Roth MarketDesignEmerged 2019

American Economic Association
How Market Design Emerged from Game Theory

Author(s): Alvin E. Roth and Robert B. Wilson
Source: The Journal of Economic Perspectives , Vol. 33, No. 3 (Summer 2019), pp. 118-
143
Published by: American Economic Association
Stable URL: https://www.jstor.org/stable/10.2307/26732324
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
American Economic Association is collaborating with JSTOR to digitize, preserve and extend
access to The Journal of Economic Perspectives
This content downloaded from

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
All use subject to https://about.jstor.org/terms
Journal of Economic Perspectives—Volume 33, Number 3—Summer 2019—Pages 118–143
How Market Design Emerged from

Game Theory: A Mutual Interview
Alvin E. Roth and Robert B. Wilson
W
e go back a fairly long way. Al was a PhD student at Stanford in Opera-
tions Research from 1971 to ’74, and Bob was his dissertation advisor.
Game theory was also young in those days; its offspring, mechanism
design, was even younger; and practical market design by economists was not yet
on the horizon.
To jog our memories about the history and development of game theory and
how it shaped and was reshaped by market design, we interviewed each other over
coffee during Fall 2018.1 We also touched on what we think has been learned about
markets and marketplaces by trying to design them.
What emerged from our discussion is that, when we learned game theory, games
were modeled either in terms of the strategies available to the players (“noncooper-
ative game theory”) or in terms of the outcomes that could be attained by coalitions
of players (“cooperative game theory”), and these were viewed as models appro-
priate for different kinds of games. In either case, the particular model was viewed
as a mathematical object that could be viewed in its entirety by the theorist. Market
design, however, has come to view these models as complementary approaches for
1
We later added publication dates for the work to which we refer, and each of us inserted footnotes to
our own comments where additional background seemed useful.
■ Alvin E. Roth is the Craig and Susan McCaw Professor of Economics and Robert B. Wilson
is the Adams Distinguished Professor of Management, Emeritus, Graduate School of Busi-
ness, both at Stanford University, Stanford, California. Their email addresses are alroth@
stanford.edu and [email protected].
†
For supplementary materials such as appendices, datasets, and author disclosure statements, see the
article page at
https://doi.org/10.1257/jep.33.3.118 doi=10.1257/jep.33.3.118

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
Alvin E. Roth and Robert B. Wilson 119
examining different ways in which marketplaces operate within their economic

environment. And, because that environment can be complex, there will be aspects
of the game that are not entirely observable.
Mathematical models themselves play a less heroic, stand-alone role in market
design than in the theoretical mechanism design literature. A lot of other kinds of
investigation, communication, and persuasion play a role in crafting a workable
design and in helping it to be adopted and implemented, and then maintained and
adapted.
How Did Game Theory Look When You Began to Learn It, and
Teach It?
Wilson: Before 1960, basic concepts of strategic analysis were established but had
slight influence on economics. Studies of parlor games (for example, Borel 1921,
1953) influenced von Neumann’s early work on minmax solutions of constant-sum
two-player games. Von Neumann and Morgenstern (1944) showed existence of such
solutions for all such games and then proposed a solution of cooperative games.2
Nash’s (1950a, 1951) definition of equilibrium for noncooperative games offered
an alternative approach. The main applications to noncooperative contexts were
military and about zero-sum two-player games until Schelling’s (1960) broad view
of “the strategy of conflict,” which was nontechnical but informed by game theory
and widely read. Axiomatic cooperative theory advanced via the “value” introduced
by Shapley (1953) and the “bargaining solution” by Nash (1950b), and these were
often invoked in theoretical economic models.3
Most influential for economists was Luce and Raiffa’s (1957) book-length
critique of game theory’s potential for advancements in the social sciences; it was
guardedly optimistic, with severe criticisms, widely read, and influenced a genera-
tion of scholars. Hayek’s (1945) article on “the use of knowledge in society” set the
stage for much-later use of models from game theory. He interpreted markets as
mechanisms for eliciting preferences and equating marginal rates of substitution
among diverse agents with local information about production and consumption
opportunities. In the early 1960s, I read most of Luce and Raiffa (1957) and portions
of von Neumann and Morgenstern (1944) and Karlin (1959).
Roth: I also found Luce and Raiffa much easier to read than von Neumann and
Morgenstern.
2
Their book was heralded as the future of economics in reviews by Hurwicz (1945), Marschak (1946),
Copeland (1945), and Wald (1947). Marschak concluded: “Ten more such books and the progress of
economics is assured.” Copeland wrote: “Posterity may regard this book as one of the major scientific
achievements of the first half of the twentieth century.” It took decades for these prospects to be realized,
albeit in a form rather different than von Neumann and Morgenstern envisaged initially.
3
An excellent complement to our discussion here is Myerson’s (1999) history of Nash’s contributions
and their subsequent impact in economic theory.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
120 Journal of Economic Perspectives
Wilson: As an MBA student in 1960, I wrote a class report on how to bid in an

auction that got a failing grade because it was not “managerial.” My studies with
Howard Raiffa were focused on decision theory, but discussions with Jacob Marschak
and Lloyd Shapley turned me to game theory, initially to generalize the Lemke and
Howson (1964) algorithm to find Nash equilibria of N-player games, then in papers
on auctions, then in advising the brilliant Armando Ortega-Reichert (1969) on
his dissertation about auctions that went far beyond Vickrey (1961). Social choice
theory and auctions were main interests until 1978, although in 1968 I developed
an MBA course on “competitive strategies” with broad coverage, and a PhD course
on “multiperson decision theory.”
Roth: When I began grad school at Stanford in 1971, there was no course
offered in game theory. But Michael Maschler visited in academic year ’72–73 and
offered one.4
In those days, game theory was thought of as being divided into two parts:
cooperative and noncooperative. These were entirely separate theories, differently
formulated and thought to apply to different economic environments—namely, those
with and without binding agreements. The idea was that for cooperative games, we
would study what (binding) agreements rational agents would reach. For noncoop-
erative games, we would study Nash (1950a) equilibria, interpreted either as the result
of players’ independent optimization in the light of others’ presumed rationality, or
as the agreements they could reach (in the absence of ways to enforce agreements)
that would be self-enforcing in the sense that no player had an incentive to break the
agreement if others were expected to follow it when they chose their strategies.
This division of game theory into two parts had its origins in von Neumann and
Morgenstern (1944), although some of the particular ideas, interpretations, and
models (including Nash’s formulation of equilibria) came later.5
Also inherited from von Neumann and Morgenstern was that the goal of game
theory should be to find the “solution” to each class of games, that would “solve”
each kind of theory. They attached great importance to the idea that a solution,
when found, would apply to all games in (at least) a very broad class, and therefore
that an important property of prospective solutions should be that they should exist
for all games. Indeed, an existence proof was often regarded as the main contribu-
tion of Nash (1950a), rather than his novel formulation of strategic equilibrium.
There were two complementary models of noncooperative games: the
“normal” or strategic form of the game represented as an n-dimensional matrix
when n players were involved, and Kuhn’s (1950, 1953) formulation of extensive-
form games. The extensive form is a tree with branches representing the actions
4
Jerusalem was the center of the game theory world at that time, with a thriving group that had grown
up around Bob Aumann, which included Maschler and a growing group of top students at the Hebrew
University and later also at Tel Aviv, which by 1972 already included Bezalel Peleg, David Schmeidler,
and Shmuel Zamir.
5
There weren’t any up-to-date advanced textbooks in the 1970s, although Owen (1968) provided an
introduction, and so Maschler taught from papers and his own notes.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
How Market Design Emerged from Game Theory: A Mutual Interview 121
available to particular players as a consequence of actions taken earlier in the tree,

and “information sets” of nodes which indicated what a player knew about earlier
decisions when it was his turn to move (with all the nodes in a given information
set being indistinguishable to that player at the time when the choice of action
was demanded). In this formulation, a strategy for a player was a complete plan
of action: a function that specified for each of that player’s information sets, what
action he would take if the game reached that information set. The more compact
“normal” or “strategic” form of the game came to be understood as specifying the
(expected utility) payoffs that each player would receive as a consequence of each
possible combination of strategy choices by the players.
The “solution concept” for predicting players’ choices of strategies was Nash
equilibrium.6 Selten (1965) had already introduced, in German, the subgame
perfect refinement of Nash equilibrium in the extensive form, but most English-
speaking game theorists learned about that only from his 1975 article that introduced
what came to be called “trembling hand perfection” as a further refinement. For
many years, the search for increasingly powerful refinements was a hot topic in
game theory. The idea behind refinements was very closely related to the concep-
tion of each game as being perfectly captured by its extensive or strategic form: if
we knew everything about a game, then perhaps we could deduce from first prin-
ciples which of the multiplicity of equilibria would be the one that would be picked
by perfectly rational agents who knew one another to be perfectly rational.7 While
this was never achieved, some refinements, including various notions of perfection,
and of sequential rationality (Kreps and Wilson 1982), have become useful tools
for modern game theorists, and new refinements continue to be proposed and
explored (for example, Milgrom and Mollner 2018; Myerson and Weibull 2015),
because many if not most games have a plethora of Nash equilibria.
Harsanyi (1967–68) extended the extensive-form model with common knowl-
edge to games of incomplete information, in which there was an initial common
knowledge move by Nature that produced “types” of players who each knew his
own type but knew only the distribution of other players’ types. The idea that every-
thing about the structure of the game (including the rationality of all the players)
was known to all the players was made clearer by specifying precisely what it meant
for something to be common knowledge (made formal by Aumann 1976 indepen-
dently of, but in the spirit of, Lewis 1969).
6
The awkward term “solution concept” was widely used once it became clear that there were not readily
going to be any perfect “solutions” forthcoming, although the word “perfect” would emerge as a term in
the equilibrium refinement literature, which continued to seek a definitive solution for noncooperative
games.
7
The hope was that the “right” refinement would be a subset of all the others, possibly a unique
equilibrium for each game that captured most fully the perfect rationality of all the players. However,
refinements of equilibria turned out to be more like onions than like olives: applying all of the attractive
refinement principles to peel away imperfect equilibria did not yield an irreducible center, but rather
nothing at all. Wilson: I view the axioms in Govindan and Wilson (2012) that characterize Mertens (1989)
stable sets as a surviving core.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
I’ll come back to this, as the idea that the entire game—all the strategies avail-
able to all the participants—was common knowledge among the participants and
completely known to the theorist seeking to analyze the game, was one of the
features of early game theory that had to be overcome for practical market design
to develop. Indeed, the idea that practical design should not depend on unrealistic
common knowledge assumptions (or on assumed knowledge involving too much
detail of players’ private information) is widely known as the “Wilson doctrine,”
after Wilson (1987).
Wilson: At issue was whether the fine detail of game-theoretic models provided
new economic insights.8 Is it sufficient to assume that markets clear at whatever
prices are required to do it, or should one examine institutional arrangements for
eliciting demands and establishing prices, or even the informational content of
prices as suggested by Hayek? It was hard to give up the beautiful welfare theorems
implied by “perfect competition” if one were to take account of a welter of detail
about agents’ incentives in imperfect markets.9
Many game theorists and other economists attended summer seminars at Stan-
ford, with prominent game theorists being among the regulars, notably Robert
Aumann and colleagues from Israel, where mathematically rigorous work was most
advanced. The focus was essentially about how to formulate and analyze economic
models in which all agents are strategic players, and whether such efforts would be
useful in economic theory and applications. Essential roles were played by Kenneth
Arrow, whose influential 1963 article on markets for medical care and insurance
recognized these ingredients as intrinsic to the problem of organizing such markets,
and Leonid Hurwicz, whose 1973 article showed the impediments posed by agents’
private information and strategic behavior, and the necessity of taking account of
them in economic analysis.
Hurwicz formulated the concept of a mechanism for implementing social
choices, specified as a procedure that uses messages received from agents to select
an outcome. The messages could be reports of privately known preferences or
information, and the outcome could be an allocation of goods and/or selection
of public projects. He invoked Nash equilibrium as a predictor of agents’ strategic
behavior, and more generally, emphasized the constraints imposed by incentive
8
Early on, economists recognized that the apparatus of game theory enabled precise description of “who
knows what when” and their available actions. Its solution concepts were problematic, but its descrip-
tive power exceeded the usual tools of microeconomics. The International Journal of Game Theory began
publication in 1971. By the late 1970s, game theory was widely adopted as a basic tool for modeling and
analysis in theoretical microeconomics.
9
In economics, the roles of private information and imperfect observability of actions (hidden informa-
tion and hidden actions in Arrow’s phrasing) were suddenly on display in Akerlof (1970) on markets for
“lemons” and Mirrlees (1971) on optimal taxation, and somewhat later in Spence (1973) on signaling in
labor markets, Rothschild and Stiglitz (1976) on insurance markets with adverse selection, and Mirrlees
(1979) and Holmstrom (1977) on optimal contracting. All posited simple behavior on one side of the
market and studied optimal strategies of the other side, a style that came to characterize information
economics.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
compatibility as key to unifying classical and game-theoretic analyses. Familiar

market institutions, such as auctions and exchanges, provided abundant examples
of mechanisms. Moreover, his perspective encompassed descriptions of existing
mechanisms (and perhaps understanding why market designs of ancient vintage
work well in many contexts), and designing efficiency-improving modifications—
the genesis of market design.
After 1976, it was well-established that game theory offered potentially useful
models and analytical tools for studies of strategic behavior, especially in contexts
with imperfect observability of agents’ information and actions. Its impact was
initially in industrial organization, where game-theoretic results rebutted some
earlier conclusions; in labor, where it offered richer theories of contracting; and in
experimental and empirical studies, where detailed structural models supplanted
reduced-form regressions. Its use grew steadily until it was widely taught to PhD
students in economics (who needed such skills to read proliferating journal articles
that relied on it), the arrival of excellent texts for economists, and a surge of young
scholars who invoked strategic analysis in their research.
Roth: Cooperative games were studied in “coalitional form” models that speci-
fied what each coalition of players could achieve on its own. The most tractable
model was the “characteristic function with transferable utility” (or “with side
payments”), which modeled a game among a set N = {1, … , n} of players by speci-
fying what numerical payoffs each coalition—that is, each subset S of N—could
assure for its members. The assumption of “transferable utility” meant that each
coalition was able to distribute the maximum sum of payoffs that it could achieve in
any way that it wished among its members. Hence a game could be represented by
a vector of real numbers, one for each coalition. The “characteristic function form”
of a game was a function v on the subsets of N, that is, a pair (v, N) with v: 2N → R +
representing how much each coalition could achieve on its own. Outcomes of the
game could then be represented as payoff vectors, one to each player in the game.
Von Neumann and Morgenstern (1944) defined how one payoff vector y could
dominate another payoff vector x via a coalition S, if the coalition S, acting on its
own, could guarantee each member i of S a payoff yi  greater than the corresponding
payoff xi at the outcome x. They defined a “solution” of the game to be a set of
feasible payoff vectors of the game, none of which dominated another element in
the solution, but at least one of whose elements dominated any element outside of
the solution. It was easy to construct games with a multiplicity of solutions, but they
conjectured that there would exist no game (v, N) for which no solution existed.
This conjecture was eventually disproved (Lucas 1969), but even before that, von
Neumann–Morgenstern solutions had not proved to be useful in understanding
many games, and fell from use in economics.10
10
They didn’t fall from use entirely, however, before I wrote my PhD dissertation on generalizations of
von Neumann–Morgenstern solutions that had better existence properties (Roth 1975, 1976). Much
progress in game theory was made by exploring and identifying dead ends.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
However, their idea of domination proved quite useful, in coalitional games of

all sorts and not just of the form (v, N), and particular attention started to be paid
to the set of undominated outcomes of a game, named the core of the game. The
core could gracefully be generalized to games without side payments, in which the
set of outcomes that each coalition could guarantee to its members might have to be
described by a set that could not be summarized by a single number.11
In either case, it is sometimes useful to think of the core as the set of outcomes
for which no “blocking coalitions” can form and produce an outcome that its
members prefer. In some of the applications to labor markets, we focus on small
blocking coalitions, consisting of just a single firm and worker, who, if not matched
to each other when they would prefer to be, can form a “blocking pair.” Matchings
of firms to workers that have no blocking pairs are called pairwise stable match-
ings, and in some simple models these coincide with the core (and of course core
outcomes have no blocking pairs, since they have no blocking coalitions of any
size). Even in applications in which the set of stable matchings is bigger than the
core, it is often the subset of outcomes of the most interest, because it is easier for
small blocking coalitions to form than it is for large ones. I’ll come back to this
when speaking of the clearinghouse through which American doctors find their
first jobs.12
Models of games without side payments, which also had a long history, gradu-
ally replaced games with side payments as the primary models for particular kinds of
cooperative games. For example, exchange economies were modeled as having each
player endowed with a vector of continuously divisible commodities, and coalitions
of players were able to trade freely among themselves. This was a model without side
payments, in that the set of outcomes a coalition could achieve on its own couldn’t
be described by a single number but was rather the set of allocations to members of
the coalition that could be reached by trade within the coalition.
Although the core is empty for many games, it is non-empty for these exchange
economies since the core contains the competitive allocations. This reinforces
the idea that, when it is non-empty, the core can be interpreted as a model of the
outcomes that would result from perfect competition. This idea is reinforced by
11
A lot of creative effort went into generalizing the core in ways that would give a non-empty set for all
games (v, N ), such as the bargaining set (Aumann and Maschler 1964) and the kernel and nucleolus
(Maschler 1992). A very different solution concept was the Shapley (1953) value, also well defined as a
unique outcome of any game (v, N ), which was meant to capture something like the expected utility
of playing the game, in each of its positions (see Roth 1988 for a collection of articles on the Shapley
value collected in honor of Shapley’s 65th birthday). Despite heroic attempts, the generalizations of the
bargaining set and Shapley value never caught on. Regarding the Shapley (1969) value for games without
transferable utility, see my exchange with Aumann in Roth (1980, 1986) and Aumann (1985, 1986), all of
which are reprinted in Aumann (2000) along with some other closely related papers.
12
The term “blocking coalition” has become standard, despite the fact that it may not be the term
that best expresses the manner in which these coalitions make outcomes outside of the core less likely.
Shapley (1973) suggested that they be called “improving” coalitions, so that the core could be defined as
the set of outcomes upon which no coalition can improve.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
the observation that if the economy grows large in appropriate ways, then the core
shrinks to become precisely the set of competitive allocations.
Models of cooperative games without side payments that became very impor-
tant in my own work were the two-sided marriage model of Gale and Shapley (1962),
in which individuals on opposite sides of the market could match to one another
if they both agreed, and the exchange economy with indivisible goods of Shapley
and Scarf (1974). Both papers demonstrated algorithms that, for any specification
of the players’ preferences, would produce an outcome (a matching of pairs, or
a redistribution of initial endowments, respectively) in the core of the game. In
this way, both papers showed constructively that, for any preferences, the core of
the game was non-empty. (Gale and Shapley concentrated primarily on a model in
which the core and the set of pairwise stable outcomes coincide.)
Despite the development of some important models and results, cooperative
game theory was in decline by the late ’70s, as more game theorists took the point
of view that came to be called the “Nash program,” which is that all games could be
modeled strategically. The idea was that if binding agreements were possible, then
how they were reached should be modeled in the extensive form, so that all games
could/should be modeled strategically, as noncooperative games.13 Of course, to
model and analyze complex games is difficult, so one consequence of this approach
was that games studied by game theorists would have strategy sets that could be
generated by a small set of rules.
What Was Missing That Was Needed for Practical Market Design?
Wilson: For market design to reach practice, the missing ingredients were theo-
retical and experimental studies that gave some confidence to predictions about
how design features affect performance. Early applications were partly guesswork,
but with accumulated experience and an increasing trove of scholarly studies, fewer
informed guesses are needed. Game theory has been the principal analytical tool
because it enables detailed modeling of agents’ information, incentives, and feasible
strategies and provides predictions about equilibrium behavior and outcomes.
Theoretical and experimental exercises rely on simplistic models but they clarify
and test the basic concepts applied in practical work where invariably the situation
is more complicated than can be modeled precisely.
Designs of auctions and matching markets evolved from the disparate branches
of noncooperative and cooperative game theory. Agents’ private information is
the main consideration in auctions, and designs focus on procedures that elicit
demands and yield good outcomes. The goal is to implement Walras and Hayek
using Hurwicz’s scheme. This is straightforward when bidders simply know their
own private values for items, but more complicated when their information includes
13
For example, the important Game Theory textbook of Fudenberg and Tirole (1991) contained no
mention of the core, or of any other solution concepts from cooperative game theory.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
estimates of unobserved factors that ex post will affect all their realized values (for
example, in an auction of spectrum licenses, customers’ ultimate demands for uses
of spectrum), and then multiple rounds of bidding can enhance implicit revelation
of bidders’ estimates via their bids. Auction designs often take the set of bidders as a
datum, but as Al describes below, a matching market presents the more formidable
challenge of yielding an outcome so good it attracts participants.
Roth: A big missing part of cooperative game theory involved how some features
of the game that could be expected to be private information, such as the prefer-
ences of the players, would become known. This concern actually fit in well with the
Nash program of modeling games strategically: for example, if we asked participants
in a game to reveal their preferences, then their strategies would include stating
preferences different from their true preferences. Under what circumstances would
a “revelation game” of this sort elicit the information we might wish to know?
Wilson: Hurwicz (1973) considered such revelation games and argued, using
the example of an Edgeworth box, that they could not be viable unless “incentive
compatibility” constraints were imposed on the mechanism.
Roth: A big missing part of noncooperative game theory was how we would know
if a game produced “bad” outcomes that some participants might wish to circumvent
by engaging in a larger game that might not be fully visible to the theorist. When I
started to study labor markets, I saw that firms and workers had very large strategy
sets that allowed them to approach each other in many ways, and at many times.
For example, when professional organizations tried to organize job markets for new
doctors, or new lawyers, they specified rules of engagement between applicants and
employers, but for many years these rules didn’t succeed in organizing those markets
because there were incentives for applicants and employers to find creative ways to
work around them, often reaching agreements well before the markets were officially
supposed to begin (Roth and Xing 1994). Analyzing those markets under the assump-
tion that everyone played by “the rules” would have yielded different outcomes than
were observed, and indeed many observed outcomes in labor markets involved strate-
gies that were either not imagined or explicitly forbidden under the official rules.
Mechanism design was in the spirit of studying fully known games—a game would
be designed, in all its parts, that would specify all possible strategies, so the players
would have no options outside of the game (except perhaps, not playing at all). That
worked well when the designer could make players play the game: for example, a
company or government that wanted to sell or buy something and could define the
rules of the auction which those who wanted to transact must participate in.14 But
14
Even in these cases, Klemperer (2004) for example emphasizes that governments’ auctions of spectrum
licenses or Treasury bonds may allow strategies outside the formal rules, such as pre-auction mergers of
firms to reduce competition in auctions, and trades in post-auction secondary markets.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
lots of markets don’t have this kind of compulsory power and must persuade users to
participate.
What cooperative game theory ideas like the core allow us to do is to see
whether the equilibrium of a game played by the rules of a particular marketplace
is an outcome in the core of the larger game in which coalitions can find ways to act
on their own outside of the marketplace we are modeling strategically. If the equi-
librium outcome of the strategic model is not in the core of the coalitional model,
then there are some coalitions (for example, of firms and workers) who might have
incentives to try to get a better outcome. Even if we don’t know all the strategies
available to them, that’s a clue that the rules may be subject to attack and evasion
by the dissatisfied parties. I’ll say more about the complementary roles played by
“noncooperative” and “cooperative” models of the same economic environment as
I talk about the clearinghouse designs that were ultimately successful in organizing
the market for new doctors, and the periodic market failures that continue to afflict
the market for new lawyers.15
This is why market designers started to ask whether the equilibrium behavior
elicited by particular rules led to outcomes that were in the core of the game. The
idea is that trying to promote rules that lead to outcomes outside the core—that is,
outcomes that leave some coalitions getting less than they might be able to get by
acting on their own—might give potential marketplace participants incentives to
transact outside the marketplace. The core and related formulations of stability give
us a way of saying something about the fact that participants have strategies outside
of the marketplace, and that successful marketplaces will be those that don’t give
participants reason to go elsewhere.
The big lesson of market design is that marketplaces are small institutions in
a big economic environment: participants have bigger strategy sets than you can
see, and there are lots of players, not all of whom may even be active participants
in the marketplace, but can influence it. So we needed a way to design mechanisms
that had both good equilibrium properties for the rules we knew about, and good
stability properties for the strategies we didn’t know about.
Thus, the connection between coalitional and strategic models as they can be
used in market design is not as models of different kinds of games, but as models
of a given game at different levels of detail, used for complementary purposes. For
parts of the game that we’re designing, we use “noncooperative” strategic models to
precisely specify actions available to players. For parts of the game that we don’t have
complete control over, we use “cooperative” coalitional models to tell us something
about the incentives that agents and coalitions of agents may have to circumvent the
rules. The idea of focusing on, say, pairwise stability in two-sided matching models
15
In Roth (1991a), I wrote of the separation of cooperative and noncooperative game theory, saying
that the less-detailed cooperative models, which try to represent a game without specifying all the rules,
aspire to a spurious generality (because the omitted details matter), while the noncooperative, strategic
models, which are analyzed as if they represented all the potential moves in a game, offer a spurious
specificity when the game in question is a model of some observable situation (because we can seldom
know all the potential moves).

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
is that if a pair of agents is eager to match with each other despite the fact that
the rules of the marketplace mechanism prevent them from doing so, then maybe
their strategy sets will be big enough to find a way to match with each other. (But if
just one of them is interested in matching with the other, it may be difficult for the
unhappy player to find a way to circumvent the marketplace and force a match with
an unenthusiastic partner.)
For example, in the job market for new doctors, before a centralized clearing-
house was adopted, and in some of the markets for new lawyers still, candidates are
often hired years before employment will begin, and before the official rules of the
market allowed hiring to begin (Roth 1984; Roth and Xing 1994; Avery, Jolls, Posner,
and Roth 2001, 2007). That is, firms and workers who were dissatisfied with the way
the official marketplace functioned were able to circumvent it by signing contracts
before it opened. This problem was effectively solved for the medical market by a
clearinghouse that produces stable matchings (Roth and Peranson 1999).
Wilson: This perspective is analogous to one in the older literature on general

equilibrium. A modern view might aim to determine whether a perfectly competi-
tive market is a mechanism yielding an allocation that is efficient, or better yet, in
the core. Because a competitive allocation is easily shown to be efficient and in
the core, a theorem establishing existence of equilibrium prices and the resulting
allocation is, in effect, an affirmative answer when competition is sufficient to justify
traders’ price-taking behavior in response to prevailing equilibrium prices. Such a
theorem typically suppresses all detail about how the market is organized and how
prices are established, summarizing it all in traders’ budget constraints.
Roth: So general equilibrium theory shares with cooperative game theory the
goal of identifying likely outcomes without focusing on all the details of how they
are achieved.
Wilson: The focus on properties of the allocation began with Edgeworth’s (1881)
informal argument that the core shrinks to the competitive allocations as the market
becomes more competitive (by replicating traders), established formally by Debreu
and Scarf (1963), and culminated in Aumann’s (1964) proof that the core consists
only of the competitive allocations when the set of traders is a non-atomic measure
space (so that no trader is large enough to have market power). These results led to
the modern view that an ideal price-mediated perfectly competitive market might
indeed be a mechanism largely immune to institutional details that yields a core
allocation, but realistically the challenge in practice remains to design a mechanism
that yields a core allocation. The focus on the core, and coalition stability more
generally, stems from the prediction that the mechanism will miss some gains from
trade if other opportunities attract away some potential participants. The design
problem is most acute in those matching markets without transfer payments, but
it is relevant whenever the mechanism’s performance depends on attracting wide
participation. Many of the most successful auction designs addressed contexts where

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
participation was mandated by a monopolist seller, such as government auctions of

spectrum licenses or a system operator’s auctions of access to power transmission,
but attracting wide participation is paramount in newer applications such as the
design of trading platforms that operate in competition with other venues.
Roth: Another way in which market design involves a bigger economic environ-
ment than a narrowly defined mechanism design problem is that it may have to
take account of players who are not intended to be, and who do not intend to be,
participants in the market. In particular, some transactions, and the markets that
serve them, are “repugnant” in the sense that some people would like to partici-
pate in them, while other people (who may not have any apparent connection to
these transactions) think that they shouldn’t be allowed (Roth 2007). But successful
markets require a degree of social support, so these concerns need to be taken into
account if a marketplace is to succeed. Widely held feelings of repugnance often
make it necessary for a market designer to study and understand the moral, ethical,
and esthetic opinions of members of the society in which the market might func-
tion, as well as their professional and social codes of conduct and courtesy.
Much of my work in facilitating kidney transplants through the design of
exchange mechanisms can be viewed as arising from the widespread repugnance to,
and laws against, the purchase of organs for transplants.16 And some of my current
work on expanding kidney exchange internationally, while gaining gratifying
support in some quarters, is also meeting with a repugnance reaction in others,
including concerns that it might expand black markets in poor countries.17
Note that it is also a market design task to think about how and whether partic-
ular kinds of markets can be effectively banned, since laws seeking to ban markets
often inadvertently serve to design illegal black markets. Like other kinds of market-
place designs, legal bans on markets also occupy a place in a larger economic
environment, and may be difficult to effectively enforce without wide social support,
or if the markets in question are available in other jurisdictions. Markets and
marketplaces that are legal in some places but banned in others include markets for
prostitution, surrogacy, marijuana, etc.
Finally, in addition to requiring an expanded view of what strategies players
may have access to, and which players may be involved, market design also has to
take into account that players may fail to coordinate on equilibrium behavior. In this
regard, experiments have played an important role in exploring the gap between
what perfectly rational players might deduce, and what ordinarily competent
16
For examples, see Roth, Sönmez, and Ünver (2004, 2005a, b), Roth, Sönmez, Ünver, Delmonico, and
Saidman (2006), Rees et al. (2009), Leider and Roth (2010), and Ashlagi, Gilchrist, Roth, and Rees
(2011a, b). Notice that this selection of papers is drawn equally from analyses appearing in economics
journals and in medical journals, which reveals something about how practical market designs are
developed.
17
Rees, Dunn et al. (2017) offer a new proposal to expand kidney exchange internationally, Delmonico
and Ascher (2017) express opposition, and Rees, Paloyo et al. (2017) and Roth et al. (2017) reply.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
humans might find difficult,18 particularly in the absence of common knowledge

that all players were perfectly rational (see Roth 2016 on experiments specifically
aimed at market design).
Why Were Auctions and Two-Sided Matching Markets Such Fertile

Ground for Market Design? How and Why Did Auction Design
Proceed Differently from the Design of Matching Markets?
Wilson: Centralized auctions and matching markets were fertile grounds for
market design due to coincidence of several features. The key design element speci-
fies rules of a game. The contexts are often sufficiently circumscribed that, if the
design yields efficient (or better, core) outcomes, then one can expect agents to
play that game rather than some larger game with myriad other possibilities. And
the contexts usually justify assumptions of rational optimizing behavior rather than
various behavioral possibilities. Moreover, the game is sufficiently simple that it can,
to a limited extent, be modeled and analyzed, or in any case rough predictions of
performance can be based on applications of basic concepts, simulations, experi-
mental evidence, and prior experience.
This simplicity accounts also for the profusion in academic journals of scholarly
studies of these kinds of markets, including theoretical, experimental, and empir-
ical studies. But the methodologies employed for studies of auction and matching
markets differ.
Auction studies usually rely on preferences represented as expectations of net
monetary values, the mechanism translates static or dynamically adjusted bids into
an allocation, the objective is an equilibrium allocation that is efficient or revenue
maximizing for the seller, and beyond that objective, the design task often focuses
on rules that suppress “gaming the system.” Except in Vickrey auctions, there is
no attempt to elicit agents’ true preferences; instead, one elicits a willingness-to-
pay that already “shades” the bid to exploit monopoly power derived from small
numbers of bidders and their private information, often called informational rents.
Efficiency is hard to assure in cases, like spectrum auctions, where agents want to
acquire packages of complementary goods, so designs aim for approximate effi-
ciency. Ex post efficiency is the actual goal, but this is tenuous due to agents’ private
information or estimates about common-value components.19
In contrast, studies of prominent matching markets rely on ordinal prefer-
ences solely about one’s assigned partner(s), the mechanism translates directly
reported preferences into recommended assignments, and the objective is a core
18
See, for example, Roth and Erev (1995) for discussion of games in which players learn quickly to play
equilibrium and others (such as the ultimatum game) in which learning may be very slow, and see Li
(2017) for discussions of how strategy-proof mechanisms may not be transparent to participants.
19
The modern state-of-the-art in auction design is presented superbly in the two books by Milgrom
(2014, 2017).

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
allocation. This stronger criterion discourages matches outside the mechanism (as
Al described above), and in simple cases, such as a “marriage market,” a core allo-
cation based on reported preferences is obtained via Gale and Shapley’s deferred
acceptance algorithm, or Shapley and Scarf’s (1974) top trading cycles. More-
over, truthful reporting is a dominant strategy for the proposing side (Dubins
and Friedman 1981; Roth 1982), so only the other side might gain from other
strategies.
Roth: “Design” is a noun as well as a verb, and market design has its origins in
the noun, in the study of the designs of existing marketplaces, and how different
designs—different marketplace institutions, rules, and customs—can induce
different strategies and produce different outcomes. Centralized marketplaces
are a good place to start the study of market designs, because, by virtue of being
centralized, a significant portion of their design may reside in well-codified rules
and procedures that are easy to observe. For the same reason, when it becomes
necessary to design new rules and procedures, the work involved in designing
centralized marketplaces can have a very mechanism-design “look and feel,” with
well-defined kinds of messages communicated and processed in precisely specified
ways that offer a concrete path to implementation in practice.
Auctions are centralized marketplaces in which the messages are bids, and
the auction rules determine the form that bids take, how they are communicated,
and how they determine the resulting payments and allocation of the items being
auctioned. Because auctions are ancient tools of demand elicitation, practical
knowledge about auctions began to be developed fairly early and game theory
allowed auction theory to be formalized and extended as one of the early successes
of the theory of mechanism design (for example, Vickrey 1961; Milgrom and Weber
1982). The view that auction rules can be designed was enhanced by Cassidy’s
(1967) survey of the vast variety of auctions used in practice, with differing incen-
tives and performance.
In addition, if the goods being sold are available only from a single seller,
then an auction satisfies the implicit assumption of mechanism design theory that
purchasers must participate in the auction if they wish to buy. (For example, oil
drilling or timber cutting licenses sold by the Department of the Interior, spectrum
licenses sold by the Federal Communications Commission, and advertisements
sold by Google connected to searches on their search engine are each sold by a
single seller.) Thus, at least to a first approximation, the strategies that the auction
designer makes available are the strategies that the bidders must use, and (some
appropriate refinement of) strategic equilibrium among those strategies may be a
good guide to designing the market and predicting the outcome.
Wilson: Of course, the seller should also abide by the rules, or have an incen-
tive to do so to the extent observable by bidders. This criterion is implied by the
definition of a credible mechanism proposed by Akbarpour and Li (2018); for an
historical application, see Engelbrecht-Wiggans (1988).

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
Mechanism design theory usually imposes an “individual rationality” constraint

that no agent is worse off from participating, but this constraint is very weak
compared to the stronger requirement that agents prefer the centralized market to
contracting outside the market, which is achieved by a mechanism yielding a core
allocation. Electricity markets suggest another paradigm: actual energy flows must
be determined by bids in the transmission operator’s centralized market, but partic-
ipants often contract bilaterally via long-term contracts or buy financial hedges
pegged to the operator’s real-time prices for energy and transmission. Decentral-
ized hedging markets ameliorate price volatility in the operator’s market, and each
relies on the other to function well.
Roth: In contrast to auctions, labor markets, kidney transplants, and other

matching markets start off very decentralized. Labor markets have many applicants
and employers, and kidney transplants in the US are performed at hundreds of
hospitals, each with considerable autonomy. So rather than being able to design
a marketplace that all participants must use, the design of marketplaces for many
matching markets involves finding designs that will entice users to try them, and
satisfy them well enough that they will accept the outcomes and continue to come
back to the marketplace. But when these designs lead to centralized clearinghouses,
the marketplace itself nevertheless has considerable mechanism design flavor (as a
stand-alone game) when one concentrates on the options available to participants
within the marketplace. So these marketplaces were also a good starting point for
market design.
The American marketplace for new doctors, the National Resident Matching
Program, developed such a clearinghouse in the early 1950s in response to wide-
spread market failures of the various decentralized market designs that had been
employed in the first half of the 20th century. When I studied it in Roth (1984), I
found that, by a process involving more than a little trial and error, the market had
become organized since 1952 by a centralized clearinghouse in which candidates
and employers submitted rank order lists of one another, and a centralized algo-
rithm produced a suggested match. The algorithm that had been settled on turned
out to be essentially equivalent to the hospital-proposing deferred acceptance algo-
rithm studied a decade later by Gale and Shapley (1962). So on the one hand, this
was a “mechanism” whose design could be studied, but on the other hand, one ques-
tion that had to be answered was why this design had been enticing enough to attract
the lion’s share of the market. This was a pressing question when the marketplace
needed to be redesigned in the mid-1990s. A critical fact, discovered by comparing
successful and unsuccessful clearinghouses (Roth 1991b) and by experimentation
(Kagel and Roth 2000), was that the stability of the resulting outcome was an impor-
tant factor in its success. This was a clear example of the complementary uses of
strategic and coalitional models in understanding the success of a marketplace.
Note how this reflects how game-theoretic ideas about the core and stable
matchings have evolved as they have been confronted by the realities of market
design. When we used to think of a game as a whole world, we often thought of the

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
core as a model of what players would coordinate on, based on complete informa-
tion about the game. So if players didn’t know each other’s preferences, blocking
coalitions would be difficult to identify, and the core (that is, the set of outcomes
for which there aren’t any blocking coalitions) might not have much predic-
tive power. But in a labor clearinghouse like the medical match, no one knows
everyone else’s preferences: it’s a game of massively incomplete information. Yet,
empirically, matching algorithms that produce unstable outcomes fail. Because the
clearinghouse is part of a larger economic environment, it’s no mystery how this
can happen. If I am matched to my third-choice residency program, I only have to
make two phone calls to find out if I am part of a blocking pair—that is, if one of the
residency programs I prefer might also prefer me to one of the doctors it has been
matched with. If in previous years many matches were found that way, then this
year people will make those phone calls too, and the clearinghouse will fail to orga-
nize the market because residency programs and individual doctors will make deals
on their own, and not accept those produced by the clearinghouse. But a stable
matching algorithm will be robust to phone calls, since there aren’t any blocking
pairs to find. Those phone calls aren’t part of the description of the clearinghouse,
they are part of the larger economic environment in which the clearinghouse is just
a small marketplace.
In contrast to the market for new doctors, the market for judicial clerks uses
rules that have been regularly designed and subsequently abandoned by judges
themselves, and have yet to find a market design that entices judges to partici-
pate according to the rules (Avery et al. 2001, 2007, and my blog post at https://
marketdesigner.blogspot.com/search/label/clerks).
Wilson: I was fascinated by the Roth and Xing (1994) article describing many
markets that work imperfectly because they fail to deter contracting outside the
market including prior contracting (to snag a good partner before others do), as
well as backup plans to try again in a decentralized aftermarket among those not
satisfied by the recommended assignment.
A novel feature of some matching markets is assignments recommended to
agents, rather than binding contracts, and most agents voluntarily accept their
recommended matches: the deferred acceptance algorithm assures that no two
agents prefer each other to their assigned partners. Some simple auctions have
this property, and there are designs that aim for it, but generally the prevalence of
private information precludes assurance of a core allocation based on revealed pref-
erences and information.20 Government agencies typically use auctions designed to
20
Day and Milgrom (2008) derive key properties of a core-selecting auction when one exists, and relate
this to the stable matching literature. Kelso and Crawford (1982) consider an auction that closely
resembles the deferred acceptance algorithm, in a labor market context in which the auction chooses
both a matching and the associated market-clearing doubly personalized wages (that is, wages for each
firm-worker pair).

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
yield an efficient allocation, but alternative designs forego efficiency to maximize

the seller’s expected revenue.
Roth: How about in electricity markets?
Wilson: Just as you said before, participants in electricity markets have big
strategy sets, and there are interested parties who aren’t participants in the market.
Firms can bid in the spectrum auctions of the Federal Communications Commis-
sion, buy licenses in secondary markets (if the FCC approves the transfer), or rent
spectrum from those with licenses. In wholesale electricity markets, firms must bid
in the system operator’s daily and hourly energy auctions to get power scheduled
for transmission, and the operator also runs various auxiliary auction markets for
transmission rights (which hedge transmission charges) and reserves of capacity
available in various time frames. But these are minor parts of the overall market
because most power is contracted long term. A typical bilateral contract for delivery
has an agreed price, and the parties settle the difference between their price and
the operator’s price.
There are also financial markets for financial instruments that hedge against
volatility of the operator’s prices. On the supply side there are further markets for
fuel, especially long-term contracts for natural gas that ensure priority when supplies
are tight. And on the demand side there are markets for demand reduction by firms
who can curtail or interrupt power usage when prices are high.
Thus, no one of these auctions is isolated; rather, each operates within a loosely
coordinated system of related markets. This system need not yield an outcome
in the core like in a matching market because no analog of the deferred accep-
tance algorithm has been found, so it relies on competitive pressures to ensure
efficient outcomes—and part of the design task is to promote vigorous competition.
Besides participants, there are other important actors who affect the system design,
most importantly the Federal Energy Regulatory Commission that prescribes stan-
dards for system operators, each state’s public utility commission, which regulates
retail distribution (and in some cases appoints the board of the system operator),
and federal agencies that regulate commodity trading and financial markets for
commodity futures contracts.
Roth: And are aspects of electricity sales repugnant?
Wilson: The repugnance factor can be traced in the history of the restructuring
of electricity markets. Some view electricity as a necessary service that is best provided
by vertically integrated utilities. Prior to restructuring, this was implemented in each
state by tight regulation that set retail service standards and prices and in return
provided utilities with an assured rate of return on capital. In most states, this regime
dissolved because high prices for retail service were attributed to distorted incen-
tives, resulting in excessive capital intensity manifest in massive power plants, and
monopolization of transmission that disadvantaged independent power producers.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
After political battles, the industry was restructured by most states requiring utili-
ties to divest their generation assets, and federal requirements for open access to
transmission via auction markets conducted by independent system operators. Not
all states restructured. But even in those that did, lesser battles continue, repre-
sented for example by newly formed municipal or cooperative power distribution
companies that opt out of utilities’ retail services by buying supplies directly from
system operators’ markets. Basically, market design for electricity markets aspires
to the ideal service that vertically integrated utilities were intended to achieve, but
now implemented in an open-access decentralized system with stronger incentives.
How Was Your Work Influenced by Your Teachers, Colleagues, and

Students?
Wilson: The interests of my advisor Howard Raiffa had turned to statistical deci-
sion theory, but I retained interest in game theory, motivated by a colleague’s study
of investment banking syndicates formed to bid for corporate bonds. After some
early consulting on corporate strategies, my practical work on market design began
in consulting with the US Department of Interior’s section on oil exploration led
by Darius Gaskins. He hired me because he had concluded that game theory was
needed to analyze their auctions of licenses. There were aspects of auction design,
but I focused mainly on algorithms for bidding strategies based on models that
included adverse selection—aka “the winner’s curse”—and methods for ex post
analyses of auction performance. This was also a focus in the late 1970s of my
consulting with oil companies, especially with George Harwell at Natomas, but in
these cases my primary job was to help them understand the effects of adverse selec-
tion. I learned a lot from being inside a company, watching how bids were derived
by interpreting geological data to obtain (vaguely probabilistic) estimates that were
then combined with data about costs and predictions of future oil prices. Equally
educational was to hear insistence on finding the minimum bid that would win.
Some dismissed adverse selection as hokum, but a few old sages insisted it was real
and claimed they had survived in a fiercely competitive industry by using rules of
thumb that severely cut engineers’ estimates to be on the safe side.
Roth: The adverse selection that Bob is referring to, the “winner’s curse,” is that
when each bidder gets an estimate of how much oil is under the ground at a given
site, the bidder who has the highest estimate is very likely to have an overestimate.
And the more bidders there are, the bigger the amount of the overestimation.
Wilson (1977) introduced the model of common-value auctions (sometimes called
the “mineral rights model”). The model and its equilibrium initiated a large body of
theoretical, experimental, and applied work. One important insight from this model
is that winning an auction contains “bad” news, since it implies in equilibrium that
the winner’s estimate is the highest. In equilibrium, rational bidders fully account
for this, but the paper raises the empirical question of the extent to which actual

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
bidders are able to fully discount for the fact that, if they win the auction, they likely
overestimated the value of winning. Thus, Wilson’s work initiated a new research
program on the winner’s curse, involving systematic overbidding compared to
equilibrium, sometimes involving losses to the winning bidder.21 The private-value
model of Vickrey (1961) and the common-value model of Wilson (1977) together
form the basis of much of modern auction theory and practice, since most auctions
have elements of both private and common value.
Wilson: Another formative experience was at the Electric Power Research Insti-
tute in a section led by Steven Peck and Hung-po Chao, working with them and my
co-consultant Shmuel Oren. It focused initially on innovative contracts for utilities’
retail services, but over the ensuing 40 years its scope expanded to include all the
issues posed by fundamental restructuring of the industry, and most relevant here,
the design of centralized markets for energy, transmission, and reserves.
I was deeply affected in the early 1990s by working with Paul Milgrom on design
of the FCC spectrum auctions. I marveled at his insights and creativity in constructing
rules for a “simultaneous ascending auction” that would have good prospects of
yielding an approximately efficient outcome in an environment afflicted with strong
complementarities, dispersed private information about market fundamentals, and
substantial market power. And we were greatly influenced by Evan Kwerel, who was
the main protagonist at the FCC seeking innovative auction designs for allocating
spectrum licenses.
Roth: I was also much influenced by Paul when we developed and co-taught
what may have been the first courses in market design, in 2000 and again in 2001
when he was on leave at Harvard and MIT.
Wilson: The strongest influences on my work in game theory came from Robert
Aumann’s articles and lectures, and later, collaborations with David Kreps and
Srihari Govindan. Even after I pursued market design, I continued my interest in
foundations, seeking the full implications of rationality in multiperson interactions.
I was also deeply influenced by superb PhD students, of which some directly affected
my work in game theory and ultimately market design. Before 1980 they included
Armando Ortega-Reichert, Robert Rosenthal, Alvin Roth, Jean-Pierre Ponssard,
Claude d’Aspremont, Paul Milgrom, and Bengt Holmstrom. Later, the chief influ-
ence was Peter Cramton.
21
In an early use of experimental economics to elucidate this issue, Bob invented the now famous “jar of
coins” experiment, in which the value of the coins in a jar is auctioned off to the highest bidder. If every
bidder forms his own estimate of the value of the coins (for example, of how many coins are in the jar),
then the high bidder almost invariably has an overestimate, and, failing to account for this, bids more
than the jar is worth. Used as a demonstration, this helps convince skeptics that the winners’ curse is real
(for a fuller account, see Roth 2016).

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
Roth: I would have had a very brief academic career if not rescued by Bob after
flunking my qualifying exams. As I’ve learned also from my students, the teacher–
student relationship can be one of life’s big ones, although less appreciated than
the other big relationships.22 I’ve learned from (and designed with) students,
postdocs, research fellows, and colleagues, some of whom fall in more than one
category (and who I refrain from enumerating here only because the list has such
ill-defined boundaries and includes so many of my students and coauthors that I
would inevitably err by omission). I’ve also profited enormously from collaborating
with practitioners. I think that virtually all of the market designs I’ve been involved
in that were adopted and successfully implemented benefited from collaboration
with someone involved in the market who became the champion of the new design.
How Have Other Domains of Market Design Developed?
Roth: I’d like to highlight two other domains I think foreshadow further ways
market design is developing into a robust part of economics. The first is the design
of school choice systems, which in its origins closely resembles the stable matching
deployed in the clearinghouse marketplaces for doctors (Abdulkadiroğlu and
Sönmez 2003; Abdulkadiroğlu, Pathak, and Roth 2005, 2009; Abdulkadiroğlu,
Pathak, Roth, and Sönmez 2005). School choice has now opened a new window for
the empirical study of schools, as econometric tools take advantage of the particular
elements of the market design to measure not only the effects of the new designs on
how students are assigned to schools, but also the importance to students of being
well matched to a school (for example, Abdulkadiroğlu, Agarwal, and Pathak 2017;
Abdulkadiroğlu, Angrist, Narita, and Pathak 2017; Agarwal and Somaini 2018).23 I
think of this as a kind of third generation of market design, since those of us initially
involved in design were game theorists, who of necessity became engineers to help
new designs be implemented and maintained, and we are now seeing those designs
and their outcomes subjected to, and enabling, sophisticated empirical scrutiny by
applied economists able to develop new econometric tools informed by the details
of the markets’ designs.
Another area of market design involves decentralized markets. Most markets
are decentralized at least to some degree, and many almost entirely. Even markets
22
A brief account of our subsequent teacher–student interactions, along with some other remembrances
related to the present essay, is included in my intellectual autobiography at the Nobel Prize website:
https://www.nobelprize.org/prizes/economics/2012/roth/auto-biography/. The rabbinic literature
does not overlook teacher–student relations. In the Talmud, for example, one is enjoined: “Provide
for yourself a teacher and get yourself a friend …” See my related blog post for more on this: https://
marketdesigner.blogspot.com/2013/06/notes-on-teachers-and-students-from.html. The martial arts also
value teacher–student relations, and I benefited from that too, as I describe at https://marketdesigner.
blogspot.com/2013/06/honorary-7th-dan-black-belt-in-jka.html.
23
Agarwal (2015) similarly uses econometric tools that leverage the stable matchings arising from the
resident match to do a demand analysis of the market for different residency programs.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
that employ centralized marketplaces may be preceded or followed by decentral-

ized interaction. For example, the market for new academic economists has a
somewhat centralized marketplace for interviews, preceded by decentralized
applications and followed by decentralized campus visits, offers, acceptances, and
rejections.24
Designers who wish to introduce more centralized marketplaces into existing
markets need to understand how this will interact with pre-existing decentralized
markets. In this respect, changing the presumptions about how (decentralized)
offers were made, and until when they should remain open, helped pave the way for
a centralized clearinghouse for gastroenterologists to become successful (Niederle
and Roth 2003, 2009; Niederle, Proctor, and Roth 2006, 2008; McKinney, Niederle,
and Roth 2005). PhD programs similarly are expected to leave offers of admission
open until April 15 (Roth and Xing 1994). Design in decentralized markets may
involve helping to form expectations and customs to promote and to guide decen-
tralized transactions among market participants, rather than crafting precise rules
and algorithms that can be rendered in computer code.
Wilson: A remarkable application is the conversion in the United States of

spectrum from television broadcast to smartphones (Leyton-Brown, Milgrom, and
Segal 2017). Ultimately this was done by one auction that bought spectrum from
broadcasters and another that re-sold it to phone companies. The major complica-
tion was development of algorithms to reassign retained broadcast rights to new
spectrum so as to avoid electromagnetic interference among broadcasting stations.
What’s Next for Game Theory and Market Design?
Wilson: The ongoing computerization of marketplaces will continue to make

market design a multidisciplinary endeavor, which already occupies computer scien-
tists as well as economists.25 And economic engineering more broadly—“design
economics”—will likely continue to grow in its ability to help structure contracts,
firms, and organizations and collaborations of all sorts.
24
There have been some modest further efforts to aid coordination through centralized signaling before
interviews and a scramble afterwards (Coles, Cawley, Levine, Niederle, Roth, and Siegfried 2010).
25
For examples of some collaborations between economics and computer science, see Anderson,
Ashlagi, Gamarnik, and Roth (2015) and Leyton-Brown, Milgrom, and Segal (2017). As computer and
communication technologies increase the proportion of transactions conducted over highly automated
and tightly coordinated platforms, we foresee the design of these platforms as a major area for market
design involving joint efforts among economists, computer scientists, and software engineers relying on
developments in “algorithmic game theory.” Already, parameters of some markets are adjusted automati-
cally by machine learning algorithms, and we are entering a time when market participants themselves
may be designed—to some extent this has already happened in the realm of high-speed algorithmic
trading of securities (as discussed in Budish, Cramton, and Shim 2015).

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
Roth: Smartphones have put marketplaces in our pockets, and as computer-

ized marketplaces become ever more ubiquitous, we will also generate data trails
that will continue to extend the reach of markets, socially and personally. We will
learn more about privacy and fairness, and there will be new opportunities, some of
which will come to be seen as repugnant, for which new market mechanisms, rules,
customs, and regulations will have to be designed.
What’s Your Last Word for Now?
Wilson: We’ve learned that maximizing gains from trade is more about partic-
ipants’ information and incentives than intersecting demand and supply curves.
So concepts from game theory have been useful guides in efforts to improve the
performance of trading platforms. But scholarly theorizing is minor compared to
hands-on engineering using knowledge of an industry’s technology and practices,
and familiarity with participants’ concerns is necessary if one is to help them obtain
better outcomes overall. Deep involvement discovers key features unanticipated by
abstract views of markets. I foresee more economists improving the allocation of
scarce resources rather than (just) studying it.
Roth: Market design is an outward-facing part of economics, so designers have

to be good listeners, prepared to learn from everyone. And learning from markets
and their participants is a great driver of economic theory. One of my favorite quotes
from Wilson (1993) is “for the theorist, the problems encountered by practitioners
provide a wealth of topics.”
■ This paper benefited from comments by Vic Fuchs, Dave Kreps, and Ben Roth, none of whom
endorse any of the views expressed here.
References
Abdulkadiroğlu, Atila, Nikhil Agarwal, and Alvin E. Roth. 2005. “The New York City High
Parag A. Pathak. 2017. “The Welfare Effects of School Match.” American Economic Review 95(2):
Coordinated Assignment: Evidence from the New 364–67.
York City High School Match.” American Economic Abdulkadiroğlu, Atila, Parag A. Pathak, and
Review 107(12): 3635–89. Alvin E. Roth. 2009. “Strategy-Proofness versus
Abdulkadiroğlu, Atila, Joshua D. Angrist, Efficiency in Matching with Indifferences: Rede-
Yusuke Narita, Parag A. Pathak. 2017. “Research signing the NYC High School Match.” American
Design Meets Market Design: Using Centralized Economic Review 99(5): 1954–78.
Assignment for Impact Evaluation.” Econometrica Abdulkadiroğlu, Atila, Parag A. Pathak, Alvin
85(5): 1373–1432. E. Roth, and Tayfun Sönmez. 2005. “The Boston
Abdulkadiroğlu, Atila, Parag A. Pathak, and Public School Match.” American Economic Review

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
95(2): 368–71. Law Review 68(3): 793–902.

Abdulkadiroğlu, Atila, and Tayfun Sönmez. Avery, Christopher, Christine Jolls, Richard A.
2003. “School Choice: A Mechanism Design Posner, and Alvin E. Roth. 2007. “The New Market
Approach.” American Economic Review 93(3): for Federal Judicial Law Clerks.” University of
729–47. Chicago Law Review 74(2): 447–86.
Agarwal, Nikhil. 2015. “An Empirical Model Borel, Emile. 1921. “La theorie du jeu et les
of the Medical Match.” American Economic Review equations integrales anoyau symmetrique gauche.”
105(7): 1939–78. Comptes Rendus de l’Academie des Sciences 173:
Agarwal, Nikhil, and Paulo Somaini. 2018. 1304–08.
“Demand Analysis Using Strategic Reports: An Borel, Emile. 1953. “The Theory of Play and
Application to a School Choice Mechanism.” Integral Equations with Skew Symmetric Kernels.”
Econometrica 86(2): 391–444. [Translation by L. J. Savage of Borel (1921)] Econo-
Akbarpour, Mohammed, and Shengwu Li. 2018. metrica 21(1): 97–100.
“Credible Mechanisms.” Stanford University. Avail- Budish, Eric, Peter Cramton, and John Shim.
able at SSRN: https://ssrn.com/abstract=3033208. 2015. “The High-Frequency Trading Arms Race:
Akerlof, George A. 1970. “The Market for Frequent Batch Auctions as a Market Design
‘Lemons’: Quality Uncertainty and the Market Response.” Quarterly Journal of Economics 130(4):
Mechanism.” Quarterly Journal of Economics 84(3): 1547–1621.
488–500. Cassidy, Ralph, Jr. 1967. Auctions and Auction-
Anderson, Ross, Itai Ashlagi, David Gamarnik, eering. University of California Press.
and Alvin E. Roth. 2015. “Finding Long Chains in Coles, Peter, John Cawley, Phillip B. Levine,
Kidney Exchange Using the Traveling Salesman Muriel Niederle, Alvin E. Roth, and John J. Sieg-
Problem.” PNAS 112(3): 663–68. fried. 2010. “The Job Market for New Economists:
Arrow, Kenneth J. 1963. “Uncertainty and the A Market Design Perspective.” Journal of Economic
Welfare Economics of Medical Care.” American Perspectives 24(4): 187–206.
Economic Review 53(5): 941–73. Copeland, Arthur H. 1945. “Review: John von
Ashlagi, Itai, Duncan S. Gilchrist, Alvin E. Roth, Neumann and Oskar Morgenstern, Theory of
and Michael A. Rees. 2011a. “Nonsimultaneous Games and Economic Behavior.” Bulletin of the
Chains and Dominos in Kidney-Paired Donation— American Mathematical Society 51(7): 498–504.
Revisited.” American Journal of Transplantation Day, Robert, and Paul Milgrom. 2008. “Core-
11(5): 984–94. Selecting Package Auctions.” International Journal
Ashlagi, Itai, Duncan S. Gilchrist, Alvin E. Roth, of Game Theory 36(3–4): 393–407.
and Michael A. Rees. 2011b. “NEAD Chains in Debreu, Gerard, and Herbert Scarf. 1963.
Transplantation.” American Journal of Transplanta- “A Limit Theorem on the Core of an Economy.”
tion 11(12): 2780–81. International Economic Review 4(3): 235–46.
Aumann, Robert J. 1964. “Markets with a Delmonico, Francis L., and Nancy L. Ascher.
Continuum of Traders.” Econometrica 32(1–2): 2017. Opposition to Irresponsible Global Kidney
39–50. Exchange. American Journal of Transplantation
Aumann, Robert J. 1976. “Agreeing to 17(10): 2745–46.
Disagree.” Annals of Statistics 4(6): 1236–39. Dubins, Lester E., and Daniel A. Friedman.
Aumann, Robert J. 1985. “On the Non-Transfer- 1981. “Machiavelli and the Gale–Shapley Algo-
able Utility Value: A Comment on the Roth–Shafer rithm.” American Mathematical Monthly 88(7):
Examples.” Econometrica 53(3): 667–78. 485–94.
Aumann, Robert J. 1986. “On the Non-Transfer- Edgeworth, Francis Ysidro. 1881. Mathematical
able Utility Value: Rejoinder.” Econometrica 54(4): Psychics: An Essay on the Application of Mathematics to
985–89. the Moral Sciences.
Aumann, Robert J. 2000. Collected Papers, vol. 2. Engelbrecht-Wiggans, Richard. 1988. “An
Cambridge, MA: MIT Press. Example of Auction Design: A Theoretical Basis
Aumann, Robert J., and Michael Maschler. 1964. for 19th Century Modifications to the Port of
“The Bargaining Set for Cooperative Games.” New York Imported Goods Market.” Unpublished
Chap. 21 in Advances in Game Theory, edited by M. paper, University of Illinois.
Dresher, L. S. Shapley, and A. W. Tucker. Annals Fudenberg, Drew, and Jean Tirole. 1991. Game
of Mathematics Studies 52. Princeton University Theory. MIT Press.
Press. Gale, David, and Lloyd S. Shapley. 1962.
Avery, Christopher, Christine Jolls, Richard A. “College Admissions and the Stability of Marriage.”
Posner, and Alvin E. Roth. 2001. “The Market for American Mathematical Monthly 69(1): 9–15.
Federal Judicial Law Clerks.” University of Chicago Govindan, Srihari, and Robert Wilson. 2012.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
“Axiomatic Equilibrium Selection for Generic Two- Luce, R. Duncan, and Howard Raiffa. 1957.
Player Games.” Econometrica 80(4): 1639–99. Games and Decisions: Introduction and Critical Survey.
Harsanyi, John C. 1967–68. “Games with Incom- New York: John Wiley and Sons.
plete Information Played by ‘Bayesian’ Players.” Marschak, Jacob. 1946. “Neumann’s and
Parts I–III. Management Science, 14 (3, 5, 7): 159–82, Morgenstern’s New Approach to Static Economics.”
320–34, 486–502. Journal of Political Economy 54(2): 97–115.
Hayek, F. A. 1945. “The Use of Knowledge in Maschler, Michael. 1992. “The Bargaining Set,
Society.” American Economic Review 35(4): 519–30. Kernel, and Nucleolus.” Chap. 18 in Handbook of
Holmstrom, Bengt. 1977. On Incentives and Game Theory with Economic Applications, vol. 1, edited
Control in Organizations. PhD dissertation, Stanford by Robert J. Aumann and Sergiu Hart. Elsevier.
University. McKinney, C. Nicholas, Muriel Niederle, and
Hurwicz, Leonid. 1945. “The Theory of Economic Alvin E. Roth. 2005. “The Collapse of a Medical
Behavior.” American Economic Review 35(5): 909–25. Labor Clearinghouse (and Why Such Failures Are
Hurwicz, Leonid. 1973. “The Design of Mecha- Rare).” American Economic Review 95(3): 878–89.
nisms for Resource Allocation.” American Economic Mertens, Jean-Francois. 1989. “Stable Equi-
Review 63(2): 1–30. libria—A Reformulation: Part I. Definition and Basic
Kagel, John H., and Alvin E. Roth. 2000. “The Properties.” Mathematics of Operations Research 14(4):
Dynamics of Reorganization in Matching Markets: 575–625.
A Laboratory Experiment Motivated by a Natural Milgrom, Paul. 2004. Putting Auction Theory to
Experiment.” Quarterly Journal of Economics 115(1): Work. Cambridge, MA: Cambridge University Press.
201–35. Milgrom, Paul. 2017. Discovering Prices: Auction
Karlin, Samuel. 1959. Mathematical Methods and Design in Markets with Complex Constraints. New York:
Theory in Games, Programming, and Economics, 2 vols. Columbia University Press.
New York: Wiley. Milgrom, Paul, and Joshua Mollner. 2018.
Kelso, Alexander S., and Vincent P. Crawford. “Equilibrium Selection in Auctions and High Stakes
1982. “Job Matching, Coalition Formation, and Games.” Econometrica 86(1): 219–61.
Gross Substitutes.” Econometrica 50(6): 1483–1504. Milgrom, Paul R., and Robert J. Weber. 1982.
Klemperer, Paul. 2004. Auctions: Theory and “A Theory of Auctions and Competitive Bidding.”
Practice. Princeton University Press. Econometrica 50(5): 1089–1122.
Kreps, David M., and Robert Wilson. 1982. Mirrlees, James A. 1971. “An Exploration in the
“Sequential Equilibria.” Econometrica 50(4): 863–94. Theory of Optimum Income Taxation.” Review of
Kuhn, Harold W. 1950. “Extensive Games.” PNAS Economic Studies 38(2): 175–208.
36(10): 570–76. Mirrlees, James A. 1979. “The Implications
Kuhn, Harold W. 1953. “Extensive Games and the of Moral Hazard for Optimal Insurance.” Paper
Problem of Information.” Chap. 11 in Contributions presented at conference in honor of Karl Borch,
to the Theory of Games, vol. 1, edited by Harold W. Bergen, Norway, 1979.
Kuhn and Albert W. Tucker. Princeton University Myerson, Roger B. 1999. “Nash Equilibrium and
Press. the History of Economic Theory.” Journal of Economic
Leider, Stephen, and Alvin E. Roth. 2010. Literature 37(3): 1067–82.
“Kidneys for Sale: Who Disapproves, and Why?” Myerson, Roger B., and Jörgen Weibull. 2015.
American Journal of Transplantation 10(5): 1221–27. “Tenable Strategy Blocks and Settled Equilibria.”
Lemke, Carlton E., and J. T. Howson, Jr. 1964. Econometrica 83(3): 943–76.
“Equilibrium Points of Bimatrix Games.” Journal of Nash, John F., Jr. 1950a. “Equilibrium Points in
the Society for Industrial and Applied Mathematics 12(2): n-Person Games.” PNAS 36(1): 48–49.
413–23. Nash, John F., Jr. 1950b. “The Bargaining
Lewis, David. 1969. Convention: A Philosophical Problem.” Econometrica 18(2): 155–62.
Study. Cambridge, MA: Harvard University Press. Nash, John F., Jr. 1951. “Noncooperative Games.”
Leyton-Brown, Kevin, Paul Milgrom, and Ilya Annals of Mathematics 54: 289–95.
Segal. 2017. “Economics and Computer Science Niederle, Muriel, Deborah D. Proctor, and Alvin
of a Radio Spectrum Reallocation.” PNAS 114(28): E. Roth. 2006. “What Will Be Needed for the New
7202–09. GI Fellowship Match to Succeed?” Gastroenterology
Li, Shengwu. 2017. “Obviously Strategy-Proof 130(1): 218–24.
Mechanisms.” American Economic Review 107(11): Niederle, Muriel, Deborah D. Proctor, and Alvin
3257–87. E. Roth. 2008. “The Gastroenterology Fellowship
Lucas, William F. 1969. “The Proof that a Game Match—The First Two Years.” Gastroenterology
May Not Have a Solution.” Transactions of the 135(2): 344–46.
American Mathematical Society 137: 219–29. Niederle, Muriel, and Alvin E. Roth. 2003.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
“Unraveling Reduces Mobility in a Labor Market: of Empirical Economics.” Economic Journal 101:
Gastroenterology With and Without a Centralized 107–114.
Match.” Journal of Political Economy 111(6): 1342–52. Roth, Alvin E. 1991b. “A Natural Experiment
Niederle, Muriel, and Alvin E. Roth. 2009. in the Organization of Entry-Level Labor Markets:
“Market Culture: How Rules Governing Exploding Regional Markets for New Physicians and Surgeons
Offers Affect Market Performance.” American in the United Kingdom.” American Economic Review
Economic Journal: Microeconomics 1(2): 199–219. 81(3): 415–40.
Ortega-Reichert, Armando. 1969. Models for Roth, Alvin E. 2007. “Repugnance as a Constraint
Competitive Bidding Under Uncertainty. PhD disserta- on Markets.” Journal of Economic Perspectives 21(3):
tion, Stanford University. 37–58.
Owen, Guillermo. 1968. Game Theory. Academic Roth, Alvin E. 2016. “Experiments in Market
Press. Design.” Chap. 5 in Handbook of Experimental
Rees, Michael A., Ty B. Dunn, Christian S. Kuhr, Economics, vol. 2, edited by John H. Kagel and Alvin
Christopher L. Marsh, Jeffrey Rogers, Susan E. E. Roth. Princeton University Press.
Rees, Alejandra Cicero, Laurie J. Reece, Alvin E. Roth, Alvin E., and Ido Erev. 1995. “Learning
Roth, Obi Ekwenna, David E. Fumo, Kimberly D. in Extensive-Form Games: Experimental Data and
Krawiec, Jonathan E. Kopke, Samay Jain, Miguel Simple Dynamic Models in the Intermediate Term.”
Tan, and Siegfredo R. Paloyo. 2017. “Kidney Games and Economic Behavior 8: 164–212.
Exchange to Overcome Financial Barriers to Kidney Roth, Alvin E., Kimberly D. Krawiec, Siegfredo
Transplantation.” American Journal of Transplantation Paloyo, Obi Ekwenna, Christopher L. Marsh,
17(3): 782–90. Alexandria J. Wenig, Ty B. Dunn, and Michael A.
Rees, Michael A., Jonathan E. Kopke, Ronald P. Rees. 2017. “People Should Not Be Banned from
Pelletier, Dorry L. Segev, Matthew E. Rutter, Alfredo Transplantation Only Because of Their Country of
J. Fabrega, Jeffrey Rogers, Oleh G. Pankewycz, Origin.” American Journal of Transplantation 17(10):
Janet Hiller, Alvin E. Roth, Tuomas Sandholm, Utku 2747–48.
Ünver, and Robert A. Montgomery. 2009. “A Non- Roth, Alvin E., and Elliott Peranson. 1999. “The
Simultaneous Extended Altruistic Donor Chain.” Redesign of the Matching Market for American
New England Journal of Medicine 360(11): 1096–1101. Physicians: Some Engineering Aspects of Economic
Rees, Michael A, Siegfredo R. Paloyo, Alvin E. Design.” American Economic Review 89(4): 748–80.
Roth, Kimberly D. Krawiec, Obi Ekwenna, Chris- Roth, Alvin E., Tayfun Sönmez, and M. Utku
topher L. Marsh, A. J. Wenig, Ty B. Dunn. 2017. Ünver. 2004. “Kidney Exchange.” Quarterly Journal of
“Global Kidney Exchange: Financially Incompatible Economics 119(2): 457–88.
Pairs Are Not Transplantable Compatible Pairs.” Roth, Alvin E., Tayfun Sönmez, and M. Utku
American Journal of Transplantation 17(10): 2743–44. Ünver. 2005a. “A Kidney Exchange Clearinghouse
Roth, Alvin E. 1975. “A Lattice Fixed-Point in New England.” American Economic Review 95(2):
Theorem with Constraints.” Bulletin of the American 376–80.
Mathematical Society 81(1): 136–38. Roth, Alvin E., Tayfun Sönmez and M. Utku
Roth, Alvin E. 1976. “Subsolutions and the Ünver. 2005b. “Pairwise Kidney Exchange.” Journal
Supercore of Cooperative Games.” Mathematics of of Economic Theory 125(2): 151–188.
Operations Research 1(1): 43–49. Roth, Alvin E., Tayfun Sönmez, M. Utku Ünver,
Roth, Alvin E. 1980. “Values for Games without Francis L. Delmonico, and Susan L. Saidman. 2006.
Sidepayments: Some Difficulties with Current “Utilizing List Exchange and Nondirected Donation
Concepts.” Econometrica 48(2): 457–65. through ‘Chain’ Paired Kidney Donations.” American
Roth, Alvin E. 1982. “The Economics of Journal of Transplantation 6(11): 2694–2705.
Matching: Stability and Incentives.” Mathematics of Roth, Alvin E., and Xiaolin Xing. 1994. “Jumping
Operations Research 7(4): 617–28. the Gun: Imperfections and Institutions Related
Roth, Alvin E. 1984. “The Evolution of the Labor to the Timing of Market Transactions.” American
Market for Medical Interns and Residents: A Case Economic Review 84(4): 992–1044.
Study in Game Theory.” Journal of Political Economy Rothschild, Michael, and Joseph Stiglitz. 1976.
92(6): 991–1016. “Equilibrium in Competitive Insurance Markets: An
Roth, Alvin E. 1986. “On the Non-Transferable Essay on the Economics of Imperfect Information.”
Utility Value: A Reply to Aumann.” Econometrica Quarterly Journal of Economics 90(4): 629–49.
54(4): 981–84. Schelling, Thomas C. 1960. The Strategy of Conflict.
Roth, Alvin E., ed. 1988. The Shapley Value: Essays Harvard University Press.
in Honor of Lloyd S. Shapley. Cambridge University Selten, Reinhard. 1965. “Spieltheoretische
Press. Behandlung eines Oligopolmodells mit Nachfrag-
Roth, Alvin E. 1991a. “Game Theory as a Part eträgheit.” Zeitschrift für die gesamte Staatswissenschaft

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00
121: 301–29, 667–89. Vickrey, William. 1961. “Counterspeculation,

Selten, Reinhard. 1975. “Reexamination of Auctions, and Competitive Sealed Tenders.”
the Perfectness Concept for Equilibrium Points Journal of Finance 16(1): 8–37.
in Extensive Games.” International Journal of Game von Neumann, John, and Oskar Morgenstern.
Theory 4(1): 25–55. 1944. Theory of Games and Economic Behavior.
Shapley, Lloyd S. 1953. “A Value for n-Person (2nd edition, 1947; 3rd edition, 1953). Princeton
Games.” Chap. 17 in Contributions to the Theory of University Press.
Games, vol. 2, edited by Harold W. Kuhn and Albert Wald, Abraham. 1947. Review of “Theory
W. Tucker. Princeton University Press. of Games and Economic Behavior” by John v.
Shapley, Lloyd S. 1969. “Utility Comparison and Neumann and Oskar Morgenstern. Review of
the Theory of Games.” In La Décision: Agrégation Economics and Statistics 29(1): 47–52.
et Dynamique des Ordres de Préférence, pp. 251–263. Wilson, Robert B. 1977. “A Bidding Model of
Paris: Editions du CNRS. Perfect Competition.” Review of Economic Studies
Shapley, Lloyd S. 1973. “Let’s Block ‘Block.’” 44(3) 511–18.
Econometrica 41(6): 1201–02. Wilson, Robert B. 1987. “Game-Theoretic Anal-
Shapley, Lloyd, and Herbert Scarf. 1974. “On yses of Trading Processes.” In Advances in Economic
Cores and Indivisibility.” Journal of Mathematical Theory: Fifth World Congress, edited by Truman F.
Economics 1(1): 23–37. Bewley, pp. 33–77. Cambridge University Press.
Spence, Michael. 1973. “Job Market Signaling.” Wilson, Robert B. 1993. Nonlinear Pricing.
Quarterly Journal of Economics 87(3): 355–74. Oxford University Press.

130.60.206.40 on Fri, 05 Jan 2024 07:33:12 +00:00

Roth MarketDesignEmerged 2019

Uploaded by

Copyright:

Available Formats

Roth MarketDesignEmerged 2019

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Roth MarketDesignEmerged 2019

Uploaded by

Copyright:

Available Formats

American Economic Association

How Market Design Emerged from Game Theory

Stable URL: https://www.jstor.org/stable/10.2307/26732324

This content downloaded from

How Market Design Emerged from

Alvin E. Roth and Robert B. Wilson

This content downloaded from

examining different ways in which marketplaces operate within their economic

This content downloaded from

Wilson: As an MBA student in 1960, I wrote a class report on how to bid in an

This content downloaded from

available to particular players as a consequence of actions taken earlier in the tree,

This content downloaded from

This content downloaded from

compatibility as key to unifying classical and game-theoretic analyses. Familiar

This content downloaded from

However, their idea of domination proved quite useful, in coalitional games of

This content downloaded from

This content downloaded from

This content downloaded from

This content downloaded from

Wilson: This perspective is analogous to one in the older literature on general

This content downloaded from

participation was mandated by a monopolist seller, such as government auctions of

This content downloaded from

humans might find difficult,18 particularly in the absence of common knowledge

Why Were Auctions and Two-Sided Matching Markets Such Fertile

This content downloaded from

This content downloaded from

Mechanism design theory usually imposes an “individual rationality” constraint

Roth: In contrast to auctions, labor markets, kidney transplants, and other

This content downloaded from

This content downloaded from

yield an efficient allocation, but alternative designs forego efficiency to maximize

Roth: How about in electricity markets?

Roth: And are aspects of electricity sales repugnant?

This content downloaded from

How Was Your Work Influenced by Your Teachers, Colleagues, and

This content downloaded from

This content downloaded from

How Have Other Domains of Market Design Developed?

This content downloaded from

that employ centralized marketplaces may be preceded or followed by decentral-

Wilson: A remarkable application is the conversion in the United States of

What’s Next for Game Theory and Market Design?

Wilson: The ongoing computerization of marketplaces will continue to make

This content downloaded from

Roth: Smartphones have put marketplaces in our pockets, and as computer-

What’s Your Last Word for Now?

Roth: Market design is an outward-facing part of economics, so designers have

This content downloaded from

95(2): 368–71. Law Review 68(3): 793–902.

This content downloaded from

This content downloaded from

This content downloaded from

121: 301–29, 667–89. Vickrey, William. 1961. “Counterspeculation,

This content downloaded from

You might also like