Fundamentals of Perpetual Futures

Fundamentals of Perpetual Futures∗
Songrun He Asaf Manela Omri Ross Victor von Wachter

arXiv:2212.06888v4 [q-fin.PR] 14 Jul 2023
First draft December 2022. This draft July 2023.
Abstract
Perpetual futures—swap contracts that never expire—are the most popular derivative traded in
cryptocurrency markets, with more than $100 billion traded daily. Perpetuals provide investors
with leveraged exposure to cryptocurrencies, which does not require rollover or direct cryptocur-
rency holding. To keep the gap between perpetual futures and spot prices small, long position
holders periodically pay short position holders a funding rate proportional to this gap. The
funding rate incentivizes trades that tend to narrow the futures-spot gap. But unlike fixed-maturity
futures, perpetuals are not guaranteed to converge to the spot price of their underlying asset at
any time, and familiar no-arbitrage prices for perpetuals are not available, as the contracts have
no expiry date to enforce arbitrage. Here, using a weaker notion of random-maturity arbitrage, we
derive no-arbitrage prices for perpetual futures in frictionless markets and no-arbitrage bounds for
markets with trading costs. These no-arbitrage prices provide a valuable benchmark for perpetual
futures and simultaneously prescribe a strategy to exploit divergence from these fundamental
values. Empirically, we find that deviations of crypto perpetual futures from no-arbitrage prices
are considerably larger than those documented in traditional currency markets. These deviations
comove across cryptocurrencies and diminish over time as crypto markets develop and become
more efficient. A simple trading strategy generates large Sharpe ratios even for investors paying
the highest trading costs on Binance, which is currently the largest crypto exchange by volume.
JEL Classification: G11, G12, G13

Keywords: Perpetual futures, crypto, random-maturity arbitrage, funding rate
∗ We are grateful to Lin William Cong,Shimon Kogan, Tim Roughgarden, as well as seminar participants
at a16z Crypto and at Reichman University for helpful comments.
Songrun He is at Washington University in St. Louis ([email protected]). Asaf Manela is at
Washington University in St. Louis and Reichman University ([email protected]). Omri Ross is at
University of Copenhagen ([email protected]), and Chief Blockchain Scientist at the eToro Group. Victor
von Wachter is at University of Copenhagen ([email protected]).
1 Introduction
Perpetual futures are, by far, the most popular derivative traded in cryptocurrency
markets, generating a daily volume of more than $100 billion. Prior to the recent collapse
of FTX, perpetual futures were among the most actively-traded products on the exchange,
with the now-bankrupt hedge fund Alameda Research taking the other side of many such
leveraged trades. Despite their central role in crypto markets, there is relatively little work
studying these derivatives. In this paper we ask: what are the theoretical fundamental
values of perpetual futures, and how large are deviations from these fundamentals
empirically?
Perpetuals are derivatives that allow investors to speculate on or hedge against cryp-
tocurrency price fluctuations using high leverage, without needing to acquire the cryp-
tocurrencies or to roll them over. Like traditional fixed-maturity futures, at initiation,
a long and short counterparty agree on an initial futures price without exchanging any
money, other than posting margin with the exchange. Both parties can enter or exit the
contract at any time, with profits or losses continuously calculated and allocated to each
side’s margin account based on the prevailing market prices.
Unlike fixed-maturity futures, perpetuals do not expire. This feature enhances the
liquidity of the contract, as there is no staggering of contracts with different maturities
traded on the exchange, and only a single perpetual futures contract per underlying asset
is listed. Moreover, this instrument does not require market participants to deal with the
rolling over of their futures positions and is traded 24/7.
Because they have no set expiration date, perpetuals are not guaranteed to converge
to the spot price of their underlying asset at any given time, and the usual no-arbitrage
prices for perpetuals are not applicable. To minimize the gap between perpetual futures
and spot prices, long position holders periodically pay short position holders a ’funding
rate’ proportional to this gap to incentivize trades that narrow it. For example, when
1
the futures price exceeds the spot price, arbitrageurs who borrow cash to long the spot
and simultaneously short the futures would receive the funding rate. Their trades would
tend to increase the spot price and decrease the futures price. A narrow gap means that
perpetual futures offer effective exposure to variation in the spot price of the underlying
asset to hedging and speculating investors. Typically, the funding rate is paid every
eight hours and approximately equals the average futures-spot spread over the preceding
eight hours. Note that the strategy just sketched, commonly referred to as ’funding
rate arbitrage,’ is not risk-free even disregarding margin requirements and trading costs,
simply because there is no predetermined expiration date when the trade would be
unwound at a profit.
We derive no-arbitrage prices for perpetual futures in frictionless markets and derive
no-arbitrage bounds in markets with trading costs. The theoretical perpetual futures price
is proportional to the spot price of the underlying, with a constant of proportionality that
increases in the ratio of the interest rate to a constant that determines the magnitude of
the funding rate. The interest rate captures the cost of borrowing cash to finance holding
the underlying, while the funding rate captures the benefit of shorting the futures. Thus,
intuitively, the future-spot spread is larger when the cash borrowing interest rate is large
relative to the funding rate.
Our derivation relies on a weaker notion of arbitrage that we call random-maturity

arbitrage. As its name suggests, unlike traditional riskless arbitrage, we allow the strategy’s
time-to-maturity to be random. At first glance, one might object that such prices are
not truly based on riskless arbitrage. Note, however, that riskless no-arbitrage pricing
is usually just a useful fiction (Pedersen, 2019). For example, in real-world futures
markets, arbitrageurs must maintain a margin account during the entire period in which
the arbitrage trade is open. Temporary worsening of apparent arbitrage opportunities can
lead to liquidations and losses. As the saying goes, an arbitrageur must remain liquid
longer than the market stays irrational. Thus, even arbitrage opportunities that appear to
2
be riskless in theory, may be risky in practice.
These no-arbitrage prices provide a useful benchmark for perpetual futures and
simultaneously prescribe a strategy to exploit divergence from these fundamental values.
Motivated by the theoretical understanding, we study the empirical deviations of the
perpetual futures price from the spot. We find that the mean deviation from our
benchmark is modest and statistically insignificant, which suggests that it is a useful
benchmark for futures prices.
Around this benchmark, however, perpetual futures often differ significantly from spot
prices. The mean absolute deviation is about 60% to 100% per year across different cryp-
tocurrencies, which is considerably larger than the deviations documented in traditional
currency markets by Du, Tepper, and Verdelhan (2018). We find strong comovement
of the futures-spot gap across different cryptocurrencies. This comovement can be due
to commonality in funding and market liquidity faced by arbitrageurs who operate in
multiple cryptocurrencies. Common sentiment could also drive the difference in futures
demand relative to the spot. Overall, the deviation’s magnitude is comparable to our
theoretic no-arbitrage bound calibrated to actual trading fees.
These deviations narrows considerably in 2022, suggesting a decrease in arbitrage

frictions in the market and an increase in competition among arbitrageurs. The narrowing
gap provides an additional perspective on the downfall of Alameda Research and Three
Arrows Capital. According to news reports and interviews, both hedge funds seem to
have pivoted from such arbitrage activity around late 2021 to early 2022 and started taking
more directional bets on cryptocurrencies, with both direct unhedged crypto holdings and
investments in crypto startups. The large declines in crypto prices in 2022 subsequently
exhausted their capital and led to their bankruptcies.1
1 See, for example, Forbes, November 19, 2022, on How Did Sam Bankman-Fried’s Alameda Research
Lose So Much Money?, Odd Lots, November 17, 2022, on Understanding the Collapse of Sam Bankman-
Fried’s Crypto Empire, and Hugh Hendry’s interview on December 3 2022 of Kyle Davies on the Collapse
of Three Arrows Capital.
3
To understand the economics of the futures-spot spread, we consider a trading strategy
motivated by the random-maturity arbitrage theory. Whenever the futures-spot spread
exceeds the theoretical bound under certain trading cost tiers, we open the trading position
and close it when the futures-spot spread returns to its theoretical relationship under no
trading costs. We find that empirically, the random maturity arbitrage strategy generates
a sizable Sharpe ratio even under high trading costs. For example, for Bitcoin perpetual
futures, the strategy generates a Sharpe ratio of 1.92 under high trading costs typical of
retail investors, and up to 3.94 for highly-active market makers who pay no such fees.
The strategy’s performance is even better for Ether and other cryptocurrencies and deliver
significant alphas relative to the 3-factor model of Liu, Tsyvinski, and Wu (2022) and the
5-factor model of Cong, Karolyi, Tang, and Zhao (2022b).
What explains these large no-arbitrage deviations? One natural explanation is that
liquidity in crypto markets is insufficient for arbitrageurs to eliminate such violations. Our
finding that the spreads decline over time is consistent with liquidity improving as these
markets develop, and leaves open the possibility that they will narrow going forward. We
also find, however, that past return momentum significantly explains the futures-spot gap
with a time-series regression R2 of more than 50%. When past returns are high, futures
tend to trade at a higher price relative to the spot. This indicates positive feedback or
momentum trading behavior in the perpetual futures market. This correlation may linger
even as crypto markets become more efficient.
The existing literature on perpetual futures mainly focuses on descriptive evidence.

See e.g. Alexander, Choi, Park, and Sohn (2020), Schmeling, Schrimpf, and Todorov
(2022), De Blasis and Webb (2022), Ferko, Moin, Onur, and Penick (2022), and Streltsov
and Ruan (2022). Angeris, Chitra, Evans, and Lorig (2022) provide a theoretical no-
arbitrage analysis of perpetuals, but they make over-simplifying assumptions by assuming
the payoff from the perpetual is a fixed function of the underlying spot price. Compared
to the existing literature, we illustrate the fundamental mechanism behind the perpetual
4
design and derive theoretical no-arbitrage prices and bounds for this instrument.
Also related is recent literature on fixed maturity futures in the crypto space. Schmel-
ing, Schrimpf, and Todorov (2022) provides a comprehensive analysis of the carry
of crypto futures, with the carry defined following the general definition of Koijen,
Moskowitz, Pedersen, and Vrugt (2018). Schmeling, Schrimpf, and Todorov (2022)
document a volatile convenience yield in the crypto space driven by high leverage from
trend-chasing small investors and the relative scarcity of arbitrage capital. Cong, He,
and Tang (2022a) provide a novel link of the volatile convenience yield to the staking,
service flow, and transaction convenience of the underlying tokens. They show that the
large deviation from uncovered interest rate parity can be reconciled with transaction
convenience. Our paper focuses on perpetual futures rather than fixed maturity futures
and extends fixed-maturity to random-maturity no-arbitrage pricing.
More broadly, our paper contributes to the understanding of the frictions and arbitrage
in cryptocurrency markets. Makarov and Schoar (2020) study price deviations across
exchanges. They find large gaps across countries, highlighting the important role played
by capital controls and slow-moving arbitrage capital as in Duffie (2010). Our analysis
focuses on the price wedge between the spot and the futures market. We find that even
within an exchange, futures prices deviate from their theoretical arbitrage-free values.
These results indicate there are significant limits to arbitrage as in Gromb and Vayanos
(2018) for cryptocurrencies in the early years.
The rest of the paper is organized as follows: Section 2 provides the institutional
details and history of perpetual futures. Section 3 presents the no-arbitrage analysis of the
perpetual futures market and derives the theoretical price of perpetual futures. Section
4 demonstrates the empirical futures-spot deviation and presents the simple theory-
motivated trading strategy that can exploit the arbitrage opportunity. Section 5 provides
some explanation for the deviation between futures and the spot. Section 6 concludes.
5
2 An Introduction to Perpetual Futures
The idea of perpetual futures was first introduced by Shiller (1993). The goal was to set
up a perpetual claim on the cash flows of an illiquid asset. For example, the underlying
illiquid asset could be a house which generates rents as cash flows. The purpose of the
perpetual futures is to enable price discovery for the underlying with an illiquid or hard-
to-measure price. Perpetual futures have no expiration date but cash is exchanged between
the long and the short side: after buying the perpetual futures, the long side is entitled
to receive the flow cash flow from the short side and they settle the price difference when
exiting the position.
Figure 1 Total trading volumes of perpetual futures across exchanges
The figure displays the 7-day moving average daily traded volume for perpetual futures across all
exchanges in blue. The median daily volume is $17.8 bn. (2020), $132.0 bn. (2021) and $101.9 bn.
(2022). This translates to a yearly volume of $8,551 bn. (2020), $51,989 bn. (2021) and $39,306
bn. (2022) respectively. Additionally, the grey line depicts the ratio between the traded volume in
perpetuals and spot markets. In 2022 the Perpetual markets are consistently trading between 2x or
3.5x the spot volume. The data is obtained from CoinGecko, a crypto data specialist. We exclude
exchanges that are known for misrepresenting data (e.g. forms of wash trading).
Perpetual futures in crypto markets similarly have no expiration date and cash is
exchanged between the long and the short side, but their purpose is different from Shiller’s
original idea. First, unlike, e.g. real estate market, crypto has no inherent dividend or
6
cash flow. Second, the price discovery argument of Shiller is most applicable to settings
where spot prices are difficult to measure. Crypto spot prices, however, can be measured
from active trading on different exchanges, and decentralized exchanges such as Bancor
(Hertzog et al., 2017) or Uniswap (Adams et al., 2021) offer price discovery for assets
with minimal liquidity. The major role perpetual futures play in the crypto space is to
offer an effective leveraged trading vehicle to hedge or speculate the underlying spot price
movement, which makes the market more complete. It also serves as an effective tax
payment optimization tool for investors.
Crypto perpetual futures were first introduced by BitMEX in 2016, which gained great
popularity in the crypto space since its inception. It initially served as an effective hedging
tool for crypto miners. It was later adopted by crypto speculators interested in leveraged
exposure. Nowadays, based on data from CoinGecko, the median total daily trading
volume of perpetual futures across all exchanges is 101.9 billion in the year 2022 which is
about 2× to 3× the total spot trading volume across these exchanges. Figure 1 presents the
7-day moving average of the total trading volume of perpetual futures across all exchanges.
We see a significant rise in trading of perpetual futures around January 2021 and the total
volume stabilizes at a level above $100 billion per day following the rise.
Cong, Li, Tang, and Yang (2022c) document significant wash trading behavior among
crypto exchanges because of competition, the ranking mechanism, and lack of regulation.
They estimate that over 70% of crypto trading volume is not real. Amiram, Lyandres, and
Rabetti (2021) confirm this conclusion using new data and an extended methodology.
Therefore, in calculating trading volume, we exclude exchanges that are known to
misrepresent data.
A key feature of crypto perpetual futures is the funding rate, which is the cash
exchanged between the long and short counterparties. Its goal is to keep the futures
price close to the underlying spot so that the futures can be an effective hedging tool
7
Figure 2 Bitcoin annualized funding rate across exchanges
The figure presents the 7-day moving average annualized funding rate for Bitcoin (BTC) across
exchanges. The data is obtained from glassnode, an analytics platform. The data covers two major
market turbulences: the COVID stock market crash from February to April 2020, and the FTX
insolvency in November 2022. The FTX collapse led to significant negative funding rates on all
solvent exchanges, thus the perpetual future prices were lower than the spot prices on the solvent
exchanges. Vice versa on the insolvent FTX. The funding rates become more volatile after the event,
indicating increased uncertainty for the market participants.
for spot price movement. The funding rate is typically paid every 8 hours. Its value is
approximately a weighted average of the prior 8 hours’ price gap of the futures and the
spot. If the futures price is above the spot, the funding rate will be positive, meaning the
long side of the futures needs to pay the short side. This incentivizes traders to short the
futures, and in doing so, to move its price back in line with the spot. On the other hand,
when the futures price is below the spot, the funding rate turns negative, which means the
short side needs to pay the long side. Therefore, the funding rate is the key mechanism
keeping the futures price close to the spot price.2 Note that perpetual futures cannot be
replicated by rolling over 8-hour maturity futures. For the latter, every 8 hours, the price
is guaranteed to converge to the underlying spot, while for the perpetual futures this is
not the case.
Figure 2 presents the annualized 7-day moving average of Bitcoin perpetual funding
2 https://www.binance.com/en/support/faq/360033525031 provides a detailed explanation of how
the funding rate is calculated.
8
rates across leading exchanges from January 2020 to December 2022. The funding rate is
positive when the futures-to-spot spread is positive, and negative otherwise. Funding
rates tend to be similar across exchanges due to cross-exchange arbitrage activity, but
can diverge during extreme liquidity episodes. During March 2020, as Covid-19 started
to spread and liquidity evaporated, funding rates turned substantially negative in most
exchanges. Funding rates turned highly positive during the crypto bull run of early 2021.
The last episode highlighted is the collapse of FTX, which was the 4th largest crypto
exchange at the time. The figure shows that Bitcoin futures prices were substantially
higher than spot prices at FTX, but the opposite was true on other exchanges. This
pattern is consistent with FTX investors liquidating their short futures positions quickly,
either voluntarily to reduce their exposure to the failing exchange, or involuntarily as the
exchange liquidated their underfunded positions.
3 Arbitrage in Perpetual Futures
We next derive no-arbitrage prices for perpetual futures prices relative to spot prices.
Unlike traditional futures, perpetual futures have no expiration date. To analyze arbitrage
in this market, we extend the traditional notion of risk-free arbitrage where arbitrageurs
have a guaranteed positive payoff at a certain time in the future. We first describe the
payoff structure of perpetual futures. We then introduce a generalized notion of an
arbitrage opportunity with a certain positive payoff but at an uncertain future time.
Definition 1 (Perpetual futures). A perpetual futures contract {Ft }t=0

∞ written on {S }∞ is
t t=0
an agreement between the long and the short side. There is 0 cost to enter the agreement. After
entering, both the long side and the short side can terminate the contract at any time t. Before
termination, for each unit of the perpetual, the long must pay the short an Fs −adapted cash-flow
κ(Fs − Ss )ds, s ∈ (0,t), referred to as the funding value. κ is a scaling parameter determining the
magnitude of the funding rate relative to the price gap. At termination, the short needs to pay the
9
long Ft − F0 for each unit shorted.
This definition is an approximation of real-world perpetual futures. In most exchanges,

the funding value for perpetual futures is paid every 8 hours and approximately equals the
average difference between the futures price and the spot over the 8 hours (480 minutes
1
or 1095 years). If we measure time units in years, the cumulative funding rate payment
during an 8-hour period is:
Z t+1/1095
κ(Fs − Ss )ds.
t
Consider the case when Ft − St = G, which is constant over the 8-hour interval. In the real
κG
world, the funding rate payment would equal G while in our model, the number is 1095 .
To equalize the two, we have κ = 1095.
The traditional notions of arbitrage typically consider a guaranteed positive payoff at

a certain future date.
Definition 2 (Riskless arbitrage). A riskless arbitrage opportunity is defined with respect to a

payoff x at a certain future time τ and its price p(x). If the following conditions are satisfied: (1)
x ≥ 0 almost surely, (2) x > 0 with some positive probability, and (3) its price satisfies p(x) ≤ 0,
then this payoff is an arbitrage opportunity (Cochrane, 2009).
There is no certain future time, however, when a perpetual futures expires and its price
is guaranteed to converge to some fixed function of the spot price. Therefore, a generalized
notion of arbitrage is required. Suppose the risk-free rate is constant. We define random-
maturity arbitrage opportunities as zero-cost strategies with a guaranteed positive payoff
at an uncertain future time. Stated formally:
Definition 3 (Random-maturity arbitrage). A random-maturity arbitrage opportunity is

defined with respect to a random payoff x at a future random time τ̃, τ̃ ∈ (0, ∞), and its price
p(x). If the following conditions are satisfied: (1) x ≥ 0 almost surely, (2) x > 0 with some positive
10
probability, and (3) its price satisfies p(x) ≤ 0, then this payoff is a random-maturity arbitrage
opportunity.
Definition 3 generalizes traditional arbitrage in the sense that there is a guaranteed

positive payoff but at an uncertain future time. The following corollary specializes this
definition for perpetual futures:
Corollary 1. In the perpetual futures market, if a strategy (1) has 0 cost at time 0, and (2) for
∞ and {S }∞ , there exists an unwinding time τ̃
any price path of the futures and the spot, {Ft }t=0 t t=0
such that its discounted payoff at time t = τ̃ is positive, then this strategy is a random-maturity
arbitrage.
We next show that when there is no random-maturity arbitrage, the gap between
futures and spot prices is bounded by a constant. We make the two following assumptions:
Assumption 1. The gap between the perpetual futures and the spot satisfies the following
condition: lim inft→∞ |Ft (ω) − St (ω)| < ∞, ∀ ω
Assumption 2. The risk-free rate r for arbitrageurs is constant.
Assumption 1 is a no-bubble condition. lim inf represents the greatest lower bound as
t → ∞. In other words, we can always find Ft and St that lie within a finite bound as t → ∞.
This condition allows for the gap between Ft and St to explode in the limit as long as it
shrinks at some subsequent time. Assumption 2 guarantees that there is no roll-over risk
for the arbitrageurs. With these two assumptions, we can state the conclusion as follows.
Proposition 1 (No-arbitrage bound). Under Assumptions 1 and 2, when there is a constant

trading cost C paid when terminating the position, arbitrageurs will trade the perpetual futures
until it lies within the following bound relative to the spot:
r r
St 1 + −C ≤ Ft ≤ St 1 + +C . (1)
κ κ
11
Proof. Consider the following two scenarios in Table 1. (1) F0 > S0 (1 + κr ) + C; (2) F0 <
S0 (1 + κr ) − C. We show in the first case, arbitrageurs would long the spot and short the
futures because this is a random-maturity arbitrage opportunity. We then show that the
opposite strategy is optimal in the second case.
F0 > S0 1 + κr +C F0 < S0 1 + κr −C

Actions Long spot, short futures Long futures, short spot

Futures 0 0
Time 0 Spot −S0 +S0
Cash +S0 −S0
Futures F0 − Ft Ft − F0
Spot +St −St
Time t
Cash −S0 ert +S0 ert
Rt Rt
Funding κ 0 (Fs − Ss )er(t−s) ds −κ 0 (Fs − Ss )er(t−s) ds
Trading Cost −C −C
e−rt F0R − e−rt (Ft − St ) − S0 e−rt (FRt − St ) − e−rt F0 + S0
Payoff +κ 0t (Fs − Ss )e−rs ds− −κ 0t (Fs − Ss )e−rs ds−
e−rt C e−rt C
Table 1 Discounted payoffs to arbitrage strategies in perpetual futures and spot markets
This table
presents the costs and benefits of two arbitrage trading strategies: (1) when F0 >
r r

S0 1 + κ + C, long the spot and short the futures; (2) when F0 < S0 1 + κ − C, long the futures
and short the spot. In the last row, the payoff from exiting the position is the discounted payoff
from future and spot price changes, proceeds from the cash market, and the funding rate.
Scenario 1: If F0 > S0 1 + κr +C, consider the strategy of longing the spot and shorting

the futures. We want to show that for any price path of the perpetual futures and the spot,
there exists a future unwinding time t such that the strategy’s payoff is positive, that is,
this is a random-maturity arbitrage. Suppose to the contrary that there exists a price path
{Ft }t=0
∞ and {S }∞ such that ∀t, the discounted payoff from the strategy is negative, or
t t=0
equivalently that:
Z t
−rt
e F0 − S0 + κ (Fs − Ss )e−rs ds ≤ e−rt (Ft − St ) + e|−rt{zC} . (2)
| {z } 0 | {z }
traditional spread | {z } spread at unwinding trading costs
funding payments
12
Denote (Ft − St )e−rt ≡ ut the discounted deviation. The inequality becomes:
Z t
ut ≥ e−rt (F0 −C) − S0 + κ us ds .
0
Rt
ut = e−rt (F0 −C) − S0 + κ 0 us ds provides a lower bound for all processes ut satisfying the
above inequality. Solving this integral equation, we have:
F0 re−rt

F0 −C κt
ut = + r − S0 e .
κ +r 1+ κ
When F0 > S0 1 + κr +C, limt→∞ ut = ∞. This contradicts Assumption 1, because it implies:

lim inft→∞ (Ft − St )e−rt ≥ limt→∞ ut → ∞. Therefore, when F0 > S0 1 + κr + C, longing the

spot and shorting the futures is a random-maturity arbitrage opportunity.
Scenario 2: Next, if F0 < S0 1 + κr −C, consider the strategy of longing the futures and

shorting the spot. As before, we want to show for this strategy, that for any price path of
the perpetual futures and the spot, there always exists a future time t such that the strategy
payoff is positive, i.e. this is a random-maturity arbitrage. Suppose not, then there exists
a price path {Ft }t=0
∞ and {S }∞ such that ∀t, the discounted payoff is negative:
t t=0
Z t
−rt
ut ≤ e (F0 +C) − S0 + κ us ds .
0
Rt
ūt = e−rt (F0 +C) − S0 + κ 0 ūs ds provides an upper bound for all processes ut satisfying the
above inequality. This is the same integral equation as we see in the first case. We have:
F0 re−rt

F0 +C
ūt = + − S0 eκt .
κ +r 1 + κr
When F0 < S0 1 + κr − C, limt→∞ ūt = −∞.

This contradicts Assumption 1, because
lim inft→∞ (St − Ft )e−rt ≥ limt→∞ (−ūt ) → ∞. Therefore, when F0 < S0 1 + κr − C, longing

the futures and shorting the spot is a random-maturity arbitrage opportunity.
13
No arbitrage prices are usually derived assuming away trading costs. For complete-
ness, the following result considers this special case and provides the fundamental value
of perpetual futures.
Proposition 2 (No-arbitrage price). When there is no trading cost (C = 0), arbitrageurs will
trade perpetual futures toward:
r
Ft = St 1 + . (3)
κ
Proof. The proof follows that of Proposition 1 by setting C = 0.
To gain intuition for this result consider the first three terms of Equation (2). The
first is the traditional spread familiar from fixed-maturity futures pricing. Equating it to
zero generates the usual no-arbitrage price for futures that pay no dividends and without
carrying or storage costs. The second term is the present value of cumulated funding
payments from initiation to unwinding. These first two terms account for the gains to the
arbitrageur in this scenario. The third term, on the right hand side of the inequality, is
the present value of the futures-to-spot spread at the random unwinding time and is the
source of tension here. For fixed-maturity futures this spread is guaranteed to be zero at
expiration. But an arbitrageur in perpetual futures faces the risk that when they wish to
unwind the trade, the spread may be arbitrarily large.
Our key insight is that a positive and even modestly large spread can still leave the
arbitrageur with a positive payoff because funding payments also increase with the spread.
As long as the spread does not diverge to infinity, there will be some future unwinding
time t when the accumulated funding rate payments overcome any finite potential losses
at unwinding. Arbitrage is absent when the accumulated benefits due to the funding rate
κ exactly balance against the accumulated costs due to the borrowing interest rate r. The
no-arbitrage price (3) captures this intuition and says that the futures-spot gap increases
with the ratio r/κ.
14
Note that if we relax the assumption that the short rate is constant for arbitrageurs, this
may introduce some additional risk to the payoff: (1) the arbitrageur will face some risk
in rolling over her borrowing in the cash market; (2) she faces some risk in reinvesting
the funding payment she receives. To map this more general case into the payoffs in Table
Rt
1, this corresponds to a cash market payoff of −S0 e 0 rs ds at time t and a funding payment
Rt Rt
s ru du
of κ 0 (Fs − Ss )e ds, where {rs }∞
s=0 is a stochastic process. Empirically, however, the
volatility in the short rate is considerably smaller than the volatility in the gap between
futures and spot prices. Moreover, the changes in the short rate have opposite effects on
payment from the cash market and the funding payment. Therefore, we abstract from
short rate randomness here, but note it is an interesting avenue for future work.
In the presence of trading costs, when the deviation of perpetual futures price from
S0 (1 + kr ), is larger than the round-trip trading costs C, arbitrageurs would have a strong
incentive to trade perpetual futures toward the price S0 (1 + κr ). This proposition also
prescribes a trading strategy to exploit the futures-spot divergence in markets with
different levels of trading costs. In the next section, we provide an empirical analysis of the
futures-spot deviation and present arbitrage trading strategies motivated by our theory.
4 Data and Empirical Analysis
We conduct an empirical analysis of perpetual futures arbitrage strategies. We first

describe the data. We then measure the deviations of the crypto futures-spot spread
from the no-arbitrage benchmark. Finally, we implement a trading strategy that exploits
deviations from random-maturity no-arbitrage bounds and quantify the gains from this
strategy net of trading costs.
15
4.1 Data
We focus on the five largest cryptocurrencies excluding stablecoins: Bitcoin (BTC), Ether
(ETH), BNB (BNB), Dogecoin (DOGE), and Cardano (ADA) with a total market cap of
$529 billion, which account for 64.15% of the total market share of the Crypto market by
November 2022.
For each token, we obtain perpetual futures and spot prices at a 1-hour frequency from
Binance. Binance is by far the leading exchange in the crypto realm. Another major benefit
of using the Binance data is: nearly every part of our trading strategies can be completed
within the same platform without delay in transferring funds. As such, it approximates
the real-world investment opportunities faced by traders.
We retrieve the perpetual funding rates from Binance. The funding rate is paid every
8 hours on Binance. So we have futures and spot prices every hour and realized funding
rate payment at 8:00, 16:00, and 0:00 GMT each day. The perpetual and spot tradings occur
24 hours per day and 7 days a week so there are no after-market hours in this market.
We get the earliest possible data on perpetual futures trading from Binance. The table
below lists the starting and ending dates of our data for each cryptocurrency. Our data
ends on 2022-11-13, covering the fallout of the FTX.
Crypto Start date End date N

BTC 2019-09-10 2022-11-13 27,895
ETH 2019-11-27 2022-11-13 25,985
BNB 2020-02-10 2022-11-13 24,184
DOGE 2020-07-10 2022-11-13 20,559
ADA 2020-01-31 2022-11-13 24,424
Table 2 Sample descriptions
This table presents the sample start and end dates for the 5 cryptocurrencies and their total number
of observations.
16
We obtain trading costs from Binance’s website3 . In general, trading costs for the spot
market are significantly larger than those for the perpetual futures for similar trading
volume because futures are typically traded with leverage. Fee tiers are attributed to the
30-day trading volume. We attribute high trading costs with a 30-day spot trading volume
above $1 million and futures trading volume above $15 million (small individual trader).
Medium trading costs attribute to a 30-day spot trading volume above $150 million and
futures trading volume above $1 billion (small funds). Low trading costs attribute to a
30-day spot trading volume above $2 billion and futures trading volume above 12.5 billion
(large funds). Zero fees can be negotiated with customized contracts, for example for
market makers. We consider trading costs for makers instead of takers because institutions
typically trade maker orders.4 Table 3 presents the specification of different trading
costs. It also shows the random-maturity arbitrage bound for the deviation between the
perpetual futures and the no-arbitrage price (ρl and ρu ). The bounds become wider as
the trading costs increase. Detailed explanations of the trading costs specifications are
also provided in appendix Figure 6 and Figure 7.
To measure the deviation from the no-arbitrage price, we also obtain the interest rate
data from Aave, a leading open-source DeFi liquidity protocol. Customers on Aave can
either be a supplier or a borrower of cryptocurrencies. Because of the anonymity and
decentralization of DeFi system, all borrowings are over-collateralized and the collateral
can serve as an additional supply to the system for borrowing. The customers can also
supply spare currencies to the system to earn an interest rate. The supply interest rate is
typically different from the borrowing interest rate. Both interest rates and their wedge
are determined algorithmically based on the market condition of supply and demand. We
use interest rates from this platform because we believe it is a good proxy for the funding
3 https://www.binance.com/en/fee/trading provides data on trading fees in perpetual futures and the
spot market
4 Takers trade market orders while makers trade limit orders. Takers take liquidity from the market while
makers make the market or provide liquidity to the market.
17
Fee tier Spot Futures ρl ρu
No 0% 0% 0.0% 0.0%
Low 0.0225% 0.0018% -53.2% 53.2%
Medium 0.045% 0.0072% -114.4% 114.3%
High 0.0675% 0.0144% -179.5% 179.2%
Table 3 Trading costs tiers
Fee tiers are assigned based on past 30-day trading volume. High fees correspond to a 30-day
trading volume above $1mn in spot and above $15mn in perpetuals, typically an individual trader.
Medium fees attribute to a 30-day trading volume above $150mn in spot and above $1bn in
perpetuals (small funds). Low fees attribute to a 30-day trading volume above $2bn in spot and
above $12.5bn in perpetuals (large funds). The no fees tier can be negotiated with customized
contracts, for example for market makers. We also report the theory implied no arbitrage bound (ρl
and ρu ) for ρ under different trading costs specifications. They are calculated using the following
formulas: ρl = κlog (1 −C), ρu = κlog (1 +C), where C is the round-trip percentage trading costs of
the long-short strategy of perpetual and spot.
condition in the crypto market. Our results are robust if we use the interest rate from the
traditional financial market.5
Our theory indicates using a risk-free rate available to the arbitrageurs. Therefore, we
consider the interest rates on the stablecoins, which are not subject to volatile spot price
movement and are the currency of denomination for perpetual futures margins. There are
three major stablecoins traded on Aave: USDT, USDC, and DAI. To get a robust measure
of the risk-free rate, we take an average of the three interest rates to arrive at our final
risk-free supply and borrowing rate for the arbitrageurs.
We plot the time-series evolvement of the interest rate in the appendix Figure 9. During
the early years, we see higher interest rate volatility due to the funding liquidity of the
DeFi platform. In later samples, interest rates are more stable and approach interest rates
in traditional financial markets.
Our interest rate data starts from 2020-01-08. Therefore, for coins with perpetual data
5 We also run our analysis using daily T-bill rates obtained from Kenneth French’s website at https:
//mba.tuck.dartmouth.edu/pages/faculty/ken.french/Data_Library/f-f_factors.html. The results
are very close to ones using crypto market supply and borrowing rates.
18
available before the time (BTC and ETH), we begin the analysis from 2020-01-08. For
other coins (BNB, DOGE, ADA), we begin the analysis from the time when they have
data available.
4.2 Deviations of perpetual futures from no-arbitrage benchmarks
The focus of our empirical analysis is the annualized deviation ρ, defined as the interest
rate spread would rationalize an observed future-spot spread:

r+ρ
F = S 1+ .
κ
For example, an estimated ρ of 1 percent would mean that borrowing costs faced by
arbitrageurs would have to be 1 percentage points higher than the prevailing interest rate
r for the the futures-spot gap to be arbitrage free.
Using f and s to denote log(F) and log(S) respectively, we obtain the following
approximate equation for ρ:

F
ρ =κ − 1 − r ≈ κ( f − s) − r . (4)
S
This definition is the same in spirit to Du et al. (2018) who measure covered interest parity
(CIP) deviations with the wedge that would equate the dollar borrowing rate and the
synthetic dollar borrowing rate.
At each hour, we calculate ρ using data on the perpetual futures price, spot price, and
the crypto risk-free interest rate. When the perpetual futures price is above the spot, an
arbitrageur would short the futures and long the spot, and would finance her position by
borrowing in the cash market (as is shown in Table 1). Therefore, we use the borrowing
rate from Aave as the risk-free rate. On the other hand, when the perpetual futures price
is below the spot, an arbitrageur would long the futures, short the spot, and invest the
19
proceeds from shorting in the cash market (as is shown in Table 1). In such cases, we use
the supply rate from Aave as the risk-free rate.
400%
BTC
ETH
300% BNB
Annualized percentage point deviation,
DOGE
200% ADA
100%
0%
-100%
-200%
-300%
2020-01 2020-05 2020-09 2021-01 2021-05 2021-09 2022-01 2022-05 2022-09
Figure 3 Deviations of perpetual futures from no-arbitrage benchmarks
This figure presents the 7-day moving averages of the annualized deviation of the perpetual
futures-spot spread from the no-arbitrage benchmark for the five cryptocurrencies.
ρ |ρ|
Crypto Mean Median Std p-value Mean Median Std p-value
BTC -0.02 -0.25 0.79 0.60 0.59 0.53 0.52 0.00
ETH 0.12 -0.09 0.89 0.13 0.67 0.56 0.61 0.00
BNB -0.07 -0.18 1.32 0.70 0.94 0.73 0.93 0.00
DOGE 0.08 -0.18 1.62 0.28 1.01 0.72 1.27 0.00
ADA 0.12 -0.06 1.25 0.18 0.91 0.74 0.87 0.00
Table 4 Summary Statistics for Deviations from No-arbitrage Prices
The table below summarizes statistics for ρ and |ρ| across various crypto tokens. pNW corresponds
to the Newey-West adjusted p-value, testing the null hypothesis that the mean ρ (or |ρ|) is zero.
The sample periods are shown in Table 2.
Table 4 reports summary statistics for ρ and |ρ| in our sample. As can be seen,
the mean future-spot spread is modest, lying between −2 and 12 percentage points.
20
The p-values show we cannot reject the hypothesis that the deviations are zero on
average. Thus, on average our theoretical prices provide a useful benchmark. Around
this benchmark, however, perpetual futures prices vary significantly from spot prices. The
mean absolute deviation is about 60% to 100% per year across different cryptocurrency
tokens. Figure 3 depicts the 7-day moving average of ρ for each of the five examined
cryptocurrencies. It reveals considerable comovement in futures-spot spreads across these
cryptocurrencies, suggesting a common factor at play that drives the observed divergence.
This observation is corroborated by Table 5, which demonstrates high correlations among
futures-spot deviations across the different cryptocurrencies. This trend underscores the
interconnectedness of cryptocurrency markets.
One hypothesis for this significant comovement in ρ across various cryptocurrencies

pertains to time-varying funding constraints encountered by arbitrageurs in the market
(Brunnermeier and Pedersen, 2009; Garleanu and Pedersen, 2011). Because arbitrageurs
often serve as marginal traders in all markets, their funding constraints are potentially
reflected in the ρ across all major cryptocurrency markets. Our theoretical and empirical
analyses underscore the importance of measuring ρ. Doing so not only sheds light on
potential market stress but also provides a gauge for the shadow costs associated with the
trading and funding constraints arbitrageurs face.
Conversely, from the demand side, shared sentiment across various cryptocurrencies
could also contribute to explaining the concurrent movement of ρ across different markets.
Our theoretical analysis suggests that arbitrageurs will only meet the market demand if
the price deviation surpasses the combined trading and funding costs. As a result, the
overall market sentiment in the futures market relative to the spot market becomes evident
in the price difference between futures and spot prices. If the sentiment across different
markets is influenced by a common factor, we would anticipate a significant comovement
in the futures-spot spread. In Section 5, we delve deeper into explaining the futures-spot
spread. Our results demonstrate that the past returns of each cryptocurrency serve as
21
Arbitrage Deviation Funding Rate Spot Return
ρET H ρBNB ρDOGE ρADA f rBTC f rET H f rBNB f rDOGE f rADA RBTC RET H RBNB RDOGE RADA
ρBTC 0.87 0.65 0.71 0.79 0.85 0.74 0.62 0.54 0.68 -0.03 0.00 0.01 0.02 0.00
ρET H 0.68 0.71 0.81 0.76 0.85 0.61 0.51 0.69 0.01 0.03 0.03 0.03 0.02
Arbitrage Deviation ρBNB 0.65 0.62 0.60 0.60 0.89 0.51 0.56 0.00 0.02 0.02 0.02 0.02
ρDOGE 0.74 0.62 0.62 0.60 0.81 0.68 0.03 0.04 0.07 -0.02 0.03
ρADA 0.68 0.67 0.56 0.55 0.84 0.03 0.04 0.06 0.05 0.03
f rBTC 0.86 0.70 0.62 0.73 -0.01 0.01 0.04 0.03 0.01
f rET H 0.67 0.57 0.73 0.03 0.05 0.07 0.06 0.04
Funding Rate f rBNB 0.59 0.64 0.02 0.04 0.06 0.05 0.04
22
f rDOGE 0.67 0.05 0.05 0.11 -0.02 0.05

f rADA 0.04 0.06 0.10 0.07 0.05
RBTC 0.96 0.94 0.89 0.94
RET H 0.93 0.88 0.96
Spot Return
RBNB 0.84 0.92
RDOGE 0.88
Table 5 Correlation of arbitrage deviation (ρ), funding rate ( f r), and spot returns (R)
This table presents the correlation among different cryptocurrencies’ futures-spot deviation ρ (rows 3 to 7), funding rate (rows 8 to 12),
and spot returns (rows 13 to 16).
significant explanatory variables for the time-series variation in the spread.
Interestingly, the correlation between deviations and spot market returns is quite
modest. Although spot market returns are highly correlated among themselves, in
line with the strong market factor results from Liu et al. (2022), and no-arbitrage price
deviations are highly correlated among themselves, it appears that different forces drive
these distinct phenomena.
Furthermore, we find that after the year 2022, futures-spot spreads become smaller in
magnitude and less volatile compared to earlier years. The 7-day moving average stays
around -50% most of the time for the 5 cryptocurrencies while larger sways in earlier
years are quite common. This suggests the market is becoming increasingly efficient. In
terms of the sign of the deviation, which appears to stabilize at the negative region, two
forces are potentially at play: (1) on the futures customer end, the relative demand in
the futures market is weaker compared to the spot; (2) for arbitrageurs, because of the
lack of infrastructure to short the cryptos in the spot market (high shorting costs), their
funding constraints would be larger in the negative region. All these forces contribute to
the stabilization of the futures spot deviation around −50 percentage points.
Last but not least, by design, ρ is also highly correlated with the funding rate. The
funding rate does not correlate perfectly with ρ because in real-world implementations:
(1) there is a clamp region, within which the funding rate is fixed at 0.01%; (2) the funding
rate is calculated as a weighted average of futures-spot price deviations, with larger weight
given to more recent observations; (3) in calculating the funding rate, Binance does not
just consider the quote price, they also use an impact margin notional to consider price
impact of the trade.6
6 See https://www.binance.com/en/support/faq/360033525031 for details of the funding rate calcula-
tion.
23
4.3 Random-maturity Arbitrage Strategy
We next provide a trading strategy motivated by our random-maturity arbitrage theory.

Table 3 reports for different trading cost tiers, the bounds (ρl and ρu ) beyond which there
exist random-maturity arbitrage opportunities. We consider a simple trading strategy:
whenever ρ enters the region outside the annualized round-trip trading costs in Table 3,
we open the position. We close the position when ρ first goes back to 0.
BTC
400%
open
close
300%
200%
100%
0%
-100%
-200%
-300%
2020-07 2020-10 2021-01 2021-04 2021-07 2021-10 2022-01 2022-04 2022-07 2022-10
Figure 4 Random-maturity Arbitrage Strategy: Bitcoin, high trading costs
Futures-spot deviations and trading thresholds of the random-maturity arbitrage strategy we

implement for BTC under high trading costs. Each blue dot in the figure represents the annualized
deviation of futures from the spot ρ = κ( f − s) − r. The orange line is the open-position threshold
and the red line is the close-position threshold.
Figure 4 provides an illustration of the trading strategy. When the deviation is beyond
the orange lines, the strategy opens a trading position. The position is closed when the
futures/spot deviation first hits the red line. We present trading thresholds across different
trading costs for different currencies in Figure 10 in the appendix. Different trading
strategies have the same close-position line which is equal to the risk-free rate while the
open-position line adjusts to the level of trading costs.
24
Since our trading strategy is a threshold trading rule, to annualize Sharpe ratios we
follow Lucca and Moench (2015), which considers a trading strategy that is only active on
FOMC announcement dates. We first calculate the mean (µ) and standard deviation (σ )
of our trading strategy during the time it is active. Next, we scale µ/σ by the number of
periods the strategy is active in a year:
µ√
SR = Na ,
σ
where µ and σ are average hourly returns and Na is the average number of hours the
strategy is active in a year. We follow the same approach to annualize the returns and
√
standard deviations: µann = µNa , σann = σ Na .
Table 6 presents the strategy’s performance under different fee tiers. As trading costs
decrease, the random-maturity arbitrage strategy engages in more active trading, leading
to an increase in the SRs and a decrease in the average duration of open-to-close positions.
The strategy also delivers highly significant alphas relative to the 3-factor model by Liu
et al. (2022) and the 5-factor model by Cong et al. (2022a). Evidently, the strategy’s
performance cannot be explained by previously suggested risk factors.
Table 7 zooms in and presents the trading performance of the strategy under high
trading costs. We consider two cases: (1) unrestricted; (2) long-spot only. We consider
the second case because, in earlier years of crypto derivatives, the infrastructure of
shorting cryptocurrencies is not well-developed. We find that, as is implied by our
theory, whenever the deviation between futures and spot is larger than the trading costs,
performing a random-maturity arbitrage strategy would generate returns with high SR.
After trading costs, the strategy generates a SR of 1.92 for BTC and much higher SRs for
other cryptocurrencies.
Table 7 also shows that as time goes by, the perpetual futures market seems to become
more and more efficient. As we can see in 2022, the deviation of crypto price from the
25
Fee tiers
No Low Medium High
BTC SR 3.94 2.50 2.31 1.92
Return 17.89 11.21 10.13 8.15
Volatility 4.54 4.48 4.39 4.25
MaxDD -4.24 -4.27 -4.34 -4.43
α 22.64 10.86 8.24 5.47
tα 5.39 2.71 2.10 1.38
Active % 100.00 84.93 43.87 22.32
OtC time 15.80 63.90 95.97 113.38
ETH SR 5.43 3.46 3.13 2.82
Return 28.03 17.49 15.17 12.68
Volatility 5.16 5.06 4.85 4.49
MaxDD -4.13 -4.21 -3.90 -3.94
α 45.55 23.09 18.33 14.42
tα 7.56 4.25 3.69 3.15
Active % 99.94 84.63 51.21 28.11
OtC time 12.10 37.57 70.75 75.77
BNB SR 12.49 8.07 6.41 5.44
Return 62.38 38.40 29.14 23.43
Volatility 4.99 4.76 4.55 4.31
MaxDD -1.12 -1.12 -1.10 -1.11
α 107.81 58.17 41.12 32.87
tα 10.53 7.79 7.08 6.50
Active % 99.97 89.25 66.67 39.94
OtC time 7.87 14.65 22.20 24.07
DOGE SR 12.61 8.34 5.96 4.42
Return 112.09 70.88 49.74 35.82
Volatility 8.89 8.49 8.35 8.10
MaxDD -8.56 -8.56 -8.56 -8.56
α 312.42 151.20 88.73 54.65
tα 7.76 6.56 5.53 4.46
Active % 99.85 90.35 68.52 36.90
OtC time 5.56 8.50 11.21 11.26
ADA SR 11.34 6.46 4.07 3.12
Return 69.58 38.05 23.33 16.72
Volatility 6.13 5.89 5.73 5.36
MaxDD -4.30 -4.26 -4.29 -4.34
α 120.52 52.28 27.52 18.11
tα 14.56 10.22 6.55 4.59
Active % 99.64 90.49 69.32 41.03
OtC time 6.57 11.26 20.80 33.77
Table 6 Performance of Random-maturity Arbitrage Strategy
Performance under different trading cost tiers. The fees for spot (futures) are 2.25 (0.18) bps, 4.5
(0.72) bps, and 6.75 (1.44) bps for the low, medium, and high trading cost levels. Statistics reported
are the annualized Sharpe ratio, return (%), volatility (%), max drawdown (%), alpha (%), t-stat of
the alpha, proportion of time the strategy is active (%), and average open-to-close (OtC) position
duration in hours.
26
Unrestricted Long-spot Only
2020 2021 2022 All 2020 2021 2022 All
BTC SR 2.26 2.39 1.23 1.92 1.98 2.28 0.55 1.75
Return 8.28 14.80 0.26 8.15 6.43 13.96 0.10 7.16
Volatility 3.67 6.18 0.21 4.25 3.25 6.12 0.18 4.10
Active % 28.66 34.43 1.01 22.32 22.18 32.07 0.03 18.95
N 8,616 8,760 7,536 24,912 8,616 8,760 7,536 24,912
ETH SR 3.08 3.52 1.39 2.82 2.57 3.37 0.87 2.50
Return 17.11 18.08 1.36 12.68 13.99 17.14 0.62 11.06
Volatility 5.55 5.13 0.98 4.49 5.44 5.09 0.71 4.42
Active % 37.80 34.91 9.12 28.11 34.40 34.54 0.09 24.07
N 8,616 8,760 7,536 24,912 8,616 8,760 7,536 24,912
BNB SR 6.04 6.60 2.61 5.44 4.45 5.00 0.68 3.94
Return 31.26 33.13 4.04 23.43 15.32 21.94 0.16 12.99
Volatility 5.17 5.02 1.55 4.31 3.44 4.39 0.24 3.29
Active % 55.07 48.03 14.84 39.94 34.96 31.27 0.03 22.70
N 7,815 8,760 7,536 24,111 7,815 8,760 7,536 24,111
DOGE SR 4.90 5.93 1.93 4.42 2.64 5.05 1.11 3.10
Return 59.81 53.51 1.94 35.82 31.16 36.34 0.29 22.02
Volatility 12.19 9.02 1.00 8.10 11.80 7.19 0.26 7.11
Active % 60.12 49.19 9.70 36.90 37.49 41.91 0.05 25.61
N 4,190 8,760 7,536 20,486 4,190 8,760 7,536 20,486
ADA SR 3.87 3.41 2.27 3.12 3.16 2.94 0.00 2.49
Return 21.32 24.18 3.15 16.72 16.03 20.36 0.00 12.62
Volatility 5.51 7.09 1.39 5.36 5.08 6.92 0.00 5.08
Active % 47.61 42.18 32.66 41.03 36.41 39.53 0.00 26.27
N 8,055 8,760 7,536 24,351 8,055 8,760 7,536 24,351
Table 7 Performance Over Time: High Trading Costs Tier
This table presents the Sharpe ratios, annualized returns (%), standard deviations (%), and
active percentages (%) of the random-maturity arbitrage trading strategies for five different
cryptocurrencies with high trading costs. We break down returns for each year and provide
summary statistics across all time. The left panel shows the performance of the unrestricted
trading strategy, where both long and short spot positions are allowed. The right panel shows
the performance of the long-spot-only trading strategy, where shorting the spot is not allowed.
arbitrage-free bound is much less frequent compared to earlier years. But when the
deviation happens, the resulting SR from the trade remains high.
Comparing the results with Du, Tepper, and Verdelhan (2018), we find that deviations
in crypto perpetual futures are considerably larger in magnitude. As a result, gains from
27
the arbitrage strategies we study are also larger. Even though the volatility of the trading
strategy also scales up, the Sharpe ratios in the crypto space are still larger than those in
the traditional foreign exchange market as reported in Du, Tepper, and Verdelhan (2018).
When a futures-spot gap opens up, gains from the trading strategy could potentially
arise from two main sources: price convergence and funding rate payments. While
industry publications usually emphasize the funding rate channel, we note that price
convergence can generate quicker gains from arbitrage if dislocations are short-lived. To
examine these two sources empirically, in Table 8 we provide a decomposition of the
trading strategy’s performance into price convergence versus funding rate payment. We
find that price convergence plays a dominant role in total trading returns, while funding
rate payments have a more minor role, which seems to diminish over time.
2020 2021 2022 All

BTC Return 21.68 29.60 -0.05 17.89
Price 14.00 14.29 2.11 10.51
Funding 7.68 15.31 -2.16 7.39
ETH Return 40.93 36.66 3.24 28.03
Price 27.58 19.07 3.75 17.38
Funding 13.35 17.59 -0.51 10.65
BNB Return 82.48 76.29 25.36 62.38
Price 66.70 60.37 20.00 49.80
Funding 15.78 15.92 5.36 12.57
DOGE Return 220.81 113.79 49.67 112.09
Price 214.72 93.24 50.52 102.37
Funding 6.09 20.55 -0.85 9.72
ADA Return 90.81 66.73 50.19 69.58
Price 77.17 49.51 49.05 58.52
Funding 13.64 17.22 1.13 11.06
Table 8 Return Decomposition: Price Convergence vs Funding Rate Payment for

Random-maturity Arbitrage Strategies
This table decomposes the portfolio return into the part due to price convergence and the part due
to funding rate payment.
The success of the trading strategy supports the theory of random-maturity arbitrage.
28
Whenever there is a deviation larger than the gap implied by the theory, betting on
convergence tends to generate a positive payoff at some uncertain future time. Therefore,
the convergence arbitrage trade generates high Sharpe ratios.
5 Explaining Futures-spot Deviations
From Figure 3, we observe a strong common comovement of the futures-spot deviation

across all crypto assets. Our goal is to understand the fundamental forces driving this
common factor. We consider two potential hypotheses: (1) the time-varying funding
constraints of arbitrageurs, and (2) the time-varying relative demand from end-users for
perpetual futures compared to the spot.
Both forces could potentially explain these patterns. Arbitrageurs will accommodate
the relative demand from end-users only until the price deviation lies within the random-
maturity no-arbitrage bound, which depends on arbitrageurs’ funding conditions. As
arbitrageurs are likely the marginal investors in all cryptocurrency markets, their time-
varying funding constraints may create common time-series variation in ρ across different
cryptocurrencies.
On the other hand, when the deviation lies within the no-arbitrage bound, variations
in demand from the perpetual futures market compared to the spot market will influence
the price deviation. It is plausible that relative demand has common factors across
different cryptocurrencies. For instance, sentiment could drive their common variation,
as perpetual futures allow for high leverage, which attracts overconfident, extrapolative,
and sentiment-driven investors.
Determining which of the two factors better explains the observed price deviation is
an empirical question. To shed light on this, we use past returns as a proxy for relative
extrapolative demand in the perpetual futures market compared to the spot market. Past
29
returns correlate with the demand from investors with extrapolative beliefs, who are
more likely to trade in the perpetual futures market given the high leverage it provides.
Therefore, past returns would correlate with the relative demand in the perpetual futures
market compared to the spot.
For the time-varying funding constraint, we consider using crypto return volatility as a
proxy, because, as in Adrian and Shin (2010), arbitrageurs likely face Value-at-Risk (VaR)
type constraints, which are more likely to bind when market volatility is high.
Dependent Variable: ρ
BTC ETH
Ret 0.28*** 0.28*** 0.20*** 0.23***
(7.79) (7.74) (3.87) (4.83)
Vol -0.02 -0.01 -0.01** 0.01
(-1.60) (-0.64) (-1.99) (1.58)
Const -0.12** 0.30 -0.03 -0.06 0.36 -0.28*
(-2.16) (1.22) (-0.16) (-0.58) (1.40) (-1.67)
R2 0.55 0.03 0.56 0.47 0.03 0.49
N 1011 1011 1011 1011 1011 1011
Table 9 Regression of the futures-spot gap against explanatory variables
We present regression results of ρ on the past four months’ annualized returns (Ret) and volatility
(Vol) for Bitcoin and Ethereum. The left panel reports results for Bitcoin, and the right panel
reports results for Ethereum. For each cryptocurrency, we consider three models: (1) regressing
ρ on past returns; (2) regressing ρ on past volatility; and (3) regressing ρ on both past returns
and volatility. The observations are at a daily frequency. The sample period spans from January
30, 2020, to November 13, 2022, totaling 1,011 days. The table also reports the R-squared values
for each regression model. We report HAC-robust t-statistics in parentheses. ∗p < .1, ∗ ∗ p < .05,
∗ ∗ ∗p < .01
Table 9 presents the regression results of ρ on past returns, volatility, and both for BTC
and ETH. The regression coefficients on past returns are positive and highly significant.
This suggests that when past returns are high, perpetual futures exhibit a more positive
deviation against the spot. Mapping this back to our second hypothesis, when past returns
are high, the demand for futures relative to the spot is also likely to be high. Consequently,
even after arbitrageurs accommodate the demand outside of trading costs, the residual
30
demand still manifests itself through the perpetual-spot deviation.
This observation on limits to arbitrage is related to Makarov and Schoar (2020) which
finds that crypto price deviations across international exchanges tend to comove with one
another. The driving force behind this comovement is investors’ buying pressure across
various countries, with cross-country capital controls serving as the primary impediment
to arbitrage capital. In the context of perpetual futures, end-user demand contributes
to the comovement of different cryptocurrencies. The limits to arbitrage predominantly
manifest in the form of trading frictions.
Liu and Tsyvinski (2021) document a significant time-series momentum pattern in

the crypto market. Given that positive past returns lead to a positive gap between
futures and spot prices, it is worth examining further whether the time-series momentum
phenomenon is driven by margin trading and price pressure from the perpetual futures
market. We leave this as an open question for future research.
We find that volatility does not significantly covary with the futures and spot deviation.
This suggests that time-varying funding constraints of arbitrageurs do not drive the
comovement in futures-spot deviations across different cryptocurrencies.
In summary, a prerequisite for the gap between futures and spot prices to occur is the
existence of trading costs for arbitrageurs. Arbitrage trading will accommodate all the
demand in the futures until the price deviation lies within a trading cost bound. Within
the bound, the relative demand of futures compared to the spot will still manifest itself in
the deviation of futures prices from the spot. Due to the comovement of the time-varying
relative demand, we observe a significant common factor in crypto future-spot deviations.
31
6 Conclusion
Perpetual futures play an important role in today’s crypto markets and could potentially be
adopted in non-crypto markets in the future. Understanding the fundamental mechanism
of this financial derivative is a crucial first step for understanding speculation and hedging
dynamics in this fast-evolving area. We provide a comprehensive analysis of the arbitrage
and funding rate payment mechanisms that underpin perpetual futures.
In an ideal, frictionless world, we show that arbitrageurs would trade perpetual futures
in such a way that a constant proportional relationship would hold between the futures
price and the spot price. In the presence of trading costs, the deviation of the futures price
from the spot would lie within a bound.
Motivated by the theory, we then empirically examine the comovement of the futures-
spot spread across different cryptocurrencies and implement a theory-motivated arbitrage
strategy. We find that this simple strategy yields substantial Sharpe ratios across various
trading cost scenarios. The evidence supports our theoretical argument that perpetual
futures-spot spreads exceeding trading costs represent a random-maturity arbitrage op-
portunity.
Finally, we provide an explanation for the common comovement in futures-spot

spreads across different crypto-currencies: arbitrageurs can only accommodate market
demand if the price deviation exceeds trading costs. As a result, the overall sentiment in
the futures market relative to the spot market is reflected in the spread. Our empirical
findings suggest that past return momentum can account for a significant portion of the
time-series variation in the futures-spot spread.
32
References
Adams, Hayden, Noah Zinsmeister, Moody Salem, River Keefer, and Dan Robinson, 2021,
Uniswap v3 core, Whitepaper .
Adrian, Tobias, and Hyun Song Shin, 2010, Liquidity and leverage, Journal of financial
intermediation 19, 418–437.
Alexander, Carol, Jaehyuk Choi, Heungju Park, and Sungbin Sohn, 2020, Bitmex bitcoin
derivatives: Price discovery, informational efficiency, and hedging effectiveness, Journal
of Futures Markets 40, 23–43.
Amiram, Dan, Evgeny Lyandres, and Daniel Rabetti, 2021, Cooking the order books:
Information manipulation and competition among crypto exchanges, SSRN Electronic
Journal .
Angeris, Guillermo, Tarun Chitra, Alex Evans, and Matthew Lorig, 2022, A primer on
perpetuals, arXiv .
Brunnermeier, Markus K, and Lasse Heje Pedersen, 2009, Market liquidity and funding
liquidity, The review of financial studies 22, 2201–2238.
Cochrane, John, 2009, Asset pricing: Revised edition (Princeton university press).
Cong, Lin William, Zhiheng He, and Ke Tang, 2022a, Staking, token pricing, and crypto
carry, SSRN Electronic Journal .
Cong, Lin William, George Andrew Karolyi, Ke Tang, and Weiyi Zhao, 2022b, Value
premium, network adoption, and factor pricing of crypto assets, SSRN Electronic Journal
.
Cong, Lin William, Xi Li, Ke Tang, and Yang Yang, 2022c, Crypto wash trading, Technical
report, National Bureau of Economic Research.
De Blasis, Riccardo, and Alexander Webb, 2022, Arbitrage, contract design, and market
structure in bitcoin futures markets, Journal of Futures Markets 42, 492–524.
Du, Wenxin, Alexander Tepper, and Adrien Verdelhan, 2018, Deviations from Covered
Interest Rate Parity, Journal of Finance 73, 915–957.
Duffie, Darrell, 2010, Presidential address: Asset price dynamics with slow-moving
capital, The Journal of finance 65, 1237–1267.
33
Ferko, Alex, Amani Moin, Esen Onur, and Michael Penick, 2022, Who trades bitcoin
futures and why?, Global Finance Journal 100778.
Garleanu, Nicolae, and Lasse Heje Pedersen, 2011, Margin-based asset pricing and
deviations from the law of one price, The Review of Financial Studies 24, 1980–2022.
Gromb, Denis, and Dimitri Vayanos, 2018, The dynamics of financially constrained
arbitrage, The Journal of Finance 73, 1713–1750.
Hertzog, Eyal, Guy Benartzi, Galia Benartzi, and Omri Ross, 2017, Bancor protocol,
continuous liquidity for cryptographic tokens through their smart contracts, Whitepaper
.
Koijen, Ralph SJ, Tobias J Moskowitz, Lasse Heje Pedersen, and Evert B Vrugt, 2018, Carry,
Journal of Financial Economics 127, 197–225.
Liu, Yukun, and Aleh Tsyvinski, 2021, Risks and returns of cryptocurrency, The Review of
Financial Studies 34, 2689–2727.
Liu, Yukun, Aleh Tsyvinski, and Xi Wu, 2022, Common risk factors in cryptocurrency, The
Journal of Finance 77, 1133–1177.
Lucca, David O, and Emanuel Moench, 2015, The pre-fomc announcement drift, The
Journal of finance 70, 329–371.
Makarov, Igor, and Antoinette Schoar, 2020, Trading and arbitrage in cryptocurrency
markets, Journal of Financial Economics 135, 293–319.
Pedersen, Lasse Heje, 2019, Efficiently inefficient: how smart money invests and market prices
are determined (Princeton University Press).
Schmeling, Maik, Andreas Schrimpf, and Karamfil Todorov, 2022, Crypto carry, SSRN
Electronic Journal .
Shiller, Robert J, 1993, Measuring asset values for cash settlement in derivative markets:
hedonic repeated measures indices and perpetual futures, The Journal of Finance 48, 911–
931.
Streltsov, Artem, and Qihong Ruan, 2022, Perpetual price discovery and crypto market
quality, SSRN Electronic Journal .
34
Appendix
A. Additional figures and tables
Figure 5 Trading view of BTC perpetual futures on Binance
This figure presents the perpetual futures trading view on Binance. The key information includes
the futures price (Mark), spot price (Index), real-time funding rate based on the rolling average
of the past 8 hours’ observations, and countdown toward funding rate payment (Funding /
Countdown) (illustrated with a red rectangle). In this example, futures are trading at a lower
price than the spot. The funding rate is negative and is to be paid in 1 hour and 12 minutes.
35
Figure 6 Trading costs tiers for the spot market
This figure presents the trading costs tiers for the crypto spot market from Binance: https://www.
binance.com/en/fee/trading. Our high, medium, and low trading cost specification corresponds
to VIP 1, VIP 5, and VIP 8 tiers as illustrated with red rectangles in the picture. The 30-day trading
volume requirements for VIP 1, 5, and 8 are 1 million, 150 million, and 2 billion respectively in the
spot market. We consider the trading costs for makers as institutions typically trade maker orders.
Binance offers a temporary discount for VIP 4-8 to have the same trading cost as VIP 9. We consider
the non-discounted trading costs to make the comparison more reliable and fair.
B. Data-driven Arbitrage Strategy
In the main text of our paper, we demonstrate the profitability of a simple theory-
motivated trading strategy. The trading threshold can also be potentially further improved
36
Figure 7 Trading costs tiers for the perp market
This figure presents the trading costs tiers for the crypto perpetual market from Binance: https:
//www.binance.com/en/fee/trading. Our high, medium, and low trading cost specification
corresponds to VIP 1, VIP 5, and VIP 8 tiers as illustrated with red rectangles in the picture. The
30-day trading volume requirements for VIP 1, 5, and 8 are 15 million, 1 billion, and 12.5 billion
respectively in the perpetual market. We consider the trading costs for makers as institutions
typically trade maker orders.
using a data-driven approach. In this part, we implement a data-driven two-threshold ar-

bitrage trading strategy in the perpetual futures market. The strategy can be characterized
with a tuple of two thresholds: (u, l), where u denotes the upper bar and l denotes the
lower bar, u > l. When ρ > u, we long the spot and short the futures. When −l < ρ < l,
we close the position. When ρ < −u, we long the futures and short the spot. Figure 15
37
Figure 8 Ether annualized funding rate across exchanges
The figure presents the 7-day moving average annualized funding rate for Ether (ETH) across
exchanges. The data is obtained from glassnode, an analytics platform. The data covers two major
market turbulences: the COVID stock market crash from February to April 2020, and the FTX
insolvency in November 2022. The FTX collapse led to significant negative funding rates on all
solvent exchanges, thus the perpetual future prices were lower than the spot prices on the solvent
exchanges. Vice versa on the insolvent FTX. The funding rates become more volatile after the event,
indicating increased uncertainty for the market participants.
presents an illustration of the strategy for Bitcoin with the trading thresholds estimated
using real-time data.
To determine the optimal (u, l), at the beginning of each month, we calculate the
returns of the two-threshold trading strategy based on the past 6 months’ data on a grid
of parameters. From Table 3, we find the theory implied bounds for price deviation across
all trading costs specifications lie within −200% to +200%. So we choose the grid as
increasing from 0% to 200% with an incremental step of 10% for u and l (u ≥ l). In total,
there are 210 model specifications. We choose the model that delivers the highest Sharpe
Ratio in the validation sample of the past 6 months.
Figure 15 presents the real-time trading thresholds of our arbitrage strategy for BTC
under high trading costs. To mitigate the trading cost, the strategy automatically chooses
a much lower close-position threshold compared to the open-position threshold. We also
present the visualization of all trading strategies under different trading costs in appendix
Figure 11 to 14. Under no-trading costs, the strategy chooses a low open-position threshold
and the open- and close-position thresholds coincide with each other. Since there is no
trading cost, the strategy no longer needs to wait for price convergence to avoid high
38
40%
supply rate
borrow rate
35%
30%
Annualized interest rate, percent
25%
20%
15%
10%
5%
0%
2020-01 2020-05 2020-09 2021-01 2021-05 2021-09 2022-01 2022-05 2022-09
Figure 9 Annualized interest rate from Aave
This figure presents the daily supply and borrowing rate from Aave. We consider three stablecoins:
USDT, USDC, and DAI. Each day, we take the average interest rate of the 3 crypto stablecoins to get
the proxy for the risk-free funding rate of the arbitrageurs. The averaging removes the idiosyncratic
noise in borrowing and supply in individual stablecoins. The sample period is from 2020-01-08 to
2022-11-13.
turnover. On the contrary, with high trading costs, the strategy chooses a larger open-
position threshold and a smaller close-position threshold. This reflects the automatic
adjustment from the algorithm for trading costs and turnovers.
Table 10 presents the return statistics for this strategy under high trading costs over
time. In the baseline ‘unrestricted’ strategy, we allow for both (1) longing the futures and
shorting the spot and (2) shorting the futures and longing the spot while in the ‘long-spot
only’ strategy, we only allow for shorting the futures and longing the spot. The reason
we consider the ‘long-spot only’ strategy is the infrastructure for shorting the spot is not
well-developed. So we want to examine the performance of the trading strategy in the
presence of such limits to arbitrage.
We find the arbitrage trading strategy has a high Sharpe ratio under high trading costs.
The annualized Sharpe ratio for Bitcoin is 1.78 in our sample. They are even higher for
other cryptos. The high Sharpe ratio of the trading strategy corroborates our theoretical
results that when the price deviation is large enough, the trading would be a random-
maturity arbitrage opportunity.
39
Unrestricted Long-spot Only
2020 2021 2022 All 2020 2021 2022 All
BTC N 4,416 8,760 7,536 20,712 7,344 8,760 7,609 23,713
Active % 26.11 18.06 0.29 13.31 30.27 52.45 0.03 28.76
Return 5.78 8.51 0.13 4.88 6.24 15.21 0.10 7.58
Volatility 3.74 3.26 0.19 2.74 3.47 6.50 0.18 4.40
SR 1.55 2.61 0.70 1.78 1.80 2.34 0.54 1.72
ETH N 4,416 8,760 7,536 20,712 5,880 8,760 7,609 22,249
Active % 22.98 23.90 12.08 19.40 44.12 54.35 0.09 33.09
Return 9.23 13.25 0.60 7.79 11.56 18.07 0.61 10.38
Volatility 2.55 3.73 0.93 2.76 4.29 6.13 0.71 4.45
SR 3.62 3.55 0.64 2.83 2.70 2.95 0.87 2.33
BNB N 3,672 8,760 7,536 19,968 3,672 8,760 7,609 20,041
Active % 41.04 32.32 10.51 25.69 57.30 48.11 1.10 31.94
Return 24.61 25.33 3.33 16.89 12.23 20.07 0.10 11.05
Volatility 4.24 4.23 1.32 3.44 3.77 4.80 0.30 3.57
SR 5.80 5.99 2.52 4.91 3.24 4.18 0.33 3.10
DOGE N 8,760 7,536 16,296 8,760 7,609 16,369
Active % 41.66 6.62 25.45 54.26 1.22 29.60
Return 47.57 0.98 26.02 32.74 0.08 17.56
Volatility 7.93 0.87 5.85 7.39 0.40 5.42
SR 6.00 1.14 4.45 4.43 0.21 3.24
ADA N 4,416 8,760 7,536 20,712 4,416 8,760 7,609 20,785
Active % 34.24 26.58 7.43 21.24 62.09 55.89 0.00 37.10
Return 14.15 19.45 0.61 11.46 15.15 18.77 0.00 11.08
Volatility 3.62 5.13 0.89 3.77 5.07 7.20 0.00 5.23
SR 3.91 3.79 0.68 3.04 2.99 2.61 0.00 2.12
Table 10 Portfolio Performance: Data-driven strategy, High Trading Cost
This table presents the annual return, standard deviations, Sharpe ratios of the two-threshold
trading strategies for the 5 different cryptocurrencies with high trading costs. We also report the
proportion of time the trading strategy has an open position. We break down returns into each
year and also provide summary stat across all time. The left panel shows the performance of the
unrestricted trading strategy where both long and short spot is allowed. The right panel shows the
performance of the long-spot-only trading strategy where shorting the spot is not allowed.
In the year 2022, the deviation between the futures and the spot becomes smaller and
less volatile. There seems to be a structural break. We indeed find the trading strategy
takes less active positions and significantly lower annualized returns. However, when
there is a large enough deviation, the strategy can still generate sizeable Sharpe ratios in
trading.
40
The conclusion and results remain similar if we consider ‘long-spot only’ trading
strategies where only shorting the futures and longing the spot is allowed. Considering
this one-sided trade slightly lowers the Sharpe ratio but the algorithm automatically
adjusts by increasing the proportion of times being active. The resulting annualized
returns increase. In the year 2022, since most of the time, the futures price is below the
spot and we don’t allow longing the futures and shorting the spot trade, the proportion of
time the strategy is active is very low.
Trading costs
No Low Medium High
BTC Active % 23.20 20.79 16.29 13.31
Return 15.98 7.42 6.02 4.88
Volatility 2.38 2.63 2.61 2.74
SR 6.72 2.82 2.31 1.78
ETH Active % 32.34 25.85 20.84 19.40
Return 23.48 11.80 9.41 7.79
Volatility 2.27 2.27 2.69 2.76
SR 10.32 5.20 3.50 2.83
BNB Active % 37.76 30.90 25.52 25.69
Return 54.50 30.40 22.07 16.89
Volatility 3.26 3.02 3.12 3.44
SR 16.72 10.07 7.07 4.91
DOGE Active % 36.62 39.81 32.46 25.45
Return 72.95 46.87 34.45 26.02
Volatility 5.65 5.60 5.81 5.85
SR 12.90 8.37 5.92 4.45
ADA Active % 40.00 43.22 24.38 21.24
Return 53.06 29.63 16.53 11.46
Volatility 3.22 3.33 3.26 3.77
SR 16.47 8.90 5.07 3.04
Table 11 Data-driven strategy: Portfolio Performance under Different Trading Costs
This table presents the portfolio performance under different trading cost specifications. The fee
for spot is 0, 1.5 bps, 3.75 bps, and 5.25 bps for the 4 trading costs specifications. The fee for the
futures are 0, 0, 0.54 bp, 1.08 bps for the 4 trading costs specifications.
Table 11 further reports the performance of the trading strategy under different trading
costs specifications laid out in Table 2. The results from Table 10 correspond to the last
column in Table 11. As trading costs increase, our trading strategy dynamically adjusts
by lowering the proportion of time being active. The annualized return decreases and
41
annualized standard deviation are of similar magnitude across different trading costs. As
a result, SR decreases as trading costs increase. Under no trading cost, we see a Sharpe
ratio of 6.72 for BTC and Sharpe ratios above 10 for all other cryptos. This also confirms
our theoretical results that when there is no trading cost, any deviation of perpetual price
from the no-arbitrage benchmark would be a random-maturity arbitrage opportunity.
Comparing Table 11 with Table 6, we find under low and no trading costs, the potential
incremental benefits from using a data-driven two-threshold trading rule are much larger
compared to higher trading costs. The intuition can be easily illustrated by comparing
the empirical real-time trading thresholds in the theory-motivated strategy and the data-
driven strategy under low trading costs (Figure 10 and Figure 11). With persistent ρ the
data-driven approach can find the optimal close-position threshold corresponding to the
level of ρ while the theory-motivated one is too conservative in setting a close-position
threshold equal to 0.
There are two sources of trading profit for our trading strategy: (1) the price conver-
gence; (2) the funding rate payment. In Table 12, we decompose the return to our trading
strategy without trading cost into the two sources. We find across different cryptos, the
return from price convergence accounts for a larger proportion of the total profit. For BTC,
the price convergence accounts for more than 2/3 of the profits, and for ETH, it accounts for
about 3/4. Additionally, the decomposition for the year 2022 generates different patterns,
the return due to funding rate plays a much smaller role across different cryptos. This
is because the deviation from perpetual to spot is much smaller during the year. The
results suggest when more sophisticated arbitrageurs enter the market, the large deviation
of perpetual from the spot would be more of an off-equilibrium outcome. As a result, the
funding rate payment will also be small.
In all, our analysis of the trading strategy demonstrates the profitability of perpetual-
spot arbitrage trade. The deviation of the futures to spot can also serve as an important
measure for the frictions, trading costs, and limits to arbitrage in the market.
42
2020 2021 2022 All
BTC Return 17.70 25.69 3.69 15.98
Price 11.77 16.10 3.58 10.62
Funding 5.93 9.59 0.11 5.36
ETH Return 22.51 30.99 15.31 23.48
Price 14.22 21.86 14.36 17.50
Funding 8.29 9.13 0.95 5.98
BNB Return 74.35 64.93 32.71 54.50
Price 61.16 56.63 28.70 46.92
Funding 13.19 8.29 4.01 7.58
DOGE Return 96.17 45.95 72.95
Price 82.50 45.58 65.43
Funding 13.68 0.36 7.52
ADA Return 44.24 56.21 54.57 53.06
Price 31.81 45.69 53.12 45.44
Funding 12.43 10.52 1.44 7.62
Table 12 Data-driven strategy: Price Convergence vs Funding Rate Payment
This table decomposes the portfolio return into the part due to price convergence and the part due
to funding rate payment.
43
Figure 10 Trading Strategy Visualization: Random-maturity Arbitrage
This figure presents the random-maturity arbitrage strategy motivated by the theory. The orange,
green and purple lines correspond to the open-position threshold under high, medium, and low
trading costs. The red line represents the close-position threshold.
44
Figure 11 Trading Strategy Visualization: No Trading Costs
This figure presents the real-time trading thresholds under no trading costs. The orange line is the
open-position threshold and the red line is the close-position threshold. The trading thresholds
are determined based on the adjusted SR from the past 6 months.
45
Figure 12 Trading Strategy Visualization: Low Trading Costs
This figure presents the real-time trading thresholds under low trading costs (2.25 bps for spot
trading and 0.18 bp for futures trading). The orange line is the open-position threshold and the red
line is the close-position threshold. The trading thresholds are determined based on the adjusted
SR from the past 6 months.
46
Figure 13 Trading Strategy Visualization: Medium Trading Costs
This figure presents the real-time trading thresholds under medium trading costs (4.5 bps for spot
trading and 0.72 bp for futures trading). The orange line is the open-position threshold and the red
47
Figure 14 Trading Strategy Visualization: High Trading Costs
This figure presents the real-time trading thresholds under high trading costs (6.75 bps for spot
trading and 1.44 bps for futures trading). The orange line is the open-position threshold and the red
48
BTC
400%
open
close
300%
200%
100%
0%
-100%
-200%
-300%
2020-07 2020-10 2021-01 2021-04 2021-07 2021-10 2022-01 2022-04 2022-07 2022-10
Figure 15 Data-driven strategy: Bitcoin, high trading costs
This figure presents real-time trading thresholds of the arbitrage strategy we implement for BTC
under high trading costs. Each blue dot in the figure represents annualized deviation of futures
from the spot ρ = κ( f − s) − r. The orange line is the open-position threshold and the red line is
the close-position threshold.
49

Fundamentals of Perpetual Futures

Uploaded by

Copyright:

Available Formats

Fundamentals of Perpetual Futures

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fundamentals of Perpetual Futures

Uploaded by

Copyright:

Available Formats

Fundamentals of Perpetual Futures∗

Songrun He Asaf Manela Omri Ross Victor von Wachter

First draft December 2022. This draft July 2023.

JEL Classification: G11, G12, G13

Our derivation relies on a weaker notion of arbitrage that we call random-maturity

These deviations narrows considerably in 2022, suggesting a decrease in arbitrage

The existing literature on perpetual futures mainly focuses on descriptive evidence.

Figure 1 Total trading volumes of perpetual futures across exchanges

the funding rate is calculated.

3 Arbitrage in Perpetual Futures

Definition 1 (Perpetual futures). A perpetual futures contract {Ft }t=0

This definition is an approximation of real-world perpetual futures. In most exchanges,

To equalize the two, we have κ = 1095.

The traditional notions of arbitrage typically consider a guaranteed positive payoff at

Definition 2 (Riskless arbitrage). A riskless arbitrage opportunity is defined with respect to a

Definition 3 (Random-maturity arbitrage). A random-maturity arbitrage opportunity is

Definition 3 generalizes traditional arbitrage in the sense that there is a guaranteed

Assumption 2. The risk-free rate r for arbitrageurs is constant.

Proposition 1 (No-arbitrage bound). Under Assumptions 1 and 2, when there is a constant

Actions Long spot, short futures Long futures, short spot

When F0 > S0 1 + κr +C, limt→∞ ut = ∞. This contradicts Assumption 1, because it implies:

spot and shorting the futures is a random-maturity arbitrage opportunity.

When F0 < S0 1 + κr − C, limt→∞ ūt = −∞.

the futures and shorting the spot is a random-maturity arbitrage opportunity.

Proof. The proof follows that of Proposition 1 by setting C = 0.

4 Data and Empirical Analysis

We conduct an empirical analysis of perpetual futures arbitrage strategies. We first

Crypto Start date End date N

Table 2 Sample descriptions

makers make the market or provide liquidity to the market.

Table 3 Trading costs tiers

4.2 Deviations of perpetual futures from no-arbitrage benchmarks

Figure 3 Deviations of perpetual futures from no-arbitrage benchmarks

Table 4 Summary Statistics for Deviations from No-arbitrage Prices

One hypothesis for this significant comovement in ρ across various cryptocurrencies

f rDOGE 0.67 0.05 0.05 0.11 -0.02 0.05

We next provide a trading strategy motivated by our random-maturity arbitrage theory.

Figure 4 Random-maturity Arbitrage Strategy: Bitcoin, high trading costs

Futures-spot deviations and trading thresholds of the random-maturity arbitrage strategy we

Table 6 Performance of Random-maturity Arbitrage Strategy

Table 7 Performance Over Time: High Trading Costs Tier

2020 2021 2022 All

Table 8 Return Decomposition: Price Convergence vs Funding Rate Payment for

5 Explaining Futures-spot Deviations

From Figure 3, we observe a strong common comovement of the futures-spot deviation

Table 9 Regression of the futures-spot gap against explanatory variables

Liu and Tsyvinski (2021) document a significant time-series momentum pattern in

Finally, we provide an explanation for the common comovement in futures-spot

A. Additional figures and tables

Figure 5 Trading view of BTC perpetual futures on Binance

B. Data-driven Arbitrage Strategy

using a data-driven approach. In this part, we implement a data-driven two-threshold ar-

Figure 9 Annualized interest rate from Aave

Table 10 Portfolio Performance: Data-driven strategy, High Trading Cost

Table 11 Data-driven strategy: Portfolio Performance under Different Trading Costs

Table 12 Data-driven strategy: Price Convergence vs Funding Rate Payment

Figure 15 Data-driven strategy: Bitcoin, high trading costs

You might also like