SIAM J. FINANCIAL MATH.

© 2023 Society for Industrial and Applied Mathematics

Vol. 14, No. 1, pp. SC17--SC30

Short Communication: A Primer on Perpetuals*

Guillermo Angeris\dagger , Tarun Chitra\ddagger , Alex Evans\dagger , and Matthew Lorig\S
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

Abstract. We consider a continuous-time financial market with no arbitrage and no transactions costs. In this
setting, we introduce two types of perpetual contracts, one in which the payoff to the long side is a
fixed function of the underlyers and the long side pays a funding rate to the short side, the other in
which the payoff to the long side is a fixed function of the underlyers times a discount factor that
changes over time but no funding payments are required. Assuming asset prices are continuous and
strictly positive, we derive model-free expressions for the funding rate and discount rate of these
perpetual contracts as well as replication strategies for the short side. When asset prices can jump,
we derive expressions for the funding and discount rates, which are semirobust in the sense that
they do not depend on the dynamics of the volatility process of the underlying risky assets, but
do depend on the intensity of jumps under the market's pricing measure. When asset prices can
jump and the volatility process is independent of the underlying risky assets, we derive an explicit
replication strategy for the short side of a perpetual contract. Throughout the paper, we illustrate
through examples how specific perpetual contracts relate to traditional financial instruments such
as variance swaps and leveraged exchange traded funds.

Key words. cryptocurrency, decentralized finance, perpetual contract

MSC code. 91-02

DOI. 10.1137/22M1520931

1. Introduction. A perpetual contract (often just perp, for short) is a type of financial
contract that enables relatively general payoffs. At a high level, a perp contract can be
described as follows: two parties, which we will call the long side and the short side, enter
into an agreement. The short side agrees to pay the long side some payoff, which is a function
of the prices of the underlying assets, at a time of the long side's choosing. In exchange for
this, the long side pays a continual cash-flow to the short side up until contract termination.
This cash-flow can be implemented in two distinct ways. First, it can be implemented directly
as a literal cash-flow, where the long side pays the short side cash at fixed time increments.
Second, a sometimes more practical approach is instead to replace the cash-flow by discounting
the payoff at contract termination.
Perps were first suggested in [10] as a way of approximately measuring the prices of
dividend-yielding assets and also as a tool to hedge certain illiquid assets. But, only the
special case where the payoff function was linear in the price of the underlying assets was

*
Received by the editors September 8, 2022; accepted for publication (in revised form) December 21, 2022;
published electronically March 30, 2023.
https://doi.org/10.1137/22M1520931
\dagger
Bain Capital Crypto, San Francisco, CA 94105 USA ([email protected], [email protected]).
\ddagger
Gauntlet Networks, Brooklyn, NY 11201 USA ([email protected]).
\S
Department of Applied Mathematics, University of Washington, Seattle, WA 98195 USA ([email protected]).
SC17

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 SC18 G. ANGERIS, T. CHITRA, A. EVANS, AND M. LORIG

considered. Perps with linear payoffs later gained widespread popularity as a way of taking
leveraged bets on cryptocurrency markets, where common derivatives markets were initially
relatively illiquid, if available at all. As of 2022, perps are some of the most actively traded
cryptocurrency derivatives, with daily volume in the tens of billions of dollars (see, e.g.,
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

[4]).
While perpetual futures never gained traction outside of cryptocurrencies, they were in-
troduced as a convenient way for miners (who produce newly minted tokens or coins) to hedge
inherent risks in cryptocurrency production. The two main risks that miners of currencies
such as Bitcoin and Ethereum face are that their future income (which can be viewed as a
dividend-yielding stream) is randomized with variance depending on their resource contri-
bution to the network [7]. To reduce this variance, miners used two tactics: mining pools
(e.g., pooling together resources and distributing dividends pro-rata) and futures contracts.
Initially, miner perpetual futures contracts were over-the-counter quanto futures, where min-
ers took premiums in cash with strike prices struck denominated in Bitcoin. The failure of
the early quanto derivatives market led to the creation of stablecoins (dollar-pegged demand
deposit assets) which then naturally led to the creation of the crypto perpetuals market in
2016 [1].
Recently, perps with payoffs that are proportional to some power of the price of an asset
have been proposed (cf. [8, 11]). This extension led to the creation of a decentralized perps
protocol on the Ethereum blockchain called Squeeth (short for ``Squared ETH""); see [9]. This
protocol allows users to take long (and short) positions on these power perps without requiring
an intermediary such as a broker or exchange.
The role of this paper is to clearly define perps, show a number of natural generalizations
to those known in the literature, and correct some of the misinformation that exists online as
to how the rate of cash-flow should be computed in a no-arbitrage setting. The rest of the
paper proceeds as follows: in section 2 we introduce a financial market in which risky assets
have continuous price paths. Next, in section 3, we define a perpetual contract in which the
long side must pay a funding rate to the short side. We derive a model-free expression for
the funding rate as well as a replication strategy for the short side. In section 4 we define a
second type of perpetual contract in which no funding payments are required but the payoff
is discounted over time. We derive a model-free expression for the discount rate as well as a
replication strategy for the short side. Last, in section 5, we consider a market with a single
risky asset whose value may jump. In this setting, we derive expressions for the funding and
discount rates of the two types of perpetual contracts introduced in sections 3 and 4. And,
under the assumption of an independent volatility process, we derive a replication strategy
for the short side of a perpetual contract.

2. Market model and assumptions. In sections 2, 3, and 4, we consider a continuous-

time financial market, defined on a filtered probability space (\Omega , F, \BbbF , \BbbP ), with no arbitrage
and no transaction costs. The filtration \BbbF = (Ft )t\geq 0 represents the history of the market
and \BbbP represents the real-world probability measure. We suppose that the market contains a
risk-free money market account, whose value in dollars is denoted by M = (Mt )t\geq 0 , as well as
n \in \BbbN risky assets (typically, tokens or cryptocurrencies), whose values in dollars are denoted
(1) (2) (n)
by S = (St , St , . . . , St )t\geq 0 .

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 A PRIMER ON PERPETUALS SC19

We assume the value of the money market account M is continuous, strictly positive, and
nondecreasing. As such, there exists a nonnegative \BbbF -measurable process r = (rt )t\geq 0 , known
as the risk-free rate, such that
dMt = rt Mt dt, M0 \geq 0.
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

We further assume that the prices of the risky assets are continuous and strictly positive.
(1) (2) (n)
As such, there exists an \BbbR n -valued \BbbF -measurable drift vector \mu = (\mu t , \mu t , . . . , \mu t ) and an
(1,1) (1,2) (n,d)
\BbbR n\times d
+ -valued \BbbF -measurable volatility matrix \sigma = (\sigma t , \sigma t , . . . , \sigma t ) with d \in \BbbN , such that,
for every i, the value S (i) of the ith risky asset is given by
d
(i) (i) (i) (i,j) (i) (j) (i)
\sum
dSt = \mu t St dt + \sigma t St dWt , S0 \geq 0,
j=1
(1) (2) (d)
where W = (Wt , Wt , . . . , Wt ) is a d-dimensional (\BbbF , \BbbP )-Brownian motion with indepen-
dent components. Last, throughout this paper, in order to avoid unnecessary complications,
we assume all local martingales are true martingales.
3. Perpetual contracts with funding. We will discuss two types of perpetual contracts in
this paper: (i) perpetual contracts with funding and (ii) perpetual contrast with discounting.
In this section, we focus on the former. We begin with a definition.
Definition 1. A perpetual contract with funding (or simply, a perp) written on S with payoff
function \varphi : \BbbR n \rightarrow \BbbR is an agreement between two parties, referred to as the long side and short
side. The long side has the right to terminate the contract at any time t \geq 0, at which point
it will receive a payment of \varphi (St ). In return, the long side must pay to the short side \varphi (S0 )
at the time t = 0 of inception as well as a continuous \BbbF -adapted cash-flow of F = (Ft )t\geq 0 per
unit time, referred to as the funding rate, up until the contract is terminated.
Remark 2. The long (short) side of a futures contract written on \varphi (S) with delivery
date T receives (pays) changes in the futures price F T = (FtT )0\leq t\leq T , which is defined as
FtT := \BbbE \widetilde t denotes the Ft -conditional expectation under the market's pricing
\widetilde t \varphi (ST ), where \BbbE
measure \BbbP \widetilde . As there is no cost to enter or exit a futures contract, if the long side enters at
time zero and exits at time T , the total from the cash-flows he receives is
\int T
dFtT = FTT - F0T = \varphi (ST ) - \BbbE
\widetilde \varphi (ST ).
0
By contrast, if the long side enters a perpetual at time zero and exits at time T , the total
from the cash-flows he receives is
\int T
\varphi (ST ) - \varphi (S0 ) - Ft dt.
0

Note that computation of the futures price FT

requires knowledge of the dynamics of S under
\widetilde . As we shall see below, the \BbbP
\BbbP \widetilde -dynamics of S are not needed to compute the funding rate F
of a perpetual contract.
The following theorem gives an expression for the funding rate.

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 SC20 G. ANGERIS, T. CHITRA, A. EVANS, AND M. LORIG

Theorem 3. Consider a perpetual contract as described in Definition 1. Suppose that the

function \varphi \in C 2 (\BbbR d , \BbbR ). Under the assumptions of section 2, the funding rate F is given by
n n n
1 \sum \sum (i) (j)
\Bigl( \sum (i) \Bigr)
(1) Ft = (\sigma t \sigma t\top )(i,j) St St \partial i \partial j \varphi (St ) - \varphi (St ) - St \partial i \varphi (St ) rt ,
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

2
i=1 j=1 i=1

where (\sigma t \sigma t\top )(i,j) denotes the of (i, j)th component of \sigma t \sigma t\top and \partial j := \partial
\partial sj .
Proof. We will show that, with Ft given by (1), the short side can create a self-financing
portfolio whose value X = (Xt )t\geq 0 satisfies

(2) Xt = \varphi (St )

for all t \geq 0. To begin, we note that the value of the short side's portfolio must be of the form

n \Bigl( n \Bigr) 1
(i) (i) (i)
\sum \sum
(3) dXt = \Delta t dSti + Xt - \Delta t St dMt + Ft dt
Mt
i=1 i=1
n \Bigl( n \Bigr)
(i) (i) (i)
\sum \sum
= \Delta t dSti + Xt - \Delta t St rt dt + Ft dt,
i=1 i=1

(i)
where \Delta t denotes the number of shares of invested in asset i at time t. Next, we have by
It\^
o's lemma that
n n n
\sum (i) 1 \sum \sum
(4) d\varphi (St ) = \partial i \varphi (St )dSt + \partial i \partial j \varphi (St )d\langle S (i) , S (i) \rangle t
2
i=1 i=1 j=1
n n n
\sum (i) 1 \sum \sum (i) (j)
= \partial i \varphi (St )dSt + (\sigma t \sigma t\top )(i,j) St St \partial i \partial j \varphi (St )dt.
2
i=1 i=1 j=1

Now, note that (2) holds for all t \geq 0 if and only if X0 = \varphi (S0 ) and dXt = d\varphi (St ). Comparing
(i)
(3) with (4), we see that the dSt terms will be equal if we set
(i)
\Delta t = \partial i \varphi (St ).
(i)
Next, using Xt = \varphi (St ) and \Delta t = \partial i \varphi (St ), we see that the dt terms in (3) and (4) will be
equal if Ft is given by (1).
Remark 4. Note that the funding rate can be positive or negative. If at time t the funding
rate is negative, then short side pays the long side at a rate of - Ft .
Remark 5. Observe that precise knowledge of r, \mu , and \sigma is not needed to determine the
funding rate Ft . Indeed, using

rt dt = d log Mt , (\sigma t \sigma t\top )(i,j) dt = d\langle log S (i) , log S (j) \rangle t ,

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 A PRIMER ON PERPETUALS SC21

we can express F in the following model-free form (within the class of models described in
section 2):
n n
1 \sum \sum (i) (j)
(5) Ft dt = S S \partial i \partial j \varphi (St )d\langle log S (i) , log S (j) \rangle t
2
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

i=1 j=1
\Bigl( n \Bigr)
(i)
\sum
- \varphi (St ) - St \partial i \varphi (St ) d log Mt .
i=1
By contrast, in order to price and replicate most traditional financial derivatives such as
European, American, Bermudan, and barrier options, one requires a parametric model for the
underlying S as well as knowledge of unobservable model parameters.
Example 6. A (continuously monitored) variance swap (VS), written on an asset S \equiv S (1) ,
is an agreement between two parties, referred to as the long and short sides. At the maturity
date T , the short side pays the long side
\int T
d\langle log S\rangle t - K,
0
where the swap rate K is determined at inception t = 0 so that the initial cost to enter the swap
is zero. Under the assumptions of section 2, the swap rate K is given by - 2\BbbE \widetilde log(ST /S0 ), where
\widetilde denotes expectation under the market's chosen pricing measure \BbbP
\BbbE \widetilde , which can be deduced by
observing implied volatilities of T -maturity European calls and puts (see, e.g., [3]). Because
implied volatilities tend to be higher than realized volatility (this is sometimes known as the
volatility premium), taking the long side of a VS is typically a losing trade. As an alternative
to entering the long side of a VS, an investor wishing to gain exposure to volatility could
take a long position in a perp as described in Definition 1 with payoff \varphi (St ) = 2 log(St /S0 ).
Like a VS, there is no cost to entering this perp because \varphi (S0 ) = 2 log(S0 /S0 ) = 0. Moreover,
assuming r \equiv 0 for simplicity, we have from (5) that the funding rate is
Ft dt = - d\langle log S\rangle t .
Therefore, if the long side chooses to terminate the contract at time T , the value of the payoff
minus funding paid is
\int T \Bigl( S \Bigr) \int T
T
\varphi (ST ) - Ft dt = 2 log + d\langle log S\rangle t .
0 S 0 0
Thus, by taking a long position in a perp, the investor can achieve the same exposure to
volatility that they would have had they taken the long side of a VS, without paying a
volatility premium.
Remark 7. It is important to recognize how a perp contract, as defined in Definition 1,
differs from the perp contracts that trade on centralized exchanges (e.g., Binance, BitMEX,
and Bybit). Such exchanges introduce a perpetual price (also called a mark price) P = (Pt )t\geq 0 ,
which is supposed to track the price of an underlyer \varphi (S) = (\varphi (St ))t\geq 0 . Tracking is achieved
through the funding rate F , which is typically given by
(6) Ft := \gamma t + Rc (\delta t , \gamma t ), \gamma t := Pt - \varphi (St ), Rc (\delta , \gamma ) := max( - c, min(c, \delta - \gamma )),

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 SC22 G. ANGERIS, T. CHITRA, A. EVANS, AND M. LORIG

where \gamma = (\gamma t )t\geq 0 is the premium and \delta = (\delta t )t\geq 0 is the interest rate delta, which is a function
of the interest rates for borrowing components of S and the interest rate r of the money
market account. The premium \gamma has a mean-reverting affect, keeping the price of P near
\varphi (St ). The payoff of a long side that enters a perp at time zero and exits at time t is given
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

by
\int t
Pt - P0 - Fs ds.
0

Were it the case that P = \varphi (S), there would be a clear arbitrage if the funding rate F given
in (6) differed from the expression for F given in (5). However, in practice, P and \varphi (S) do
not perfectly coincide. Moreover, funding payments are typically made discretely in eight-
hour intervals. Thus, the fact that (5) and (6) differ may not necessarily lead to a real-world
arbitrage opportunity.
4. Perpetual contracts with discounting. One of the problems with a perp with funding
is that execution of the contract requires the long side to place a deposit (e.g., on an exchange
or into a smart contract) at inception in order to pay the funding rate. If the time-integral of
the funding rate ever exceeds the deposit, the contract is automatically terminated. One way
to avoid automatic termination of the contract is to consider, instead, a perpetual contract
with discounting, whose mechanics are described in the following definition.
Definition 8. A perpetual contract with discounting (or simply, a perp) written on S with
payoff function \varphi : \BbbR n \rightarrow \BbbR is an agreement between two parties, referred to as the long side
and short side. The long side has the right\int to terminate the contract at any time t \geq 0, at
t
which point it will receive a payment of e - 0 Ds \mathrm{d}s \varphi (St ), where D = (Dt )t\geq 0 is an \BbbF -adapted
process known as the discount rate. In return, at the time of inception t = 0, the long side
must pay to the short side a premium \varphi (S0 ).
The following theorem gives an expression for the discount rate.
Theorem 9. Consider a perpetual contract with discounting as described in Definition 8.
Suppose that the function \varphi \in C 2 (\BbbR d , \BbbR ) and is either strictly positive or strictly negative.
Then, under the assumptions of section 2, the discount rate D is given by

Ft
(7) Dt = ,
\varphi (St )

where Ft is given by (1).

Proof. We will show that, with Dt given by (7), the short side can create a self-financing
portfolio whose value X = (Xt )t\geq 0 satisfies
\int t
(8) Xt = exp( - Ds ds)\varphi (St )
0

for all t \geq 0. To begin, we note that the dynamics of the short side's portfolio X must be of
the form

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 A PRIMER ON PERPETUALS SC23

n \Bigl( n \Bigr) 1
(i) (i) (i)
\sum \sum
(9) dXt = \Delta t dSti + Xt - \Delta t St dMt
Mt
i=1 i=1
n \Bigl( n \Bigr)
(i) (i) (i)
\sum \sum
= \Delta t dSti + Xt - \Delta t St rt dt,
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

i=1 i=1
(i)
where \Delta t denotes the number of shares of invested in asset i at time t. Next, we have by
It\^
o's lemma that
n n n
\Bigl( \int t
- 0 Ds \mathrm{d}s
\Bigr)
- 0t Ds \mathrm{d}s
\int \Bigl( \sum
(i) 1 \sum \sum \Bigr)
(10) d e \varphi (St ) = e \partial i \varphi (St )dSt + \partial i \partial j \varphi (St )d\langle S (i) , S (i) \rangle t
2
i=1 i=1 j=1
\int t
- Ds \mathrm{d}s
- Dt e 0 \varphi (St )dt
n n n
\int t \Bigl( \sum (i) 1 \sum \sum (i) (j)
\Bigr)
= e - 0 Ds \mathrm{d}s \partial i \varphi (St )dSt + (\sigma t \sigma t\top )(i,j) St St \partial i \partial j \varphi (St )dt
2
i=1 i=1 j=1
\int t
- Ds \mathrm{d}s
- Dt e 0 \varphi (St )dt.
\int t
Now, note that (8) will hold for all t \geq 0 if and only if X0 = \varphi (S0 ) and dXt = d(e - 0
Ds \mathrm{d}s \varphi (S
t )).
(i)
Comparing (9) with (10), we see that the dSt terms will be equal if we set
\int t
(i)
\Delta t = e - 0
Ds \mathrm{d}s
\partial i \varphi (St ).

\int t (i) \int t

Next, using Xt = exp( - 0 Ds ds)\varphi (St ) and \Delta t = exp( - 0 Ds ds)\partial i \varphi (St ), we see that the dt
terms in (9) and (10) will be equal if Dt is given by (7).
Example 10. Let L = (Lt )t\geq 0 be the value of a leveraged exchange traded fund (LETF)
with underlyer S \equiv S (1) and leverage ratio \gamma . The manager of such an LETF seeks to multiply
the returns of S by a factor of \gamma by holding \gamma Lt /St shares of the underlyer for all t \geq 0 and
borrowing from the bank to finance the position (see, e.g., [6]). Thus, the dynamics of L are
as follows:

1
dLt = \Delta t dSt + (Xt - \Delta t St ) dMt , \Delta t = \gamma Lt /St .
Mt

Solving for Lt one obtains the following expression:

\Bigl( S \Bigr) \gamma \Bigl( M \Bigr) (1 - \gamma ) \Bigl( \gamma (1 - \gamma ) \Bigr)
t t
Lt = L0 exp \langle log S\rangle t .
S0 M0 2

Now, consider a perp as described in Definition 8 with payoff \varphi (St ) = L0 (St /S0 )\gamma . We have
from (5) and (7) that

\gamma (\gamma - 1)
Dt dt = d\langle log S\rangle t - (1 - \gamma )d log Mt .
2

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 SC24 G. ANGERIS, T. CHITRA, A. EVANS, AND M. LORIG

If the long side terminates the perp at time t it will receive

\gamma (\gamma - 1) t
\int t \Bigl( S \Bigr) \gamma \Bigl( \int \int t \Bigr)
- Ds \mathrm{d}s t
\varphi (St )e 0 = L0 exp - d\langle log S\rangle s + (1 - \gamma ) d log Ms
S0 2 0 0
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

\Bigl( S \Bigr) \gamma \Bigl( M \Bigr) (1 - \gamma ) \Bigl( \gamma (1 - \gamma ) \Bigr)
t t
= L0 exp \langle log S\rangle t .
S0 M0 2
Thus, an LETF written on S with leverage ratio \gamma can be viewed as a special case of a perp with
payoff function \varphi (St ) = L0 (St /S0 )\gamma . Such perps trade widely on the Ethereum blockchain.
Indeed, Squeeth, which trades on the Decentralized Finance (DeFi) protocol Opyn, is simply
a perp with payoff function \varphi (St ) = St2 , where St is the value on dollars of Ethereum.
Example 11. If, at time t = 0, one deposits two tokens into a geometric mean constant
function market maker (CFMM), then, ignoring fees collected by the CFMM, the value V =
(Vt )t\geq 0 of this deposit at time t \geq 0 is

\Bigl( pV qV0 (2) \Bigr) (S (1) /S (1) )p (S (2) /S (2) )q

0 (1) t 0 t 0
Vt = (1) t
S + (2) St (1) (1) (2) (2)
S0 S0 p(St /S0 ) + q(St /S0 )
\Bigl( S (1) \Bigr) p \Bigl( S (2) \Bigr) q
= V0 t(1) t
(2)
,
S0 S0

where p and q are constants satisfying p, q > 0 and p + q = 1 (cf., [2, 5]).
Now, consider a perp as described in Definition 8 with payoff

\Bigl( S (1) \Bigr) p \Bigl( S (2) \Bigr) q

t t
(11) \varphi (St ) = V0 (1) (2)
.
S0 S0

We have from (5) and (7) that

p(p - 1) q(q - 1)
Dt dt = d\langle log S (1) \rangle t + d\langle log S (2) \rangle t + pqd\langle log S (1) , log S (2) \rangle t .
2 2
If the perp is terminated at time t the value of the payoff to the long side is

\int t \Bigl( S (1) \Bigr) p \Bigl( S (2) \Bigr) q

\varphi (St )e - 0
Ds \mathrm{d}s
= V0 t
(1)
t
(2)
S0 S0
\Bigl( p(1 - p) q(1 - q) \Bigr)
\times exp \langle log S (1) \rangle t + \langle log S (2) \rangle t - pq\langle log S (1) , log S (2) \rangle t .
2 2

One can show that the term in the exponent is positive along every path of (S (1) , S (2) ).1 Thus,
ignoring fees, rather than deposit tokens into a CFMM it would always be better to take a
long position in a perp with payoff (11).

1
This follows from p(1 - p)dx2 + q(1 - q)dy 2 - 2pqdxdy = pqdx2 + pqdy 2 - 2pqdxdy = (pdx - qdy)2 \geq 0.

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 A PRIMER ON PERPETUALS SC25

5. Extension to models with jumps. In this section, we derive funding and discount rates
for perpetual contracts as well as replication strategies for the short side when asset prices are
allowed to jump. For simplicity, we will assume the risk-free rate of interest is zero (rt = 0)
and we will consider perpetuals written on a single risky asset S. The extension to nonzero
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

interest rates and multiple assets it straightforward, but tedious.

Fix a filtered probability space (\Omega , F, \BbbF , \BbbP
\widetilde ), where the filtration \BbbF represents the history
of the market and \BbbP
\widetilde denotes the market's chosen pricing measure. We suppose that the
dynamics of the risky asset S are of the form

\int
(12) dSt = \sigma t - St - dW
\widetilde t + St - (e\gamma t - (z) - 1)N
\widetilde (dt, dz),

where \sigma = (\sigma t )t\geq 0 and \gamma (z) = (\gamma t (z))t\geq 0 for every z \in \BbbR are scalar \BbbF -adapted processes,
W
\widetilde = (W
\widetilde t )t\geq 0 is a scalar (\BbbP \widetilde (dt, dz) = N (dt, dz) - \nu (dz)dt is a
\widetilde , \BbbF )-Brownian motion, and N
compensated Poisson random measure on \BbbR . Observe that S is a (\BbbP \widetilde , \BbbF )-martingale, as it must
be in the absence of arbitrage. The following theorem gives the funding rate F and discount
rate D for the perpetual contracts described in Definitions 1 and 8, respectively, when the
dynamics of the underlying S are given by (12).
Theorem 12. Suppose the dynamics of a single underlying risky asset are of the form (12)
and the payoff function \varphi of a perpetual satisfies \varphi \in C 2 (\BbbR , \BbbR ) and \varphi \not = 0. Then, the funding
rate F of the perpetual contract described in Definition 1 is given by

\int \Bigl(
1 \Bigr)
(13) Ft = \sigma t2 St2 \varphi \prime \prime (St ) + \varphi (St e\gamma t (z) ) - \varphi (St ) - St (e\gamma t (z) - 1)\varphi \prime (St ) \nu (dz),
2

and the discount rate D of a perpetual contract described in Definition 8 is given by Dt =

Ft /\varphi (St ).
Proof. Consider first a perpetual contract with funding, as described in Definition 1. The
infinitesimal change in value of the long side is

(14) d\varphi (St ) - Ft dt = At \varphi (St )dt + \sigma t St \varphi \prime (St )dW \widetilde t
\int \Bigl( \Bigr)
+ \varphi (St - e\gamma t - (z) ) - \varphi (St - ) N \widetilde (dt, dz) - Ft dt,
\int \Bigl(
1 2 2 \prime \prime \Bigr)
At \varphi (St ) := \sigma t St \varphi (St ) + \varphi (St e\gamma t (z) ) - \varphi (St ) - St (e\gamma t (z) - 1)\varphi \prime (St ) \nu (dz).
2

In the absence of arbitrage, the value of the long side must be a (\BbbP
\widetilde , \BbbF )-martingale. As such,
the sum of the dt terms in (14) must be zero, which leads to the expression (13) for F . Next,
consider a perpetual contract with discounting, as described in Definition 8. The infinitesimal

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 SC26 G. ANGERIS, T. CHITRA, A. EVANS, AND M. LORIG

change in value of the long side is given by

\Bigl( \int t \Bigr) \int t \int t
(15) d e - 0 Ds \mathrm{d}s \varphi (St ) = - Dt e - 0 Ds \mathrm{d}s \varphi (St )dt + e - 0 Ds \mathrm{d}s d\varphi (St )
\int t \int t
= - Dt e - \varphi (St )dt + e - 0 Ds \mathrm{d}s At \varphi (St )dt
Ds \mathrm{d}s
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

0
\int t
+ e - 0 Ds \mathrm{d}s \sigma t St \varphi \prime \prime (St )dW \widetilde t
\int t
\int \Bigl( \Bigr)
+ e - 0 Ds \mathrm{d}s \varphi (St - e\gamma t - (z) ) - \varphi (St - ) N \widetilde (dt, dz).

Once again, in the absence of arbitrage, the value of the long side must be a (\BbbP
\widetilde , \BbbF )-martingale.
As such, the sum of the dt terms in (15) must be zero, which leads to Dt = Ft /\varphi (St ), where
F is given by (13).
Remark 13. Note that, unlike the proofs of Theorems 3 and 9, which provide expressions
for the funding rate F and discount rate D of the perpetual contracts described in Definitions 1
and 8 as well as replication strategies for the short side, the proof of Theorem 12 provides only
the funding and discount rates for perpetual contracts but says nothing about a replication
strategy for the short side. In order to derive a replication strategy for the short side when
the underlying asset S can jump, we will need to make some additional assumptions.
Henceforth, assume that the dynamics of the risky asset S are of the form
\int
(16) dSt = \sigma t - St - dWt + St - (ez - 1)N
\widetilde \widetilde (dt, dz),

where the L\'evy measure \nu associated with N

\widetilde is a Dirac comb

n
\sum
(17) \nu (dz) = \lambda j \delta zj (z)dz,
j=1

and the volatility process \sigma evolves independently of the Brownian motion W
\widetilde and the Poisson
random measure N that appear in (16),
\widetilde

(18) \sigma \bot \bot W

\widetilde , \sigma \bot \bot N
\widetilde .

Assume further that one can trade call and put options with any strike at a fixed maturity T .
This is equivalent to assuming one can trade any T -maturity European option whose payoff
can be written as the difference of convex functions, as these payoffs can be synthesized from
call and put payoffs.
Theorem 14. Suppose the dynamics of S satisfy (16), (17), and (18) and consider a per-
petual contract as described in Definition 1. From Theorem 12 the funding rate F is given by

\int \Bigl(
1 \Bigr)
(19) Ft = \sigma t2 St2 \varphi \prime \prime (St ) + \varphi (St ez ) - \varphi (St ) - St (ez - 1)\varphi \prime (St ) \nu (dz).
2

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 A PRIMER ON PERPETUALS SC27

(p)
Let P (p) = (Pt )0\leq t\leq T denote the value of a European power contract, which pays STp at time
T , where p \in \BbbR . Fix (p1 , p2 , . . . , pn ) \in \BbbR n such that the (n + 1) \times (n + 1) stochastic matrix
H = (Ht )0\leq t\leq T with entries given by
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

(j,i) (p ) p\=i pi zj (\=

p ) pi p\=i zj
Ht - = e\psi (\=pi )(T - t) Pt - i St - - 1) - e\psi (pi )(T - t) Pt - i St -
\right\}
(e (e - 1), j, i \leq n,
(j,n+1)
Ht - = St - (ezj - 1), j \leq n,
(20) (n+1,i)
\Bigl(
(p ) p\=i (\=
p ) pi
\Bigr)
Ht - = \sigma t - pi e\psi (\=pi )(T - t) Pt - i St - - p\=i e\psi (pi )(T - t) Pt - i St - , i \leq n,
(n+1,n+1)
Ht - = \sigma t - St -

is invertible for all t \in [0, T ], where p\=i := 1 - pi for all i and the function \psi is defined as
follows:
\int \Bigl( \Bigr)
\psi (p) = (epz - 1) - p(ez - 1) \nu (dz).

Let X = (Xt )0\leq t\leq T be the value of a self-financing portfolio with dynamics of the form
n
(p ) (pi )
\sum
(21) dXt = \Delta t - dSt + \Gamma t - i dYt + Ft - dt, X0 = \varphi (S0 ),
i=1
(pi ) pi )(T - t) p\=i (p ) pi (\=
pi )
(22) dYt = e \psi (\=
St - dPt i - e\psi (pi )(T - t) St - dPt ,

where \Delta and \Gamma (p1 ) , . . . , \Gamma (pn ) are given by

\left[ \right]
(p )
\varphi (St - epz1 ) - \varphi (St - )
\left[ \right]
\Gamma t - 1
.. ..
(23) . - 1
= Ht - . .
(pn ) \varphi ( S e pzn ) - \varphi (S )
\Gamma t - t - t -
\Delta t - \sigma t - St - \varphi \prime (St - )

Then the portfolio X replicates the perpetual payoff. That is, the following holds:

(24) Xt = \varphi (St )

for all 0 \leq t \leq T .

Proof. We begin by computing the dynamics of \varphi (S). Using (16) and It\^
o's lemma we have

2
(25) d\varphi (St ) = 12 \sigma t - S 2 \varphi \prime \prime (St - )dt + \sigma t - St - \varphi \prime (St - )dW
\widetilde t
\int \Bigl( t - \Bigr)
+ \varphi (St - ez ) - \varphi (St - ) - St - (ez - 1)\varphi \prime (St - ) \nu (ds)dt
\int \Bigl( \Bigr)
+ \varphi (St - ez ) - \varphi (St - ) N \widetilde (dt, dz).

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 SC28 G. ANGERIS, T. CHITRA, A. EVANS, AND M. LORIG

Next, it will be helpful to introduce Z = log S. We will separate Z into a continuous component
Z c and a jump component Z j . Using It\^ o's lemma, we have
\int \int
dZt = dZtc + dZtj , dZtc = - 12 \sigma t -
2 \widetilde t , dZ j = - (ez - 1 - z)\nu (dz)dt + z N
dt + \sigma - t dW t
\widetilde (dt, dz).
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

Note that, conditional on the path of \sigma , the random variable ZTc - Ztc is distributed as a normal
random variable. Additionally, note that Z j is a L\'evy process with characteristic exponent
\psi (\tti \cdot ). Thus, conditioning on the path of \sigma and using the L\'evy--Kintchine formula, we have
1
(26) \BbbE \widetilde t e 2 (p2 - p)(\langle Z c \rangle T - \langle Z c \rangle t ) , \langle Z c \rangle T - \langle Z c \rangle t
\widetilde t ep(ZTc - Ztc ) = \BbbE
\int T
= \widetilde t ep(ZTj - Ztj ) = e(T - t)\psi (p) ,
\sigma s2 ds, \BbbE
t

where we have introduced the short-hand notation \BbbE \widetilde ( \cdot | Ft ). Now, using (26) as well as
\widetilde t \cdot := \BbbE
c j
Z \bot \bot Z , we find that the value of a European power option satisfies
(p) j j 1 2
\widetilde t S p = epZt \BbbE c
\widetilde t ep(ZT - Zt ) = S p e\psi (p)(T - t) \BbbE
\widetilde t ep(ZT - Zt ) \BbbE c
\widetilde t e 2 (p - p)(\langle Z c \rangle T - \langle Z c \rangle t )
Pt = \BbbE T t .
Next, observe that
1
\widetilde t e 2 (p
d\BbbE
2
- p)(\langle Z c \rangle T - \langle Z c \rangle t ) \widetilde (p) ,
= (. . .)dt + dM t

where M\widetilde (p) is a (\BbbP \widetilde , \BbbF )-martingale, which is independent of W

\widetilde and N
\widetilde . Thus, using the fact
(p) (p)
that P is a (\BbbP , \BbbF )-martingale, and thus dPt has no dt terms, we have
\widetilde
(p) 1 2 1
- p)(\langle Z c \rangle T - \langle Z c \rangle t ) 2 c c
(27) dPt = e\psi (p)(T - t) \BbbE
\widetilde t - e 2 (p p \psi (p)(T - t) \widetilde
dStp + St - e d\BbbE t e 2 (p - p)(\langle Z \rangle T - \langle Z \rangle t )
1 2 c c
+ St - \BbbE t e 2 (p - p)(\langle Z \rangle T - \langle Z \rangle t ) de\psi (p)(T - t)
p \widetilde
\int
1 2 c c
\Bigl( \Bigr)
\psi (p)(T - t) \widetilde (p - p)(\langle Z \rangle T - \langle Z \rangle t ) p \widetilde p
=e \BbbE t - e 2 p\sigma t - St - dWt + St - (epz - 1)N
\widetilde (dt, dz)
p \psi (p)(T - t) \widetilde (p)
+ St - e dM
\int
(p) \widetilde (p) \widetilde (dt, dz) + e\psi (p)(T - t) S p dM\widetilde (p) .
= p\sigma t - Pt - dW t + Pt - (epz - 1)N t - t

Now, defining p\= := 1 - p and noting that p2 - p = p\=2 - p, \= we have M \widetilde (p) = M
\widetilde (\=p) . Therefore, we
have from (27) that
\int
(\=
p) (\=
p) \widetilde (\=
p) \= \widetilde (dt, dz) + e\psi (\=p)(T - t) S p\= dM \widetilde (p) .
(28) \= t - Pt - dWt + Pt - (epz
dPt = p\sigma - 1)N t - t

Thus, using (27) and (28), the process Y (p) , defined in (22), is a self-financing portfolio that
satisfies
\Bigl( \int \Bigr)
(p) p\= (p) \widetilde (p) pz
(29) dYt = e\psi (\=p)(T - t) St - p\sigma t - Pt - dW t + P t - (e - 1) N
\widetilde (dt, dz)
\Bigl( \int \Bigr)
p (\=
p) \widetilde (\=
p) pz
- e\psi (p)(T - t) St - p\sigma
\= t - Pt - dW t + P t - (e \=
- 1) N\widetilde (dt, dz) .

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 A PRIMER ON PERPETUALS SC29

Now, we wish to create a self-financing portfolio whose value X satisfies (24). As there are
at least (n + 1) sources of uncertainty (due to the Brownian motion W \widetilde and the n possible
j
jump-sizes \Delta Zt \in \{ z1 , z2 , . . . , zn \} ) the portfolio will need at least (n + 1) hedging assets; we
will use the underlying S as well as ``shares"" of Y (pi ) for i \in \{ 1, 2, . . . , n\} . Thus, the dynamics
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

of X are of the form (21), where, for the moment, the processes \Delta and \Gamma (p1 ) , . . . , \Gamma (pn ) are
unknown. Using (16), (21), and (29), we have
n \Bigl(
(p ) (p ) p\=i
\sum
(30) dXt = \Delta t - \sigma t - St - dW
\widetilde t + \Gamma t - i \sigma t - pi e\psi (\=pi )(T - t) Pt - i St -
i=1
\Bigr) \int
\psi (pi )(T - t) (\=
p ) pi \widetilde t + \Delta t - St - (ez - 1)N
- p\=i e Pt - i St - dW \widetilde (dt, dz)
n
\int \sum \Bigl(
(p ) (p ) p\=i pi z
+ \Gamma t - i e\psi (\=pi )(T - t) Pt - i St - (e - 1)
i=1
\Bigr)
\psi (pi )(T - t) (\=
p ) pi p\=i z
- e Pt - i St - (e - 1) N \widetilde (dt, dz)+Ft - dt.

Equation (24) will be satisfied if and only if the dt, dW \widetilde t and N \widetilde (dt, dz) terms in (25) and
(30) are equal. As such, the funding rate F must be given by (19) and the processes \Delta and
\Gamma (p1 ) , . . . , \Gamma (pn ) must satisfy
n \Bigl( \Bigr)
(p ) (p ) p\=i (\=
p ) pi
\sum
\prime
\sigma t - St - \varphi (St - ) = \Delta t - \sigma t - St - + \Gamma t - i \sigma t - pi e\psi (\=pi )(T - t) Pt - i St - - p\=i e\psi (pi )(T - t) Pt - i St - ,
i=1
zj zj
\varphi (St - e ) - \varphi (St - ) = \Delta t - St - (e - 1)
n \Bigl( \Bigr)
(p ) (p ) p\=i pi zj (\=
p ) pi p\=i zj
\sum
+ \Gamma t - i e\psi (\=pi )(T - t) Pt - i St - (e - 1) - e\psi (pi )(T - t) Pt - i St - (e - 1) ,
i=1

where the last equation must hold for all zj \in \{ z1 , z2 , . . . , zn \} . In matrix form, we have
\left[ \right]
(p )
\varphi (St - ez1 ) - \varphi (St - )
\left[ \right]
\Gamma t - 1
.. ..
. = Ht - . ,
\varphi (St - ezn ) - \varphi (St - ) (pn )
\Gamma t -
\sigma t - St - \varphi \prime (St - ) \Delta t -
where the entries of H are given by (20). Using the fact that H is invertible, we find that (24)
will hold if \Delta and \Gamma (p1 ) , . . . , \Gamma (pn ) are given by (23).
Remark 15. The replication strategy described in Theorem 14 works only up until the
maturity date T of the European power contracts. However, at time T one can continue the
replication strategy by trading European power contracts with a maturity date T > T .
(p )
Remark 16. Note that St - , Pt - i for all i and \sigma t - = \mathrm{d}\mathrm{d}t \langle Z c \rangle t are observable. Thus,
no assumptions about the dynamics of the volatility process \sigma are needed for the replication
strategy to work. We do, however, require knowledge of the possible jump-sizes \{ z1 , z2 , . . . , zn \}
and jump intensities under the pricing measure \{ \lambda 1 , \lambda 2 , . . . , \lambda n \} as these appear in \nu and \psi .

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.

 SC30 G. ANGERIS, T. CHITRA, A. EVANS, AND M. LORIG

REFERENCES

[1] C. Alexander, J. Choi, H. Park, and S. Sohn, Bitmex Bitcoin derivatives: Price discovery, informa-
tional efficiency, and hedging effectiveness, J. Futures Markets, 40 (2020), pp. 23--43.
[2] G. Angeris and T. Chitra, Improved price oracles: Constant function market makers, in Proceedings of
the 2nd ACM Conference on Advances in Financial Technologies, AFT '20, Association for Computing
Downloaded 07/12/24 to 154.59.124.59 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

Machinery, New York, 2020, pp. 80--91.

[3] P. Carr and D. Madan, Towards a theory of volatility trading, in Option Pricing, Interest Rates
and Risk Management, Handb. Math. Finance 22, Cambridge University Press, Cambridge, 2001,
pp. 458--476.
[4] CoinGecko, Cryptocurrency Derivatives (Perpetual Contract) by Volume, 2022, https://urldefense.com/
v3/ https://www.coingecko.com/en/derivatives ;!!CX2e5RWqcItE!rVO khNmJf1jE4zB1F vuwG9K
MLnrJEidUwh0VKjoHPVPIepeeTSTr77jfMJLJHBFVwMpYnF2DMgaKF4aw4$ (26 August 2022).
[5] A. Evans, Liquidity Provider Returns in Geometric Mean Markets, preprint, arXiv:2006.08806, 2020.
[6] T. Leung and M. Santoli, Leveraged Exchange-Traded Funds: Price Dynamics and Options Valuation,
Springer, New York, 2016.
[7] Y. Lewenberg, Y. Bachrach, Y. Sompolinsky, A. Zohar, and J. S. Rosenschein, Bitcoin mining
pools: A cooperative game theoretic analysis, in Proceedings of the 2015 International Conference on
Autonomous Agents and Multiagent Systems, 2015, pp. 919--927.
[8] llllvvuu, Raw Moment Derivatives, https://urldefense.com/v3/ https://llllvvuu.dev/blog/raw-
moments ;!!CX2e5RWqcItE!rVO khNmJf1jE4zB1F vuwG9KMLnrJEidUwh0VKjoHPVPIepeeTST
r77jfMJLJHBFVwMpYnF2DMgIlDFowY$ (2021).
[9] Opyn, Squeeth, https://urldefense.com/v3/ https://squeeth.opyn.co/ ;!!CX2e5RWqcItE!rVO khNmJf1
jE4zB1F vuwG9KMLnrJEidUwh0VKjoHPVPIepeeTSTr77jfMJLJHBFVwMpYnF2DMgHZCTfxE$
(2021).
[10] R. J. Shiller, Measuring asset values for cash settlement in derivative markets: Hedonic repeated mea-
sures indices and perpetual futures, J. Finance, 48 (1993), pp. 911--931, https://doi.org/10.1111/
j.1540-6261.1993.tb04024.x.
[11] D. White, D. Robinson, Z. Koticha, A. Leone, A. Gauba, A. Krishnan, Power Perpetuals,
https://urldefense.com/v3/ https://www.paradigm.xyz/2021/08/power-perpetuals ;!!CX2e5RWqc
ItE!rVO khNmJf1jE4zB1F vuwG9KMLnrJEidUwh0VKjoHPVPIepeeTSTr77jfMJLJHBFVwMpYnF
2DMglJFe8lI$ (August 2021).

Copyright © by SIAM. Unauthorized reproduction of this article is prohibited.