is rejected, then these departments will make an offer to the applicants that
rank number two, etc. Hence, during the algorithm, the expectation of the department are lowered as it makes
offers to students further down its preference ordering. They proved that since the deferred-acceptance
algorithm is stable, it will always producing a stable matching. The algorithm also proves that if a department
wants to hire more than one intern, it will still generate a stable matching. Gale and Shapley (1962) also
described that interests are polarized in the sense that different stable results favor one or other side of
markets. Example 2 demonstrates how the algorithm benefits the applicant’s interest (they agree is the best),
instead of the employer (they agree is the worst), and how stable institutions can be designed to systematically
favor one side of the market. However, in practice designs tend to favor the applicants. Therefore, the
algorithm operated mostly for the benefits of the participants. Based on previous, the deferred-acceptance
algorithm was explained in decentralized procedures, when in fact, it is run by a clearinghouse in a centralized
fashion. Based on the submitted preferences, the clearinghouse runs a revelation mechanism (a kind of virtual
market that does not suffer from the problems experienced in real-world markets). This mechanism makes a
simultaneous-move game (where all participants submit their preference rankings), giving a full view of how
the algorithm sets the submitted rankings into an allocation. A revelation mechanism is an incentive-
compatible if truth-telling is a dominant strategy, so the participants always find it optimal to submit their true
preference orderings. The employer-proposing algorithm (viewed as a revelation mechanism) is incentive
compatible for the employees (Dubins and Freedman, 1981, Roth 1982a). However, this mechanism is not
incentive compatible for the applicant and proves that truth-telling is not a dominant strategy for the applicants.
Roth (1982) also proved that no stable matching mechanism exists (stating the true preferences) is a dominant

2
strategy for every agent. Despite this, the final matching is stable under the true preferences and undominated
Nash equilibrium outcome. This illustrates a general fact about the Gale-Shapley algorithm and proved by Roth
(1984). However, the usefulness of this result is limited by the fact that it may be difficult for applicants to
identify their best answers (as required by Nash equilibrium) if they have very limited information about the
preference of others as proven by Roth and Rothblum (1999). It also may be profitable for her to pretend that
some acceptable employers are unacceptable. Even so, Roth (1989) proved that the applicant-proposing-
deferred-acceptance mechanism is incentive compatible for the applicants. Thus if the employers do not
behave strategically, then the Bayesian-Nash equilibrium still exists which produces a stable matching. As for
the adjustable prices and wages, when side payments are presented, the deferred-acceptance algorithm can
be regarded as a simultaneous multi-object English auction, where no object is allocated until biding stops on
all objects.

3.2 One-sided matching

The one-sided market is studied by Shapley and Scarf (1974), where a set of trade exchange indivisible objects
(i.e. plots of land), without the ability to use side-payments. They proceed that top-trading cycle algorithm
always produces a stable allocation : no coalition can reallocate their initial endowments to make all members
better off. The revelation mechanism computes by the top-trading cycle algorithm from submitted preference
orderings and given endowments, is dominant strategy incentive compatible for all participants (Roth,1982),
as well the only mechanism that Pareto efficient, individually rational and incentive compatible (Ma, 1994).
Real-world allocation problems have been formalized using the model of Shapley and Scarf (1974) such as the
Kidney exchange in Section 5.3. Another example with the Gale-Shapley algorithm will be discussed also in
Section 5.2

4. Evidence : Market for Doctors

The work on stable allocations and stable algorithms was acknowledged as an important theoretical
contribution in the 60s and 70s, but the practical relevance was not discovered until early 1980s, with the key
contribution from Roth (1984. ) He documented the evolution of the market for new doctors in the U.S and
argued that a stable algorithm improved the functioning of the market.

4.1 The U.S market for new doctors

Roth (1984) studied the evolution of the U.S market for new doctors. The market suffered from many negative
consequences , i.e. matches were made before the student could produce evidence of how qualified they
become. The market also suffered from congestion : when an offer was rejected, it was often too late to make
other offers. It has resulted in failure as not enough offers can be made in time to ensure mutually beneficial
trades. Similar problems of unravelling and congestions were found to affect many markets in the U, Canada
and the UK, as well as the market for Japanese university graduates (Roth and Xing, 1994). In response to the
failures of US market for new doctors, a centralized clearinghouse was introduced in the early 1950s called the
National Resident Matching Program (NRMP) The institution matched doctors with hospitals using an
algorithm which Roth found to be equivalent with the employer-proposing deferred-acceptance algorithm.
Roth argued that the success of NRMP was due to the fact of a stable matching produced by its algorithm.
Therefore, due to the successfully-designed market, many agents are persuaded to participate, thereby
creating a "thick" market with many trading opportunities.

3
4.2 Regional medical market

Roth (1990, 1991) studied that British regional medical markets suffered from similar problems in the 1960s
that affected U.S medical market in 1940s. Each region introduced its own matching algorithm : some were
stable but some were not.

4.3 Experimental evidence

Various conditions influence the success or failure factors of market institutions. The objective of market
designers is to isolate the role of the mechanism itself and compare the performance of different mechanisms
under the same conditions, yet this is difficult to be applied in reality. When the mechanism used priority
matching, the experimental market tended to unravel and many matches were made outside the mechanism.
The deferred mechanism did not suffer from the same kind of unravelling. Thus provided that the matching
algorithm itself (especially its stability) contribute greatly to the functioning of the market.

5. Market Design

Section 5 discusses the three sets of real-world applications : the market for doctors in the U.S, the design of
school- admission procedures and a case of one-sided matching : kidney exchange.

5.1 Redesigning the market for new doctors

As described by Roth and Peranson (1999), the changing structure of the medical labor market caused new
problems to occur, which led the NRMP to modify its algorithm. Previously Roth proved that in a market where
some agents are couples, there may not any stable matching existed. In the 1900s, the very legitimacy of the
NRMP algorithm was challenged, particularly argued for how an employer-proposing algorithm favored
hospitals at the student expenses. The basic theory of two-sided matching, outlined in Section 3.1, shows that
the algorithm is not incentive compatible for the applicants. However, the applicant proposing version is
incentive compatible for applicants. Therefore, there seemed to be strong reasons to switch to an applicant-
proposing algorithm. Then in 1995, Alvin Roth was hired to direct the design of a new algorithm with the main
objective: to construct an algorithm that would produce stable matching as favorable as possible to applicants,
while meeting the specified constraints of the media market. Computer simulations proposed that the Roth-
Peranson algorithm would be better for applicants than the old NRMP employer-proposing algorithm (Roth
and Peranson, 1999). Since the NRMP adopted the new algorithm, the empirical evidence suggests that the
result is stable despite the presence of couples.

5.2 School admission

In the Balinski and Sonmez (1999) and Abdulkadiroglu and Sonmez (2003) models, classroom slots are allocated
among a set of applicants but the school are not considered strategic agents. Stability captures the ideas that
if student 1 has higher priority than student 2 at school S and student 2 attend school S, then student 1 must
attend a school that she likes at well as S. The main difference between this model and two-side model (in
Section 3.1) is objective criteria can form the priority ranking of students. These arguments proposed using the
applicant-proposing-deferred-acceptance algorithm that is not fully incentive-compatible with the applicants
but also applicant optimal. The New York City public high school started using this algorithm in 2003 and the
Boston public school system used a different version in 2005 (Roth, 2008). This process suffered from
congestion as the applicants did not have enough opportunities to make offers and the market failed to clear
about 30,000 students (In New York City. While in Boston, applicants who simply reported their true
preferences suffered from unnecessarily poor outcomes and every Nash equilibrium would have been Pareto
dominated by the truthful equilibrium of the applicant-proposing-deferred-acceptance mechanism (Ergin and

4
Sonmez, 2006). Furthermore, the process was not incentive-compatible. In 2003, Roth and his colleagues A.
Abdulkadiroglu and P.A Pathak helped to re-design the school admission process that uses an applicant-
proposing-deferred-acceptance algorithm which modified to comply with the regulations and customs of New
York City. This algorithm is also incentive compatible for the applicants, resulting in a 90% reduction compared
to the previous year. In 2003, Roth and his colleagues, A. Abdulkadiroglu, P.A Pathak and T.Sonmez, also were
asked to provide advice on the design of a new BPS clearinghouse algorithm. Since it is incentive-compatible
for the applicants, it eliminates the need for strategizing.

5.3 Kidney exchange

The kidney exchange emphasizes an essential aspect of market design. Roth, Sonmez and Unver (2004)
observed the similarity in kidney exchange and the Shapley-Scarf one-sided matching model (described in
Section 3.2). One important difference is while all objects in this matching model can be assigned
simultaneously, some kidney patients must be set to a waiting list. Roth, Sonmez and Unver (2004) applied the
top trading cycle algorithm to allow waiting list options and as a result, constructed an incentive-compatible
chain and efficient selection rules. Another problem lies not only in the waiting list but also due to logical
constraints. Roth, Sonmez and Unver (2005) demonstrated how efficient results with good incentives can be
founded efficiently, only when bilateral exchanges are feasible. The new problems, i.e., the waiting list and
logistical problems of kidney exchange, stimulate new theoretical research and in turn, lead to the new
application.

6 Other Contributions

6.1 Lloyd Shapley

Shapley's contributions in the non-cooperative theory in the form of numerous innovative studies for dynamic
games. However, mostly Shapley played a major role and be an important researcher in cooperative game
theory. For example , Shapley and Shubik (1969) characterized the class of transferable-utility market games.
In addition, in 1971, Shapley proved that Shapley value occupies a central position in the core of convex games.
The Shapley value has played a significant role with its vast variety of applications in cooperative game theory,
despite originally intended as a prediction for a player to receive from a game.

6.2 Alvin Roth

The book from Roth and Sotomayor in 1990 describes the state of the two-sided matching theory that includes
many important outcomes. Roth (1991) also documents how laboratory experiments and field observations
can interact with game theory and establish an economy as a satisfactory empirical science. Then, Roth with
Erev in 1995 developed a reinforcement learning model where players tend to repeat choices that produce
good results and turned out to be consistent with the behaviors of experimental gamers. After that in 1998,
they showed that a reinforcement learning model can predict behavior ex-ante. Consequently, it has shown
that if realistic cognitive limitations are considered, game theory can increase its predictive power and
explanatory.

7. Conclusion

Lloyd Shapley has led the development of the game theory that not only strengthened the fundamental
theories but also improving the usefulness of the theories for the policymaking and applied for work. The work
from Alvin Roth also enhances our understanding of the works in market industries, such as understanding
how the institutions improve the market performance and therefore, clarifying the need for stability and
incentive compatibility.