1. Introduction
This paper shows fragility of stable matchings in decentralized matching markets. Most of
the existing literature focuses on fully centralized markets, in which a central clearinghouse
matches agents by using their submitted lists of preferences. Nonetheless, many real-life
matching markets are decentralized or not fully centralized, e.g. college admissions, markets
for junior economists, marriage markets, and so on.
Furthermore, decentralized interactions often precede or follow centralized markets. In
particular, unstable clearinghouses (e.g., the Boston school-choice mechanism, see Abdulka-
diroğlu and Sönmez, 2003) might lead to unstable outcomes, and therefore create rematching
opportunities. However, even for stable centralized clearinghouses (including the Deferred
Acceptance mechanism of Gale and Shapley, 1962), unstable outcomes might emerge due to
ex-post preferences shocks, changes in market composition, small implementation errors, or
ex-ante incomplete information (Fernandez, Rudov, and Yariv, 2022).
Stability is a central solution concept for two-sided matching markets.
1
A matching is
stable if there is no pair of agents who prefer each other to their current partners; such
a pair is usually called a blocking pair. In practice, centralized mechanisms often seek to
implement stable outcomes. One of the main reasons is that stable systems often succeed,
while unstable ones collapse through pre- or post-market decentralized interactions (Roth
and Xing, 1994; Roth, 2002; McKinney, Niederle, and Roth, 2005).
It is commonly believed that if an unstable outcome is realized, decentralized interactions
eventually lead to stability. Theoretically, Roth and Vande Vate (1990) proved that in the
one-to-one matching problem, starting from any unstable matching, there is a sequence of
blocking pairs that can be iteratively satisfied to get a stable matching.
2
As a corollary, any
1
See Roth and Sotomayor (1992) for an early survey of the matching literature.
2
Similar results were later obtained in alternative settings, e.g. the roommate problem (Chung, 2000;
Diamantoudi, Miyagawa, and Xue, 2004; Inarra, Larrea, and Molis, 2008), the many-to-one matching problem
with couples (Klaus and Klijn, 2007), the many-to-many matching problem without contracts (Kojima and
Ünver, 2008) and with contracts (Millán and Pepa Risma, 2018), and matching markets with incomplete
information (Lazarova and Dimitrov, 2017; Chen and Hu, 2020).
2