Centralised exchanges & proof-of-solvency: The guardians of trust

David Vidal-Tomása,b,c,
a Department of Computer Science, University College London, Gower Street, WC1E 6EA - London, UK.
b UCL Centre for Blockchain Technologies, London, United Kingdom
c Department of Economics, Universitat Jaume I, Campus del Riu Sec, 12071 - Castellón, Spain.

Abstract

This paper provides (i) a review of existing literature on proof-of-solvency concepts, including proof-of-assets, proof-of-

liabilities and proof-of-reserves and (ii) an empirical analysis of proof-of-assets using data disclosed by leading centralized

exchanges. From an empirical perspective, our contribution to the literature includes three key dimensions: (i) the eval-

uation of the exchange asset composition, (ii) an examination of centralized exchanges’ resilience during extreme events

that occurred in 2022, and (iii) an exploration of the significance of impersonal trust, within the framework proposed

by Shapiro (1987). On the one hand, concerning exchange asset composition, we observed a high heterogeneity among

centralized exchanges based on their nature, distinguishing between multi-asset and Bitcoin-only exchanges. On the

other hand, analysing centralized exchanges’ stability: the FTX bankruptcy event had the most detrimental impact on

all centralized exchanges, while Terra-Luna and Three Arrows Capital collapses were not so relevant, underscoring the

robustness of centralised exchanges in the face of external shocks; additionally, stablecoins acted as a double-edged sword

since they improved the resilience of the exchanges, however, if a large stablecoin were to collapse, the exchange would

immediately go into bankruptcy. Finally, we observed that centralised exchanges are characterised by the emergence of a

complex chain of impersonal trust, i.e. users must trust not only the exchange itself and its management, but also any

related crypto entity. Within Shapiro’s context, to create a trustworthy and reliable environment for participants in the

cryptocurrency market, exchanges and policymakers will need to establish even more “guardians of trust”, in the form

of new regulations or organizational forms. Paradoxically, the cryptocurrency ecosystem, which philosophically aimed to

minimize the reliance on trust, now requires even more “guardians of trust” than traditional finance.

JEL codes: G10 · G11 · G40

Keywords: Centralised exchanges · Proof-of-solvency · Proof-of-reserves · Cryptocurrency · Stablecoins

∗ Correspondingauthor
Email address: [email protected] // [email protected] (David Vidal-Tomás)

Preprint submitted to Elsevier February 16, 2024

1. Introduction

In an era of increasingly complex economic systems and digital transformation (Murinde et al., 2022), individuals lack

the ability to personally control trust relationships that are not rooted in personal connections. Indeed, people walk into

an unfamiliar branch office of their bank or log into the mobile banking application to deposit money without having

any additional insight into where and how that money will ultimately be utilized. Consequently, nowadays trust is not

based on direct personal contact but mediated by a social organization or structure, i.e. impersonal trust (Shapiro, 1987;

Williamson, 1993).

According to the seminal paper written by Shapiro (1987), impersonal trust develops in situations where social-control

measures derived from social ties and direct interactions between the “principal” (trustor) and “agent” (trustee) are

unavailable. It occurs when faceless and easily interchangeable individual or organizational agents (e.g., banks, insurance

companies, brokerage firms and government agencies) exercise substantial delegated power and privilege on behalf of

principals who are unable to specify, scrutinize, evaluate, or restrict their performance. Therefore, despite interacting

with strangers and being vulnerable, people do expose their lives and wealth to significant risks due to their impersonal

trust on the trustees or “guardians of trust”, who are represented by a supporting social-control framework of procedural

norms, organizational forms, and social-control specialists (Zucker, 1986).

Paradoxically, Shapiro (1987) underlined that in their efforts to ensure agents fidelity, guardians created new issues.

This occurs because the various rules, limitations, entry requirements, monitoring methods, specialists in social control,

and insurance-like arrangements ultimately not only increased opportunities for misuse but also encouraged trustees to

behave in less desirable ways. The most notable example can be seen in the misbehaviour of lenders, investment banks,

and credit rating agencies during the 2007 subprime mortgage crisis (Ryan, 2008; Demyanyk and Van Hemert, 2011).1

The subsequent Great Recession was a wake-up call about the essential role well-functioning trustees play in economic

growth, and also emphasized how impersonal trust is crucial for agents involved in the economy (Fungáčová et al., 2019).

In response to the crisis, authorities took steps with a battery of measures to protect consumers and maintain financial

stability in the future (Gertler and Gilchrist, 2018). The most impactful initiative was the Dodd-Frank Wall Street Reform

and Consumer Protection Act (Congress, 2010), which aimed to address the sectors of the financial system that were

responsible for the financial crisis.2 Both the components of the Dodd-Frank Act and other international frameworks, like

Basel III, shared a common ultimate goal: to maintain society’s trust in the financial system.

However, while international standards for bank capital adequacy, stress testing, and liquidity requirements were

discussed by policymakers to restore lost confidence, a completely different approach was considered to address the

aftermath of the 2008 financial crisis: Nakamoto (2008) introduced the concept of blockchain through Bitcoin (BTC). This

technology proposed a novel model for coordinating and governing economic interactions between parties. Essentially,

1 Lenders provided loans to individuals with poor credit and a high risk of default. Investment banks purchased these mortgages from lenders,

securitized them into bonds, and subsequently sold them to investors via collateralized debt obligations (CDOs). Credit rating agencies should
have assigned to these CDOs much lower ratings than the AAA-rating, given to the higher quality tranches (Christiano et al., 2015).
2 The legislation encompassed numerous provisions, spanning a total of 848 pages. For instance, The Financial Stability Oversight Council

and the Orderly Liquidation Authority were established to oversee the financial stability of significant financial institutions considered “too
big to fail”. The Consumer Financial Protection Bureau was founded to combat predatory mortgage lending. The Volcker Rule imposed
limitations on banks’ investment activities, and the Office of Credit Ratings was tasked with ensuring the reliability of credit ratings.

2
blockchain technology allowed for the creation of innovative exchange networks that eliminate the necessity for trust

within the digital ecosystem when conducting financial transactions. As stated by Nakamoto (2008) “what is needed is

an electronic payment system based on cryptographic proof instead of trust, allowing any two willing parties to transact

directly with each other without the need for a trusted third party”. Consequently, blockchain allowed participants in

digital ecosystems to engage without requiring a trusted third party or a central institution (like a traditional bank) to

oversee and secure digital relationships. Indeed, it facilitated the exchange of value and the transfer of ownership within

an environment that lacks inherent trust on trustees.3 In this context, policymakers aimed to rebuild trust in the financial

system through new legislation, while Nakamoto (2008) introduced a novel approach to dismantle the trust element.

Compared with traditional banks that act as intermediaries and agents of economic trust between exchange parties,

blockchain system similarly represents impersonal trust in the exchange relationship. In essence, blockchain technology

itself represents a guardian of institutionalized impersonal trust as blockchain technology hardly needs any social-control

strategies, such as internal social control, private surveillants, or even industry self-regulation for guarding trust in these

relationships (Tan and Saraniemi, 2023). However, the recent growth in blockchain-based cryptocurrency ecosystem may

ultimately lead to similar trust issues as those found in traditional financial systems.

As described in Aramonte et al. (2021) and Financial Stability Board (2023), the blockchain ecosystem, while aiming to

mimic aspects of the traditional financial system, inherit and exacerbate its well-known vulnerabilities. These vulnerabil-

ities include familiar issues, such as operational fragilities, liquidity, maturity mismatches, leverage, mismanagement and

increasing interconnectedness. The systemic dependence is clearly represented in the current blockchain ecosystem, which

relies largely on different infrastructures, such as existing blockchain networks, centralised and decentralised crypto-asset

exchanges, oracles, bridges, decentralised autonomous organisations (DAOs), lending/borrowing companies and stable-

coins, among other third-party technology providers (Financial Stability Board, 2023). Consequently, similarly to the

2007 subprime mortgage crisis, the failure in a single element could generate adverse spillovers that can further propagate

in the system. As a manifestation of this risk, the Terra-Luna collapse in June 2022 (Briola et al., 2023; Liu et al., 2023)

extended its impact well beyond the closely associated Decentralised Finance (DeFi) lending protocol, Anchor. It trig-

gered a domino effect of bankruptcies within the cryptocurrency sphere and decreased the trust in the system, giving rise

to the crypto-winter (Clarke, 2022). Moreover, even though the blockchain ecosystem aims to provide financial services

without traditional intermediaries, using automated protocols on blockchains and stablecoins to facilitate fund transfers,

the system exhibits a “decentralization illusion” since it unavoidably relies on centralized governance and tends to cen-

tralize power via blockchain consensus mechanisms (Aramonte et al., 2021).4 In the same line, most of the centralised

exchanges operate in non-compliance with or outside of supervisory and regulatory frameworks, raising the chances of

concentrated ownership, inadequate custody arrangements, liquidity issues, the potential for price manipulation, fraud,

and other forms of misconduct. The mismanagement of customer assets in the case of FTX exchange (Vidal-Tomás et al.,

3 For example, in the context of decentralized finance, the parties involved in exchanges, like individuals seeking to borrow or lend money,

typically remain unaware of each other’s identities. Thus, principals must rely solely on blockchain technology as their primary point of
interaction and trust.
4 According to Financial Stability Board (2023), the percentage of DeFi governance tokens supply owned by top 100 addresses, as of June

2022, in Uniswap, PancakeSwap, Aave, Compound and Ampleforth was 86.74%, 94.64%, 85.89%, 91.46% and 92.19%, respectively.

3
2023; Conlon et al., 2023, 2024), the third-largest centralised exchange in November 2022, exemplified these governance

issues. As stated by Ray III (2022), “Never in my career have I seen such a complete failure of corporate controls and such

a complete absence of trustworthy financial information as occurred here”. As a result, the growing interconnections and

the concentration of power in the crypto ecosystem ultimately result in the same fundamental issue within the traditional

financial system, i.e. the problem of trust.

The issue of mismanagement in many centralised exchanges is especially a significant concern for the future of the

blockchain ecosystem. Centralised exchanges, known for their user-friendly interfaces, play a crucial role in making the

crypto ecosystem and DeFi more accessible to a broader range of investors and improving product liquidity (Financial

Stability Board, 2023). These exchanges also engage in activities that go beyond traditional exchange operations and often

form connections with various counterparties in the crypto-asset/DeFi ecosystem, such as through lending or investments

in DeFi protocols (Bank for International Settlements, 2023). Consequently, centralised exchanges serve as the primary

entry point for investors and users in the blockchain space, acting as a bridge to DeFi protocols.5 In other words,

they can be seen as the first “guardians of trust” from the customer/user perspective, in the form of organisational

structures. The lack of trust in these platforms would impede the growth of the crypto-ecosystem. Indeed, the Securities

and Exchange Commission (SEC) increased uncertainty regarding the future of centralised exchanges by filing 13 charges

against Binance, the world’s largest crypto exchange, and its founder, Changpeng Zhao (Goswami, 2023).6 The problems

related to the transparency of centralised exchanges worsened when Binance Holdings agreed in 2023 to pay $4.3 billion

to settle with the Department of Justice and Commodity Futures Trading Commission over illicit financial breaches, and

Zhao pleaded guilty to breaking U.S. anti-money laundering laws. However, the SEC’s case is still pending against the

exchange (Prentice et al., 2024).

Within Shapiro’s context, the bankruptcy cascade of various crypto entities and the growing uncertainty surrounding

centralised exchanges forced these platforms to advocate for additional “guardians of trust”, in the form of procedural

norms and blockchain implementations, which should ensure consumer protection, increase transparency and prevent

malicious behavior. These safeguards are referred to as proof-of-assets, proof-of-liabilities, proof-of-reserves, and proof-of-

solvency. While these measures are, indeed, not new (Wilcox, 2014), the majority of centralised exchanges have adopted

them, after the FTX’s collapse, to maintain customer trust and prevent the loss of deposits. In short, proof-of-assets

(Pines et al., 2022) and proof-of-liabilities (Ji and Chalkias, 2021) are methods that use the unchangeable nature of a

public blockchain and inherent cryptographic mechanisms to establish evidence of the existence and control of digital

assets -and corresponding liabilities as costumers’ deposits- held by centralized entities on behalf of their clients. Proof-of-

reserves represents the combination of proof-of-assets and liabilities. It denotes a particular process in which a custodian

openly confirms the presence of on-chain assets and subsequently offers corresponding evidence, assisted by an auditor, to

ensure that the outstanding liabilities do not exceed those assets (Buterin, 2022; Buxton et al., 2022). Proof-of-solvency,

5 Proof of this, at the time of writing, 75% of the crypto transactions involve centralised exchanges, while only the 15% are related to

decentralised exchanges (The Block, 2023).

6 The allegations include the commingling of billions of dollars worth of user funds and their transfer to a European company controlled by

Zhao.

4
otherwise, encompasses a broader concept that takes into account the company’s liabilities beyond the distributed ledger,

as well as any other factors that could impact its ability to operate as a going concern in the future (Buxton et al., 2022),

i.e. it includes proof-of-reserves and any relevant off-chain factor. Interestingly, a definitive consensus is currently lacking

regarding the precise definitions and mechanisms of the aforementioned terms, even though some authors have already

discussed about them (e.g., Ji and Chalkias, 2021; Pines et al., 2022; Buterin, 2022; Buxton et al., 2022).

In this framework, our contributions to the literature will unfold in three aspects. On the one hand, we provide a

brief literature review, coupled with in-depth discussions on the subjects of proof-of-assets, proof-of-liabilities, proof-of-

reserves, and proof-of-solvency. This exposition will additionally include an explication of the evolutionary trajectory

of proof-of-reserves, commencing with the proposition of the Maxwell/Todd Merkle approach (Wilcox, 2014) for sub-

stantiating liabilities. On the other hand, using the available data disclosed by the most relevant centralized exchanges

(Binance, Bitfinex, BitMEX, Bybit, CheckSig, Crypto.com, Deribit, Huobi, KuCoin, OKX, and SwissBorg) and reported

by Glassnode, we conduct an empirical analysis of proof-of-assets. This analysis aims to scrutinize the asset composition

of these exchanges and explore its potential implications for the future of the crypto ecosystem. Additionally, we evaluate

the stability of centralized exchanges in the face of extreme events. This evaluation involves (i) an event study focused

on occurrences in 2022, such as the collapses of Terra-Luna and FTX, utilizing an AR-GARCH framework, and (ii) a

Value at Risk (VaR) assessment to quantify the downside risk associated with these exchanges. This empirical framework

also holds relevance for policymakers, providing insights to establish capital buffers that ensure a more resilient global

crypto system. Lastly, we engage in a discussion regarding the implications of our study with respect to proof-of-solvency,

considering the philosophical framework outlined by Shapiro (1987) concerning the “guardians of trust”.

From an empirical perspective, regarding exchange asset composition, our findings revealed (i) a significant hetero-

geneity among centralized exchanges based on their nature, distinguishing between multi-asset and BTC-only exchanges,

(ii) the increasing relevance of stablecoins since 2021, with some exchanges holding over 50% of total customer assets in

stablecoins, (iii) Binance emerging as the favourite platform for users trading in BTC, Ethereum (ETH), and stablecoins.

On the other hand, analysing centralized exchanges’ stability: (i) the FTX bankruptcy event had the most adverse impact

on all centralized exchanges, while Terra-Luna and Three Arrows Capital collapses were less significant, underlining the

resilience of centralised exchanges in the face of external shocks, (ii) stablecoins acted as a double-edged sword, given

that they improved the resilience of the exchanges; however, if a large stablecoin were to collapse, the exchange would

immediately face bankruptcy, (iii) exchanges should maintain a capital conservation buffer ranging from 6% to 14% to

face negative events. Finally, we observed that centralised exchanges are characterised by the emergence of a complex

chain of impersonal trust. When users select a specific centralized exchange, they must trust multiple layers of entities

and components. This includes trust in the exchange itself and its management, but it extends far beyond that, encom-

passing any other crypto entity associated with the exchange. Within Shapiro (1987) context, establishing a trustworthy

environment for participants in the cryptocurrency market will require the implementation of even more “guardians of

trust”, through new regulations or organizational structures. Paradoxically, while the cryptocurrency ecosystem initially

aimed to minimize reliance on trust, it now requires even more “guardians of trust” than traditional finance.

5
2. Literature review on proof-of-solvency mechanisms

In recent years, exchanges have been working towards establishing solvency proofs to build trust among users. This

also involves minimizing the disclosure of additional information that could compromise client data, thereby preserving

the characteristics of “(pseudo)-anonymity”. Unavoidably, in response to the bankruptcy cascades during 2022, we are

now observing a noticeable increase in the demand for standardizing proof-of-solvency within the digital assets industry,

which also includes proof-of-assets, proof-of-liabilities, and therefore, proof-of-reserves (Chalkias et al., 2020; Buxton et al.,

2022). Nevertheless, these novel technologies are currently undergoing scrutiny in terms of their concepts, mechanisms,

security assurances, and specific considerations related to potential weaknesses in their implementation (Chalkias et al.,

2022). Given this fact, within this section, we briefly discuss (i) the primary papers addressing the conceptualizations and

mechanisms behind the aforementioned terms, in Sec.(2.1), and (ii) their historical evolution leading up to the present,

in Sec.(2.2).

2.1. Conceptualization

2.1.1. Proof-of-reserves: proof-of-assets & proof-of-liabilities

Acting as “guardians of trust”, centralised exchanges facilitate user deposits through methods such as bank transfers

or other conventional means. These exchanges retain users’ private keys and offer authentication through usernames

and passwords to verify the customer’s identity. Additionally, they provide recovery options in case customers forget or

lose their authentication details. Consequently, users are relieved from the responsibility of storing private keys for their

cryptocurrencies in exchange for a user-friendly interface that facilitates cryptocurrency trading. However, as previously

mentioned, there is a potential risk of loss of customer assets managed by exchanges.

The platform’s digital assets, which are based on blockchain technology, are relatively easy to verify in terms of

quantity and ownership. To report proof-of-assets, the exchange simply has to consistently disclose the complete list of

digital asset addresses under its control, mainly, cold storage wallet addresses. As a result, any external party can track

the balances of coins and tokens within these addresses reported by the exchange by aggregating the amounts visible

at a specific block height (Pines et al., 2022). More specifically, the existence and ownership of digital assets can be

verified through cryptographic signatures, send to self transactions, hierarchical deterministic wallets, multi-signature

wallets and/or secure multi-party computation.7 Given these facts, any user can find data providers that report the most

accurate estimation of assets held by centralized exchanges, including platforms like Glassnode and Nansen, among others.

Similarly, researchers have the ability to independently determine the quantity of tokens held by exchanges, as outlined in

Saggese et al. (2023). In this study, authors extract on-chain information for BTC and ETH using established clustering

heuristics (Meiklejohn et al., 2013) to link addresses controlled by the same cryptocurrency entity.

7 As an additional safeguard, external auditors should authenticate cryptographic signatures representing the overall balance of customer

assets. Third-party assurance reports on the internal control environments for financial reporting and IT systems within digital asset platforms
do provide users with a high level of assurance. Their duty is to verify through a formal attestation that the custodian is genuinely holding
the specified amount of assets (Wade, 2023).

6
 However, proof-of-assets is only one aspect of the equation, as stated by Carter (2022a). Simply verifying the existence

and ownership of a certain amount of assets held by exchanges is not enough on its own. In the interest of transparency and

customer protection, exchanges must (i) demonstrate existence and ownership of client assets and (ii) present outstanding

liabilities owed to clients, in the form of deposits (Carter, 2022b). The former refers to proof-of-assets, while the latter

constitutes proof-of-liabilities. Proof-of-reserves is the outcome obtained by applying both processes.

In contrast to demonstrating proof-of-assets, validating proof-of-liabilities is more challenging, primarily due to the

increasing prevalence of fraud and mismanagement. As illustrated by Buxton et al. (2022), if a digital asset platform has

incurred a loss or if management is engaged in fraudulent activities, there may be an inclination to under-report liabilities.

This tactic is employed to create the illusion that the digital asset platform is fully reserved. As a result, it is imperative

for both the scientific community and the industry to identify a method that ensures the inclusion of customers’ deposits

in the exchange balance. The primary objective of proof-of-liabilities is, indeed, to demonstrate the amount of funds a

centralised exchange owes to its customers, while preserving privacy.

Recently, decentralized solutions (Wilcox, 2014, Hu et al., 2019, Chalkias et al., 2019, Camacho, 2014, Dagher et al.,

2015, Chalkias et al., 2020, Ji and Chalkias, 2021, Buterin, 2022, Buterin et al., 2023) have been proposed as an alter-

native or supplementary approach to traditional auditing in the crypto space. These solutions involve customers actively

participating in the auditing process, reducing reliance on auditors that could collude with the exchange. Decentralized

auditing is considered more promising because it enables customers to verify their own balances, addressing a limitation

of centralized auditing. Existing schemes generally follow a similar principle: a prover aggregates user balances using an

accumulator, and consumers can then verify the inclusion of their balances in the reported total (Ji and Chalkias, 2021).

In 2013, BTC developer Greg Maxwell explored schemes for establishing proof-of-reserves, particularly discussing the

“merkleized approach” to address this issue. In the same discussion, Maxwell explained how aggregating hashed user

information in a Merkle tree could allow users on a digital asset platform to efficiently confirm their membership in the

set without exposing themselves to the entire contents of the liability set. Maxwell acknowledged that while this process

would not prevent fractional reserve or theft, it could discourage the covering of thefts, act as a preventive measure against

prolonged fraudulent activities and detect platforms that were insolvent for a long period of time.

Expanding on these discussions, in February 2014, Zak Wilcox (Wilcox, 2014) formalized the ideas proposed by BTC

developers Greg Maxwell and Peter Todd, with a specific focus on the proof-of-liabilities concept and a detailed exploration

of the Merkle approach. Wilcox (2014) proposed a summation Merkle tree construction to prove total liabilities. Merkle

trees serve as a data structure that allows a set owner to demonstrate the inclusion of an element in the set with efficiency.

In their construction, each participant is depicted by a leaf node within the tree, aligning with the squares in the bottom

row (e.g, 1, 2, 3, 4), as illustrated in Figure (1). The branches, or internal nodes (e.g., 5, 6), are utilized to reach the root,

integrating information from the leaf nodes while safeguarding privacy. In the Merkle tree, each node contains a hash

field (h) and a value field (v). The vroot denotes the total liabilities, representing the collective deposits of consumers

on the platform. Furthermore, v5 and v6 indicate the sums of values in their respective child nodes, specifically v1 − v2

and v3 − v4 , which reflect the deposits of individual customers. For the hash field, Maxwell-Todd includes the value field

7
Figure. 1: Example of Merkle tree following Chalkias et al. (2022).

Balance 𝑣𝑅𝑜𝑜𝑡 = 27

Hash ℎ𝑅𝑜𝑜𝑡 = 𝐻(27| ℎ5 | ℎ6 )

Balance 𝑣5 = 15 Balance 𝑣6 = 12

Hash ℎ5 = 𝐻(15| ℎ1 | ℎ2 ) Hash ℎ6 = 𝐻(12| ℎ3 | ℎ4 )

Balance 𝑣1 = 5 Balance 𝑣2 = 10 Balance 𝑣3 = 4 Balance 𝑣4 = 8

Hash ℎ1 = 𝐻(5| 𝐴𝑙𝑖𝑐𝑒| 𝑛𝑜𝑛𝑐𝑒1 ) Hash ℎ2 = 𝐻(10| 𝐵𝑜𝑏| 𝑛𝑜𝑛𝑐𝑒2 ) Hash ℎ3 = 𝐻(4| 𝐶𝑎𝑟𝑜𝑙| 𝑛𝑜𝑛𝑐𝑒3 ) Hash ℎ4 = 𝐻(8| 𝐷𝑎𝑣𝑒| 𝑛𝑜𝑛𝑐𝑒4 )

and the hashes of child/leaf nodes at hashing, e.g. h5 = H(15||h1 ||h2 ). For each leaf node mapped to a user, the hash

value is the SHA256 (a common cryptographic hashing algorithm) digest of the concatenated (i) users’ deposit v, (ii)

user information ui (e.g. Alice) and (iii) and a random “nonce”, i.e., h = H(v||ui||nonce).8 Through the application of a

hash summary function, the Merkle tree enables users to verify their inclusion in the complete tree. Users are unable to

reconstruct the entire tree from their partial information unless they acquire more than half of the total number of users

(Bybit, 2022).

Nevertheless, this scheme presents a relevant issue, as it allows the exchange to claim lower liabilities while still

generating inclusion proofs and effectively responding to user queries (Chalkias et al., 2022). Furthermore, the public

disclosure of the total liabilities value enables anyone to deduce both the population and individual liabilities through a

series of inclusion proofs. To address the issues presented by the original Merkle tree proposal, authors have introduced

supplementary schemes or modifications to offer exchanges and users more appropriate approaches. Readers can find a

comprehensive description of these schemes in Ji and Chalkias (2021).

Hu et al. (2019) [Maxwell+] addressed the vulnerability of underreporting liabilities by directly modifying each node

field to include both values and hashes of child nodes. This adjustment securely ties the value of each node to its

parent, preventing the manipulation of the tree. However, it is crucial to highlight that this scheme does not rectify

the privacy concerns. In the work presented by Chalkias et al. (2019) [Maxwell++], an extension of Maxwell+, the

concealment of both population and individual liabilities is achieved. This is accomplished through the fragmentation

and shuffling of values into smaller units. However, it is essential to note that despite these privacy enhancements, the

total liabilities of the exchange are still observable. In Camacho (2014) [Camacho], privacy of liabilities is achieved by

substituting the values in Maxwell-Todd with Pedersen commitments accompanied by zero-knowledge range proofs. The

hiding and binding properties of Pedersen commitments ensure privacy while preventing manipulation of liabilities, and

8 The nonce serves as a one-time random number employed as a privacy-preserving measure, akin to the use of salt in password encryption.

Its role is to guarantee that customers cannot deduce information about other nodes along their path to the Merkle root. The nonce should
only be accessible to both the digital asset platform and the respective customer.

8
their homomorphic properties allow summation in commitments. Total liabilities can only be disclosed through voluntary

commitment opening, or alternatively, the exchange can prove the range of total liabilities using zero-knowledge proofs. In

short, Camacho (2014) ensures the privacy of individual and total liabilities but does not extend this privacy protection to

the number of users. Additionally, it does not rectify the security flaw present in Maxwell-Todd’s summation Merkle tree.

In Chalkias et al. (2020) [DAPOL] and Ji and Chalkias (2021) [DAPOL+], enhancements to Camacho are introduced by

incorporating the proposed fix from Hu et al. (2019) and implementing sparse Merkle trees (SMT) for a more efficient

concealment of the number of users.

In Dagher et al. (2015) [Provisions], the protection of liability privacy is achieved through the utilization of homomor-

phic commitments and zero-knowledge proofs, omitting the use of a Merkle tree. The prover publishes commitments to

their liabilities for each user, accompanied by a range proof on the public bulletin board. Each user autonomously verifies

the validity of their commitment and associated range proofs. In this scheme, no information about customer holdings is

disclosed, and the total liabilities of the digital asset platform remain concealed and secure.

In a recent proposal by Buterin (2022), the utilization of ZK-SNARK technology is suggested to significantly enhance

and simplify privacy in proof-of-liabilities protocols. In this approach, all users’ deposits are incorporated into a Merkle

tree (or, alternatively, a KZG commitment), and a ZK-SNARK is employed to demonstrate that all balances in the tree

are non-negative and collectively amount to a claimed value. Adding a layer of hashing for privacy ensures that the Merkle

branch (or KZG proof) provided to each user discloses no information about the balance of any other user.

As discussed by Carter (2022b), at present, the prevailing approach used by most of the exchanges in proof-of-liabilities

involves the use of the Merkle proof method, where liabilities are revealed only on a per-client basis. This method opens up

more possibilities for concealing liabilities (see also Chalkias et al., 2022). However, future advancements may address this

limitation through next-generation schemes that leverage Zero Knowledge proofs, enabling the disclosure of the complete

liability set without compromising privacy.9

Irrespective of the chosen scheme, the involvement of auditors remains crucial in the proof-of-reserve process, partic-

ularly in the stages of proof-of-assets and proof-of-liabilities. In essence, the audit procedure should at least encompass

(i) an initial audit analysis and (ii) independent consumer verification (Bitpanda, 2022) conditional to the exchanges’

activities. The third-party auditor initiates the process by generating a snapshot of the crypto company’s liabilities,

representing the total customer deposits. Additionally, the auditor verifies the ownership of the company’s crypto assets,

wherein the company provides a digital signature from the private keys corresponding to all its addresses. Ultimately,

the auditor ensures that the amount of assets surpasses the number of liabilities. In the subsequent stage, exchanges

must consider independent customer verification to enhance transparency and assure customers that their accounts are

included in the liabilities covered by the held assets. To achieve this, the exchange constructs a Merkle tree where each

customer account serves as a leaf. The auditor verifies that the Merkle tree encompasses all liabilities and provides each

customer with their leaf information and a means to verify the path from their leaf to the Merkle root.

9 In this paper, we refrain from proposing a definitive solution, recognizing the existence of various options, each with its own set of strengths

and weaknesses.

9
 The necessity for an external auditor should be contingent upon the complexity inherent in the exchange platform. In

instances where the platform deals with multi-assets or intricate financial products within the crypto-ecosystem (such as

yield farming, staking, derivatives), the complexity could compel the exchange to engage an auditor for a more thorough

analysis. This is because a proof-of-reserves procedure alone might not provide sufficient assurances to customers in such

intricate financial scenarios.

As the reader can observe, aligning with Carter (2022a,b), proof-of-reserves implicitly incorporates proof-of-assets and

proof-of-liabilities in the process, making it challenging to completely isolate an individual part from the whole. Broadly

speaking, as discussed earlier, proof-of-assets involves demonstrating ownership and existence of the exchange’s assets,

proof-of-liabilities entails the accurate identification of all liabilities in the exchange, and proof-of-reserves involves an

analysis to ascertain whether the exchange has sufficient assets to cover its existing liabilities. It is important to note that

all these processes are executed on-chain, meaning assets or liabilities outside the blockchain are not considered, as they

are included in the proof-of-solvency stage.

2.1.2. Proof-of-solvency

As emphasized in the introduction, proof-of-solvency involves a more comprehensive concept, taking into considera-

tion not only proof-of-reserves (on-chain assets and liabilities) but also the company’s assets and liabilities beyond the

distributed ledger and any other factors that might affect its proper function in the future. In other words, proof-of-

solvency offers insights into the overall financial health of the centralised exchange. Conversely, the primary objective of

proof-of-reserves is to demonstrate on-chain that customer liabilities are either less than or equal to the assets held by the

exchanges on behalf of customers (Buxton et al., 2022). Therefore, proof-of-reserves alone does not guarantee the solvency

of the exchange or address the risks tied to its assets and liabilities. Even if an exchange demonstrates that the value of

its assets is equal to or greater than its liabilities, insolvency could still occur due to external shocks or spillover effects

affecting its assets. For instance, a cryptocurrency entity with Terra-Luna tokens in its possession could have undergone

a bank run or faced bankruptcy due to its association with this ecosystem, even if its assets were valued higher than its

liabilities before the Terra-Luna collapse.

The proof-of-solvency requirements for a digital asset platform providing services across various crypto-assets (e.g.,

Binance) should significantly differ from those applicable to a platform exclusively offering institutional investor services

for specific assets, such as BTC (e.g., BitMEX). A proof-of-solvency approach should encompass the evaluation of assets

and liabilities held by the exchange, considering their potential impact on the entity’s future solvency. Such an assessment

is crucial for both the “parent exchange” and any associated branches. The relevance of this point also became especially

apparent during the collapse of FTX, where its activities concealed a conglomerate comprising over 100 companies as

integral parts of the FTX group (Ray III, 2022).

In the empirical sections that follow, we examine proof-of-solvency with a focus on the composition of exchange assets

(i.e., proof-of-assets), emphasizing the importance of disclosed wallet addresses within the exchange.

10
2.2. Historical evolution

As discussed in Sec.(2.1), blockchain schemes designed to enhance transparency and ensure customer protection have

been present since 2013. However, the significant surge in the implementation of proof-of-reserves by exchanges commenced

in 2022 (CNBC, 2022). In other words, despite the existence of transparency-enhancing options, it required about nine

years and the collapse of various protocols (e.g. Terra-Luna) and exchanges (e.g. FTX) for many platforms to prioritize the

safeguarding of customers’ interests. Some of the exceptions that implemented different schemes to enhance transparency

included Coinfloor, Bitbuy, ShakePay, HBTC, and Gate.io, among others. From 2014 to 2021, the BTC exchange Coinfloor

performed regular monthly “Provable Solvency Audits” without relying on a third-party firm to verify the completeness

of its liability accounting. Nevertheless, the initiative demonstrated by Coinfloor was noteworthy and demonstrated

foresight. In 2019 and 2020, digital asset platforms Bitbuy and ShakePay issued third-party memorandums outlining

their custody processes. These reports, which presented CipherBlade’s summary of their on-premises analysis, solely

represented the opinion of a third party and did not include a detailed listing of liabilities for customers to independently

verify. Moreover, the concerned digital asset platforms did not furnish cryptographic evidence of their ownership of client

assets to the general public. In 2020, HBTC released a guide detailing the process of proving full reserves for their BTC,

ETH, and Tether (USDT) balances, utilizing the Merkleized liability approach. In the same year, Gate.io collaborated

with Armanino LLP to conduct a proof-of-reserves assessment, featuring a user-friendly verification dashboard. Both

proposals enabled customers to verify the inclusion of their balances in the liability set, however, neither was carried out

on a continuous basis (see Buxton et al., 2022; Pines et al., 2022).

After the collapse of FTX in 2022, the majority of exchanges initiated the implementation of proof-of-reserves. However,

there is considerable heterogeneity among the platforms in how they approach this practice. Carter (2023) categorized the

quality of various proof-of-reserve schemes based on different criteria, including the existence of cryptographic verification

of assets, the percentage of assets included, frequency of implementation, user verification of liabilities, consideration of

margin and lending, and the presence of auditor oversight. The resulting classification, ordered by scores, was as follows:

BitMEX, Deribit, Kraken, KuCoin, OKX, Gate.io, Crypto.com, Bybit, Luno, and Binance. BitMEX and Deribit were

categorised as the exchanges with the best proof-of-reserves schemes given two main reasons. First, they stand out as

the only platforms implementing proof-of-reserve procedures with a high frequency.10 Second, BitMEX and Deribit are

unique in publishing the complete liability set and allowing third parties to independently verify their liabilities. While

most exchanges permit clients to individually verify if their liabilities are included, creating a form of “herd immunity”,

the majority of these platforms do not publish the complete liability set.

10 For BitMEX and Deribit, Carter (2023) asserts that users receive sufficient assurances without the involvement of a third-party auditor.

Through the process, users can verify that BitMEX controlled a specific amount of BTC, and their account balance is included in the final Merkle
tree of balances. If a significant number of users conduct the analysis, it provides robust assurances that BitMEX isn’t selectively excluding
liabilities, thereby accurately representing their solvency. This setup primarily attests to BTC in a relatively straightforward full-reserve model.
However, in more intricate scenarios, the incorporation of an auditor is necessary.

11
3. Data

In this section, we use proof-of-assets data reported by Glassnode (Glassnode, 2023) for the most relevant centralized

exchanges. Our purpose in conducting this empirical analysis of proof-of-assets is to closely examine the asset composition

of these exchanges and to explore the potential implications for the future of the cryptocurrency ecosystem. In Sec.(3.1),

we provide a static representation of the asset composition within centralized exchanges as of March 1, 2023. In Sec.(3.2),

we report the dynamic evolution of asset composition over time, starting from the initial data point for each exchange

and tracking it until March 1, 2023. Finally, in Sec.(3.3), we construct portfolios for each centralized exchange based on

their respective asset compositions, enabling us to assess the stability of these exchanges in the face of extreme events. In

particular, to evaluate the resilience of the Centralised Finance (CeFi) system, we will apply the methodologies described

in Sec.(4) to these portfolios.

3.1. Database & static representation

To analyse the composition of exchange assets held on behalf of customers, we utilize the Glassnode database. This

database offers insights into the assets stored in disclosed cold wallets by exchanges, serving as proof-of-assets. In Table 1,

we present the featured exchanges along with their respective sample periods. Specifically, data is available for Binance,

Bitfinex, BitMEX, Bybit, CheckSig, Crypto.com, Deribit, Huobi, KuCoin, OKX, and SwissBorg. We retrieve data from

the earliest available data point up to March 1, 2023.

Table 1: Information on exchanges sourced from Glassnode.

Exchange Initial date Final date Exchange Initial date Final date
Binance 28/09/2018 01/03/2023 Deribit 01/06/2017 01/03/2023
Bitfinex 27/04/2016 01/03/2023 Huobi 23/05/2017 01/03/2023
BitMEX 22/11/2014 01/03/2023 KuCoin 26/04/2020 01/03/2023
Bybit 25/02/2021 01/03/2023 OKX 17/03/2018 01/03/2023
CheckSig 01/10/2020 01/03/2023 SwissBorg 10/12/2020 01/03/2023
Crypto.com 18/03/2019 01/03/2023

In constructing this dataset, Glassnode has prioritized the most resilient networks and ecosystems within the space.

As a result, it encompasses the BTC and ETH ecosystems, including all corresponding ETH-based tokens (Schultze-

Kraft, 2022). Despite not encompassing all existing assets, the resulting dataset represents approximately 66.89% of the

total assets held by the exchanges, as reported in Table 2.11 The disparity between proof-of-assets and total assets is

understandable, given that most exchanges extend their services to a multitude of crypto assets beyond the BTC and

ETH blockchains. In Table 2, Binance, Huobi and KuCoin serve as examples showcasing this fact. Nevertheless, it is also

relevant to underline that due to the elevated expenses associated with implementing proof-of-reserves across multiple

assets, certain exchanges opt to provide proof only for the most substantial assets. For instance, Binance discloses reserves

11 The total assets data is sourced from Defillama (Defillama, 2023), with the exception of BitMEX (BitMEX, 2023) and CheckSig (CheckSig,

2023), which can be directly reviewed by any user.

12
for only 31 tokens, a practice viewed as a vulnerability by the community (Thomas, 2023, Carter, 2023).12

Table 2: Proof-of-assets and total assets of exchanges, expressed in billions of USD. The percentage reflects the proportional amount of
USD represented by proof-of-assets compared to the overall total assets, as of March 1, 2023.

Exchange Proof-of-assets Total assets Percentage Exchange Proof-of-assets Total assets Percentage
Binance 34.6864 62.3360 55.64% Deribit 1.6929 1.6500 102.60%
Bitfinex 7.8398 7.8430 99.96% Huobi 1.7850 3.1830 56.08%
BitMEX 1.5209 1.4963 101.65% KuCoin 1.6405 2.7490 59.68%
Bybit 1.9681 2.6460 74.38% OKX 7.6816 8.0110 95.89%
CheckSig 0.0132 0.0128 103.11% SwissBorg 0.4633 0.5054 91.67%
Crypto.com 3.5441 3.5050 101.11% Aggregated 62.8358 93.9374 66.89%

Glassnode provides data for the 10 largest asset positions in each exchange, with the remaining assets aggregated

under the label “Other”. Table 3 presents an overview of the aggregated cryptocurrencies held by all exchanges as

of March 1, 2023, with the corresponding sectors according to Messari (Messari, 2023). We observe that 94.85% of

assets are represented by 35 assets in the exchanges, while the remaining 5.15% encompasses the rest of the crypto

assets. This highlights a significant concentration of investor attention on a select group of assets within the BTC and

ETH ecosystems. Specifically, BTC (33.63%), ETH (20.08%), and stablecoins (30.36%)—comprising especially BUSD

(13.12%), USDT (9.55%), and USDC (7.64%)—account for 84.03% of the assets held in the exchanges for our dataset.13

The asset composition on exchanges, as emphasized by Barucci et al. (2022, 2023) and Adachi et al. (2021), reflects

the significance of stablecoins within the crypto space. These digital assets play a critical role in the crypto ecosystem,

serving as a secure haven for mitigating volatility, facilitating crypto-asset trading, and providing liquidity in decentralized

finance.

Table 4 also reveals notable percentages for various sectors, including exchange tokens (5.62%), memecoins (2.86%),

data management (0.95%), and gaming (0.76%). The presence of exchange tokens is inherent in centralized exchanges,

where they function as utility tokens, providing users with benefits such as reduced fees (Briola and Aste, 2022; Conlon

et al., 2023). Memecoins represent a particularly speculative aspect of cryptocurrency trading, involving tokens not

necessarily associated with robust projects (Rosenberg, 2023). Blockchain projects focused on data management aim to

facilitate secure connections between smart contracts and real-world data, APIs, and payment systems. The associated

tokens typically serve to compensate node operators (or oracles) for data retrieval and provision, as well as other network

services. The rise of gaming tokens reflects the increasing impact of metaverse and play-to-earn initiatives in the crypto

ecosystem. These tokens primarily act as currency within their respective metaverses, facilitating the acquisition of

in-game items like land, cards, or weapons (Dowling, 2021a,b; Vidal-Tomás, 2023).

12 Considering the novelty of proof-of-reserves procedures, undisclosed wallets may still exist, potentially contributing to the disparity between

proof-of-assets and total assets. However, this issue is expected to diminish over time as exchanges progressively disclose all cold wallets. It
is also important to note that, in comparison to wallets containing proof-of-assets, total assets are often more dynamic as they encompass a
broader set of addresses.(Schultze-Kraft, 2022).
13 Since our sample constitutes 66.89% of the total assets, we can infer that BTC, ETH, and stablecoins collectively account for approximately

56.17% of the total assets across these exchanges, i.e., 84.03% of 66.89%.

13
Table 3: Constituents of proof-of-assets, considering all the exchanges, as of March 1, 2023.

Crypto Sector Percentage Crypto Sector Percentage Crypto Sector Percentage

BTC Currency 33.6302% MANA Gaming 0.2684% DAI Stablecoin 0.0051%
ETH Smart contract platform 20.0771% CRO Exchange token 0.2408% USDP Stablecoin 0.0040%
BUSD Stablecoin 13.1228% HOT Storage 0.1657% OMG Scaling 0.0039%
USDT Stablecoin 9.5549% STETH Staking 0.1523% SAN Asset management 0.0033%
USDC Stablecoin 7.6412% MATIC Scaling 0.1202% APE Gaming 0.0030%
Other Other 5.1467% TEL Payment platform 0.0675% UNI Decentralised exchange 0.0023%
LEO Exchange token 3.4659% QNT Interoperavility 0.0400% AAVE Defi 0.0021%
SHIB Memecoin 2.8602% ENJ Gaming 0.0336% FTT Exchange token 0.0013%
HT Exchange token 1.6136% OCEAN Data management 0.0335% HUSD Stablecoin 0.0000%
LINK Data management 0.9190% USDK Stablecoin 0.0271% RSR Asset management 0.0000%
SAND Gaming 0.4546% UTK Payment platform 0.0252% EURS Stablecoin 0.0000%
CHSB Exchange token 0.2962% WBTC Wrapped-token 0.0180% WETH Wrapped-token 0.0000%

Table 4: Allocation of assets, considering all the exchanges, based on the sector classification provided by Messari, as of March 1, 2023.

Sector Percentage Sector Percentage Sector Percentage

Currency 33.6302% Data management 0.9524% Interoperavility 0.0400%
Stablecoin 30.3551% Gaming 0.7597% Wrapped-token 0.0180%
Smart contract platform 20.0771% Storage 0.1657% Asset management 0.0033%
Exchange token 5.6179% Staking 0.1523% Decentralised exchange 0.0023%
Other 5.1467% Scaling 0.1241% Defi 0.0021%
Memecoin 2.8602% Payment platform 0.0928%

In Table 5, we display the relative weights of each cryptocurrency, and in Table 6, we show the relative weights of

each cryptocurrency sector on their respective exchanges. From these tables, three key observations can be highlighted.

First, the disparity in available crypto-assets across exchanges is notable, particularly as BitMEX and CheckSig pre-

dominantly focus on BTC, while other platforms provide the opportunity to trade a broader range of alternative assets.

Indeed, Binance and KuCoin distinguish themselves by providing an extensive list of alternative cryptocurrencies. This

is exemplified by the “other” category, representing 7.65% and 8.96% on these respective exchanges. Second, consistent

with the findings in Table 3 and 4, the dominance of BTC, ETH, and stablecoins is evident across all exchanges. Ex-

cept for SwissBorg, BTC constitutes at least 20% of proof-of-assets on every exchange, and similarly, ETH represents

a minimum of 9% across the board. Notably, stablecoins play a significant role in the asset composition of Binance

(39.31%), Bybit (51.28%), KuCoin (43.15%), and OKX (37.30%), surpassing even BTC and ETH percentages. This

underscores policymakers’ concerns regarding the potential risks associated with stablecoins in shaping the future of the

crypto-ecosystem. Specifically, it is observed that USDT is the primary stablecoin in Bybit, KuCoin, and OKX, while

Binance utilizes BUSD, as of the time of writing this paper. Thirdly, we emphasize the inclusion of the exchange tokens

LEO (27.78%), HT (56.8%), and CHSB (40.17%) in Bitfinex, Huobi, and SwissBorg, respectively. As mentioned earlier,

the utilization of these utility tokens within centralized exchanges is inherent for internal services and offerings. Hence, it

is not rare to find a concentration of these tokens within their respective centralized exchanges. However, following the

misuse of the exchange token FTT by the FTX exchange, there is a crucial need for the crypto community, policymakers,

and management to promote a regulatory framework for exchange tokens that prioritizes transparency and safeguards

customer protection.

14
Table 5: Relative weight of cryptocurrencies on respective exchanges, as of March 1, 2023.

Binance Bitfinex BitMEX Bybit CheckSig Crypto.com Deribit Huobi KuCoin OKX SwissBorg

AAVE - - - - - - - - - - 0.2822%
APE - - - 0.0963% - - - - - - -
BTC 25.2405% 58.7266% 100% 32.8495% 100% 29.5007% 59.9404% 22.8490% 22.0388% 34.9398% 16.9574%
BUSD 23.7724% - - - - - - - - - -
CHSB - - - - - - - - - - 40.1737%
CRO - - - - - 4.2700% - - - - -
DAI - - - 0.1619% - - - - - - -
ENJ - - - - - 0.4602% - - - - 1.0378%
ETH 21.7235% 9.1727% - 13.0562% - 24.2421% 37.8091% 9.5467% 15.4957% 27.7601% 10.4013%
EURS - - - - - - - - - 0.0000% -
FTT - 0.0105% - - - - - 0.0005% - - -
HOT 0.3002% - - - - - - - - - -
HT - - - - - - - 56.8019% - - -
HUSD - - - - - - - 0.0007% - - -
LEO - 27.7793% - - - - - - - - -
LINK 1.4487% 0.1221% - 0.3563% - 1.0391% - 0.3145% 0.9699% - -
MANA 0.4695% 0.0743% - - - - - - - - -
MATIC - - - 0.3381% - 1.7500% - - - - 1.4763%
OCEAN - - - - - - - - 1.2823% - -
OMG - 0.0316% - - - - - - - - -
Other 7.6537% 1.2205% - 1.3871% - 6.0425% - 4.7033% 8.9604% - 2.3926%
QNT - - - - - - - - 1.5323% - -
RSR - - - - - - - - - 0.0000% -
SAN - 0.0261% - - - - - - - - -
SAND 0.8107% - - 0.2280% - - - - - - -
SHIB 3.0463% - - 0.4015% - 19.0825% - - 3.4375% - -
STETH - - - - - - - - - - 20.6584%
TEL - - - - - - - - 2.5864% - -
UNI - - - - - - - 0.0795% - - -
USDC 11.3377% 0.2241% - 3.5835% - 11.5944% 2.2505% 2.2608% 6.1008% 2.1947% 4.8767%
USDK - - - - - - - - - 0.2220% -
USDP - - - - - - - - - 0.0326% -
USDT 4.1967% 2.6122% - 47.5417% - 1.6986% - 3.2539% 37.0537% 34.8507% 0.9686%
UTK - - - - - - - 0.1893% 0.5421% - 0.7751%
WBTC - - - - - 0.3199% - - - - -
WETH - - - - - - - - - 0.0000% -

Table 6: Relative weight of cryptocurrency sectors on respective exchanges, as of March 1, 2023.

Sector Binance Bitfinex BitMEX Bybit CheckSig Crypto.com Deribit Huobi KuCoin OKX SwissBorg

Asset management - 0.0261% - - - - - - - 0.0000% -

Currency 25.2405% 58.7266% 100% 32.8495% 100% 29.5007% 59.9404% 22.8490% 22.0388% 34.9398% 16.9574%
Data management 1.4487% 0.1221% - 0.3563% - 1.0391% - 0.3145% 2.2522% - -
Decentralised exchange - - - - - - - 0.0795% - - -
Defi - - - - - - - - - - 0.2822%
Exchange token - 27.7898% - - - 4.2700% - 56.8024% - - 40.1737%
Gaming 1.2801% 0.0743% - 0.3243% - 0.4602% - - - - 1.0378%
Interoperavility - - - - - - - - 1.5323% - -
Memecoin 3.0463% - - 0.4015% - 19.0825% - - 3.4375% - -
Other 7.6537% 1.2205% - 1.3871% - 6.0425% - 4.7033% 8.9604% - 2.3926%
Payment platform - - - - - - - 0.1893% 3.1285% - 0.7751%
Scaling - 0.0316% - 0.3381% - 1.7500% - - - - 1.4763%
Smart contract platform 21.7235% 9.1727% - 13.0562% - 24.2421% 37.8091% 9.5467% 15.4957% 27.7601% 10.4013%
Stablecoin 39.3069% 2.8363% - 51.2871% - 13.2929% 2.2505% 5.5154% 43.1545% 37.3001% 5.8453%
Staking - - - - - - - - - - 20.6584%
Storage 0.3002% - - - - - - - - - -
Wrapped-token - - - - - 0.3199% - - - 0.0000% -

In Table 7, we present the relative weights of individual cryptocurrencies, and Table 8 reports the relative weights of

cryptocurrency sectors across the 11 exchanges. In contrast to Tables 5 and 6, our emphasis in this instance is not on the

weights of cryptocurrencies within individual exchanges. Instead, we are investigating the distribution of cryptocurrencies

across the broader ecosystem among the 11 exchanges. As expected from Table 2, Binance is the preferred choice for users

to deposit and trade cryptocurrencies. Indeed, it stands out as the primary custodian of BTC and ETH, with holdings

of 40.43% and 59.73%, respectively, across the 11 exchanges available to us. In the same line, despite the well-known

15
dominance of Binance in the cryptocurrency market, it is still noteworthy that it holds 71.48% of all the stablecoins

represented in our dataset. Additionally, its multi-asset nature is emphasized in Table 8, where we observe that Binance

holds 82.09% of all the assets aggregated in the “Other” category across the 11 exchanges.

Given the absolute dominance of Binance, controlling a significant portion of the crypto capital, it is unsurprising

that the SEC is scrutinizing all procedures and potential transactions associated with the exchange. This scrutiny is

particularly driven by concerns regarding potential fund mismanagement. As a matter of fact, even though Binance and

Zhao agreed to plead guilty to breaking U.S. Anti-Money Laundering laws in November 2023 as part of a $4.3 billion

settlement with the U.S. Justice Department, Treasury Department and the Commodity Futures Trading Commission, the

SEC is still looking for evidence that Binance.US had a backdoor to potentially control customer assets in a similar style

to FTX (Lindrea, 2023). Under these circumstances, the prevailing dominance of Binance could ultimately pose a threat

to the stability of the system, considering that the exchange serves as the primary custodian across all cryptocurrency

sections.
Table 7: Distribution of cryptocurrencies across the 11 exchanges, as of March 1, 2023.

Binance Bitfinex BitMEX Bybit CheckSig Crypto.com Deribit Huobi KuCoin OKX SwissBorg

AAVE - - - - - - - - - - 100%
APE - - - 100.0000% - - - - - - -
BTC 41.4304% 21.7873% 7.1973% 3.0595% 0.0624% 4.9476% 4.8018% 1.9301% 1.7109% 12.7009% 0.3718%
BUSD 100% - - - - - - - - - -
CHSB - - - - - - - - - - 100%
CRO - - - - - 100% - - - - -
DAI - - - 100% - - - - - - -
ENJ - - - - - 77.2312% - - - - 22.7688%
ETH 59.7283% 5.7003% - 2.0369% - 6.8102% 5.0735% 1.3508% 2.0151% 16.9030% 0.3820%
EURS - - - - - - - - - 100% -
FTT - 98.9581% - - - - - 1.0419% - - -
HOT 100% - - - - - - - - - -
HT - - - - - - - 100% - - -
HUSD - - - - - - - 100% - - -
LEO - 100% - - - - - - - - -
LINK 87.0230% 1.6571% - 1.2144% - 6.3778% - 0.9722% 2.7555% - -
MANA 96.5449% 3.4551% - - - - - - - - -
MATIC - - - 8.8125% - 82.1303% - - - - 9.0572%
OCEAN - - - - - - - - 100% - -
OMG - 100% - - - - - - - - -
Other 82.0911% 2.9586% - 0.8442% - 6.6219% - 2.5961% 4.5455% - 0.3427%
QNT - - - - - - - - 100% - -
RSR - - - - - - - - - 100% -
SAN - 100% - - - - - - - - -
SAND 98.4291% - - 1.5709% - - - - - - -
SHIB 58.7932% - - 0.4396% - 37.6294% - - 3.1378% - -
STETH - - - - - - - - - - 100%
TEL - - - - - - - - 100% - -
UNI - - - - - - - 100% - - -
USDC 81.9067% 0.3659% - 1.4689% - 8.5582% 0.7935% 0.8405% 2.0845% 3.5112% 0.4706%
USDK - - - - - - - - - 100% -
USDP - - - - - - - - - 100% -
USDT 24.2458% 3.4110% - 15.5846% - 1.0026% - 0.9674% 10.1247% 44.5890% 0.0747%
UTK - - - - - - - 21.2989% 56.0644% - 22.6367%
WBTC - - - - - 100% - - - - -
WETH - - - - - - - - - 100% -

16
Table 8: Distribution of cryptocurrency sectors across the 11 exchanges, as of March 1, 2023.

Sector Binance BitMEX Bitfinex Bybit CheckSig Crypto.com Deribit Huobi KuCoin OKX SwissBorg

Asset management - - 99.9881% - - - - - - 0.0119% -

Currency 41.4304% 7.1973% 21.7873% 3.0595% 0.0624% 4.9476% 4.8018% 1.9301% 1.7109% 12.7009% 0.3718%
Data management 83.9641% - 1.5989% 1.1717% - 6.1536% - 0.9380% 6.1737% - -
Decentralised exchange - - - - - - - 100.0000% - - -
Defi - - - - - - - - - - 100.0000%
Exchange token - - 61.7175% - - 4.2869% - 28.7232% - - 5.2724%
Gaming 93.0183% - 1.2208% 1.3371% - 3.4165% - - - - 1.0072%
Interoperavility - - - - - - - - 100.0000% - -
Memecoin 58.7932% - - 0.4396% - 37.6294% - - 3.1378% - -
Other 82.0911% - 2.9586% 0.8442% - 6.6219% - 2.5961% 4.5455% - 0.3427%
Payment platform - - - - - - - 5.7958% 88.0444% - 6.1598%
Scaling - - 3.1769% 8.5326% - 79.5210% - - - - 8.7695%
Smart contract platform 59.7283% - 5.7003% 2.0369% - 6.8102% 5.0735% 1.3508% 2.0151% 16.9030% 0.3820%
Stablecoin 71.4808% - 1.1658% 5.2921% - 2.4699% 0.1997% 0.5162% 3.7117% 15.0218% 0.1420%
Staking - - - - - - - - - - 100.0000%
Storage 100.0000% - - - - - - - - - -
Wrapped-token - - - - - 100.0000% - - - 0.0000% -

3.2. Dynamic representation

Compared to the static representation of proof-of-assets depicted in Table 5 and 6 for the centralized exchanges as

of March 1, 2023, Fig. (2) and (3) present a dynamic perspective, illustrating the evolution of asset composition over

time. Specifically, Fig. (2) displays the absolute value in USD and the weights of individual cryptocurrencies on their

respective exchanges over time, while Fig. (3) outlines the progression of USD amounts and weights for each particular

cryptocurrency sector on the corresponding exchange. These figures allow us to emphasize two general facts regarding the

evolution of proof-of-assets over time. First, it is noticeable that centralized exchanges experienced a surge in popularity

towards the conclusion of 2021, aligning with the peak of the cryptocurrency market’s last bubble. The growth in assets

held by exchanges might be associated with the surge in cryptocurrency advertising during the 2021 bull run, where

crypto ads became omnipresent on various surfaces within the realm of sports. This encompassed naming agreements

for stadiums, team uniforms, and even liveries for Formula One racing cars. According to a Messari report, Crypto.com

invested more than $200 million in marketing during the initial half of 2021 alone (Medium, 2023). Second, since

2021, there has been a noticeable increase in the prevalence of stablecoins across various exchanges. Notably, instances

involving Binance, Huobi, KuCoin, Bybit, and OKX reveal periods when these platforms held more than 50% of total

customer assets in stablecoins. These two points emphasize the existing literature where policymakers, institutions, and

the scientific community underline the necessity to analyze the risks associated with stablecoins and centralized exchanges

in the crypto ecosystem (Bullmann et al., 2019; Arner et al., 2020; ECB, 2020; Adachi et al., 2021; Adachi et al., 2022;

Bank for International Settlements, 2022; Asmakov, 2022, European Council, 2022; Financial Stability Board, 2023; Beau,

2023).

In the contemporary crypto landscape, stablecoins play a pivotal role, contributing to nearly 75% of the overall

trading activity (Adachi et al., 2021). Interestingly, as demonstrated by Barucci et al. (2022), the trading volumes of

markets involving USDT against major cryptocurrencies have consistently risen, establishing USDT as the cryptoasset

with the highest trading volume since the second quarter of 2019. Additionally, Barucci et al. (2023) highlighted that

stablecoins, such as USDT, serve as a secure haven and/or store of value, facilitating cryptocurrency trading without the

need for traditional currencies, which hold a marginal role in the crypto space. Similarly, the growing significance of the

stablecoin BUSD within Binance (see Fig. (2)) drew the scrutiny of US policymakers for legal reasons. Specifically, in

17
February 2023, the United States Securities and Exchange Commission raised concerns about BUSD being an unregistered

security in a Wells notice directed at Paxos—the entity responsible for issuing the stablecoin. The New York Department

of Financial Services issued an order instructing Paxos to cease the issuance of BUSD. In this context, it becomes

evident that stablecoins could introduce risks to financial stability through various contagion channels unless a regulatory

framework is established. These channels include: (i) financial sector exposures; (ii) wealth effects; (iii) confidence effects;

and (iv) the extent of crypto-assets’ use in payments and settlements (Adachi et al., 2022; Financial Stability Board,

2023). Beyond financial stability, stablecoins present challenges and risks in areas such as consumer protection, taxation,

cybersecurity, money laundering, terrorism financing, market price manipulation, and legal certainty. The potential for

certain stablecoins to achieve substantial international scale in the future raises concerns related to domains such as

monetary sovereignty and monetary policy. Exemplifing these concerns, Blackrock disclosed ‘exposure to stablecoin risks’

on spot BTC ETF: “a large portion of the digital asset market still depends on stablecoins such as USDT and USDC,

there is a risk that a disorderly de-pegging or a run on USDT or USDC could lead to dramatic market volatility in digital

assets more broadly” (Barbosa, 2023).

On the other hand, the surge in crypto-asset service activity facilitated by centralized exchanges introduces risks

akin to those prevalent in mainstream financial services, encompassing credit, liquidity, and market risks. Notably, these

activities lack the regulatory safeguards designed to mitigate these risks and ensure their proper recognition, monitoring,

and management (Beau, 2023). The collapse of FTX has thrust the issue of cryptocurrency exchange failures into the

spotlight, emphasizing the imperative need for regulatory oversight in this rapidly evolving industry. Despite the challenges

associated with regulating a swiftly changing market, governments and regulators are actively striving to adapt to these

transformations. Nevertheless, the application of existing financial regulations to cryptocurrencies varies by country and

may not be sufficiently comprehensive to address all pertinent issues (Asmakov, 2022).

To address the various challenges associated with the increasing prevalence of stablecoins and the dominant role of

centralized exchanges, as depicted in Fig. (2) and (3), governments around the world are proposing diverse regulatory

frameworks (Asmakov, 2022). An exemplary initiative is the European Union’s Markets in Crypto Assets regulation

(Official Journal of the European Union, 2023), MiCA, which stands as the first major jurisdiction globally to introduce

comprehensive and tailored rules for the crypto sector. Under MiCA, stablecoins, referred to as “e-money tokens” (EMTs)

when linked to the value of a fiat currency or “asset-referenced tokens” (ARTs) otherwise, are mandated to maintain

appropriate reserves and adhere to robust governance standards. The regulatory constraints become more stringent as

the tokens gain wider usage, with stablecoins not pegged to an EU currency facing an outright ban if they exceed 1 million

transactions per day and EUR 200 million, as authorities seek to safeguard the euro’s status.

Under MiCA framework, exchanges intending to offer crypto services within the EU, spanning custody, trading,

portfolio management, or advice, must also secure authorization from one of the EU’s 27 national financial regulators

(Official Journal of the European Union, 2023). In alignment with the conduct of business rules under MiFID II, MiCA

dictates that exchanges -refereed as Crypto-Asset Service Providers (CAS-Providers)- must operate honestly, fairly, and

professionally in the best interests of their clients and prospective clients (Article 66(1)). However, it is important to note

18
that MiCA only partially covers crypto-asset service providers engaging in multiple functions and activities across different

jurisdictions, often referred to as crypto conglomerates. To enhance regulatory coverage, it is imperative to restrict and

oversee the concentration of activities in these conglomerates through stringent governance requirements. Additionally,

efforts should be directed at bolstering investor protection by addressing conflicts of interest and reinforcing redemption

rights, along with fortifying the protection of reserves (Beau, 2023).

Figure. 2: Proof-of-assets on centralised exchanges, presented in both USD and percentage terms.

Binance Binance
1.00

6e+10 Crypto Crypto

BTC BTC
BUSD 0.75 BUSD
Cryptos (USD)

ETH ETH

Percentage
4e+10 HOT HOT
LINK LINK
0.50
MANA MANA
Other Other
SAND SAND
2e+10 SHIB SHIB
0.25
USDC USDC
USDT USDT

0e+00 0.00
2019 2020 2021 2022 2023 2019 2020 2021 2022 2023
Date Date
BitMEX BitMEX
1.00

6e+09

0.75
Cryptos (USD)

Percentage

4e+09 Crypto Crypto

0.50
BTC BTC

2e+09
0.25

0e+00 0.00
2016 2018 2020 2022 2016 2018 2020 2022
Date Date
Bitfinex Bitfinex
1.00

Crypto Crypto
1.5e+10 BTC BTC
ETH 0.75 ETH
Cryptos (USD)

FTT FTT
Percentage

LEO LEO
1.0e+10 LINK LINK
0.50
MANA MANA
OMG OMG
Other Other
5.0e+09 SAN SAN
0.25
USDC USDC
USDT USDT

0.0e+00 0.00
2016 2018 2020 2022 2016 2018 2020 2022
Date Date

19
Figure. 2: Proof-of-assets on centralised exchanges, presented in both USD and percentage terms.

Bybit Bybit
1.00

Crypto Crypto
APE APE
2e+09 BTC 0.75 BTC
Cryptos (USD)

DAI DAI

Percentage
ETH ETH
LINK LINK
0.50
MATIC MATIC
Other Other
1e+09
SAND SAND
SHIB SHIB
0.25
USDC USDC
USDT USDT

0e+00 0.00
2021−07 2022−01 2022−07 2023−01 2021−07 2022−01 2022−07 2023−01
Date Date
CheckSig CheckSig
1.00

1.5e+07
0.75
Cryptos (USD)

Percentage
1.0e+07 Crypto Crypto
0.50
BTC BTC

5.0e+06 0.25

0.0e+00 0.00
2021−01 2021−07 2022−01 2022−07 2023−01 2021−01 2021−07 2022−01 2022−07 2023−01
Date Date
Crypto.com Crypto.com
1.00
1.0e+10

Crypto Crypto
BTC BTC
7.5e+09 CRO 0.75 CRO
Cryptos (USD)

ENJ ENJ
Percentage

ETH ETH
LINK LINK
5.0e+09 0.50
MATIC MATIC
Other Other
SHIB SHIB
USDC USDC
2.5e+09 0.25
USDT USDT
WBTC WBTC

0.0e+00 0.00
2020 2021 2022 2023 2020 2021 2022 2023
Date Date
Deribit Deribit
6e+09
1.00

0.75
4e+09
Cryptos (USD)

Crypto Crypto
Percentage

BTC BTC
ETH 0.50 ETH
USDC USDC
USDT USDT
2e+09

0.25

0e+00 0.00
2018 2020 2022 2018 2020 2022
Date Date

20
Figure. 2: Proof-of-assets on centralised exchanges, presented in both USD and percentage terms.

Huobi Huobi
1.00

4e+09 Crypto Crypto

BTC BTC
ETH 0.75 ETH
Cryptos (USD)

FTT FTT
3e+09

Percentage
HT HT
HUSD HUSD
0.50
LINK LINK
2e+09 Other Other
UNI UNI
USDC USDC
0.25
1e+09 USDT USDT
UTK UTK

0e+00 0.00
2018 2020 2022 2018 2020 2022
Date Date
KuCoin KuCoin
1.00

Crypto Crypto
3e+09
BTC BTC
ETH 0.75 ETH
Cryptos (USD)

LINK LINK

Percentage
OCEAN OCEAN
2e+09 Other Other
0.50
QNT QNT
SHIB SHIB
TEL TEL
1e+09 USDC USDC
0.25
USDT USDT
UTK UTK

0e+00 0.00
2021 2022 2023 2021 2022 2023
Date Date
OKX OKX
1.00

7.5e+09 Crypto Crypto

BTC 0.75 BTC
ETH ETH
Cryptos (USD)

EURS EURS
Percentage

5.0e+09 RSR RSR

UNI 0.50 UNI
USDC USDC
USDK USDK
USDP USDP
2.5e+09
USDT 0.25 USDT
WETH WETH

0.0e+00 0.00
2018 2020 2022 2018 2020 2022
Date Date
SwissBorg SwissBorg
8e+08
1.00

Crypto Crypto
AAVE AAVE
6e+08 0.75
BTC BTC
Cryptos (USD)

CHSB CHSB
Percentage

ENJ ENJ
4e+08 ETH ETH
0.50
MATIC MATIC
Other Other
STETH STETH
2e+08 USDC USDC
0.25
USDT USDT
UTK UTK

0e+00 0.00
2021−01 2021−07 2022−01 2022−07 2023−01 2021−01 2021−07 2022−01 2022−07 2023−01
Date Date

21
Figure. 3: Proof-of-assets on centralised exchanges, categorised by sector and presented in both USD and percentage terms.

Binance Binance
1.00

6e+10

Sector 0.75 Sector

Currency Currency
Sector (USD)

Data management Data management

Percentage
4e+10 Gaming Gaming
Memecoin 0.50 Memecoin
Other Other
Smart contract platform Smart contract platform
2e+10 Stablecoin Stablecoin
Storage 0.25 Storage

0e+00 0.00
2019 2020 2021 2022 2023 2019 2020 2021 2022 2023
Date Date
BitMEX BitMEX
1.00

6e+09

0.75
Sector (USD)

Percentage
4e+09 Sector Sector
0.50
Currency Currency

2e+09
0.25

0e+00 0.00
2016 2018 2020 2022 2016 2018 2020 2022
Date Date
Bitfinex Bitfinex
1.00

1.5e+10 Sector Sector

Asset management 0.75 Asset management
Currency Currency
Sector (USD)

Percentage

Data management Data management

1.0e+10 Exchange token Exchange token
0.50
Gaming Gaming
Other Other
Scaling Scaling
5.0e+09 Smart contract platform Smart contract platform
0.25
Stablecoin Stablecoin

0.0e+00 0.00
2016 2018 2020 2022 2016 2018 2020 2022
Date Date

22
Figure. 3: Proof-of-assets on centralised exchanges, categorised by sector and presented in both USD and percentage terms.

Bybit Bybit
1.00

Sector 0.75 Sector

2e+09
Currency Currency
Sector (USD)

Data management Data management

Percentage
Gaming Gaming
Memecoin 0.50 Memecoin
Other Other
1e+09 Scaling Scaling
Smart contract platform Smart contract platform
Stablecoin 0.25 Stablecoin

0e+00 0.00
2021−07 2022−01 2022−07 2023−01 2021−07 2022−01 2022−07 2023−01
Date Date
CheckSig CheckSig
1.00

1.5e+07
0.75
Sector (USD)

Percentage
1.0e+07 Sector Sector
0.50
Currency Currency

5.0e+06 0.25

0.0e+00 0.00
2021−01 2021−07 2022−01 2022−07 2023−01 2021−01 2021−07 2022−01 2022−07 2023−01
Date Date
Crypto.com Crypto.com
1.00
1.0e+10

Sector Sector
Currency 0.75 Currency
7.5e+09
Data management Data management
Sector (USD)

Exchange token Exchange token

Percentage

Gaming Gaming
5.0e+09 Memecoin 0.50 Memecoin
Other Other
Scaling Scaling
Smart contract platform Smart contract platform
2.5e+09 Stablecoin 0.25 Stablecoin
Wrapped−token Wrapped−token

0.0e+00 0.00
2020 2021 2022 2023 2020 2021 2022 2023
Date Date
Deribit Deribit
6e+09
1.00

0.75
4e+09
Sector (USD)

Percentage

Sector Sector
Currency Currency
0.50
Smart contract platform Smart contract platform
Stablecoin Stablecoin
2e+09

0.25

0e+00 0.00
2018 2020 2022 2018 2020 2022
Date Date

23
Figure. 3: Proof-of-assets on centralised exchanges, categorised by sector and presented in both USD and percentage terms.

Huobi Huobi
1.00

4e+09
Sector 0.75 Sector
Currency Currency
Sector (USD)

3e+09 Data management Data management

Percentage
Decentralised exchange Decentralised exchange
Exchange token 0.50 Exchange token
2e+09 Other Other
Payment platform Payment platform
Smart contract platform Smart contract platform
Stablecoin 0.25 Stablecoin
1e+09

0e+00 0.00
2018 2020 2022 2018 2020 2022
Date Date
KuCoin KuCoin
1.00

3e+09
Sector 0.75 Sector
Currency Currency
Sector (USD)

Data management Data management

Percentage
2e+09 Interoperavility Interoperavility
Memecoin 0.50 Memecoin
Other Other
Payment platform Payment platform
Smart contract platform Smart contract platform
1e+09
Stablecoin 0.25 Stablecoin

0e+00 0.00
2021 2022 2023 2021 2022 2023
Date Date
OKX OKX
1.00

7.5e+09

0.75
Sector Sector
Sector (USD)

Asset management Asset management

Percentage

5.0e+09 Currency Currency

Decentralised exchange 0.50 Decentralised exchange
Smart contract platform Smart contract platform
Stablecoin Stablecoin
Wrapped−token Wrapped−token
2.5e+09
0.25

0.0e+00 0.00
2018 2020 2022 2018 2020 2022
Date Date
SwissBorg SwissBorg
8e+08
1.00

Sector Sector
6e+08 Currency 0.75 Currency
Defi Defi
Sector (USD)

Exchange token Exchange token

Percentage

Gaming Gaming
4e+08 Other 0.50 Other
Payment platform Payment platform
Scaling Scaling
Smart contract platform Smart contract platform
2e+08 Stablecoin 0.25 Stablecoin
Staking Staking

0e+00 0.00
2021−01 2021−07 2022−01 2022−07 2023−01 2021−01 2021−07 2022−01 2022−07 2023−01
Date Date

24
3.3. Portfolio construction

To assess the stability of centralized exchanges during extreme events, we construct a portfolio for each platform by

considering the assets (i) held on behalf of the customers, i.e. proof-of-assets. The portfolio returns of the centralised

exchanges (rCEX,t ) are computed using the standard formula:

N
X
rCEX,t = wi,t · ri,t (1)
i=1

where N is the number of assets held by the exchange, wi,t denotes the proportion of assets on each day t, as depicted

in Fig.(2) and ri,t represents each cryptocurrency log-return, which is calculated using the prices from the CoinMarketCap

(2023) database (see Vidal-Tomás, 2022).

Table 9 provides descriptive statistics for the portfolios created from February 25, 2021, to March 1, 2023. These

statistics are computed starting on February 25, 2021, as it represents the earliest date for which comprehensive data is

accessible for all exchanges. Specifically, Bybit is the exchange with the most recent initial date, as indicated in Table 1.

The heterogeneous composition of proof-of-assets is apparent in Table 9, where Binance, Bybit, Crypto.com, Huobi, and

KuCoin exhibit positive means, while Bitfinex, BitMEX, Checksig, Deribit, OKX, and Swissborg report negative means.

Despite the differences in means, the standard deviation is relatively consistent across exchanges. However, Binance and

Huobi show the lowest standard deviation values, a pattern we hypothesize is linked to the substantial proportion of

stablecoins held by these exchanges.

Table 9: Descriptive statistics from February 25, 2021, to March 1, 2023.

Exchange Mean Std. Dev. Skewness Kurtosis Min. Max.

Binance 0.00007 0.02528 -0.65843 5.51846 -0.18433 0.11933
Bitfinex -0.00048 0.03633 -0.45878 3.23527 -0.20621 0.15519
BitMEX -0.00088 0.03564 -0.39194 2.83974 -0.17405 0.13576
Bybit 0.00033 0.03410 -0.60170 10.51725 -0.29092 0.20859
Checksig -0.00088 0.03564 -0.39194 2.83974 -0.17405 0.13576
Crypto.com 0.00000 0.03597 -0.64387 6.10186 -0.26125 0.20057
Deribit -0.00047 0.03845 -0.51574 3.23788 -0.20955 0.15559
Huobi 0.00037 0.02322 -0.10198 3.76991 -0.12231 0.12043
KuCoin 0.00086 0.04059 -0.33396 10.80428 -0.31524 0.23305
OKX -0.00010 0.03080 -0.45518 3.69460 -0.18929 0.13700
Swissborg -0.00035 0.03983 -0.74299 9.26603 -0.32543 0.22066

In Table 10, we present the descriptive statistics spanning from January 1, 2022, to March 1, 2023. In contrast

to Table 9, it is noteworthy that all exchanges exhibit negative means during this period. This observation might be

linked to the cryptocurrency bubble crash in 2021 and subsequent bankruptcies in the crypto sphere since January 2022,

leading to a cascade of bankruptcies within the system. Notably, Binance, Bybit, Huobi, KuCoin, and OKX stand

out for reporting relatively low standard deviations. As depicted in Fig. (3), these lower standard deviations could be

attributed to a significant rise in stablecoin holdings by most exchanges in 2022. It is plausible to assume that users have

been exchanging their volatile cryptocurrencies for stablecoins, using the latter as a “parking space” to mitigate crypto

volatility (Adachi et al., 2022).

25
Table 10: Descriptive statistics from January 1, 2022, to March 1, 2023.

Exchange Mean Std. Dev. Skewness Kurtosis Min. Max.

Binance -0.00088 0.01906 -0.42634 2.76838 -0.08453 0.07691
Bitfinex -0.00183 0.03206 -0.45416 3.65671 -0.16447 0.12533
BitMEX -0.00158 0.03260 -0.55398 4.76835 -0.17405 0.13576
Bybit -0.00143 0.02609 -0.55293 4.47166 -0.13321 0.10257
Checksig -0.00158 0.03260 -0.55398 4.76835 -0.17405 0.13576
Crypto.com -0.00139 0.03004 -0.56263 3.30586 -0.14969 0.12098
Deribit -0.00162 0.03540 -0.55781 3.97914 -0.17416 0.12901
Huobi -0.00056 0.02372 0.11639 4.23565 -0.11874 0.12043
KuCoin -0.00085 0.01964 -0.51040 2.50345 -0.08415 0.06770
OKX -0.00103 0.02353 -0.38868 3.29239 -0.10448 0.09809
Swissborg -0.00199 0.03446 -0.54164 4.87586 -0.20375 0.17092

To evaluate the stability of centralized exchanges during extreme events, we use the portfolio returns calculated from

January 1, 2022, to March 1, 2023. This specific timeframe is chosen as it corresponds to the period wherein all identified

bank runs and bankruptcies have occurred in 2022 (Clarke, 2022).

4. Methodology

In this section, we outline the methodology employed for analyzing the stability of the CeFi system using the previously

constructed portfolios. To be more precise, in Sec.(4.1), we introduce the AR-GARCH framework, which is utilized to

evaluate how exchanges responded to extreme negative events that occurred in 2022. In Sec.(4.2), we describe the modified

Value at Risk (VaR) approach, which is utilized to evaluate the down risk associated with depositing users’ assets in a

given exchange.

4.1. Event study: AR-GARCH framework

In our pursuit of evaluating the robustness of centralized exchanges, we aim to analyze how the portfolio returns,

computed between January 1, 2022, and March 1, 2023, reacted to significant adverse events (d) within the cryptocurrency

market. In particular, we focus on the most significant bankruptcies within the crypto ecosystem in 2022: (d1 ) Terra-

Luna on May 1114 , (d2 ) Three Arrows Capital on June 1615 , (d3 ) Voyager Digital on July 116 , (d4 ) Celsius on July 15,

(d5 ) FTX on November 817 , and (d6 ) Blockfi on November 2818 . With this aim, we employ the traditional AR-GARCH

framework. This methodology allows us to observe the market response according to each event, quantifying the effect on

each centralised exchange and determining its significance. We model the dynamic behaviour of the total assets held by

14 From May 11, 2022, UST’s price sharply declined to 10 cents, and LUNA tumbled to “virtually zero”, marking a substantial drop from

its previous all-time high of $119.51 (Briola et al., 2023).

15 The Financial Times reported on June 16, 2022, that Three Arrows had not met its margin calls (Financial Times, 2022).
16 The company suspended “trading, deposits, withdrawals, and loyalty rewards” on July 1, 2022 (Ge Huang, 2022).
17 On November 8, 2022, FTX experienced a significant decrease in the price of FTT. According to Khoo et al. (2022), creditors initiated the

recall of Alameda Research loans backed by FTT as collateral. As Alameda Research was unable to repay these loans, Sam Bankman-Fried
was compelled to request Binance’s intervention in acquiring the firm. As a result, on the same day, Binance made an announcement regarding
a non-binding letter of intent to acquire FTX (Coghlan, 2022). See also Vidal-Tomás et al. (2023); Conlon et al. (2023, 2024); Yousaf and
Goodell (2023).
18 On November 28, BlockFi filed for Chapter 11 bankruptcy protection with more than 100,000 creditors (The Guardian, 2022).

26
the exchanges, and represented by the portfolios, with the AR-GARCH model along with the corresponding events (d) as

exogenous variables, resulting in the methodological form:

n
X
rCEX,t = c + bj rCEX,t−j + ϕdt + ut ,
j=1 (2)
ut = σt zt , zt ∼ i.i.d(0, 1)

where ϕt denotes the impact of each event (d) on each exchange, d represents the event dummies, ut is the error

term, zt is a white noise process, and ht denotes the conditional standard deviation. Concerning the dummy variables,

we take into account two scenarios: (i) “Period [0]”, in which we examine the event’s effect solely on the specific day it

occurs, and (ii) “Period [-1,1]”, where we assess the event’s impact while also considering the potential effects on the day

before and the day after it occurs.19,20 In line with existing literature21 , we adopt the AR(1)-GARCH(1,1) specification

as our baseline model, and we present its findings in Sec.(5.1). Additionally, for the sake of ensuring the robustness of our

analysis, we employ various GARCH-type models. Table 11 provides a comprehensive list of all the GARCH-type variants

utilized in this study, which encompass GARCH, EGARCH, IGARCH, APARCH, CGARCH, and GJR. We elaborate on

the robustness of our results in Sec.(5.2).22

Table 11: GARCH-type models.

Model Volatility equation Proposed by

GARCH σt2 = ω + αu2t−1 + βσt−1
2
Bollerslev (1986)
2 2 2



EGARCH log σt = ω + αϵt−1 + γ (|ϵt−1 | − E (|ϵt−1 |)) + β log σt−1 Nelson (1991)
IGARCH σt2 = ω + αu2t−1 + (1 − α)σt−1
2
Engle and Bollerslev (1986)
APARCH σt = ω + α (|ut−1 | − γut−1 )δ + βσt−1
δ δ
Ding et al. (1993)
2 2 2



CGARCH σt = qt + α ut−1 − qt−1 + β σt−1 − qt−1
GJR σt2 = ω + αu2t−1 + γI (rt−1 < 0) u2t−1 + βσt−1
2
Glosten et al. (1993)

4.2. Modified VAR

Gaussian VaR is one of the most standard methods in the literature to measure the downside risk of a portfolio,

whose returns are characterised by the normal distribution. With this methodology, any practitioner can examine the

maximum potential loss incurred by an investment during a time period, given a certain probability level (Boudt et al.,

2008). Mathematically, it is defined as

VaRp (1 − α) = µp − σp z(α), (3)

where µp and σp denote the mean and standard deviation of the portfolio and z(α) is the α-quantile of the standard

normal distribution (Wilmott, 1998). However, this VaR measure does not consider the empirical stylised facts that are

19 In Period [0], we utilize a dummy variable set to 1 for the same day of the event and 0 otherwise. In Period [-1,1], we employ dummy

variables set to 1 for the same day, the day before, and the day after the event, with 0 otherwise.
20 Results are consistent using other time ranges. Material upon request.
21 GARCH frameworks have been widely used in the literature not only for event studies (e.g. Vidal-Tomás and Ibañez, 2018; Conlon et al.,

2023) but also for volatility analysis (Katsiampa, 2017; Sensoy et al., 2021) or forecasting (Trucı́os, 2019), among other financial subjects.
22 Bollerslev (1986) argued for restrictions on the parameters for positivity, ω > 0, α ≥ 0 and β ≥ 0, and the wide-sense stationarity condition,

α + β < 1.

27
already described in the literature (Cont, 2001) (e.g. skewness, kurtosis and fat tails), which reduces its accuracy (Giot

and Laurent, 2003). Given this fact, Zangari (1996) proposed an alternative estimator for VaR, namely modified VaR

(MVaR), in order to correct the Gaussian VaR for skewness and kurtosis in the return series.23 More specifically, it adjusts

the Gaussian quantile function for skewness and kurtosis, by means of the Cornish-Fisher expansion (Cornish and Fisher,

1938). As a result, the modified VaR is computed as follows:

1 1 1
z(α)2 − 1 Sp + z(α)3 − 3z(α) Kp − 2z(α)3 − 5z(α) Sp2 ,




Ẑ (α, Sp , Kp ) = z(α) + (4)
6 24 36

where Sp and Kp denote the skewness and kurtosis of a given portfolio. Therefore, the MVaR is defined substituting in

Eq. (3), z(α) by Ẑ (α, Sp , Kp ), i.e.,

MVaRp (1 − α) = µp − σp Ẑ (α, Sp , Kp ) . (5)

Using this approach, our objective is to assess the risks involved in depositing users’ assets in a given exchange. A

higher downside risk indicates that the exchange is less resilient to adverse events, making it potentially vulnerable to

bank runs. To achieve this objective, we compute the MVaR for the portfolios associated with each exchange. In contrast

to the event study previously described, this method focuses on assessing downside risk across the entire available sample

rather than analyzing specific individual events.

5. Empirical results

In this section, we present the empirical findings of our study. Sec.(5.1) includes (i) AR-GARCH outcomes based

on the selected negative events (d) in 2022 and (ii) MVaR(5%) & MVaR(1%) for the portfolios associated with each

centralized exchange. Sec.(5.2) provides a robustness analysis of different GARCH variants, including GARCH, EGARCH,

IGARCH, APARCH, CGARCH, and GJR variants. Sec.(5.3) explores the impact of stablecoins on our study’s results.

Lastly, in Sec.(5.4), we replicate the primary empirical exercises of this study using an alternative dataset from Glassnode,

specifically, exchange balance data.

5.1. Event study & MVaR(1%-5%) for proof-of-assets

Table 12 and 13 report the AR-GARCH estimates for the event dummies related to the most significant collapses in

2022: (d1 ) Terra-Luna on May 11, (d2 ) Three Arrows Capital on June 16, (d3 ) Voyager Digital on July 1, (d4 ) Celsius on

July 15, (d5 ) FTX on November 8, and (d6 ) Blockfi on November 28. Table 12 and 13 consider the Period [0] and [-1,1],

respectively.

We can observe in Table 12 that the most significant negative events for the Period [0] occurred chronologically as

follows: (d1 ) Terra-Luna, (d2 ) Three Arrows Capital, (d5 ) FTX, and (d6 ) Blockfi. When considering their impact on

the exchanges, the most severe events were (d5 ) FTX, (d2 ) Three Arrows Capital, (d1 ) Terra-Luna, and (d6 ) Blockfi.

23 See Conlon and McGee, 2020, for a BTC application of this methodology during the COVID-19 pandemic.

28
In general, it is noteworthy that the most resilient exchanges to the impact of these events were Binance and KuCoin.

Additionally, the stability of Huobi, OKX and Bitfinex was also remarkable.

Focusing on the Period [-1,1] as shown in Table 13, a shift in perspective becomes evident. Remarkably, the sole

significant negative event impacting all the centralized exchanges during this period was the bankruptcy of FTX (d5 ).

Whereas the collapses of Terra-Luna and Three Arrows Capital had a pronounced impact on centralized exchanges on the

day of the events, this impact diminishes when also considering the day before and after. This observation is consistent

with the findings of Briola et al. (2023), who observed that certain major cryptocurrencies, such as BTC, experienced

significant recoveries in the days following the Terra-Luna collapse. In other words, the assets held by centralized exchanges

exhibited resilience in the face of turbulence within the DeFi ecosystem and the immediate spillover effect of Three Arrows

Capital’s bankruptcy. Indeed, we can also state that users did not withdraw enough amounts of cryptocurrencies to cause a

sufficient negative effect on the total value of assets held by centralized exchanges.23 However, compared to the events more

related to the Defi ecosystem, centralised exchanges did suffered from the FTX collapse. The decline in cryptocurrency

prices following rumors of a potential collapse of FTX (Vidal-Tomás et al., 2023; Conlon et al., 2023, 2024) likely triggered

a withdrawal of funds from the majority of centralized exchanges. This withdrawal, in turn, exacerbated the negative

impact on the value of assets held by these exchanges. Considering that FTX was the third-largest centralized exchange

at the time of the events, it is rational to assume that users preferred safer options, such as cold wallets, to deposit their

cryptocurrencies. This precautionary measure aimed to avoid the negative effects of a cascade of bankruptcies among

other centralized exchanges within the crypto-ecosystem. At any rate, in a similar line to Table 13, the most resilient

centralised exchanges to the collapse of FTX were Binance, KuCoin and OKX. Certainly, Binance emerged as a winner

in this context, since its monthly trading activity increased by 30% just after the FTX collapse (Pan, 2022).24

Taking into account the findings from Tables 12 and 13, it can be affirmed that centralized exchanges demonstrate

resilience to external shocks not directly tied to the CeFi ecosystem. While DeFi and other cryptocurrency-related

disruptions may influence centralized exchanges, their adverse effects are transient and typically dissipate after the same

day of occurrence.

Table 12: AR-GARCH results based on the selected negative events (d) in 2022. The dummy variables are included for the exact day on
which the events occurred, i.e., Period [0].

Period [0] Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi KuCoin OKX Swissborg
d1 -0.0616*** -0.0760*** -0.0682*** -0.0656*** -0.0682*** -0.1050*** -0.0876*** -0.0548*** -0.0623*** -0.0634*** -0.1343***
d2 -0.0461*** -0.1019*** -0.1010*** -0.0853*** -0.1010*** -0.0777*** -0.1134*** -0.0443*** -0.0585*** -0.0671*** -0.0906***
d3 -0.0104 -0.0214 -0.0268 -0.0180 -0.0268 -0.0202 -0.0203 -0.0036 -0.0106 -0.0106 -0.0254
d4 0.0093 0.0215 0.0144 0.0162 0.0144 0.0181 0.0219 0.0126 0.0121 0.0146 0.0189
d5 -0.0586*** -0.0878*** -0.1060*** -0.0929*** -0.1060*** -0.1296*** -0.1270*** -0.1191*** -0.0598*** -0.0691*** -0.0902***
d6 -0.0059 -0.0443*** -0.0132 -0.0074 -0.0132 -0.0112 -0.0160 -0.0569*** -0.0085 -0.0085 -0.0168

23 The collapse of Terra-Luna shattered investor confidence and exacerbated the downward spiral of cryptocurrencies, which was already in

motion as part of a broader risk-averse trend. An inundation of margin calls from Three Arrows Capital’s creditors clamored for repayment,
yet the firm found itself unable to satisfy these demands due to a shortage of funds. Ultimately, the crypto hedge fund collapse after taking on
major directional trades and borrowing from over 20 institutions, and the founders defaulted on its payments.
24 In 2022, Binance’s volume market share increased from 48.7%, in the first quarter, to 66.7%, in the last quarter (CryptoCompare, 2023).

29
Table 13: AR-GARCH results based on the selected negative events (d) in 2022. The dummy variables are included, considering the day
before, the specific day of the events, and the day after, i.e., Period [-1,1].

Period [-1,1] Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi KuCoin OKX Swissborg
d1 -0.0161 -0.0095 -0.0016 -0.0156 -0.0016 -0.0391 -0.0203 -0.0230* -0.0127 -0.0171 -0.0152
d2 -0.0051 -0.0072 -0.0074 -0.0111 -0.0074 -0.0105 -0.0056 -0.0074 -0.0046 -0.0069 -0.0231
d3 -0.0055 -0.0133 -0.0144 -0.0111 -0.0144 -0.0087 -0.0131 -0.0034 -0.0058 -0.0073 -0.0125
d4 0.0132** 0.0304** 0.0164* 0.0231** 0.0164* 0.0214* 0.0330*** 0.0222*** 0.0163** 0.0222** 0.0164
d5 -0.0527*** -0.0867*** -0.1063*** -0.0873*** -0.1063*** -0.1116*** -0.1272*** -0.0917*** -0.0567*** -0.0658*** -0.0769***
d6 0.0006 0.0019 0.0000 0.0004 0.0000 0.0011 0.0001 0.0107 0.0029 0.0002 0.0006

On the other hand, Table 14 and Table 15 present the MVaR(5%) and MVaR(1%) values for the portfolios associated

with each centralized exchange. Table 14 lists the values for the exchanges in alphabetical order, consistent with the

arrangement in other tables. To enhance clarity, Table 15 displays the same values but organizes the exchanges based on

the down-risk experienced during the sample period. In comparison to Table 12 and Table 13, which focus on specific

cryptocurrency events, MVaR(5%) and MVaR(1%) consider the entire sample period. Consequently, they complement

the previous results by accounting for any potential events that may have been omitted earlier.

Consistent with our event study findings presented in Table 14 and Table 15, it remains evident that Binance and

KuCoin stand out as the most resilient centralized exchanges. Consequently, these exchanges exhibit a lower level of risk,

indicating a reduced likelihood of encountering substantial losses compared to other participants in the CeFi system. In

essence, based on these results, Binance and KuCoin can be considered stable and suitable choices for risk-averse investors,

as they are less susceptible to significant value fluctuations or potential bank runs due to a loss of confidence.

Table 14: MVaR(5%) & MVaR(1%) for the portfolio corresponding to each centralised exchange. Centralised exchanges are ordered alpha-
betically.

Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi KuCoin OKX Swissborg
MVaR(5%) -0.0334 -0.0561 -0.0569 -0.0459 -0.0569 -0.0534 -0.0623 -0.0367 -0.0349 -0.0406 -0.0603
MVaR(1%) -0.0623 -0.1122 -0.1234 -0.0971 -0.1234 -0.1034 -0.1274 -0.0772 -0.0635 -0.0793 -0.1315

Table 15: MVaR(5%) & MVaR(1%) for the portfolio corresponding to each centralised exchange. Centralised exchanges are ordered accord-
ing to MVaR(5%) & MVaR(1%) values.

Binance KuCoin Huobi OKX Bybit Crypto.com Bitfinex BitMEX Checksig Swissborg Deribit
MVaR(5%) -0.0334 -0.0349 -0.0367 -0.0406 -0.0459 -0.0534 -0.0561 -0.0569 -0.0569 -0.0603 -0.0623

Binance KuCoin Huobi OKX Bybit Crypto.com Bitfinex BitMEX Checksig Deribit Swissborg
MVaR(1%) -0.06229 -0.06353 -0.07716 -0.07934 -0.09711 -0.10344 -0.11217 -0.12343 -0.12343 -0.12742 -0.13152

These results are relevant from a policymaker’s perspective. Table 14 and Table 15 can also be interpreted as the

additional capital buffer that exchanges should retain to guarantee the resilience and proper functioning of the exchange

during turbulent times. This could be a crucial factor enabling centralized exchanges to absorb losses while continuing

to provide key services to the crypto ecosystem. In general, we observe that the CeFi system should maintain a capital

conservation buffer in the range of 6% to 14%, depending on the particular characteristics of each centralized exchange.

In fact, the inclusion of a capital buffer framework for traditional banks stands out as one of the key innovations within

the Basel III regulatory framework. Introduced in response to the global financial crisis of 2007-09, Basel III addresses

30
numerous shortcomings in the pre-crisis regulatory landscape, forming a solid foundation for a resilient banking system

capable of supporting the real economy throughout economic cycles. Capital buffers play a pivotal role in achieving this

goal, serving as shock absorbers during periods of stress to mitigate procyclicality.

Given the significant volatility within the crypto ecosystem and the inherent instability of certain financial entities

within it, it is prudent for the crypto ecosystem to implement a minimum capital buffer, akin to the regulatory framework

present in traditional finance. This would enable members of the CeFi system to utilize these buffers as safeguards against

shocks.

5.2. GARCH-type variants

In Sec. (5.1), we employed the conventional GARCH framework to examine the impact of the major collapses in 2022 on

centralized exchanges. To ensure the robustness of our analysis, in Figures (4) and (5), we present boxplots illustrating the

estimated dummy variables derived from a comprehensive set of GARCH-variants, which include GARCH, EGARCH,

IGARCH, APARCH, CGARCH, and GJR. As can be observed, the consistency in our results is apparent due to the

similarity in the estimates obtained from the various GARCH types. Consequently, the results obtained in Sec. (5.1)

remain robust even when utilizing different GARCH-variants.25

25 The primary exception to this consistency can be observed in Figure (5), where we note a greater variation among the estimates for the

Three Arrows Capital event. Nevertheless, it is worth highlighting that in this case, no significant estimate is observed, and the impact of the
Three Arrows Capital collapse was generally uniform across exchanges.

31
Figure. 4: GARCH-type variant results based on the selected negative events (d) in 2022. The boxplot includes the dummy estimates ob-
tained using GARCH, EGARCH, IGARCH, APARCH, CGARCH, and GJR variants. The dummy variables are included for the exact day
on which the events occurred, i.e., Period [0].

d 1: Terra−Luna d 2: Three Arrows Capital

−0.06

−0.06

−0.08
Value

Value
−0.08
−0.10

−0.12 −0.10

Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg
Exchange Exchange

d 3: Voyager digital d 4: Celsius

0.024

0.020
−0.01
Value

Value

0.016

−0.02

0.012

Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg
Exchange Exchange

d 5: FTX d 6: Blockfi

−0.06

−0.08 −0.02
Value

Value

−0.10

−0.04

−0.12

−0.14 −0.06
Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg
Exchange Exchange

32
Figure. 5: GARCH-type variant results based on the selected negative events (d) in 2022. The boxplot includes the dummy estimates ob-
tained using GARCH, EGARCH, IGARCH, APARCH, CGARCH, and GJR variants. The dummy variables are included, considering the
day before, the specific day of the events, and the day after, i.e., Period [-1,1].

d 1: Terra−Luna d 2: Three Arrows Capital

0.00 0.00

−0.01

−0.01

−0.02
Value

Value
−0.03
−0.02

−0.04

Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg
Exchange Exchange

d 3: Voyager digital d 4: Celsius

−0.003

0.030
−0.006

0.025
Value

Value

−0.009

0.020

−0.012

0.015

−0.015
Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg
Exchange Exchange

d 5: FTX d 6: Blockfi
−0.05
0.025

0.020
−0.07

0.015
Value

Value

−0.09
0.010

0.005
−0.11

0.000

−0.13
−0.005
Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg Binance Bitfinex BitMEX Bybit Checksig Crypto.com Deribit Huobi Ku−coin OKX Swissborg
Exchange Exchange

5.3. The relevance of stablecoins

As previously detailed in Sec.(3), particularly in Figures (2) and (3), the persistent growth of stablecoins within

centralized exchanges over time can be regarded as one of the most significant characteristics of the cryptocurrency and

CeFi ecosystem. Its relevance and risks have been underlined from different perspectives by policymakers and the scientific

community. As previously described, stablecoins are used by traders as (i) a bridge between fiat currencies and crypto-

assets and (ii) as a “parking space” for crypto volatility (Adachi et al., 2022). Nonetheless, it is important to acknowledge

that they also pose a systemic risk because the failure of a major stablecoin could potentially endanger the entire system.

33
 In this study, we highlight an additional role for stablecoins within the CeFi system. As shown in Fig. (6), the results

observed for MVaR can be linked to the proportion of stablecoins held by the exchanges. Notably, Binance, KuCoin, and

Huobi demonstrated greater resilience among the centralized exchanges, primarily due to their substantial holdings of

stablecoins. Given their nature, stablecoins are stable cryptocurrencies that mainly fluctuate around 1 USD. Consequently,

during the emergence of negative shocks that lead to a price decline in all the cryptocurrency prices, stablecoins tend to

stay near the 1 USD level. This characteristic implies that the presence of significant stablecoin deposits within a given

exchange helps mitigate the overall negative impact on the centralised exchange. Nevertheless, stablecoins could also have

an adverse effect on the overall value of assets held by exchanges in two scenarios: (i) when users withdraw their deposits

in stablecoins, and (ii) in the event of a stablecoin depegging and collapse, as exemplified by the Terra-Luna case.

The results observed in the previous sections imply that, in the case of Binance, KuCoin, and Huobi, stablecoins did

indeed mitigate the negative shock on the total value of the assets held without experiencing significant withdrawals in

stablecoins. However, at the same time, stablecoins in these exchanges can be considered a double-edged sword. The

large deposits of stablecoins could improve the resilience of the system during times of stress; however, if one of the large

stablecoins collapses, the centralized exchange will immediately go into bankruptcy. In other words, paradoxically, the

most resilient exchanges could become the most unstable overnight due to the large deposits in stablecoins.

Figure. 6: Scatter plots reporting the relationship between the proportion of stablecoin on centralized exchanges since 2022 and MVaR
(5%) & MVaR (1%) values.

−0.03 −0.06 Binance

Binance
Ku-Coin
Ku−coin
OKX Huobi
OKX Huobi
−0.04 −0.08

Bybit
MVaR(5%)

MVaR(1%)

Bybit
−0.05 −0.10 Crypto.com
Crypto.com
CheckSig Bitfinex Bitfinex
/BitMEX
−0.06 Swissborg −0.12 CheckSig/BitMEX
Deribit Deribit
Swissborg
−0.07 −0.14
0.0 0.2 0.4 0.0 0.2 0.4
% Stablecoin % Stablecoin

34
5.4. Exchange balance data

In Secs. (5.1), (5.2), and (5.3), we have utilized proof-of-assets data, as previously described in Sec.(3). This metric

exclusively focuses on the balances of a predefined set of addresses publicly disclosed by the respective exchange. This set

of addresses remains static and only changes when an exchange publicly updates its proof-of-assets addresses.

Alternatively, Glasssnode also provides a different methodology to identify the assets held by exchanges, namely,

exchange balance data (Schultze-Kraft, 2022). It is relevant to consider that exchanges operate on-chain in a highly

intricate manner, involving labyrinthine and dynamic transaction patterns and wallet management practices, each tailored

to the specific exchange. A complex network of cold, hot, and one-time wallets constantly evolves, occasionally undergoing

complete redesigns. In this context, data providers like Glassnode compute exchange balance metrics that are the result

of sophisticated data acquisition systems relying on a combination of publicly available information, clustering algorithms,

and exchange-specific heuristics. Exchange balance metrics do not necessarily align with the officially reported figures

provided by the exchanges themselves, nor can they guarantee an accurate quantification of on-chain exchange funds.

Instead, these metrics represent Glassnode’s best approximation of the actual balance held by each exchange, quantified

as closely and as verifiably accurate as possible, taking into account the intricate and ever-changing network of cold, hot,

and one-time wallets. In contrast, proof-of-assets data simply reflects the total coins held in addresses that exchanges

officially disclose.

With both of these datasets available, this section conducts a robustness analysis by applying the same methodologies

to exchange balance data as previously utilized for proof-of-assets data.

Exchange balances are frequently more dynamic as they encompass a broader range of addresses, whereas proof-of-

assets addresses typically constitute only a small portion of (cold) wallets. Depending on the comprehensiveness of the

officially disclosed proof-of-assets addresses, the total balance can closely align with the exchange balance data reported

by Glassnode. In Table 16, we present the ratios derived from dividing the proof-of-assets and exchange balance data

for Binance, Bitfinex, BitMEX, Bybit, Crypto.com, Deribit, Huobi, KuCoin and OKX.26 The outcome predominantly

underscores the similarity between both datasets, with a few exceptions. Indeed, as mentioned earlier, it is not uncommon

to encounter differences between the two datasets. This can be attributed to two primary factors: (i) the dynamic nature

of exchange balance data and (ii) the possibility that some addresses have not yet been disclosed due to the relatively

novel nature of these metrics. Related to the latter point, it is important to emphasize that due to the significant costs

associated with implementing a proof-of-reserves framework for all the assets held by platforms, some exchanges choose

to disclose only a subset of their available addresses for a selected sample of cryptocurrencies.

26 Exchange balance data were not available for CheckSig and Swissborg.

35
Table 16: Ratios computed by dividing proof-of-assets and exchange balance values.

Binance Bitfinex BitMEX Bybit Crypto.com Deribit Huobi KuCoin OKX

APE - - - 100.6196% - - - - -
BTC 59.8560% 64.6663% 100.0578% 99.2261% 97.9110% 100.1497% 37.6670% 100.1266% 83.8662%
BUSD 89.0674% - - - - - - - -
CRO - - - - 100.1735% - - - -
DAI - - - 100.1102% - - - - -
ENJ - - - - 100.8207% - - - -
ETH 97.5803% 24.2144% - 101.0294% 99.6957% 101.0294% 95.9603% 100.9480% 100.3616%
EURS - - - - - - - - 0.2030%
FTT - 100.3361% - - - - 1.1252% - -
HOT 97.2581% - - - - - - - -
HT - - - - - - 100.1753% - -
HUSD - - - - - - 0.1773% - -
LEO - 100.0437% - - - - - - -
LINK 95.5967% 100.4543% - 100.4543% 100.1267% - 68.0065% 100.4509% -
MANA 88.5609% 100.5305% - - - - - - -
MATIC - - - 101.0500% 100.3270% - - - -
OCEAN - - - - - - - 507.3492% -
OMG - 98.3720% - - - - - - -
QNT - - - - - - - 99.9669% -
RSR - - - - - - - - 0.0075%
SAN - 100.5445% - - - - - - -
SAND 97.2636% - - 100.6510% - - - - -
SHIB 90.8924% - - 100.5801% 100.4863% - - 100.5786% -
TEL - - - - - - - 101.4644% -
UNI - - - - - - 23.3748% - -
USDC 97.5492% 100.0976% - 100.0976% 95.9986% 100.0976% 100.0949% 99.8058% 93.6034%
USDK - - - - - - - - 71.0101%
USDP - - - - - - - - 99.4648%
USDT 97.8875% 2.0116% - 100.0837% 78.6462% - 41.7746% 100.0716% 98.2400%
UTK - - - - - - 609.6168% 381.9803% -
WBTC - - - - 97.2292% - - - -
WETH - - - - - - - - 0.2949%

Following the same approach as in Sec.(5.1), Table 17 and 18 report the AR-GARCH estimates for the event dummies

related to the most significant collapses in 2022. Table 17 and 18 consider the Period [0] and [-1,1], respectively. We can

generally observe consistent results: (i) the FTX bankruptcy event (d5 ) had the most negative impact on all centralized

exchanges, (ii) the collapses of Terra-Luna (d1 ) and Three Arrows Capital (d2 ) also had significant adverse effects, (iii)

Binance, KuCoin, OKX, Huobi and Bitfinex displayed considerable resilience in the face of these events, (iv) in the subset

Period [-1,1] (Table 18), the negative impact of Terra-Luna (d1 ) and Three Arrows Capital (d2 ) events disappears, while

the significant and negative impact of the FTX bankruptcy event remains, underlining the robustness of the CeFi system

to external shocks not directly tied to this ecosystem.

Table 17: AR-GARCH results based on the selected negative events (d) in 2022. The dummy variables are included for the exact day on
which the events occurred, i.e., Period [0].

Period [0] Binance Bitfinex BitMEX Bybit Crypto.com Deribit Huobi KuCoin OKX
d1 -0.0643*** -0.0679*** -0.0682*** -0.0669*** -0.1024*** -0.0881*** -0.0660*** -0.0557*** -0.0661***
d2 -0.0520*** -0.0548*** -0.1010*** -0.0870*** -0.0765*** -0.1136*** -0.0674*** -0.0530*** -0.0736***
d3 -0.0104 -0.0086 -0.0268 -0.0183 -0.0199 -0.0203 -0.0280 -0.0092 -0.0130
d4 0.0105 0.0086 0.0144 0.0160 0.0180 0.0218 0.0119 0.0119 0.0142
d5 -0.0609*** -0.0613*** -0.1060*** -0.0942*** -0.1285*** -0.1276*** -0.1202*** -0.0565*** -0.0753***
d6 -0.0059 -0.0169 -0.0132 -0.0073 -0.0105 -0.0160 -0.0506*** -0.0079 -0.0087

36
Table 18: AR-GARCH results based on the selected negative events (d) in 2022. The dummy variables are included, considering the day
before, the specific day of the events, and the day after, i.e., Period [-1,1].

Period [-1,1] Binance Bitfinex BitMEX Bybit Crypto.com Deribit Huobi KuCoin OKX
d1 -0.0162 -0.0193 -0.0016 -0.0154 -0.0363 -0.0203 -0.0156 -0.0084 -0.0158
d2 -0.0067 -0.0134 -0.0074 -0.0116 -0.0116 -0.0056 -0.0001 -0.0065 -0.0080
d3 -0.0059 -0.0074 -0.0144 -0.0112 -0.0087 -0.0130 -0.0139 -0.0051 -0.0082
d4 0.0144** 0.0164** 0.0164* 0.0224** 0.0199* 0.0324*** 0.0185** 0.0145** 0.0208**
d5 -0.0562*** -0.0586*** -0.1063*** -0.0888*** -0.1107*** -0.1278*** -0.1014*** -0.0549*** -0.0705***
d6 0.0005 0.0002 0.0000 0.0004 0.0012 0.0001 0.0051 0.0028 0.0003

Table 19 and 20 continue to exhibit consistent results when compared to Sec.(5.1). While some minor changes may

be noticeable, Binance and KuCoin remain the most resilient exchanges, as indicated by the MVaR values. They are

followed by Huobi, Bitfinex, and OKX.27

Table 19: MVaR(5%) & MVaR(1%) for the portfolio corresponding to each centralised exchange. Centralised exchanges are ordered alpha-
betically.

Balance Binance Bitfinex BitMEX Bybit Crypto.com Deribit Huobi KuCoin OKX
MVaR(5%) -0.0357 -0.0387 -0.0569 -0.0468 -0.0535 -0.0625 -0.0445 -0.0329 -0.0436
MVaR(1%) -0.0680 -0.0846 -0.1234 -0.0995 -0.1035 -0.1277 -0.0846 -0.0620 -0.0851

Table 20: MVaR(5%) & MVaR(1%) for the portfolio corresponding to each centralised exchange. Centralised exchanges are ordered accord-
ing to MVaR(5%) & MVaR(1%) values.

KuCoin Binance Bitfinex OKX Huobi Bybit Crypto.com BitMEX Deribit

MVaR(5%) -0.0329 -0.0357 -0.0387 -0.0436 -0.0445 -0.0468 -0.0535 -0.0569 -0.0625

KuCoin Binance Huobi Bitfinex OKX Bybit Crypto.com BitMEX Deribit

MVaR(1%) -0.0620 -0.0680 -0.0846 -0.0846 -0.0851 -0.0995 -0.1035 -0.1234 -0.1277

6. Discussion: The problem of impersonal trust

As described in the introduction, the 2007 subprime mortgage crisis served as a stark reminder of the importance

of responsible behavior among lenders, investment banks, and credit rating agencies. It underscored the crucial role

that effective trustees play in fostering economic growth and highlighted the significance of impersonal trust within the

economy (Fungáčová et al., 2019). In response to this crisis, regulatory authorities implemented a comprehensive set of

measures to safeguard consumers and ensure the stability of the financial system in the years ahead (Gertler and Gilchrist,

2018).

As an alternative response to the aftermath of the 2008 financial crisis, Nakamoto (2008) introduced the concept of

blockchain through BTC as a novel approach to eradicating the need for trust. However, the subsequent development

of the cryptocurrency ecosystem has diverged significantly from this original vision, and trust-related problems persist.

After our empirical analysis, within Shapiro’s framework, we can even state that the issue of impersonal trust in the

27 Bitfinex’s higher resilience when using exchange balance data can be attributed to the presence of substantial stablecoin deposits in

exchange balance information, as opposed to the proof-of-assets dataset.

37
crypto ecosystem is far more significant than in traditional finance. A user investing in a centralized exchange does need

to trust all the interconnected and unregulated elements of the crypto ecosystem, such as existing blockchain networks,

decentralised exchanges, oracles, bridges, DAOs, lending/borrowing companies, and stablecoins, among other third-party

technology providers (Financial Stability Board, 2023). On a local level, the user must also place trust in the resilience

and stability of all the cryptocurrency assets held by the exchange. This is crucial because the failure of a significant

cryptocurrency could potentially lead to the exchange’s bankruptcy. As a result, any user who invests in cryptocurrencies

via centralized exchanges must consider three fundamental factors: (i) the inherent nature of multi-asset exchanges, (ii)

stablecoins, and (iii) the chain of impersonal trust.

6.1. Multi-asset exchange vs BTC-only exchange

Buxton et al. (2022) suitably stated that “a digital asset platform that offers multiple types of digital assets to a

globally distributed retail customer base will likely have a high number of counterparties with varying account balances

across multiple legal and tax jurisdictions. The risk assessment for that type of digital asset platform would look very

different from a digital asset platform that only offers institutional investors custodial services for BTC”. Indeed, our

proof-of-asset analysis highlights significant heterogeneity among the various centralized exchanges available in the market.

On one hand, exchanges such as Binance and KuCoin offer users a wide range of assets for investment. On the other hand,

BitMEX and CheckSig primarily focus on BTC. Interestingly, as observed in previous empirical sections, Binance and

KuCoin appear to be safer options than BitMEX and CheckSig from a financial perspective due to their lower exposure

to downside risks. This reduced risk is closely linked to their asset composition. BitMEX and CheckSig are significantly

influenced by BTC’s volatility, whereas Binance and KuCoin mitigate downside risks by maintaining a diversified exchange

with considerable deposits of stablecoins. At this point, a fundamental question arises: from a trust perspective, which

exchange is more secure and stable? If a user’s primary focus is investing in BTC and they place significant importance

on trust, they may find BitMEX or CheckSig to be more suitable options compared to Binance and KuCoin. Depositing

funds in BitMEX primarily requires trust in its management and BTC’s stability. In contrast, investing in Binance

entails trusting not only its management but also all the associated cryptocurrency assets and services, along with the

corresponding third-party management. Indeed, as observed in Sec.(3), Binance hold 82.09% of all the assets aggregated

in the “Other” category across the 11 exchanges, which underlines its multi-asset nature. This rationale aligns with Carter

(2023)’s proof-of-reserve classification, as BitMEX provides the best proof-of-reserve mechanism, evidenced by its frequent

application (weekly), full disclosure of liabilities, coverage of all BTCs, and the ability for users to verify their inclusion

in the liability set. Consequently, exchange size and the number of cryptoassets offered play a significant role, given

that the smaller the exchange, the easier it is to apply proper proof-of-reserves mechanisms. However, a large exchange

characterized by a multi-asset nature will face the need to apply additional regulations to guarantee the proper operation

and implementation of proof-of-reserves mechanisms. It is at this juncture that the significant deposits of stablecoins

within exchanges and the concept of the chain of impersonal trust also become particularly relevant.

38
6.2. Stablecoins

As explained in Sec.(5.3), stablecoins within centralised exchanges can be likened to a double-edged sword. While the

substantial deposits of stablecoins can enhance the system’s resilience during periods of stress, it also implies that if one

of the major stablecoins were to collapse, the centralized exchange would immediately face insolvency. Within the context

outlined by Shapiro (1987), this signifies that users must also place impersonal trust in the management responsible for

stablecoins and their ability to maintain the stablecoin’s parity with the corresponding fiat currency. In essence, if a user

intends to deposit their funds in multi-asset exchanges for trading across a diverse range of cryptocurrencies, they should

assess the proportion of stablecoins within the exchange and the reliability of the third entity acting as a stablecoin issuer.

From a policymaker’s perspective, the risks outlined in this paper concerning stablecoins align with the findings of

reports issued by the European Central Bank. Adachi et al. (2021) stated that “given the potential risks and cross-

border nature of stablecoins, a granular and robust global regulatory approach is essential”. In the same line, Born and

Vendrell Simón (2023) affirmed: “given that the largest stablecoins serve a critical function for crypto-asset markets’

liquidity, this could have wide-ranging implications for crypto-asset markets if there is a run on or failure of one of the

largest stablecoins. In turn, this could have contagion effects for the financial system if at some point in the future crypto-

asset markets pose a risk to financial stability. A run on a stablecoin could also have contagion effects for the financial

system through large-scale redemptions of reserve assets, which usually comprise traditional assets such as government

bonds or commercial paper”. Consequently, due to the vital role that stablecoins play in the cryptocurrency ecosystem,

there is an urgent need to bring them under the regulatory umbrella, a task that can be addressed through the MiCA

framework and additional regulatory measures.

6.3. Chain of impersonal trust

The points mentioned above emphasize the presence of an intricate chain of impersonal trust within centralized

exchanges; a factor that cannot be overlooked by users, policymakers, and the scientific community. When a user relies

on a specific centralized exchange, they are required to trust not only the exchange itself and its management but also

all the interconnected components, forming a complex network that is challenging to assess due to the involvement of

numerous third parties in the cryptocurrency space. The absence of a regulatory framework to guarantee transparency

for most entities in the crypto industry exacerbates this complexity, resulting in somewhat of a black box. Consequently,

users must place their trust not only in the centralized exchanges but also in the entire cryptocurrency ecosystem as a

whole. In other words, users must have blind trust in the crypto ecosystem.

6.4. Proof-of-reserves & Proof-of-solvency

The proof-of-reserves and proof-of-solvency approaches can be considered among the first concepts aimed at unveiling

the black box and reducing the necessity for users to blindly trust centralized exchanges and the crypto system. However,

they do not represent the final solution but rather an initial step.

As explained in earlier sections, proof-of-reserves is the combination of proof-of-assets and liabilities. It refers to a

specific process in which a custodian openly verifies the presence of on-chain assets and subsequently provides supporting

39
evidence, often with the assistance of an auditor, to ensure that the outstanding liabilities do not surpass the value

of those assets (Buterin, 2022; Buxton et al., 2022). In contrast to traditional financial statement audits, which tend

to be slow, costly, and infrequent, proof-of-reserve mechanisms offer a promising application of blockchain technology.

These mechanisms enable exchanges to provide irrefutable evidence, either daily or weekly (such as BitMEX), to their

customers, government authorities, or the general public, confirming the presence of all the assets they have claimed

to hold. However, there are essential factors beyond the scope of this process, including (i) the presence of inadequate

management, (ii) key loss, (iii) fraud, (iv) concealed off-chain instruments and liabilities, and (v) evaluation of the asset

composition (Carter, 2023). This is the reason why proof-of-reserves approach does not provide assurances regarding the

solvency of the centralised exchange.

Despite the presence of several relevant factors beyond the scope of proof-of-reserves, this paper has placed significant

emphasis on the importance of evaluating the asset composition. Indeed, the positive effects of implementing the proof-of-

reserves approach may be limited or entirely absent if there is not a thorough assessment of the assets held by exchanges.

In other words, even if a particular exchange correctly implements the existing proof-of-reserves framework, it could still

face insolvency due to the collapse of some of its assets. For example, certain stablecoins might be susceptible to failures

due to inadequate blockchain frameworks, as seen in the case of Terra Luna; exchange tokens might be launched by

insolvent exchanges, as in the case of FTX; or tokens associated with unviable projects could also experience collapses.

Thus, when considering an approach that encompasses all the interconnections within a specific exchange, it becomes

crucial for proof-of-solvency to include an examination of the assets listed on the company’s balance sheet. These

assets have a direct impact on solvency and the exchange’s ability to fulfill customer fund obligations. From a practical

perspective, exchanges should implement additional “guardians of trust”, in the form of complementary information

about the potential risks that their users might encounter when holding certain types of crypto-assets. For example: (i)

if there is a substantial deposit of stablecoins, the exchange should clarify whether the stablecoin issuers fully backs the

digital dollars and offers the necessary transparency; (ii) if there is a substantial deposit of exchange tokens, the exchange

should specify whether the associated exchanges implement proof-of-reserves mechanisms as part of the proof-of-solvency

framework; (iii) if a significant deposit of meme coins exists, especially those without robust underlying projects, the

exchange should advise users about the inherent risks associated with these tokens and the subsequent risks they pose to

the exchange. This rationale should be applied to the different assets held by centralised exchanges.

On the other hand, considering the FTX collapse, which involved a conglomerate of over 100 companies prior to its

downfall, it is also imperative for exchanges to disclose details about the formation of such conglomerates. This should

involve explicit information regarding the interrelationships among the various constituent companies, their assets and

liabilities. Moreover, it should be mandatory for exchanges to disclose the addresses included in the proof-of-reserves.

This practice not only allows third parties to self-verify asset balances presented in the assessment but also could enable

policymakers to formulate more robust regulations aimed at ensuring transparency and safeguarding customer interests.

40
7. Conclusion

The prevailing literature has largely overlooked the importance of proof-of-solvency mechanisms. To address this gap,

our paper provided a brief review of existing literature, followed by an exploration of proof-of-assets, proof-of-liabilities,

proof-of-reserves, and proof-of-solvency concepts. We also reported an in-depth examination of the evolutionary path of

proof-of-reserves, starting with the Maxwell/Todd Merkle approach proposed by Wilcox (2014) for validating liabilities.

Moreover, we performed an empirical analysis of proof-of-assets using data disclosed by leading centralized exchanges

such as Binance, Bitfinex, BitMEX, Bybit, CheckSig, Crypto.com, Deribit, Huobi, KuCoin, OKX, and SwissBorg. This

analysis aimed at analysing the asset composition of these exchanges and exploring its potential implications for the

future of the cryptocurrency ecosystem. Additionally, we assessed the stability of centralized exchanges in the face of

extreme events. From an empirical perspective, our contribution to the literature encompassed three key dimensions:

(i) the evaluation of exchange asset composition, (ii) an examination of centralized exchanges’ resilience during extreme

events in 2022, and (iii) an exploration of the significance of impersonal trust in the crypto sphere, within the framework

proposed by Shapiro (1987).

On the one hand, when assessing asset composition, we have observed relevant patterns among various exchanges.

Firstly, we have highlighted the heterogeneity among centralized exchanges, notably with BitMEX and CheckSig primarily

focusing on BTC, while other platforms offer a broader selection of alternative assets. Binance and KuCoin, for instance,

stand out by providing a wide range of alternative cryptocurrencies. Secondly, in terms of asset composition, we observed

that BTC generally constitutes at least 20% of proof-of-assets on all exchanges, and similarly, ETH represents a minimum

of 9% across the board. Remarkably, stablecoins play a substantial role in the asset composition of exchanges, even

surpassing the percentages of BTC and ETH. In fact, there were instances when Binance, Huobi, KuCoin, Bybit, and

OKX held over 50% of total customer assets in stablecoins during certain periods. This highlights policymakers’ concerns

about potential risks associated with stablecoins in shaping the future of the cryptocurrency ecosystem. Thirdly, Binance

emerges as the preferred platform for users to deposit and trade cryptocurrencies. It notably stands out as the primary

custodian of BTC, ETH, and stablecoins holding 40.43%, 59.73% and 71.48% respectively, across the 11 exchanges in

our dataset. Given Binance’s overwhelming dominance and control over a significant portion of crypto capital, it is

unsurprising that the SEC is closely scrutinizing all procedures and potential transactions associated with the exchange.

On the other hand, our evaluation of centralized exchanges’ stability during extreme events involved two approaches:

(i) an event study focused on incidents in 2022, such as the collapses of Terra-Luna and FTX, using the AR-GARCH

framework, and (ii) a Value at Risk (VaR) assessment to quantify the downside risk associated with these exchanges.

Our general findings were as follows: (i) The FTX bankruptcy event had the most detrimental impact on all centralized

exchanges; (ii) The collapses of Terra-Luna and Three Arrows Capital also had significant adverse effects; (iii) Binance,

KuCoin, OKX, Huobi, and Bitfinex exhibited substantial resilience in the face of these events; (iv) When considering the

day of the event, the day before, and the day after, the negative impact of the Terra-Luna and Three Arrows Capital

events dissipated, whereas the significant and negative impact of the FTX bankruptcy event persisted, underscoring the

robustness of the CeFi system in the face of external shocks; (v) Stablecoins in these exchanges can be considered a double-

41
edged sword, given that the large deposits held in the exchange could improve the resilience of the system during times

of stress; however, if one of the large stablecoins collapses, the centralized exchange will immediately go into bankruptcy;

(vi) In light of the results obtained from the AR-GARCH framework and a MVaR assessment, it appears that the CeFi

system should maintain a capital conservation buffer ranging from 6% to 14%, depending on the specific characteristics

of each centralized exchange.

Within the framework proposed by Shapiro (1987), the previously mentioned points emphasize the presence of a

complex chain of impersonal trust within centralized exchanges. This factor is of paramount importance for users,

policymakers, and the scientific community to consider. When users select a specific centralized exchange, they must place

their trust in multiple layers of entities and components. This includes trust in the exchange itself and its management,

but it extends far beyond that. Users must also trust in all interconnected elements, forming a complex network that

poses challenges in assessment due to the involvement of numerous third parties in the cryptocurrency realm.

The concepts of proof-of-reserves and proof-of-solvency represent some of the initial steps taken to enhance trans-

parency and reduce the uncertainty regarding centralized exchanges and the broader cryptocurrency ecosystem. However,

to improve the credibility of CeFi and the cryptocurrency ecosystem as a whole, two key actions are necessary. First,

centralized exchanges should offer additional information about their asset holdings and provide greater transparency

regarding their overall corporate structure and operations. Second, policymakers should propose further regulations, par-

ticularly focusing on off-chain elements that cannot be directly tracked through the blockchain. These regulations would

help ensure accountability and security in the cryptocurrency space.

By taking these measures, it could be possible to work towards establishing a more trustworthy and reliable environment

for participants in the cryptocurrency market, reducing the risks associated with trust and enhancing the overall integrity

of the industry. However, at present, the cryptocurrency ecosystem, which philosophically aimed to minimize the reliance

on trust, paradoxically requires more “guardians of trust” than traditional finance.

8. Acknowledgements

The author acknowledges the financial support from the Margarita Salas contract MGS/2021/13 (UP2021-021) financed

by the European Union-NextGenerationEU, and the financial support from Generalitat Valenciana under the project

CIGE/2022/132.

References

Adachi, M., Born, A., Gschossmann, I., and Van der Kraaij, A. (2021). The expanding functions and uses of stablecoins.

Financial Stability Review (ECB).

Adachi, M., Pereira Da Silva, P. B., Born, A., Cappuccio, M., Czák-Ludwig, S., Gschossmann, I., Paula, G., Pellicani, A.,

Philipps, S.-M., Mirjam, P., Rossteuscher, I., and Zeoli, P. (2022). Stablecoins’ role in crypto and beyond: functions,

risks and policy. European Central Bank.

42
Aramonte, S., Huang, W., and Schrimpf, A. (2021). Defi risks and the decentralisation illusion.

Arner, D. W., Auer, R., and Frost, J. (2020). Stablecoins: risks, potential and regulation. Bank for International

Settlements.

Asmakov, A. (2022). Japan passes stablecoin bill aimed at protecting crypto investors. Decrypt.

Bank for International Settlements (2022). Application of theprinciples for financial market infrastructures tostablecoin

arrangements. Technical report. 22 July 2022.

Bank for International Settlements (2023). The crypto ecosystem:key elements and risks. Bank for International Settle-

ments.

Barbosa, V. (2023). Blackrock discloses ‘exposure to stablecoin risks’ on spot bitcoin etf. Finbold.

Barucci, E., Giuffra Moncayo, G., and Marazzina, D. (2023). Market impact and efficiency in cryptoassets markets.

Digital Finance, pages 1–44.

Barucci, E., Moncayo, G. G., and Marazzina, D. (2022). Cryptocurrencies and stablecoins: a high-frequency analysis.

Digital Finance, 4(2-3):217–239.

Beau, D. (2023). After the crypto-winter, the spring of crypto-assetsregulation and supervision. Bank for international

settlements.

BitMEX (2023). Proof of reserves and liabilities. https://www.bitmex.com/app/porl.

Bitpanda (2022). Proof of reserves andliquidity in crypto explained. Crypto security lessons.

Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. Journal of econometrics, 31(3):307–327.

Born, A. and Vendrell Simón, J. M. (2023). A deep dive into crypto financial risks: stablecoins, defi and climate transition

risk. Macroprudential Bulletin. European Central Bank.

Boudt, K., Peterson, B. G., and Croux, C. (2008). Estimation and decomposition of downside risk for portfolios with

non-normal returns. Journal of Risk, 11(2):79–103.

Briola, A. and Aste, T. (2022). Dependency structures in cryptocurrency market from high to low frequency. Entropy,

24(11):1548.

Briola, A., Vidal-Tomás, D., Wang, Y., and Aste, T. (2023). Anatomy of a stablecoin’s failure: The terra-luna case.

Finance Research Letters, 51:103358.

Bullmann, D., Klemm, J., and Pinna, A. (2019). In search for stability in crypto-assets: are stablecoins the solution?

ECB Occasional Paper, (230).

Buterin, V. (2022). Having a safe cex: proof of solvency and beyond. Vitalik.ca.

43
Buterin, V., Illum, J., Nadler, M., Schär, F., and Soleimani, A. (2023). Blockchain privacy and regulatory compliance:

Towards a practical equilibrium. Available at SSRN.

Buxton, N., Carter, N., Davine Kim, A., South, P., and Ternullo, S. (2022). Proof of reserves: The practitioner’s guide

to an emerging standard for increasing trust and transparency in digital asset platform services. Chamber of Digital

Commerce.

Bybit (2022). What is a merkle tree & what is its role in blockchain? Bybit Learn.

Camacho, P. (2014). Secure protocols for provable security. Working Paper.

https://es.slideshare.net/philippecamacho/protocols-for-provable-solvency-38501620.

Carter, N. (2022a). Proof of reserves for policymakers. Medium.

Carter, N. (2022b). The status of proof of reserve as of year end 2022. Medium.

Carter, N. (2023). Nic’s por. wall of fame. niccarter.info. https://niccarter.info/proof-of-reserves/.

Chalkias, K., Chatzigiannis, P., and Ji, Y. (2022). Broken proofs of solvency in blockchain custodial wallets and exchanges.

Cryptology ePrint Archive.

Chalkias, K., Lewi, K., Mohassel, P., and Nikolaenko, V. (2019). Practical privacy preserving proofs of solvency. Amster-

dam ZKProofCommunity Event.

Chalkias, K., Lewi, K., Mohassel, P., and Nikolaenko, V. (2020). Distributed auditing proofs of liabilities. Cryptology

ePrint Archive.

CheckSig (2023). The proof of reserves. https://www.checksig.com/por.

Christiano, L. J., Eichenbaum, M. S., and Trabandt, M. (2015). Understanding the great recession. American Economic

Journal: Macroeconomics, 7(1):110–167.

Clarke, A. (2022). 7 biggest crypto collapses of 2022 the industry would like to forget. Cointelegraph.

CNBC (2022). Crypto exchanges that have published their proof-of-reserves and what these audits tell us. CNBC.

Coghlan, J. (2022). Ftx’s ongoing saga: Everything that’s happened until now. Cointelegraph.

CoinMarketCap (2023). Coinmarketcap: Historical data.

Congress, U. (2010). Dodd-frank wall street reform and consumer protection act.

Conlon, T., Corbet, S., and Hou, Y. G. (2024). Contagion effects of permissionless, worthless cryptocurrency tokens:

Evidence from the collapse of ftx. Journal of International Financial Markets, Institutions and Money, 91:101940.

Conlon, T., Corbet, S., and Hu, Y. (2023). The collapse of the ftx exchange: The end of cryptocurrency’s age of innocence.

The British Accounting Review, page 101277.

44
Conlon, T. and McGee, R. (2020). Safe haven or risky hazard? bitcoin during the covid-19 bear market. Finance Research

Letters, page 101607.

Cont, R. (2001). Empirical properties of asset returns: stylized facts and statistical issues. Quantitative Finance, 1(2):223–

236.

Cornish, E. A. and Fisher, R. A. (1938). Moments and cumulants in the specification of distributions. Revue de l’Institut

international de Statistique, pages 307–320.

CryptoCompare (2023). Centralised exchange retrospective. CryptoCompare.

Dagher, G. G., Bünz, B., Bonneau, J., Clark, J., and Boneh, D. (2015). Provisions: Privacy-preserving proofs of solvency

for bitcoin exchanges. In Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications Security,

pages 720–731.

Defillama (2023). Cex transparency. https://defillama.com/cexs.

Demyanyk, Y. and Van Hemert, O. (2011). Understanding the subprime mortgage crisis. The review of financial studies,

24(6):1848–1880.

Ding, Z., Granger, C. W., and Engle, R. F. (1993). A long memory property of stock market returns and a new model.

Journal of empirical finance, 1(1):83–106.

Dowling, M. (2021a). Fertile land: Pricing non-fungible tokens. Finance Res. Lett., page 102096.

Dowling, M. (2021b). Is non-fungible token pricing driven by cryptocurrencies? Finance Res. Lett., page 102097.

ECB, C.-A. T. F. (2020). Stablecoins: Implications for monetary policy, financial stability, market infrastructure and

payments, and banking supervision in the euro are. Occasional Paper Series.

Engle, R. F. and Bollerslev, T. (1986). Modelling the persistence of conditional variances. Econometric reviews, 5(1):1–50.

European Council (2022). Digital finance: agreement reached on european crypto-assets regulation (mica). Technical

report. 22 July 2022.

Financial Stability Board (2023). The financial stability risks of decentralised finance.

Financial Times (2022). Crypto hedge fund three arrows fails to meet lender margin calls. Financial Times.

Fungáčová, Z., Hasan, I., and Weill, L. (2019). Trust in banks. Journal of Economic Behavior & Organization, 157:452–

476.

Ge Huang, V. (2022). Crypto broker voyager digital suspends withdrawals. Wall Street Journal.

Gertler, M. and Gilchrist, S. (2018). What happened: Financial factors in the great recession. Journal of Economic

Perspectives, 32(3):3–30.

45
Giot, P. and Laurent, S. (2003). Value-at-risk for long and short trading positions. Journal of Applied Econometrics,

18(6):641–663.

Glassnode (2023). https://glassnode.com/.

Glosten, L. R., Jagannathan, R., and Runkle, D. E. (1993). On the relation between the expected value and the volatility

of the nominal excess return on stocks. The journal of finance, 48(5):1779–1801.

Goswami, R. (2023). Sec sues binance and ceo changpeng zhao for u.s. securities violations. CNBC.

Hu, K., Zhang, Z., and Guo, K. (2019). Breaking the binding: Attacks on the merkle approach to prove liabilities and its

applications. Computers & Security, 87:101585.

Ji, Y. and Chalkias, K. (2021). Generalized proof of liabilities. IACR LogoCryptology ePrint Archive.

Katsiampa, P. (2017). Volatility estimation for bitcoin: A comparison of garch models. Economics Letters, 158:3–6.

Khoo, Y. L., Leow, S., Choe, L., Polk, N., and Chia, D. (2022). Blockchain analysis: The collapse of alameda and ftx.

Nansen.

Lindrea, B. (2023). Sec still looking for potential ftx-style fraud at binance.us: Report. Cointelegraph.

Liu, J., Makarov, I., and Schoar, A. (2023). Anatomy of a run: The terra luna crash. Technical report, National Bureau

of Economic Research.

Medium (2023). The risks of trusting centralized exchanges: Lessons from the ftx crash. Medium.

Meiklejohn, S., Pomarole, M., Jordan, G., Levchenko, K., McCoy, D., Voelker, G. M., and Savage, S. (2013). A fistful

of bitcoins: characterizing payments among men with no names. In Proceedings of the 2013 conference on Internet

measurement conference, pages 127–140.

Messari (2023). Messari: Messari crypto research, data, and tools. https://messari.io.

Murinde, V., Rizopoulos, E., and Zachariadis, M. (2022). The impact of the fintech revolution on the future of banking:

Opportunities and risks. International Review of Financial Analysis, 81:102103.

Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system.

Nelson, D. B. (1991). Conditional heteroskedasticity in asset returns: A new approach. Econometrica: Journal of the

econometric society, pages 347–370.

Official Journal of the European Union (2023). European union’s markets in crypto assets regulation. Official Journal of the

European Union. REGULATION (EU) 2023/1114 OF THE EUROPEAN PARLIAMENT AND OF THE COUNCILof

31 May 2023.

Pan, D. (2022). Binance sees 30% surge in trading activity on ftx implosion. Bloomberg.

46
Pines, M., Zell, D., McCarty, G., and Jason, B. (2022). Proof of assets: A summary analysis. Bitcoin Policy Institute.

Prentice, C., Godoy, J., and Lang, H. (2024). Binance to face off against us securities regulator in court. Reuters.

Ray III, J. J. (2022). Declaration of john, j. ray iii. In support of chapter 11 petitions and first day pleadings.

Rosenberg, E. (2023). Meme coin explained. Investopedia.

Ryan, S. G. (2008). Accounting in and for the subprime crisis. The accounting review, 83(6):1605–1638.

Saggese, P., Segalla, E., Sigmund, M., Raunig, B., Zangerl, F., and Haslhofer, B. (2023). Assessing the solvency of virtual

asset service providers: Are current standards sufficient? arXiv preprint arXiv:2309.16408.

Schultze-Kraft, R. (2022). Introducing proof-of-reserve exchange metrics. Glassnode insights.

Sensoy, A., Silva, T. C., Corbet, S., and Tabak, B. M. (2021). High-frequency return and volatility spillovers among

cryptocurrencies. Applied Economics, 53(37):4310–4328.

Shapiro, S. P. (1987). The social control of impersonal trust. American journal of Sociology, 93(3):623–658.

Tan, T. M. and Saraniemi, S. (2023). Trust in blockchain-enabled exchanges: Future directions in blockchain marketing.

Journal of the Academy of marketing Science, 51(4):914–939.

The Block (2023). Decentralised finance. The Block. https://www.theblock.co/data/decentralized-finance/dex-non-

custodial.

The Guardian (2022). Crypto lender blockfi files for bankruptcy after ftx collapse. The Guardian.

Thomas, D. (2023). Binance releases new proof-of-reserves report with some tokens missing. Be(in)crypto.

Trucı́os, C. (2019). Forecasting bitcoin risk measures: A robust approach. International Journal of Forecasting, 35(3):836–

847.

Vidal-Tomás, D. (2023). The illusion of the metaverse and meta-economy. International Review of Financial Analysis,

86:102560.

Vidal-Tomás, D., Briola, A., and Aste, T. (2023). Ftx’s downfall and binance’s consolidation: the fragility of centralized

digital finance. Physica A, 625:129044.

Vidal-Tomás, D. and Ibañez, A. (2018). Semi-strong efficiency of bitcoin. Finance Research Letters, 27:259–265.

Vidal-Tomás, D. (2022). Which cryptocurrency data sources should scholars use? International Review of Financial

Analysis, 81:102061.

Wade, J. (2023). Proof of reserves: Could it have avoided the ftx meltdown? Investopedia.

Wilcox, Z. (2014). Proving your bitcoin reserves. Internet Archive.

47
Williamson, O. E. (1993). Calculativeness, trust, and economic organization. The journal of law and economics, 36(1,

Part 2):453–486.

Wilmott, P. (1998). Derivatives: The theory and practice of financial engineering. John Wiley & Son Limited.

Yousaf, I. and Goodell, J. W. (2023). Reputational contagion and the fall of ftx: Examining the response of tokens to the

delegitimization of ftt. Finance Research Letters, 54:103704.

Zangari, P. (1996). A var methodology for portfolios that include options. RiskMetrics Monitor, 1:4–12.

Zucker, L. G. (1986). Production of trust: Institutional sources of economic structure, 1840–1920. Research in organiza-

tional behavior.

48