An Empirical Study of Cross-Chain Arbitrage in Decentralized Exchanges

Abstract

Blockchain interoperability refers to the ability of blockchains to share information with each other. Decentralized Exchanges (DEXs) are peer-to-peer marketplaces where traders can exchange cryptocurrencies. Several studies have focused on arbitrage analysis within a single blockchain, typically in Ethereum. Recently, we have seen a growing interest in cross-chain technologies to create a more interconnected blockchain network. We present a framework to study cross-chain arbitrage in DEXs. We use this framework to analyze cross-chain arbitrages between two popular DEXs, PancakeSwap and QuickSwap, within a time frame of a month. While PancakeSwap is implemented on a blockchain named BNB Chain, QuickSwap is implemented on a different blockchain named Polygon. The approach of this work is to study the cross-chain arbitrage through an empirical study. We refer to the number of arbitrages, their revenue as well as to their duration. This work lays the basis for understanding cross-chain arbitrage and its potential impact on the blockchain technology.

Full text

An Empirical Study of Cross-chain Arbitrage in
Decentralized Exchanges
Ori Mazor
Technion - Israel Institute of Technology
orimazor@campus.technion.ac.il
Ori Rottenstreich
Technion - Israel Institute of Technology
or@technion.ac.il
Abstract—Blockchain interoperability refers to the ability of
blockchains to share information with each other. Decentralized
Exchanges (DEXs) are peer-to-peer marketplaces where traders
can exchange cryptocurrencies. Several studies have focused
on arbitrage analysis within a single blockchain, typically in
Ethereum. Recently, we have seen a growing interest in cross-
chain technologies to create a more interconnected blockchain
network. We present a framework to study cross-chain ar-
bitrage in DEXs. We use this framework to analyze cross-
chain arbitrages between two popular DEXs, PancakeSwap and
QuickSwap, within a time frame of a month. While PancakeSwap
is implemented on a blockchain named BNB Chain, QuickSwap
is implemented on a different blockchain named Polygon. The
approach of this work is to study the cross-chain arbitrage
through an empirical study. We refer to the number of arbitrages,
their revenue as well as to their duration. This work lays the basis
for understanding cross-chain arbitrage and its potential impact
on the blockchain technology.
I. Introduction
Decentralized Finance (DeFi) [1], [2] is an emerging
technology for various financial services that run as smart
contracts, often implemented over the Ethereum blockchain
network [3]. The DeFi market size was estimated to be over
180 billion USD in late 2021 and 63.1 billion USD in Novem-
ber 2023. Typical DeFi applications include blockchain-based
lending: The ability to lend and borrow cryptocurrencies in
exchange for interest [4] as well as insurance services. A
primary DeFi application is decentralized exchanges (DEXs)
that allow the exchange of cryptocurrencies as non-centralized
marketplaces connecting cryptocurrency traders.
As the popularity and usage of blockchain increased, so
did the challenges and limitations. Some of the most common
challenges are scalability, decentralization, and security. The
mutual dependency is known as the blockchain trilemma as
often the improvement of one compromises others. Numerous
projects have implemented various solutions to address these
issues, forming their own blockchains as indicated in Table I.
Within this table, many blockchains, including Polygon and
BNB Chain, are EVM (Ethereum Virtual Machine) compatible
and can run smart contracts. Traditionally, such blockchains
are independent without the mutual ability to communicate.
Arbitrage refers to the simultaneous purchase and sale of
the same asset in different markets in order to profit from
differences in the asset price. It exploits variations in the price
of identical assets in different markets. DEXs have an inherent
potential for arbitrage as several of them refer to a similar
set of major cryptocurrencies with dynamic exchange rates
influenced by the liquidity of each cryptocurrency. A range
of past works study arbitrages in DEXs implemented over
Ethereum as a single blockchain [5],[6],[7].
Recently, several methods for facilitating communication
between blockchains have been developed and studied, with a
focus on addressing the security risks and challenges presented
by this technology [8],[9],[10],[11],[12]. Such interconnectiv-
ity paves the way for arbitrage opportunities across DEXs that
operate on distinct blockchains, capitalizing on the variations
in pricing across these separate networks. Such an arbitrage
that spans multiple blockchains is referred to as cross-chain
arbitrage. Cross-chain arbitrage is more difficult to detect as
it spans multiple chains that are not updated simultaneously.
It presents a higher risk as it cannot be performed atomi-
cally within a single block on a single chain [13],[14]. The
loss of atomicity increases the risk for the token exchange
rates to change between transactions, which may cause the
arbitrage to fail. Moreover, an arbitrageur (user who conduct
arbitrages) must use assets on different chains, increasing the
arbitrage complexity. Cross-chain arbitrage is significant in
the blockchain ecosystem as it presents many arbitrage possi-
bilities, exploiting price discrepancies between blockchains.
Beyond the existing studies on arbitrage within a single
blockchain, the literature on cross-chain arbitrage is lacking
and primarily concentrates on the formalization of cross-chain
arbitrage. Our work aims to fill this gap.
In this paper, we provide a framework to study cross-chain
arbitrage empirically. We demonstrate its potential by using
the framework to analyze in practice cross-chain arbitrage
opportunities between two DEXs: PancakeSwap and Quick-
Swap, which are implemented on two major blockchains BNB
Chain and Polygon (Polygon is a side chain to Ethereum). We
focus on these blockchains for two reasons: (i) Being among
the blockchains with the highest total value locked in them
(see Table I). (ii) The ability to access and analyze their data
and exchange values. Within these blockchains, the particular
DEXs PancakeSwap and QuickSwap were chosen as being
dominant DEXs with respect to their value as of September
2023 in their respective blockchains.
II. Background and Related Work
In this section, we introduce some basic terms and concepts
from DeFi and overview related work.

A. Glosarry
Smart contracts are programs stored on a blockchain that
execute when predefined conditions are met. Smart contracts
can be used to create fungible tokens (ERC-20 tokens) beyond
the native token Ether. For example, WETH refers to the ERC-
20 compatible version of Ether, allowing it to be swapped
with other ERC-20 tokens [15], or implemented on other
blockchains. We refer to ERC-20 tokens as tokens.
Side chains concept (first proposed in [16]) are separate
blockchains that function in parallel to the main blockchain,
referred to as the mainnet. The side chain and the mainnet are
connected via a two-way-sided bridge that enables the transfer
of assets smoothly and securely between the blockchains.
Sidechains can have separate block parameters, design, and
consensus protocols that mainly aim to solve scalability issues
in the mainnet.
BNB Chain comprises BNB Beacon Chain (BC) and BNB
Smart Chain (BSC). The BC is responsible for governance,
staking, and voting, vital elements in the blockchain opera-
tion. The BSC is responsible for the consensus layers and
compatibility with different blockchains [17].
Liquidity pools are implemented inside a DEX to enable
users to swap tokens. Each pool allows users to swap between
a pair of tokens. To swap between two tokens, A and B, first,
a liquidity provider (LP) needs to add any amount of token
A and a proportional relative amount of token B to a pool
so other users can swap between them. In PancakeSwap, the
commission fee is 0.25% of the value of every swap made
using a pool [18], while it is slightly higher with a fee of
0.3% in QuickSwap [19].
Bonding curve is a concept describing through a formula
the inverse relationship between the supply of an asset and its
price. This concept underlines that reduced asset quantities
lead to price increases [20]. In this paper, we focus on a
common bonding curve for two tokens indicating a relation
between the amounts of them. For tokens Aand Bwith
amounts α > 0 and β > 0 (respectively), it indicates that
α·β=k. For a commission fee λwe denote γ=(1 −λ). A
transaction trading ∆α > 0 of token Afor ∆β > 0 of token B
with λfee, must satisfy
(α+ ∆α·γ)·(β−∆β)=k,
implying ∆β=β·γ·∆α
α+γ·∆α. DEXs using the bonding curve,
maintaining a constant product of token amounts in pools,
are known as Constant Product Market Makers (CPMMs).
PancakeSwap and QuickSwap are CPMMs.
PancakeSwap and QuickSwap are well-known DEXs on
BNB Chain and Polygon, respectively. They have an auto-
mated market maker design and follow the bonding curve
concept. The DEXs enable the creation of pools that can be
used to trade any pair of fungible tokens and provide fees for
liquidity provides (LPs). Consequently, they are extensively
used and constitute a significant portion of their respective
blockchains.
We present some approximations of statistics for Pan-
cakeSwap and QuickSwap to demonstrate their capacity in
TABLE I
10 Largest Blockchains in Crypto Ranked by total value locked as of
September 2023 [23]
Blockchain Total Value Locked
Ethereum $47.09B
Tron $6.54B
BNB Chain $3.43B
Arbitrum $1.81B
Polygon $930.23M
Avalanche $612.37M
Optimism $608.7
Cronos $297.97M
Kava $267.04M
TABLE II
Basic details of the Analyzed DEXs.
DEX TVL Blockchain Number of pools
PancakeSwap $2.61B BNB Chain 17394
QuickSwap $166.5M Polygon 2328
Table II. TVL (total value locked) refers to the total amount
of money in all pools, and the number of pools as of Jan.
13, 2023. PancakeSwap and QuickSwap are exclusively im-
plemented on BNB Chain and Polygon, respectively.
B. Related Work
CPMMs overview. There is extensive research regarding
arbitrage in blockchains and CPMMs. In particular, a few
papers analyze theoretical and practical properties of constant
product market makers. Angeris et al. [21] analyze constant
function market makers, which CPMMs are part of. They show
under reasonably general assumptions that constant function
market makers tend to report the correct price of assets.
They also provide lower bounds, guaranteeing no user can
drain the reserves of assets of a given DEX. An analysis of
Uniswap is presented in [22]. The paper shows that CPMMs
must closely track the reference market price under some
common conditions. It also indicates that Uniswap is stable
under various market conditions. The DEXs in our paper are
CPMMs and, therefore, have the properties as described.
Theoretical and empirical arbitrage analysis in DEXs.
Berg et al. [6] perform an empirical analysis to assess the effec-
tiveness of Uniswap and SushiSwap in tracking the reference
market. Their analysis involves identifying optimizable trades
and cyclic arbitrage opportunities arising from market price
inaccuracies. They show that market insufficiencies were es-
pecially common in the summer of 2020 and disappeared with
time. They suggest that the market becomes less efficient when
the prices of cryptocurrencies are highly volatile. Hansson [7]
examines how arbitrageurs contribute to price discovery and
efficiency in decentralized markets. The study identifies cross-
exchange and triangular arbitrages by analyzing transaction
data from Ethereum. The findings indicate that arbitrage profits
are swiftly realized following price anomalies, showcasing the
market’s ability to adjust rapidly. Overall, Hansson under-
scores the role of arbitrageurs in enhancing efficiency within
blockchain-based systems.
2

Boonpeam et al. [24] investigate the potential profits gained
through DEXs and suggest an arbitrage system that can iden-
tify profit opportunities by trading token routes on different
DEXs using statistical techniques. Khetan et al. [25] examine
cryptocurrency exchange arbitrage opportunities and propose
various strategies for arbitrage. They analyze bitcoin pricing
data from Binance, Kucoin, and Coinbase to identify the buy
and sell markets for different timeframes and quantify profits
from implementing an arbitrage strategy.
Danos et al. [26] present a proposal for modeling the global
money market of DeFi using an abstract notion of networks of
DEXs. They formalize routing and arbitrage on these networks
as convex optimization problems and provide bounds with
closed formulas in the case of Uniswap. They propose a the-
oretical framework for studying cyclic arbitrage and analyze
profitability conditions and optimal trading strategies for cyclic
transactions. They examine exploitable arbitrage opportunities
and market size with transaction-level data from Uniswap V2.
Wang et al. [5] introduce a theoretical framework for examin-
ing cyclic arbitrage and provide an analysis of the profitability
conditions and optimal trading strategies involved. They also
investigate the exploitable arbitrage opportunities and market
size of cyclic arbitrages by analyzing the transaction-level data
from Uniswap V2.
Sjursen et al. [27] try to identify cross-chain arbitrage
conducted by users using transaction data between Uniswap
on four different blockchains: Ethereum, Arbitrum, Optimism,
and Polygon. In contrast, we do not try to identify cross-
chain arbitrages done by users. Using a theoretical framework,
we analyze cross-chain arbitrage opportunities between the
exchanges that users did not utilize.
Overall, previous works studied arbitrages in DEXs within
a single blockchain. In contrast, we study arbitrages between
DEXs on different blockchains. To the best of our knowledge,
our paper is the first to empirically study opportunities for
cross-chain arbitrage between DEXs. Cross-chain arbitrage
has been mostly formalized, but no work has been done to
understand its existence and potential revenue, excluding [27].
Our work aims to provide a theoretical framework to study
cross-chain arbitrage empirically and to provide insights into
its profitability and the ability to utilize it.
III. A Framework for Cross-chain Arbitrage
This section formalizes the concepts and definitions for the
theoretical framework to study cross-chain arbitrage. We use
the framework in Section IV to collect cross-chain arbitrages
between PancakeSwap and QuickSwap.
A DEX runs over a particular blockchain Cand supports a
collection of tokens T={t1,...,tn}where nis the number of
tokens. It supports trading between selected pairs of tokens in
Tsuch that the set of pairs is P⊆T×T. For p={ti,tj} ∈ P,
one can trade values of token tifor values of token tjand
vice versa. The trading is supported through a liquidity pool
of tokens dedicated to this pair. The amount of the two tokens
in the pool can vary over time. The initial values are based on
the amounts at the pool establishment.
Definition 1 (Swap): is the exchange of niof tlfor njof tr
over blockchain Cas follows:
ni·tl
C
−→ nj·tr
Definition 2 (Cyclic swap sequence): CSS in short, is a
sequence of m∈N+token swaps satisfying:
1 : n1·t1
C1
−−→ n2·t2
2 : n2·t2
C2
−−→ n3·t3
. . .
m:nm·tm
Cm
−−→ nm+1·t1
A CSS is a type of arbitrage that is considered token-
profitable where the arbitrageur earns an additional amount
from the value of the token it started with nm+1>n1.
In blockchain systems, each transaction incurs gas fees for
execution. These fees are on top of the commission charges
imposed by DEXs. To determine the gas fees for a transaction,
one must multiply the gas price with the transaction gas
limit. The gas limit reflects the work needed to execute the
transaction.
A CSS is profitable when the token revenue value (nm+1−
n1)·t1exceeds the fees paid for the CSS. To allow such
arbitrage, each token referred on multiple chains must have
an implementation on them; for instance, t1should have an
implementation on Cmand C1. We consider the different
implementations of the same token on different chains to have
the same value as they are entwined.
Cyclic cross-chain arbitrage involves more than one
chain. There are several types of arbitrages. We focus on
cyclic cross-chain arbitrage because of the gap in the literature
on this type of arbitrage. We consider cyclic cross-chain
arbitrage between two chains. Assume we have mtokens
{ti|i∈[1,m],m≥5}, every two consecutive tokens share a
pool, and two separate chains C1,C2where t1and tkfor some
k∈[4,m−1] are implemented on both chains. We denote it
as an m-cycle cross-chain. The following CSS is considered
an m-cycle cross-chain arbitrage with two chains:
1 : n1·t1
C1
−−→ n2·t2
2 : n2·t2
C1
−−→ n3·t3
. . .
k−1 : nk−1·tk−1
C1
−−→ nk·tk
k:nk·tk
C2
−−→ nk+1·tk+1
. . .
m:nm·tm
C2
−−→ nm+1·t1.
We denote cyclic cross-chain arbitrage as cross-chain ar-
bitrage. We calculate the profitability from such an arbitrage
by calculating the optimal input amount n1that maximizes
the difference between the output amount to the input amount
nm+1−n1(we use the results from [5] to do so). We then use
n1to calculate the revenue. To calculate the profit, we need
3

0-0.09
0.1-0.19
0.2-0.29
0.3-0.39
0.4-0.49
0.5-0.59
0.6-0.69
0.7-0.99
1.0-1.89
1.9+
0
20
40
Transaction fee (USD)
Percentage (%)
swaps number =1
swaps number =2
swaps number =3
4≤swaps number
Fig. 1. Distribution of the transaction fee per swaps number in PancakeSwap
to approximate the gas fees for the transaction. We denote it
as transaction fee for simplicity.
We use Bitquery [28] to obtain all the gas fees for trans-
actions involving trades on PancakeSwap and QuickSwap
between Dec. 12, 2022, and Jan. 13, 2023. Subsequently, This
data is used to estimate the gas fees an arbitrager would incur
for conducting cross-chain arbitrage during this period. The
analysis of cross-chain arbitrage for these dates is presented
in Section V.
In Fig. 1, we show the transaction fee by the amount of
swaps in the transaction. A higher number of swaps requires
more work and, therefore, a higher transaction fee. The
transactions with one swap have the lowest average fee, with
0.3 USD per transaction. About 50% of them are between
0.1-0.19 USD. The low average is because they require the
lowest amount of work with one swap. The transactions with
two swaps have an average fee of 0.42 USD per transaction,
and about 40% of them are between 0.2-0.29 USD. The
transactions with three swaps have an average fee of 0.68
USD. About 33% of them are between 0.3-0.49 USD. The
more swaps, the higher the transaction fee. In total, the average
fee of all transactions is 0.36 USD.
We note that cross-chain arbitrage consists of swaps from
PancakeSwap and QuickSwap. The transaction fee in Quick-
Swap is negligible. Most transactions have fees between 0.005-
0.049 USD, as shown in Fig. 2. Therefore, the transaction fee
for the cross-chain arbitrage is determined by the transaction
fee in PancakeSwap. We retrieved the transactions in Quick-
Swap to approximate the transaction fee for arbitrages that
are solely in QuickSwap, which we analyze in Section V. The
transactions with three swaps have an average fee of 0.035
USD. About 57% of them are between 0.01-0.049 USD.
IV. Data Collection
In this section, we present the data we collect for the
data analysis in Section V of arbitrages in PancakeSwap,
QuickSwap, and between them. We note that the arbitrages
we collect are arbitrage opportunities in the blockchains at
the analyzed timestamps that arbitrageurs did not utilize. To
0-0.0009
0.001-0.0049
0.005-0.009
0.01-0.049
0.05-0.09
0.1+
0
20
40
60
Transaction fee (USD)
Percentage (%)
swaps number =1
swaps number =2
swaps number =3
4≤swaps number
Fig. 2. Distribution of the transaction fee per swaps number in QuickSwap
DEX X DEX Y
A
B
C
D
F
(a) Length =3
DEX X DEX Y
A
B
C
D
E
F
(b) Length =4
Fig. 3. Cross-chain arbitrages with lengths three and four. A, B, C, D, E,
F refer to various tokens. Solid arrows refer to swaps as part of the cyclic
arbitrage. Dashed lines indicate the same token in two DEXs. The length of
an arbitrage equals its number of swaps.
collect arbitrages, we need the state of each DEX at each
timestamp. As discussed, each DEX is made of pools. Thus,
the state of the DEX at a timestamp is determined by the
state of the pools constituting it. To retrieve all the pools of
each DEX, we utilize their factory smart contract, primarily
used for deploying the smart contracts of pools. The factory
smart contract emits a PoolCreated event whenever a pool is
created with its deployment address. The sync event is emitted
when there is a change in the pool. It includes the amount of
each token in the pool, block number, and timestamp. In other
words, it provides for the state of the pool.
We collect all the pools created until Jan. 13, 2023. For
each pool, we extract its state between Dec. 12, 2022 and Jan.
13, 2023. We only include pools with a sync event during this
month to avoid deprecated pools. We denote chain1 and chain2
as the numbers of chains of the analyzed respected DEXs. We
use several algorithms to acquire the arbitrages, summarized
as Algorithms 1-3.
Algorithm 1 describes the collection of arbitrages.
startChain1Time indicates the start time of the analysis.
GetChain2Time receives a timestamp and returns the times-
tamp of a block that was either created at the same time or at
the nearest earlier time on chain2. endChain1Time indicates
the end time of the analysis. NextTime receives a timestamp
4

Algorithm 1: Collect arbitrages between two DEXs on
separate chains
chain1T ime ←startChain1T ime
chain2T ime ←GetChain2T ime(chain1T ime)
while chain1T ime <endChain1T ime do
G←BuildGra ph(chain1T ime,chain2T ime)
FindArbitrages(G,MAX(chain1T ime,chain2T ime))
chain1T imeN ext ←N extT ime(chain1T ime,chain1)
chain2T imeN ext ←N extT ime(chain2T ime,chain2)
if chain1T imeN ext ≤chain2T imeN ext then
chain1T ime ←chain1T imeNe xt
if chain1T imeN ext == chain2T imeN ext then
chain2T ime ←chain2T imeNe xt
else
chain2T ime ←chain2T imeNe xt
Algorithm 2: BuildGraph (chain1T ime,chain2T ime)
Init an empty graph G
for pool in Pools do
if pool in chain1Pools then
e←GetClo sestS tate(pool,chain1T ime,chain1)
else
e←GetClo sestS tate(pool,chain2T ime,chain2)
Add eto G
return G
and a chain number and returns the subsequent timestamp
of a block created on the specified chain after the provided
timestamp. For each timestamp, we describe the state of the
two DEXs as a graph. Vertices are tokens and edges indicate
the amount of each of the two connected tokens in the pool
between them.
Algorithm 2 describes the building of the graph. Pools refers
to all the pool contracts in the DEXs and chain1Pools refers
to the pool contracts in the DEX on chain1. GetClosestState
receives a pool contract, timestamp, and chain number and
returns the closest smallest state of the pool using the sync
events (the amounts of each token in the pool).
Algorithm 3 receives a graph and saves all arbitrages by
timestamp. sharedTokens refers to tokens that have an imple-
mentation on both chains. PathsUpLen2 returns all paths, with
a maximum length of two, between vertices on a given chain.
Arbitrage returns whether the arbitrage is profitable and valid
(involving more than two swaps). SaveArbitrage receives the
arbitrage, timestamp, and the starting chain of the arbitrage,
and based on these parameters, it saves the arbitrage. Using
these algorithms, we can acquire up to 4-cycle cross-chain
arbitrages between two DEXs.
Fig. 3 illustrates the structure of the cross-chain arbitrages
between the two DEXs. To understand the relationship be-
tween cross-chain arbitrage and single-chain arbitrage, we
collect arbitrages of length three from each DEX separately.
The considered arbitrages are in the form A→B→C→
A,A∈sharedT oken s. They were considered for their joint
tokens with the cross-chain arbitrages and similar structure.
We use public APIs from BscScan and polygonscan to get the
addresses of the pools, Bitquery to query for the timestamps,
Algorithm 3: FindArbitrages (G,timestamp)
for tokenS in sharedTokens do
for tokenM in sharedTokens do
for Ch in [chain1,chain2] do
Ch′=chain1
if Ch == chain1then
Ch′=chain2
pathsS t art ←
PathsU p Len2(tokenS,tokenM,Ch)
pathsEnd ←
PathsU p Len2(tokenM,tokenS ,Ch′)
for pathS in pathsStart do
for pathE in pathsEnd do
if Arbitrage(pathS , pathE,C h) then
SaveArbitrage(pathS ,pathE,
timestamp,Ch)
Dec 12
Dec 17
Dec 23
Dec 28
Jan 2
Jan 8
Jan 14
0.96
0.98
1
1.02
1.04
Date
DAI/USDT
PancakeSwap
QuickSwap
(a) The ratio between DAI to USDT
Dec 12
Dec 17
Dec 23
Dec 28
Jan 2
Jan 8
Jan 14
2
4
·10−3
Date
USDC/WBNB
(b) The ratio between USDC to WBNB
Fig. 4. Ratios between DAI to USDT and USDC to WBNB in both DEXs
respectively
and Moralis to get the sync events and metadata on the tokens.
V. Data analysis
This section provides an analysis of the arbitrage opportu-
nities data that we collect. We use our framework to collect
cross-chain arbitrage opportunities between PancakeSwap and
QuickSwap. For simplicity, we divide this section into three
subsections. In the first subsection V-A, we present an intuition
for arbitrage possibilities between PancakeSwap and Quick-
Swap. The second subsection V-B analyzes the profitability
and number of arbitrages, while the third subsection V-C
analyzes the duration of cross-chain arbitrages.
5

Dec 12
Dec 17
Dec 23
Dec 28
Jan 2
Jan 8
Jan 14
0
50
100
Date
Number of arbitrages
0.1 ≤revenue <0.42
0.42 ≤revenue <1.9
revenue ≥1.9
(a) Number of arbitrages by date between PancakeSwap and QuickSwap
Dec 12
Dec 17
Dec 23
Dec 28
Jan 2
Jan 8
Jan 14
1
10
100
1000
10000
Date
Revenue (USDC)
(b) Total potential revenue over all arbitrages by date between PancakeSwap
and QuickSwap
Fig. 5. The number and revenue of arbitrages between PancakeSwap and
QuickSwap
Along the section, the revenue amounts are calculated based
on arbitrages that do not share a pool. This is to avoid summing
up revenues that utilize the same price discrepancy more than
once. In contrast, we count all possible arbitrage opportunities
because price discrepancies can be utilized in different ways.
The revenue is measured in USDC at the exchange rate at the
time of the arbitrage within the exchange. Transaction fees are
measured in USD. USDC is a cryptocurrency that aims to have
the same value as USD in close proximity. This allows easy
joint consideration of fees (in USD) and revenues (in USDC).
A. Tokens comparison
The following figures present the ratios of tokens imple-
mented at each chain and traded in the DEXs. Fig. 4(a) shows
over time the price of a stablecoin named DAI in terms of
another stablecoin USDT. In each DEX, the price remains
close to 1, implying a slight difference in the prices over the
two DEXs. For most of the time period, the price ratios of the
two stay within a range of 0.99 and 1.01.
In Fig. 4(b) the ratios between the tokens are highly
correlated following the high trade rate in both DEXs. We use
the difference between the exchange rates of tokens shared
between the chains, combined with different exchange rates
of tokens not necessarily shared to find cross-chain arbitrages.
Dec 12
Dec 17
Dec 23
Dec 28
Jan 2
Jan 8
Jan 14
0
100
200
300
Date
Number of arbitrages
0.1 ≤revenue <0.42
0.42 ≤revenue <1.9
revenue ≥1.9
(a) Number of arbitrages by date between PancakeSwap and QuickSwap
Dec 12
Dec 17
Dec 23
Dec 28
Jan 2
Jan 8
Jan 14
0.1
1
10
100
1000
Date
Revenue (USDC)
(b) Total potential revenue over all arbitrages by date between PancakeSwap
and QuickSwap
Fig. 6. The number and revenue of arbitrages between PancakeSwap and
QuickSwap, which do not share pools with existing arbitrages in each DEX
B. Revenue and number of arbitrages
For each of the following figures, we present the revenue
with respect to the approximation of the transaction fee
required to complete the arbitrage. We divide the fee into
three categories: lower bound, average, and high bound. Those
categories are based on Fig. 1 and Fig. 2. Arbitrages in
PancakeSwap and QuickSwap require three swaps to complete.
While, between PancakeSwap and QuickSwap, two swaps are
required at most at each exchange to complete the arbitrage.
As discussed, the number of swaps affects the transaction fee.
Therefore, the values of the categories are appropriate to the
number of swaps.
We denote APas a token Ain PancakeSwap and BQas token
B in QuickSwap. In Fig. 5 until Dec 17, the revenue spans
between 50 to 80 USDC. This is mainly due to discrepancies in
UP−US DCP,UP−US DTP. On Dec 17, the revenues surpassed
1,000 USDC and peaked at approximately 10,000 USDC.
Arbitrages that use the pool WBN BQ−US DCQcombined
with pools in PancakeSwap produce high revenue, thousands
of USDCs. For example, US DCP→BU S DP→WBN BP−
WBN BQ→US DCQproduce a revenue of 10,877 USDC.
This amount of revenue is not reached in PancakeSwap and
QuickSwap even though the discrepancy exists in one of the
6

Dec 12
Dec 17
Dec 23
Dec 28
Jan 2
Jan 8
Jan 14
0
50
100
Date
Number of arbitrages
0.1 ≤revenue <0.68
0.68 ≤revenue <1.9
revenue ≥1.9
(a) Number of arbitrages by date in PancakeSwap
Dec 12
Dec 17
Dec 23
Dec 28
Jan 2
Jan 8
Jan 14
1
10
100
500
1000
Date
Revenue (USDC)
(b) Total potential revenue over all arbitrages by date in PancakeSwap
Fig. 7. The number and revenue of arbitrages in PancakeSwap
DEXs as this arbitrage does not exist in Fig. 6. Following Dec
17, the revenue stabilizes at around 40 USDC mainly due to
the discrepancy in UP−US DCP,UP−U S DTP.
In Fig. 7 on Dec 19, there is a spike in revenues. This is due
to a discrepancy involving GOT C HIP. The spike with revenue
of around 1,000 USDC on Dec 22 is due to a discrepancy
involving AT PP, and the one with revenue of around 2,000
USDC on Dec 26 is due to a discrepancy involving JT S P.
After Dec 30, the revenue stabilizes around 500 USDC mainly
due to discrepancy involving FLDP.
The similarity between PancakeSwap and the cross-chain
is reasonable because the cross-chain arbitrage utilizes price
differences in each DEX. The cross-chain revenue is much
higher than the revenue in QuickSwap, as shown in Fig. 8.
This is due to the fewer resources and pools QuickSwap
has, as shown in Table II, and the potential revenue within
is much smaller than the high-resource PancakeSwap. In
Fig. 8 until Dec 26, the high revenue is primarily due to
discrepancy involving KI JQ. From Jan 9, it is primarily due
to a discrepancy involving 4.0Q.
In Fig. 5, the revenue does not surpass the revenues in
PancakeSwap at several time frames. This is mainly due to
the limitation in the length of the arbitrage and the resources
in QuickSwap. For example, on Jan 13, the revenue in Pan-
cakeSwap reaches approximately 500 USDC mainly for the
Dec 12
Dec 17
Dec 23
Dec 28
Jan 2
Jan 8
Jan 14
0
50
100
150
Date
Number of arbitrages
0.005 ≤revenue <0.035
0.035 ≤revenue <0.1
revenue ≥0.1
(a) Number of arbitrages by date in QuickSwap
Dec 12
Dec 17
Dec 23
Dec 28
Jan 2
Jan 8
Jan 14
0.1
1
10
Date
Revenue (USDC)
(b) Total potential revenue over all arbitrages by date in QuickSwap
Fig. 8. The number and revenue of arbitrages in QuickSwap
arbitrage US DT P→BUS DP→FLDP→U S DTP. In cross-
chain, such revenue is not reached on Jan 13. To reach such
revenue by utilizing this arbitrage in cross-chain, we need
to analyze its form as it was in cross-chain. The cross-chain
arbitrage has a path in QuickSwap and a path in PancakeSwap,
where each path constitutes two swaps at most. Therefore,
for the arbitrage to appear as cross-chain arbitrage, the path
in PancakeSwap must be BUS DP→FLDP→US DT P,
meaning BUS D and US DT must be shared with QuickSwap.
However, in all the pools in QuickSwap, there is only 10.57
BUSD, thus deterring the full utilization of the arbitrage. If
the cross-chain arbitrage had been longer, then the arbitrage
could have been fully utilized because there is not a shortage of
US DT and U S DC in QuickSwap in that timeframe, making
US DCP→BUS DP→F LDP→US DTPpossible where
US DC and US DT are shared.
Fig. 6 shows arbitrages that do not share a pool with existing
arbitrages within each DEX. This indicates a potential addition
for revenue that is not utilized in each DEX separately. The
revenue spans primarily between 0.1 to 10 USDC, where even
a value over 1,000 USDC is reached.
The ability of cross-chain arbitrage to utilize the price
discrepancies depends on the arbitrage complexity and the
amount of resources in each DEX. We show that it offers a
way to gain profits that do not necessarily exist in each DEX
separately and enhance existing ones.
7

1-9
10-59
60-99
100-999
1000+
102
104
106
Number of consecutive blocks
Number of arbitrages
0.1 ≤revenue <0.42
0.42 ≤revenue <1.9
1.9 ≤revenue <100
100 ≤revenue <1000
revenue ≥1000+
Fig. 9. Distribution of arbitrage duration (in units of consecutive blocks) as
a function of its revenue. between PancakeSwap and QuickSwap
0.1-0.419
0.42-1.89
1.9-99.9
100-999.9
1000+
5
10
Revenue (USDC)
Mean duration (blocks)
Fig. 10. Distribution of the mean arbitrage duration (in units of blocks) per
revenue between PancakeSwap and QuickSwap
C. Duration of arbitrages
We define the duration of cross-chain arbitrage as the
number of consecutive blocks that the arbitrage exists with the
same amount of revenue. This definition is reasonable because
we want to measure the duration an arbitrageur can conduct
a cross-chain arbitrage and gain the arbitrage revenue. We
denote the amount of time before a transaction is added to the
blockchain as transaction time. We approximate the transaction
times on BNB Chain based on [29] and on Polygon based on
[30]. In [29] and [30], the amount of time for a transaction
to be included in the blockchain is presented with three
categories: standard, fast, and rapid. Each category describes
the gas price compared to the transaction time, where a lower
transaction time requires higher gas price. As discussed, the
transaction fee is calculated by multiplying the gas price with
the gas limit, and therefore, we can deduct the transaction fee
compared to the transaction time.
On the BNB Chain, transaction times typically range from
5 to 60 seconds; this correlates to 2 to 20 blocks, with a
new block every 3 seconds. On Polygon, transaction times
are similar, but with blocks every 2 seconds; this correlates
to 2 to 30 blocks. Transaction time has a high variance and
mainly depends on network traffic and gas prices. Generally,
higher transaction fee can result in lower transaction time.
In Fig. 9, the total number of arbitrages with revenue
between 0.1 to 0.42 USDC is the largest in every duration
group. This is due to the proximity of those arbitrages to the
transaction fee, making them highly volatile to market changes
on the lower end of the duration group and less likely to be
extracted on the higher end. Although the average number
of consecutive blocks is relatively low, 11.34 as shown in
Fig. 10, there are 76,980 arbitrages with a duration longer
than 60 blocks. In 60 consecutive blocks, there are 36 polygon
blocks (60% of these blocks) and 24 BNB Chain blocks. This
correlates to the longest approximation of 20 blocks in BNB
Chain and 30 blocks on Polygon.
The arbitrages with revenue 0.42 to 1.9 USDC follow the
same principles as those with revenues of 0.1 to 0.42 USDC.
However, as the achievable revenue tends to be higher than
the transaction fee, they are more worthwhile to utilize. A
higher transaction fee can be paid and still gain a profit. The
arbitrages with revenue between 100 to 1,000 USDC with
average consecutive blocks of 5.09, as shown in Fig. 10 have
high revenue. Therefore, a high transaction fee can be paid
to obtain a profit from arbitrages with a duration of 1-10
blocks. In summary, high revenues are associated with shorter
transaction times, but this can be mitigated by paying a higher
transaction fee. Moreover, there are approximately 118,127
arbitrage opportunities that span a duration of more than 60
blocks.
VI. Conclusions And Further research
The blockchain ecosystem is evolving to contain multiple
chains. Multiple DEXs are being developed across these
chains, leading to high availability of assets across different
chains. This development offers diverse trading methods and
paves the way for new arbitrage opportunities.
Through an experimental study, we show that cross-chain
arbitrage potentially provides opportunities for arbitrageurs to
utilize and gain profits that do not necessarily exist in each
DEX separately and enhance existing ones. The properties of
the DEXs, the arbitrage complexity, and the network state play
a critical role in establishing cross-chain arbitrage. As far as
we know, we are the first to show an empirical analysis of
cross-chain arbitrage opportunities in terms of quantity and
revenue. We lay the foundations for further research on cross-
chain arbitrages. Open questions for further research are:
•Is there a correlation between the properties of DEXs
and the amount and revenue that can be established from
cross-chain arbitrages?
•How can we investigate cross-chain arbitrages in DEXs
that are not CPMMs?
•To what extent can we minimize the risks associated with
cross-chain arbitrage?
•Can we further research or find different methods to
extract value between different chains?
8

References
[1] F. Sch¨
ar, “Decentralized finance: On blockchain-and smart contract-
based financial markets,” FRB of St. Louis Review, 2021.
[2] M. Truchet, “Decentralized finance (DeFi): Opportunities, challenges
and policy implications,” Eurofi, 2022.
[3] G. Wood et al., “Ethereum: A secure decentralised generalised
transaction ledger,” Ethereum project yellow paper, 2014. [Online].
Available: https://gavwood.com/paper.pdf
[4] T. Azoulay, U. Carl, and O. Rottenstreich, “Allowing blockchain loans
with low collateral,” in IEEE International Conference on Blockchain
and Cryptocurrency (ICBC), 2023.
[5] Y. Wang, Y. Chen, H. Wu, L. Zhou, S. Deng, and R. Wattenhofer, “Cyclic
arbitrage in decentralized exchanges,” in ACM Web Conference, 2022.
[6] J. A. Berg, R. Fritsch, L. Heimbach, and R. Wattenhofer, “An empirical
study of market inefficiencies in Uniswap and SushiSwap,” CoRR, vol.
abs/2203.07774, 2022.
[7] M. Hansson, “Arbitrage in crypto markets: An analysis of primary
Ethereum blockchain data,” SSRN, 2022. [Online]. Available: http:
//dx.doi.org/10.2139/ssrn.4278272
[8] H. Mao, T. Nie, H. Sun, D. Shen, and G. Yu, “A survey on cross-
chain technology: Challenges, development, and prospect,” IEEE Access,
vol. 11, pp. 45 527–45 546, 2023.
[9] L. Li, J. Wu, and W. Cui, “A review of blockchain cross-chain technol-
ogy,” IET Blockchain, 2023.
[10] P. Han, Z. Yan, W. Ding, S. Fei, and Z. Wan, “A survey on cross-chain
technologies,” Distrib. Ledger Technol., vol. 2, no. 2, 2023.
[11] K. Ren, N.-M. Ho, D. Loghin, T.-T. Nguyen, B. C. Ooi, Q.-T. Ta, and
F. Zhu, “Interoperability in blockchain: A survey,” IEEE Transactions on
Knowledge and Data Engineering, vol. 35, no. 12, pp. 12 750–12 769,
2023.
[12] G. Wang, Q. Wang, and S. Chen, “Exploring blockchains interoperabil-
ity: A systematic survey,” ACM Computing Surveys, 2023.
[13] A. Obadia, A. Salles, L. Sankar, T. Chitra, V. Chellani, and P. Daian,
“Unity is strength: A formalization of cross-domain maximal extractable
value,” CoRR, vol. abs/2112.01472, 2021.
[14] C. McMenamin, “SoK: Cross-domain MEV,” CoRR, vol.
abs/2308.04159, 2023.
[15] A. Dolmatov, “Wrapped Ether (WETH),” https://academy.binance.com/
en/glossary/wrapped-ether.
[16] A. Back, M. Corallo, L. Dashjr, M. Friedenbach, A. Miller, A. Poelstra,
J. Tim´
on, and P. Wuille, “Enabling blockchain innovations with
pegged sidechains,” 2014. [Online]. Available: https://blockstream.com/
sidechains.pdf
[17] “BNB smart chain white paper.” [Online]. Available: https://github.
com/bnb-chain/whitepaper
[18] https://docs.pancakeswap.finance/products/pancakeswap-exchange/
trade.
[19] S. Singh, “What is quickswap, and how does it
work.” [Online]. Available: https://cointelegraph.com/explained/
what-is- quickswap-and-how-does-it- work
[20] J. Xu, K. Paruch, S. Cousaert, and Y. Feng, “SoK: Decentralized
exchanges (DEX) with automated market maker (AMM) protocols,”
ACM Comput. Surv., vol. 55, no. 11, pp. 238:1–238:50, 2023.
[21] G. Angeris and T. Chitra, “Improved price oracles: Constant function
market makers,” in ACM Conference on Advances in Financial Tech-
nologies, 2020.
[22] G. Angeris, H. Kao, R. Chiang, C. Noyes, and T. Chitra, “An analysis
of Uniswap markets,” CoRR, vol. abs/1911.03380, 2019.
[23] https://coinmarketcap.com/chain-ranking/.
[24] N. Boonpeam, W. Werapun, and T. Karode, “The arbitrage system on
decentralized exchanges,” in International Conference on Electrical En-
gineering/Electronics, Computer, Telecommunications and Information
Technology (ECTI-CON), 2021.
[25] I. Khetan, P. Sheth, S. Dalal, S. Mistry, V. Sharma, and N. Katre,
“Arbitrage in cryptocurrency: A survey,” in International Conference
on Information Systems and Computer Networks (ISCON), 2021.
[26] V. Danos, H. E. Khalloufi, and J. Prat, “Global order routing on exchange
networks,” in Financial Cryptography and Data Security Workshops,
2021.
[27] J. H. Sjursen, W. Meng, and W.-Y. Chiu, “Towards quantifying cross-
domain maximal extractable value for blockchain decentralisation,” in
Information and Communications Security, 2023.
[28] https://ide.bitquery.io/.
[29] https://bscscan.com/gastracker.
[30] https://polygonscan.com/gastrackerr.
9