Financial analytics of inverse BTC options in a

stochastic volatility world*

Huei-Wen Teng„

Wolfgang Karl Härdle

October 18, 2022

Abstract
Bitcoin (BTC) has attracted a plethora of investors and professional traders and
becomes an almost inevitable asset class in today’s financial markets. Deribit, the
largest exchange for crypto options, offers European-typed inverse options, which
target to BTC in USD but have payoff denominated in BTC. However, analytical
insights to inverse options remain scarce. The dynamics of the underlying BTC is
well described by a stochastic volatility model but in pricing inverse options one
meets numerical difficulties in calibration and hedging. Financial analytics for the
practically useful stochastic volatility with correlated jumps model is provided and
comparison with simpler nested models both in in-sample and out-of-sample pricing
is given. A dynamic Delta hedging exercise yields that - surprisingly - the portfolio
of nested models gives almost identical hedging errors.
Keywords: cryptocurrency, crypto options, stochastic volatility models, corre-
lated jumps, simulation, Delta hedge, CRIX, crypto index.
JEL Codes: G13, C15, C58, E37

* Financial support of the European Union’s Horizon 2020 research and innovation program ”FIN-
TECH: A Financial supervision and Technology compliance training programme” under the grant agree-
ment No 825215 (Topic: ICT-35-2018, Type of action: CSA), the European Cooperation in Science
& Technology COST Action grant CA19130 - Fintech and Artificial Intelligence in Finance - Towards
a transparent financial industry, the Deutsche Forschungsgemeinschaft’s IRTG 1792 grant, the Yushan
Scholar Program and the Higher Education Sprout Project of National Yang Ming Chiao Tung Uni-
versity by the the Ministry of Education of Taiwan, the Ministry of Science and Technology of Taiwan
under Grants 110-2118-M-A49-003 and 111-2118-M-A49-007, and the Czech Science Foundation’s grant
no. 19-28231X / CAS: XDA 23020303 are greatly acknowledged.
„ Department of Information Management and Finance, National Yang Ming Chiao Tung University,
Taiwan. E-mail: [email protected].
Blockchain Research Center, Humboldt-Universität zu Berlin, Germany. Wang Yanan Institute for
Studies in Economics, Xiamen University, China. Sim Kee Boon Institute for Financial Economics,
Singapore Management University, Singapore. Faculty of Mathematics and Physics, Charles University,
Czech Republic. National Yang Ming Chiao Tung University, Taiwan. E-mail: [email protected].

Electronic copy available at: https://ssrn.com/abstract=4238213

1 Introduction

Bitcoin (BTC) was the first open-source distributed cryptocurrency (CC) after the white
paper by Nakamoto (2008). Other CC’s followed and together now they represent a
market cap more than that of the top twenty stocks worldwide in Q2 2022. Financial
institutions have therefore constructed tailor-made products for investors involing CCs
for portfolio diversification (Härdle et al., 2020). BTC has become therefore an inevitable
asset requiring analytical insight into its derivatives market. As the largest exchange for
crypto options, Deribit offers the “inverse option”, where BTC denominated in USD is
the underlying asset, but the payoff is denominated in BTC. As a consequence, this option
type allows professional traders to avoid exchanging frequently between fiat money and
CC.
An intriguing feature of this inverse option is its non-linear payoff. Let us look at
two examples. Suppose that the current BTC is traded at 20,000 USD and focus on
an at-the-money (ATM) option with strike 20,0000 USD. If the BTC raises to 40,000
at maturity, the payoff for the inverse BTC call option with strike 20,000 USD is worth
simply 0.5 BTC. On the contrary, if the BTC drops to 5,000 BTC at maturity, the payoff
for the inverse BTC put option with strike 20,000 USD is valued at 3 BTCs. These two
examples display this striking asymmetry of the payoff.
Earlier work include Hu et al. (2019), Siu and Elliot (2021) focussing e.g. on various
GARCH models. Chen and Huang (2021) works with jump-diffusion models, and Cao
and Celik (2021) apply mixed jump-diffusion process. Madan et al. (2019) compare the
Black-Scholes, Laplace model, five variance gamma-related models, and the Heston model
for BTC prices. Cretarola et al. (2020) includes market attention as an exogenous factor
as an extension to the geometric Brownian motion. Hou et al. (2020) consider stochastic
volatility with correlated jumps models. These papers estimate dynamics using BTC
prices, then produce prices for the plain vanilla options. But, we calibrate complex
models using real options data directly.
With simulated option prices of the plain vanilla type, Matic et al. (2021) simulate
complex dynamics for CC to study hedging performance. An increasing number of stud-

Electronic copy available at: https://ssrn.com/abstract=4238213

ies start to investigate crypto options with real data, but they also apply the vanilla
option. For example, Li et al. (2019) combines a multiple input LSTM-based model with
the Black-Scholes model; Zulfiqar and Gulzar (2021) calculate implied volatility under
the pricing formulae for vanilla options; Jalan et al. (2021) compare the Heston-Nandi
GARCH (1,1) asymmetric multiple volatility-regime model with Black-Scholes model.
Alexander and Imeraj (2021) is the first paper highlighting that the inverse option
type is dominating (rather than the vanilla option) in the CC market. Considering inverse
options under the Black-Scholes model, Alexander et al. (2022) contrast IV curves of tick-
level Deribit option price data with those of S&P 500 options, and Alexander and Imeraj
(2022) analyse robust dynamic delta hedging of BTC options. However, these studies are
restricted to the Black-Scholes model. Because stylized features in the dynamics of BTC
include SV and jumps from numerical empirical studies, this paper aims at understanding
whether a more complex dynamics helps to price and hedge crypto options.
A first difficulty in model calibration is the lack of closed-form pricing formulae for
a complex dynamics. Even for the relatively simple Heston’s Stochastic Volatility (SV)
model, calibration using plain vanilla options remains a challenge (Cui et al., 2017; Mrázek
et al., 2016). Dynamics of CCs have been documented to be extremely volatile and to
contain frequent jumps (Hacioglu, 2019; Scaillet et al., 2020; Saef et al., 2021). To incor-
porate these features, this paper aims at investigating if the practically useful SV with
correlated jumps outperforms simpler model in pricing and hedging for the inverse option.
Although Monte Carlo simulation can be used to price the inverse option, it is notoriously
known for its slow convergence. Worse, its variability in each Monte Carlo simulation
deteriorates the iteration in the optimization process. To overcome this difficulty in
model calibration, we first provide a feasible simulation scheme for model calibration
using common random variables (Ross, 2022).
Once calibrated, we calculate the Greeks for implementing hedging routines (Hull,
2014). Greeks are price sensitivities of option price with respect to certain parameters
and are critical measures for risk management in derivatives market. Among Greeks, the
Delta ∆ is the first-order partial differentiation of option price with respect to the initial

Electronic copy available at: https://ssrn.com/abstract=4238213

price of the underlying asset and is used to hedge the change of the underlying asset. The
calculation of Greeks is not straightforward, particularly when there exists no closed-form
pricing formulae for a specific option with complex dynamics (Glasserman, 2004). As a
second numerical contribution of this paper, we apply the parameter derivative in Lyuu
et al. (2019) to provide an unbiased Delta formula under various stochastic volatility
models, which coincides with the Delta formula in Alexander and Imeraj (2021) for the
Black-Scholes case. The Delta formula allows traders dynamic hedging for the described
setting and their helps in creating stable portfolio with predefined risk structure.
The rest of this paper is organized as follows. Section 2 explores the inverse options
based on data of the Blockchain-Research-Center.com and demonstrates a puzzle for
implied volatility slippery. Section 3 calibrates and investigates implied parameters of the
stochastic volatility with correlated jumps model and its nested models. It also compares
in-sample and out-of-sample pricing errors. Section 4 provides the Delta formula and
implements dynamic Delta hedging. The last section concludes. All code for this paper
is available on quantlet.com. A presentation video of this paper is on quantinar.com.

2 Inverse options

In the following, Section 2.1 contrasts the payoff function for vanilla options and inverse
options, Section 2.2 explores the data, and Section 2.3 presents the slippery of implied
volatility for inverse options.

2.1 The payoff function

Let K be the strike price, and T be the time to maturity in days. We use an indicator
ω = 1 and ω = −1 for a call and put option, respectively. In addition, let r denote the
discount rate, and y denote the dividend yield. Let St denote the BTC price denominated
in USD at time t. Recall the European plain vanilla option with payoff,

℘vanilla (ω, K, T ) = max {ω(ST − K), 0} , ω = ±1. (1)

Electronic copy available at: https://ssrn.com/abstract=4238213

Vanilla options have piecewise linear payoff functions, which guarantees the put-call par-
ity.
A distinct feature of the inverse option is that its strike price corresponds to the value
of a BTC denominated in USD, but the payoff is converted back to BTC. Let St−1 = 1
St

denote the inverse of the price St . The inverse option has therefore the payoff

℘inverse (ω, T, K) = ST−1 max {ω(ST − K), 0} , ω = ±1. (2)

In contrast to the payoff (1), inverse options have piecewise nonlinear payoff functions,
and thus the standard put-call parity does not hold and hence also the implied volatility
of put and call are different.
Figure 1 depicts the payoff function against ST for vanilla and inverse BTC call and
put options. Suppose the strike price is K USD. Recall that the payoff for a vanilla
option is denominated in dollars, but the payoff for an inverse option is denominated in
BTC. Although vanilla options may have unlimited positive payoff, this is not the case
for inverse options. Specifically, the inverse BTC call option has a payoff capped with
one BTC, but the inverse put option has unlimited positive payoff. This indicates that
inverse BTC call options should be priced less than one BTC, otherwise, an arbitrage
opportunity will exist. However, an inverse BTC put option could be worth more than
one BTC in the real market. For example, an inverse BTC put option has a payoff of five
BTCs if BTC drops to one-fifth of the strike price at maturity. As indicated already, all
numerical examples can be replayed by checking the Quantlet.com link.

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 1: Payoff functions of vanilla and inverse options. This figure contrasts pay-
off against ST between vanilla and inverse options. Suppose the strike price is K USD. The
first and second columns depict payoff for vanilla and inverse options, respectively. The
first and second rows depict payoff for call and put options, respectively. It is shown that
vanilla options have piecewise linear functions, whereas inverse options have nonlinear
payoff functions. In addition, the inverse call option has payoff capped with one BTC, but
the inverse put option has unlimited positive payoff. Deribit inverse BTC options

2.2 Data

In this research, we collect Deribit intraday transaction data for inverse options through
Blockchain-Research-Center.com during December 1, 2021 to February 28, 2022. The
dataset includes 736,563 observations, a total of 90 trading days. As of this writing
(September 2, 2022), among the total BTC options with open interest at 243.60K BTC,
Deribit takes a dominating proportion at 88% (215.90K BTC), followed by CME at
8% (16.44K BTC), and Okex at 3% (8.44K BTC). Table 5 in the Appendix summarize
columns and explanations for each column of the data. The intraday transaction data
includes information such as timestamps, BTC prices in USD, implied volatility, types,
strikes, and maturities of inverse options. This information allows us to obtain a full
picture of the BTC options market.

Electronic copy available at: https://ssrn.com/abstract=4238213

 Figure 2 sketches the time series of daily accumulative numbers of trades and daily
accumulative volume from the intraday transaction for inverse options. One sees that
both daily numbers of trades and volume are stable for the defined time period. The
average daily number of trades is 8,184, and the average volume is 21,155.

Figure 2: Daily accumulated trades and quantities. For intraday transaction

on inverse options, the upper panel provides time series plot for daily accumulative
number of trades, and the lower panel provides time series plot for daily accumulated
quantities. The study period spans from December 1, 2021 to February 28, 2022.
Deribit inverse BTC options

Figure 3 depicts strike prices of intraday transaction for inverse BTC call and put
options during the study period, and overlays the intraday BTC prices in a black bold
line. First, it may be noted that out-of-the-money (OTM) options are more frequently
traded compared with in-the-money (ITM) options: Inverse BTC call options are more
frequently traded at strike prices above current BTC prices, whereas inverse BTC put
options are more frequently traded at strike prices below current BTC prices. Second, one
discovers that the range of strike price is substantially larger than the range for classical
option market. The average BTC is 43,874.72 USD, but the minimum and maximum
strike prices are 12,000 USD and 400,000 for the inverse option, respectively.

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 3: Visualizing strike prices for intraday transaction of inverse op-
tions. The left panel and right panel depict strike prices for intraday transaction of
inverse BTC call and put options, respectively. The BTC prices are depicted in a
black bold line. The study period spans from December 1, 2021 to February 28, 2022.
Deribit inverse BTC options

Figure 4 depicts a three-dimensional scatter plot for prices of inverse options traded
on January 1, 2022. It is clear that inverse options are more frequently traded with short-
term maturity and at a wide range of moneyness. In addition, excluding the highest 0.1%
prices of BTC inverse options, all options have prices less than 0.2 BTC.

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 4: Visualizing prices of inverse options. This figure depicts a 3-D scatter
plot for prices of inverse options traded on January 18, 2022. Call and put options are
depicted in blue and red colors, respectively. We remove the highest 0.1% prices of BTC
inverse options to have a better demonstration. Deribit inverse BTC options

Table 1 provides percentiles of time to maturity (T ), moneyness (K/S0 ), and prices

for intraday transaction. About 50% of the options are short-term: Inverse call and put
options have a median of time maturity 9 and 7, respectively. Second, more than 90%
of the options are traded OTM. However, 5% are traded at extremely high or low strike
prices: The 5th percentile of moneyness for call options is 0.51, and the 95th percentile
of moneyness for put options is 1.55.

Electronic copy available at: https://ssrn.com/abstract=4238213

Table 1: Distributions of option characteristics. This table reports percentiles for
the time to maturity (T ), moneyness (S0 /K), and prices (p) of inverse BTC call and put
options. Here, Qk denotes the k-th percentile. The study period spans from December
1, 2021 to February 28, 2022. Deribit inverse BTC options

T S0 /K p

Percentile Call Put Call Put Call Put

Min 0 0 0.09 0.12 0.0001 0.0002

Q1 0 0 0.27 0.85 0.0005 0.0005
Q5 1 0 0.51 0.96 0.0005 0.0010
Q10 1 1 0.64 0.99 0.0010 0.0015
Q20 1 1 0.78 1.00 0.0030 0.0035
Q30 2 2 0.86 1.02 0.0055 0.0055
Q40 6 4 0.91 1.03 0.0080 0.0090
Q50 9 7 0.94 1.05 0.0115 0.0130
Q60 15 12 0.96 1.08 0.0170 0.0190
Q70 22 18 0.98 1.12 0.0250 0.0280
Q80 42 30 0.99 1.21 0.0380 0.0435
Q90 84 64 1.00 1.38 0.0630 0.0740
Q95 149 105 1.02 1.55 0.0954 0.1095
Q99 291 259 1.06 2.11 0.1750 0.2275
Q99.9 355 347 1.34 2.91 0.3260 0.6375
Q99.99 363 364 2.63 3.38 0.6255 4.2101
Max 365 365 3.90 4.04 0.7465 7.3356

Since, as noted above, the options are traded on a much wider range of moneyness
than for standard options, we define maturity-moneyness categories in Table 2. We report
the proportions of traded options in each maturity-moneyness category. Call options
represent 55.6% of the trades, and 82.3% of the options are OTM. For the rest of this

10

Electronic copy available at: https://ssrn.com/abstract=4238213

section, we summarize the basic empirical findings.

Table 2: Maturity-moneyness categories. This table defines nine maturity-

moneyness categories. For maturity, short-term, mid-term, and long-term options
are defined according to time to maturity (T ). Definitions for OTM, ITM, and
deep ITM options are according to the moneyness (S0 /K). This table also reports
proportions (%) of trades for inverse options at each maturity-moneyness category.
Deribit inverse BTC options

Inverse BTC call option Deep OTM OTM ITM

S0 S0 S0
K
≤ 0.86 0.86 < K
≤1 1< K
Sum

Short-term: T ≤ 5 0.3% 16.3% 3.7% 20.3%

Mid-term: 5 < T ≤ 28 6.0% 12.9% 1.8% 20.7%
Long-term: 28 < T 10.9% 3.0% 0.7% 14.6%

Sum 17.2% 32.2% 6.2% 55.6%

Inverse BTC put option ITM OTM Deep OTM

S0 S0 S0
K
≤1 1< K
≤ 1.12 1.12 < K
Sum

Short-term: T ≤ 5 3.5% 13.5% 1.1% 18.1%

Mid-term: 5 < T ≤ 28 2.5% 7.8% 6.4% 16.7%
Long-term: 28 < T 1.3% 2.3% 6.0% 19.6%

Sum 7.3% 23.6% 13.5% 44.4%

Table 3 reports average prices and standard deviations (in parenthesis) for call and
put options at each maturity-moneyness category. Average prices for call options are less
than one BTC at all maturity-moneyeness categories. In addition, it is consistent with
intuitions that inverse options are more expensive for longer-term and deeper ITM, for
both call and put options.

11

Electronic copy available at: https://ssrn.com/abstract=4238213

Table 3: Average prices in BTC and standard deviations of inverse options.
This table reports average prices and standard deviations (in parenthesis) for options at
each maturity-moneyness category. Deribit inverse BTC options

Inverse BTC call option Deep OTM OTM ITM

S0 S0 S0
K
≤ 0.86 0.86 < K
≤1 1< K

Short-term: T ≤ 5 0.0006 0.0054 0.0232

(0.0003) (0.0049) (0.0190)
Mid-term: 5 < T ≤ 28 0.0059 0.0224 0.0683
(0.0059) (0.0154) (0.0468)
Long-term: 28 < T 0.0358 0.0865 0.1614
(0.0391) (0.0477) (0.0946)

Inverse BTC put option ITM OTM Deep OTM

S0 S0 S0
K
≤1 1< K
≤ 1.12 1.12 < K

Short-term: T ≤ 5 0.0172 0.0041 0.0013

(0.0558) (0.0052) (0.0011)
Mid-term: 5 < T ≤ 28 0.0695 0.0234 0.0071
(0.0815) (0.0145) (0.0062)
Long-term: 28 < T 0.1884 0.0816 0.0342
(0.3674) (0.0373) (0.0289)

Figure 5 provides histograms of prices (p), days to maturity (T ), and moneyness ( SK0 )
for inverse put options with prices larger than one. It is fascinating that a put option
is likely to be worthy more than one BTC. In fact, there are 130 put options, or about
0.04% of put options that have prices larger than one BTC. If we see the distribution
of these extremely high priced put options, they are mostly long-term and deep ITM
(with moneyness between 0.5 to 0.1). The data indicates that investors believe that
there is a chance that the BTC price drops to one-half or one-tenth of current BTC price,
respectively.

12

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 5: Extremely high prices of inverse BTC put options. This figure provides
histograms of prices (p), days to maturity (T ), and moneyness ( SK0 ) for inverse put options
with prices larger than one. The study period spans from December 1, 2021 to February
28, 2022. Deribit inverse BTC options

2.3 Slippery of implied volatility

Suppose for a moment the dynamics of S follows a geometric Brownian motion (GBM)
with volatility σ. Alexander and Imeraj (2021) provide pricing formulae

n 2
o
mBS (σ; ω, T, K) = ω e−rT Φ(ωd2 ) − e(y−r+σ )T S0−1 KΦ(ωd3 ) , (3)

where
S0
+ (r − y + 12 σ 2 )T √ √
K
d1 = √ , d2 = d1 − σ T , d3 = d2 − σ T .
σ T
def
In practice, the implied volatility (IV) σBS = σBS (ω, T, K) is calibrated by solving

mBS (σBS ; ω, T, K) = p(ω, T, K). (4)

13

Electronic copy available at: https://ssrn.com/abstract=4238213

In the introduced quantlet links, fsolve (a Matlab function) is used to calculate σBS .
Discrepancies between iv communicated from Deribit and σBS are displayed in Figure 6.
However, one discovers substantial differences and a discernible curvature.

Figure 6: Discrepancies between iv and σBS . This figure plots iv against σBS ,
which solves (4) from our algorithm, and overlays a 45 degree line in red color.
The data is on January 18, 2022. Similar patterns can be found in other dates.
Deribit inverse BTC options

To check which IV is representing the risk structure of this data analysis, plug iv and
σBS in (3):

def
mBS (iv) = mBS (iv; ω, T, K),
def
mBS (σBS ) = mBS (σBS ; ω, T, K).

Figure 7 investigates if iv and σBS can successfully reproduce the prices, in the upper
and lower panels, respectively. One sees that a large proportion of mBS (σBS ) lays on the
45 degree line. This indicates that σBS is able to reproduce the market price of inverse
options. However, still a small proportion of mBS (σBS ) is in the upper part of the 45
degree line. This hints the inappropriateness of BS model in practice. On the other hand,
mBS (iv) deviates from price p. Thus, the reported iv indeed posts a pricing puzzle for
inverse options.

14

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 7: Pricing inverse options using the BS formulae through iv and
σBS . The left and right panels plot market prices of inverse option (p) against model
prices mBS (iv) and mBS (σBS ), respectively, using the intraday transaction of inverse
options on January 18, 2022. Both panels overlay a 45 degree line in red color.
Deribit inverse BTC options

Figure 8 visualizes iv and σBS calibrated on January 18, 2022, in a 3-D scatter plot.
These two implied volatilities have discernibly different patterns. Given a time to matu-
rity, σBS appears to expose higher curvature across moneyness, yet iv is relatively flat.
This higher curvature pattern in comparison with iv is a visible fact also on other trading
days.

15

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 8: Implied volatility iv and σBS . The upper and lower panels depict scatter
plots for iv and σBS , respectively. Data is on January 18, 2022. σBS is more skewed.
Deribit inverse BTC options

16

Electronic copy available at: https://ssrn.com/abstract=4238213

3 Pricing inverse options

To start, Section 3.1 introduces the stochastic volatility with correlated jumps model
and its nested ones. Then, Section 3.2 proposes a feasible simulation scheme in the
stochastic optimisation for model calibration. Section 3.3 summarizes implied parameters
and compares in-sample and out-of-sample pricing errors among different models.

3.1 The stochastic volatility with correlated jumps model

We turn now to pricing and explaining a stochastic volatility with correlated jumps
(SVCJ) model, advocated as appropriate stochastic dynamics by Hou et al. (2020). The
SVCJ model reads as:

dSt p
= rdt + Vt dWts + Zty dNt (5)
St
p
dVt = κ(θ − Vt )dt + σv Vt dWtv + Ztv dNt (6)

Cov(dWts , dWtv ) = ρdt (7)

P (dNt = 1) = λdt (8)

Zty |Ztv ∼ N (µy + ρj Ztv , σy2 ), (9)

Ztv ∼ Exp(µv ), (10)

Since we are looking at short-term options in an almost r = 0 world, we set r = 0 as in

Matic et al. (2021).
Parameters in the SVCJ model can be interpreted as follows. Between the return
process (5) and variance process (6), ρ is the correlation between two Brownian motions
Wts and Wtv . In the variance process, κ is the speed for conversion to long-term variance,
θ is the long-term variance, and σv is the volatility of the variance process. The jump
components in return and variance precesses has a frequency of jumps of λ as shown
in (8). To be positive for the variance process, it has a jump size Ztv which is assumed
to be exponentially distributed with positive mean µv . And, given the jump size of the
variance process Zty , the jump size in the return process Zty is assumed to be normally

17

Electronic copy available at: https://ssrn.com/abstract=4238213

distributed with mean µy + ρj Ztv and variance σy2 .
The SVCJ includes a rich family of models. If constraints on the parameters of the
SVCJ models are imposed, a variety of stochastic volatilities or jump diffusions could be
obtained. For example, if we set Ztv in (10) as zero, jumps are only present in prices
and the model reduces to the SVJ model of Bates (1996); If we set λ = 0, the model
does not exhibit jumps and reduces to the original SV model of Heston (1993); If we set
κ = θ = σv = 0 and define Ztv = 0, the model reduces to the pure jump diffusion of
Merton (1976); The pure jump diffusion of Merton (1976) reduces to the celebrated BS
model (Black and Scholes, 1973) with Zty = 0.
Let θ = (θ1 , . . . , θp ) denote the model parameter with size p. Because stochastic
volatility is a prominent feature in financial time series data, we focus on the SV, SVJ,
and SVCJ models. Specifically, we have θ = (µ, ρ, α, β, V0 , σv ) with p = 6 for the SV
model, θ = (µ,ρ, α, β, V0 , σv , λ, µy , σy ) with p = 9 for the SVJ model, and θ = (µ, ρ, α,
β, V0 , σv , λ, µy , ρj , σy , µv ) with p = 11 for the SVCJ model.

3.2 Model calibration and stochastic optimisation

Let p(ω, T, K) denote the observed market option price, and let m(θ; ω, T, K) denote an
option price with parameter θ. Following Bakshi et al. (1997), we calibrate the model
parameter by solving the problem:

X
min L(θ) = {p(ωi , Ti , Ki ) − m(θ; ωi , Ti , Ki )}2 . (11)
i

Thus, model calibration using inverse options is essentially an optimisation.

When there exists no closed-form formula for pricing inverse options under stochastic
volatility models, we replace it by Monte Carlo estimate, m(θ;
b ωi , Ti , Ki ). In this case,
parameter θ is calibrated by solving

M
X
min L(θ)
b = b ωi , Ti , Ki )}2 ,
{p(ωi , Ti , Ki ) − m(θ; (12)
i=1

where M is the number of inverse options. Thus, the model calibration becomes a stochas-

18

Electronic copy available at: https://ssrn.com/abstract=4238213

 tic optimisation, because m(θ;
b ω, T, K) is calculated using stochastic simulation method.
However, for stochastic volatility models, there exist no closed-form formulae for pric-
ing inverse options. We must employ Monte Carlo simulation. To simulate a path of
the stock price, we use the Euler-Maruyama discretization as follows (Belaygoro, 2005).
Define Yt+1 = log(St+1 /St ) as the log-return for t = 1, . . . , T . Let α = κθ and β = 1 − κ.
Then, we have

Vt−1 εyt + Zty Jt ,

p
Yt = r + (13)
p
Vt = α + βVt−1 + σv Vt−1 εvt + Ztv Jt , (14)

where εyt and εvt are the N (0, 1) variables with correlation ρ, Jt is a Bernoulli random
variable with P (Jt = 1) = λ, and jump sizes follow Zty |Ztv ∼ N (µy + ρj Ztv , σy2 ) and
Ztv ∼ Exp(µv ). Then, the path of asset price under the SVCJ model is obtained through

St = S0 exp(Y1 +···+Yt ) for t = 1, . . . , T, (15)

where S0 denotes the initial stock price.

Now, we are ready to outline procedures to price inverse options using Monte Carlo
simulation. We use N to denote the simulation sample size.

Step 1. Generate the following random variables:

(n)
(a) Jt ∼ Ber(λ),
v,(n)
(b) Zt ∼ Exp(µv ),
y,(n) v,(n)
(c) Zt |Zt ∼ N (µy + ρj Ztv , σy2 ),
     
v,(n)
 εt  0   1 ρ 
(d)   ∼ N   ,  ,

y,(n)
εt 0 ρ 1

for t = 1, . . . , T and n = 1, . . . , N .

(n) (n)
Step 2. Calculate (Yt , Vt ) using (13) and (14) for t = 1, . . . , T and n = 1, . . . , N .

(n)
Step 3. Calculate St using (15) for t = 1, . . . , T and n = 1, . . . , N .

19

Electronic copy available at: https://ssrn.com/abstract=4238213

Step 4. Price the inverse option price using sample mean,

N
e−rT X 1 n
(n)
o
m(θ;
b ω, T, K) = max ω(ST − K), 0 .
N n=1 ST(n)

The magnitude of Monte Carlo simulation error in L(θ)

b may dominate the increase
in the loss function when updating a parameter. Consequently, the optimisation process
is failed. To tackle this problem, we provide an approach with common random numbers
for variance reduction (Glasserman, 2004). The idea is to generate one set of random
samples and stick to this set of random samples through the whole process of solving
stochastic optimisation (12). See also Teng (2022).
To apply common random numbers, one generates samples having the Bernoulli and
exponential distributions (Ross, 2022). The quantile function for the Bernoulli random
variable with parameter λ is FB−1 (u, λ) = I{u>(1−λ)} (u), and the quantile function for the
exponential random variable with mean µ is FE−1 (u, µ) = − log(u)
µv
. In addition, we generate
v,(n) y,(n)
the correlated normal random variables εt and εt from two independent normal
random variable through the Cholesky decomposition. Thus, in calculating m(θ;
b ω, T, K),
we simply replace Step 1 by Step 1∗ :

(n) i.i.d. (n) i.i.d.

Step 1∗ . Generate Ui,t ∼ U (0, 1) for i = 1, 2 and Zj,t ∼ N (0, 1) for j = 1, 2, 3 for
t = 1, . . . , T and n = 1, . . . , N . Then, set
 
(n) (n)
(a) Jt = I{U (n) >(1−λ)} U1,t ,
1,t
 
(n)
v,(n) log U2,t
(b) Zt =− µv
,
y,(n) v,(n) (n)
(c) Zt = (µy + ρj Zt ) + σy Z1,t ,
v,(n) (n)
(d) εt = Z2,t ,
y,(n) (n) (n)
p
(e) εt = ρZ2,t + 1 − ρ2 Z3,t ,

for t = 1, . . . , T and n = 1, . . . , N .

20

Electronic copy available at: https://ssrn.com/abstract=4238213

3.3 Implied parameters

For each trading day, we include all intraday transaction of inverse options and calibrate
a set of model parameter for each day. As a remark, all analysis through out this paper is
carried out using Matlab (R2021a) in a desktop (Intel Core i9 CPU with 64 GB RAM).
We set N = 10, 000 as Monte Carlo sample size. To calibrate the BS model, we use
the Matlab solver fminsearch. To calibrate the SV, SVJ, SVCJ models, we use the
Matlab solver lsqnonlin for optimisation with Levenberg-Marquardt algorithm and set
’FunctionTolerance’ as 10−4 . In addition, to ensure the positiveness of the stochastic
volatility models, instead of using the Feller condition, 2κθ > σv2 (or 2α > σv2 by re-
parametrization), we impose constraints on parameters α > 0, β < 1, V0 > 0. We
impose a natural constraint on the correlation coefficient between two Brownian motions,
−1 ≤ ρ ≤ 1. For the SVJ and SVCJ models, λ refers to the probability of jump, thus it
has to satisfy 0 ≤ λ ≥ 1. For the SVCJ model, the average jump size for the volatility
needs to satisfy µv ≥ 0.
Figures 9 to 12 provide times series plots for the implied parameters for the BS, SV,
SVJ, and SVCJ models, calibrated on a daily basis, respectively. For the BS model, the
calibrated parameter σ is volatile across time. For the SV, SVJ, and SVJ models, some
implied parameters appear to be constant yet some have huge fluctuations across time.

21

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 9: Implied parameters for the BS model. This figure pro-
vides time series plot for daily calibrated implied parameters for the BS model.
Deribit inverse BTC options

Figure 10: Implied parameters for the SV model. This figure pro-
vides time series plot for daily calibrated implied parameters for the SV model.
Deribit inverse BTC options

22

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 11: Implied parameters for the SVJ model. This figure pro-
vides time series plot for daily calibrated implied parameters for the SVJ model.
Deribit inverse BTC options

Figure 12: Implied parameters for the SVCJ model. This figure pro-
vides time series plot for daily calibrated implied parameters for the SVCJ model.
Deribit inverse BTC options

23

Electronic copy available at: https://ssrn.com/abstract=4238213

 Moreover, suppose the implied parameter is denoted as θ.
b To measure if the model

fits inverse options well, we calculate the in-sample root-mean-squared-error (RMSE) by

v
u
u1 X M n o2
In-sample RMSE = t p(ωi , Ti , Ki ) − m(
b θ;
b ωi , Ti , Ki ) . (16)
M i=1

Figure 13 provides boxplots and time series plots to compare the in-sample RMSE of
in-sample fit among these four models. Unlike the fluctuating time series plots for the
implied parameters in Figures 9 to 12, the time series plots for the in-sample RMSE
among the four models are rather stable. Furthermore, the SVCJ model is dominating
because it produces the least in-sample RMSE, followed by the SVJ, SJ, and BS model.
Although SVCJ and SVJ models are indistinguishable in most cases, the SVJ model
produces extremely high in-sample RMSE at some times. This is possibly because the
SVJ model includes simply jump components for the return process, but the SVCJ model
includes jumps for both return and volatility processes. Thus, the SVJ model is less
stable in fitting model prices than the SVCJ model. The contrast between the volatile
implied parameters and stable in-sample RMSE is an numerical evidence that multiple
parameters produce the same option prices. This observation is consistent with the
addressed statistical difficulties in fitting a stochastic volatility model with option prices
as in Cont and Tankov (2004) and He et al. (2006).

24

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 13: In-sample root-mean-squared errors. The upper and lower panels provide
boxplots and time-series plots for RMSEs in calibrating intrady inverse option prices
under the BS, SV, SVJ, and SVCJ models. The study period spans from December 1,
2021 to February 28, 2022. Deribit inverse BTC options

Table 4 summarizes the average implied parameters from December 1, 2021 to Febru-
ary 28, 2022. We also convert α and β to κ and θ for better presentations. It is clear that
the estimated parameters differ from those obtained in Heston (1993). This is underlined
by current empirical findings that historical returns of an underlying and options data
yield different parameters. For the BS model, the average implied volatility is 0.0272,
which equals 51.97% on a yearly basis. For comparisons, the average implied volatility
for options in S&P 500 stocks is about 30%.
In the following, our analysis addresses the issue that multiple combinations of pa-
rameters may produce the same price of inverse options. First, we focus on implied
parameters for the SV, SVJ, and SVCJ models. Recall that ρ refers to the correlation
between the Wiener processes of the return and volatility processes. However, the SV,
SVJ, and SVCJ models do not agree the same values for ρ. In fact, the average implied
ρ are 0.9798, 0.2001, and -0.9633 for the SV, SVJ, and the SVCJ models, respectively.

25

Electronic copy available at: https://ssrn.com/abstract=4238213

Second, θ refers to the value of long term volatility, but it is interesting that these three
models do not agree in the value of θ: the average implied θ are 0.0010, 0.0000, and
0.0006 for the SV, SVJ, and SVCJ models. On the other hand, the value of θ can be
compromised by the value of σv , which refers to the volatility of the volatility process:
The higher θ is, the lower σv is. The SV, SVJ, and SVCJ models yield the average values
of σv at 0.0043, 0.3301, 0.0469, respectively.
Between the SVJ and SVCJ models, recall that λ refers to the frequency of jump.
There is a trade-off between the value of θ and λ. The average values of λ are 0.2410
and 0.0077 for the SVJ and SVCJ models, respectively. Compared with the SVJ model,
the SVCJ has higher long-term probability, but it has lower jump frequency. Next, recall
that µy refers to the average jump size of the return process. And there is also a trade-off
between the value of λ and µy . The average values of the µy are 0.0040 and 0.1272 for the
SVJ and SVCJ models, respectively. Compared with the SVJ model, the SVCJ model
has higher frequency of jumps, but has smaller average jump size.

26

Electronic copy available at: https://ssrn.com/abstract=4238213

Table 4: Average implied parameters. This table reports average implied parameters
for the BS, SV, SVJ, and SVCJ models using intraday inverse options calibrated on
a daily basis. The study period spans from December 1, 2021 to February 28, 2022.
Deribit inverse BTC options

BS SV SVJ SVCJ

σ 0.0272
ρ 0.9798 0.2001 -0.9633
α 0.0002 0.0001 0.0017
β 0.7982 -3.5021 -1.8620
V0 0.0013 0.0001 0.0009
σv 0.0043 0.3301 0.0469
λ 0.2410 0.0077
µy 0.0040 0.1272
ρj 0.5722
σy 0.0774 0.1109
µv 0.0465
Reparametrisation
κ 0.2018 4.5021 2.8620
θ 0.0010 0.0000 0.0006

To investigate if the SVCJ model suffers an overfitting problem, we calculate the out-
of-sample RMSE at day t by replacing the parameters implied from day t to those implied
from day (t − 1) in (16). We provide boxplots and time series plots for the out-of-sample
RMSE in Figure 14, and find the SVCJ remains dominating and is followed by the SVJ,
SV, and BS model.

27

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 14: Out-of-sample root-mean-squared errors. The upper and lower panels
provide boxplots and time-series plots for RMSEs in calibrating intrady inverse option
prices under the BS, SV, SVJ, and SVCJ models. The study period spans from December
1, 2021 to February 28, 2022. Deribit inverse BTC options

4 Dynamic delta hedging

In this section, Section 4.1 provides a general Delta formula that is applicable under
complex dynamics. With this Delta formula, we implement dynamic delta hedging using
inverse options and compare whether the SVCJ model outperforms its nested ones in
Section 4.2.

4.1 The Delta formula

The Delta is price sensitivity with respect to the initial underlying price, specifically,

∂
∆= m(S0 , ω, K, T, θ),
∂S0

28

Electronic copy available at: https://ssrn.com/abstract=4238213

where m(S0 , ω, K, T, θ) follows the notations introduced before. When a closed-form
formula of an option price exists, the Delta can be calculated by direct differentiation.
For example, under the BS model, Alexander and Imeraj (2021) provide the Delta formula

2
ωeσ T S −2 KΦ(ωd3 ). (17)

However, the calculate of Delta under a more complicated model is challenging and has
to be calculated with more sophisticated numerical procedures.
In the following, we apply the parameter derivative in Lyuu et al. (2019) to provide
the Delta formula for the inverse option. As an extension to Lyuu and Teng (2011), the
parameter derivative approach is generally applicable and simple to implement. Specif-
ically, it allows us to provide a formula to calculate an unbiased Delta, as long as the
stock prices at maturity can be generated as a function of S0 and a random vector χ, i.e.,
ST = S0 g(χ). Let IA denote the indicator function for set A. Then, we rewrite the payoff
function of the inverse option as a product of smooth function and an indicator function,

℘inverse (ω, K, T ) = ST−1 max {ω(ST − K), 0}

= ST−1 ω(ST − K)I{ω(ST −K)>0}

 
1 K
= ω 1− I{ω(ST −K)>0} .
S0 g(χ)

With the parameter derivative, we obtain the Delta formula:

 
∂℘inverse (ω, K, T )
−rT
∆ = e E
∂S0
   
−rT 1 K
= e E ω I{ω(ST −K)>0}
S02 g(χ)
   
−rT 1 K
= e E ω I{ω(ST −K)>0} . (18)
S0 ST

As a note, the form of the Delta formula is the same under the BS, SV, SVJ, and SVCJ
models, but the differences come from how ST is generated. Under the BS model, when
the expectation is carried out through standard calculus, (18) reduces to (17).

29

Electronic copy available at: https://ssrn.com/abstract=4238213

 For the inverse BTC call option, the Delta can be approximated by

    
−rT 1 K −rT K 1
∆ = e E I{(ST −K)>0} =e E I{(ST −K)>0}
S0 ST S0 ST
K 1
≈ e−rT P r {(ST − K) > 0}
S0 S0
K 1
≤ e−rT , (19)
S0 S0

where P r(E) denotes the probability that an event E occurs. Because S0 is the value of
a BTC denominated in USD, which has the average price of 43,875.72 USD during the
study period, thus Delta is positive but less than about 1
43,375.72
= 2.3 × 10−5 . This is an
extremely small value. Similarly, the inverse BTC put option has an approximate Delta:

K 1
∆ ≈ −e−rT P r {(ST − K) < 0}
S0 S0
K 1
≥ −e−rT . (20)
S0 S0

Thus, an inverse BTC put option has negative Delta but its absolute value is of the scale
of 10−5 .
Figure 15 depicts Deltas under the BS, SV, SVJ, and SVCJ models for different time
to maturity. Each panel presents Deltas with T = 7 (one week), T = 28 (four weeks),
T = 90 (a quarter), and T = 180 (half year), respectively. We use implied parameters
calibrated on January 1, 2022 as the parameter for each model. In addition, we set
S0 = 47128.85, which is the last price of BTC on January 1, 2022. The Delta under the
SV, SVJ, and SVCJ models are calculated using Monte Carlo simulation with a sample
size of N = 10, 000. We overlay a horizontal line in each panel. Recall the Delta formula
(18):

      
−rT 1 K −rT 1 K
∆=e E ω I{ω(ST −K)>0} = ωe E I{ω(ST −K)>0} . (21)
S0 ST S0 ST

Because the expectation in the right-hand side of (21) is positive (assuming BTC would
never has a negative price), the sign of inverse BTC call and put options depend on ω.

30

Electronic copy available at: https://ssrn.com/abstract=4238213

Thus, inverse BTC call and put options have positive and negative Deltas, respectively.
First, Deltas are of extremely small values at a magnitude of 10−5 . This observation
is consistent with our approximations in (19) and (20). Second, for short-term and mid-
term inverse options, i.e., T = 7 and T = 28, Deltas under the four models are extremely
close. For long-term options, i.e., T = 90 and T = 180, Deltas under different models
deviate. However, BS and SV models produce close Deltas, and also the SVJ and SVCJ
models produce almost identical Deltas.

Figure 15: Deltas under the BS, SV, SVJ, and SVCJ models. Each panel presents
Deltas with different time to maturity: T = 7 (one week), T = 28 (four weeks), T = 90 (a
quarter), and T = 180 (half year). We overlay a horizontal line in each panel. Lines above
and below the horizontal line are Deltas for inverse BTC call and put options, respectively.
The Delta under the BS model is in a black-solid line, the Delta under the SV model is in a
orange dashed line, the Delta under the SVJ model is in a red dashed-dotted line, and the
Delta under the SVCJ model is in a cyan dotted line. Deribit inverse BTC options

31

Electronic copy available at: https://ssrn.com/abstract=4238213

4.2 Hedging performance

Now, we describe how to implement the dynamic Delta hedging for inverse options.
Suppose we are hedging on a daily basis, t = 1, . . . , T . Given an option contract, let P (t)
denote its price denoted in BTC at time t. Let ∆(t) denote the Delta of this option at
time t. Consider a Delta-neutral portfolio which is composed of three ingredients:

1. A short position in an inverse option of value P (t)

2. A long position of ∆(t) number of BTC

3. An amount B(t) in the money market denoted in USD

To start, the Delta-neutral portfolio has value denoted in USD:

Π(1) = −P (1)S(1) + ∆(1)S(1) + B(1), (22)

where we set
B(1) = P (1)S(1) − ∆(1)S(1)

to form a self-rebalancing portfolio with zero initial value. Then, the dynamic hedge
procedures are repeated on a daily basis as follows. At day t, we adjust the position of
BTC from ∆(t − 1) to ∆(t) and obtain

Π(t) = −P (t)S(t) + ∆(t)S(t) + B(t),

where value changes in BTC is added to the money market account,

B(t) = er(t−(t−1)) B(t − 1) − S(t)(∆(t) − ∆(t − 1)).

Repeat the above procedures until the time to maturity T . The hedging portfolio value at
maturity is Π(T ). To measure the performance of hedging, we define the relative hedging
error, denoted by π, as
e−rT Π(T )
π= .
P (1)S(1)

32

Electronic copy available at: https://ssrn.com/abstract=4238213

 To fully explore dynamic Delta hedging under various models with intraday transac-
tion, we avoid creating non-existing prices which are not traded. As a result, we rearrange
the data as follows. We pick up the last traded inverse options for each trading day, then,
fixing a trading day in the study period and a maturity-strike combination, we check if
this inverse option has consecutive trades to maturity. If this is the case, we record this
inverse option and implement dynamic delta hedging. Otherwise, we simply ignore it.
We repeat the above procedure for each trading during the study period, and we calculate
the relative hedging errors at various maturity-money categories at each trading day in
the study period.
Figure 16 summarizes the relative hedging error for inverse BTC call options with
boxplots across nine maturity-moneyness categories. First, deep OTM inverse BTC call
options do not have consecutive trades to maturity during our study period. As result,
panels for deep OTM inverse BTC call options are empty. However, for other maturity-
moneyness categories, the distributions of relative hedging errors among the BS, SV,
SVJ, and SVCJ model are indistinguishable. The same phenomena for inverse BTC put
options are observed in Figure 17: Distributions of relative hedging errors with dynamic
Delta hedging for inverse BTC put options under the BS, SV, SVJ, and SVCJ models,
are indistinguishable. Panels for deep OTM inverse put options are empty.
The empirical results that dynamic Delta hedging for inverse options with a more
complicated model does not necessarily outperform that with the simple BS model. These
results are in fact consistent with existing literature for vanilla options Bakshi et al.
(1997). The prediction error caused by a model is one reason why hedging becomes
indistinguishable across models. In the case of inverse options, as shown in Section 4.1,
Deltas for short-term and mid-term options are extremely close. In fact, according to
Table 1, about 80% inverse options are short-term or mid-term. Thus, the majority
inverse options actually use indistinguishable Deltas across models in dynamic Delta
hedging. Thus, we observe indistinguishable relative hedging errors.

33

Electronic copy available at: https://ssrn.com/abstract=4238213

Figure 16: Relative hedging errors for inverse BTC call options. This figure
compares boxplots of relative hedging errors for inverse BTC put options at each maturity-
monyeness cateogory. Panels for deep OTM inverse call options are empty, because there
are no consecutively traded put options. The study period spans from December 1, 2021
to February 28, 2022. Deribit inverse BTC options

Figure 17: Relative hedging errors for inverse BTC put options. This figure
compares boxplots of relative hedging errors for inverse BTC put options at each maturity-
monyeness cateogory. Panels for deep OTM inverse put options are empty, because there
are no consecutively traded put options. The study period spans from December 1, 2021
to February 28, 2022. Deribit inverse BTC options

34

Electronic copy available at: https://ssrn.com/abstract=4238213

5 Conclusion

In this paper, we analyse intraday transaction of inverse BTC optoins traded in Deribit.
We first summarize stylized features of inverse options and find a puzzle of implied volatil-
ity slippery in inverse options. Then, we consider complex models including stochastic
volatility and jumps. We provide a feasible simulation scheme for model calibration. Our
empirical analysis finds that the SVCJ model out-performs its nested models in terms of
in-sample and out-of-sample pricing analysis. Furthermore, we provide the Delta formula
of the inverse option under complex models. The Delta formula allows us to implement
the dynamic hedging routines. In dynamic Delta hedging, the SVCJ model is indistin-
guishable to simpler models in terms of relative hedging errors.
Understanding the cropto currencies and its derivatives is of considerable importance.
Along this line of research, portfolio management, trading, and investor’s behaviour,
deserve further research. It is also worthy of exploring arbitrage opportunity from the
implied volatility slippery and investigating put-call parity for the inverse option. On the
other hand, it is found that multiple parameters may produce the same prices for inverse
options. Thus, it would be interesting to see if the parameter from historical data could
be incorporated when calibrating a model. Second, fast and accurate pricing for inverse
options and higher-order Greeks requires further attention.

References

Alexander, C., J. Deng, J. Feng, and H. Wan (2022). Net buying pressure and the
information in bitcoin option trades. Journal of Financial Markets.

Alexander, C. and A. Imeraj (2021). Inverse options in a Black-Scholes world. Available

at arXiv: https://arxiv.org/abs/2107.12041.

Alexander, C. and A. Imeraj (2022). Delta hedging bitcoin options with a smile. Available
at SSRN: https://ssrn.com/abstract=4097909.

35

Electronic copy available at: https://ssrn.com/abstract=4238213

Bakshi, G., C. Cao, and Z. Chen (1997). Empirical performance of alternative option
pricing models. The Journal of Finance 52 (5), 2003–2049.

Bates, D. S. (1996). Jumps and stochastic volatility: Exchange rate processes implicit in
deutsche mark options. The Review of Financial Studies 9 (1), 69–107.

Belaygoro, A. (2005). Solving continuous time affine jump-diffusion models for econo-
metric inference. Available at https://citeseerx.ist.psu.edu/viewdoc/download
?doi=10.1.1.475.9517&rep=rep1&type=pdf.

Black, F. and M. Scholes (1973). The pricing of options and corporate liabilities. Journal
of Political Economy 81 (3), 637–654.

Cao, M. and B. Celik (2021). Valuation of bitcoin options. Journal of Futures Mar-
kets 41 (7), 1007–1026.

Chen, K.-S. and Y.-C. Huang (2021). Detecting jump risk and jump-diffusion model for
bitcoin options pricing and hedging. Mathematics 9 (20), 2567.

Cont, R. and P. Tankov (2004). Nonparametric calibration of jump-diffusion option

pricing models. The Journal of Computational Finance 7, 1–49.

Cretarola, A., G. Figà-Talamanca, and M. Patacca (2020). Market attention and bitcoin
price modeling: Theory, estimation and option pricing. Decisions in Economics and
Finance 43 (1), 187–228.

Cui, Y., S. del Bano Rollin, and G. Germano (2017). Full and fast calibration of the
heston stochastic volatility model. European Journal of Operational Research 263 (2),
625–638.

Glasserman, P. (2004). Monte Carlo Methods in Financial Engineering. New York:

Springer.

Hacioglu, U. (2019). Blockchain economics and financial market innovation: Financial

innovations in the digital age. Switzerland: Springer.

36

Electronic copy available at: https://ssrn.com/abstract=4238213

Härdle, W. K., C. R. Harvey, and R. C. Reule (2020). Understanding cryptocurrencies.
Journal of Financial Econometrics 18 (2), 181–208.

He, C., J. S. Kennedy, T. F. Coleman, P. A. Forsyth, Y. Li, and K. R. Vetzal (2006).

Calibration and hedging under jump diffusion. Review of Derivatives Research 9 (1),
1–35.

Heston, S. L. (1993). A closed-form solution for options with stochastic volatility with
applications to bond and currency options. The Review of Financial Studies 6 (2),
327–343.

Hou, A. J., W. Wang, C. Y. Chen, and W. K. Härdle (2020). Pricing cryptocurrency

options. Journal of Financial Econometrics 18 (2), 250–279.

Hu, Y., S. T. Rachev, and F. J. Fabozzi (2019). Modelling crypto asset price dynamics,
optimal crypto portfolio, and crypto option valuation. Available at arXiv: https:
//arxiv.org/abs/1908.05419.

Hull, J. (2014). Options, Futures, and Other Derivatives (9 ed.). New York: Prentice
Hall.

Jalan, A., R. Matkovskyy, and S. Aziz (2021). The bitcoin options market: A first look
at pricing and risk. Applied Economics 53 (17), 2026–2041.

Li, L., A. Arab, J. Liu, J. Liu, and Z. Han (2019). Bitcoin options pricing using LSTM-
based prediction model and blockchain statistics. In 2019 IEEE international confer-
ence on Blockchain (Blockchain), pp. 67–74. IEEE.

Lyuu, Y.-D. and H.-W. Teng (2011). Unbiased and efficient Greeks of financial options.
Finance and Stochastics 15 (1), 141–181.

Lyuu, Y.-D., H.-W. Teng, Y.-T. Tseng, and S.-X. Wang (2019). A systematic and efficient
simulation scheme for the Greeks of financial derivatives. Quantitative Finance 19 (7),
1199–1219.

37

Electronic copy available at: https://ssrn.com/abstract=4238213

Madan, D. B., S. Reyners, and W. Schoutens (2019). Advanced model calibration on
bitcoin options. Digital Finance 1 (1), 117–137.

Matic, J. L., N. Packham, and W. K. Härdle (2021). Hedging cryptocurrency options.

Revised and resubmitted to Review of Derivatives Research. Available at arXiv: https:
//arxiv.org/abs/2112.06807.

Merton, R. C. (1976). Option pricing when underlying stock returns are discontinuous.
Journal of Financial Economics 3 (1-2), 125–144.

Mrázek, M., J. Pospı́šil, and T. Sobotka (2016). On calibration of stochastic and fractional
stochastic volatility models. European Journal of Operational Research 254 (3), 1036–
1046.

Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system. Decentralized

Business Review , 21260.

Ross, S. M. (2022). Simulation (6 ed.). London, UK: Academic Press.

Saef, D., O. Nagy, S. Sizov, and W. K. Härdle (2021). Understanding jumps in high
frequency digital asset markets.

Scaillet, O., A. Treccani, and C. Trevisan (2020). High-frequency jump analysis of the
bitcoin market. Journal of Financial Econometrics 18 (2), 209–232.

Siu, T. K. and R. J. Elliot (2021). Bitcoin option pricing with a SETAR-GARCH

model. The European Journal of Finance 27 (6), 564–595.

Teng, H.-W. (2022). Importance sampling for calculating the value-at-risk and expected
shortfall of the quadratic portfolio with t-distributed risk factors. Computational Eco-
nomics.

Zulfiqar, N. and S. Gulzar (2021). Implied volatility estimation of bitcoin options and
the stylized facts of option pricing. Financial Innovation 7 (1), 1–30.

38

Electronic copy available at: https://ssrn.com/abstract=4238213

Appendix

Table 5: Data descriptions. This table lists columns and their descriptions for the intra-
day transaction of Deribit inverse options through the Blockchain-Research-Center.com.

Index Column Type Description

1 q number Quantity
2 p number Option price denoted in BTC
3 s string
4 t integer Timestamp (13 digit UNIX format, millisec-
onds since the UNIX epoch)
5 d string Date (YYYY/mm/dd)
6 trade seq integer The sequence number of the trade within in-
strument
7 trade id string Unique (per currency) trade identifier
8 tick direction integer Direction of the ”tick” (0=Plus Tick, 1 =
Zero-Plus Tick, 2 =Minus Tick, 3 = Zero-
Minus Tick)
9 iv number implied volatility
10 instrument name string UNDERLYING-MATURITY-
STRIKE(Call/Put)
11 index price number the price of the underyling BTC denoted in
USD
12 direction string

39

Electronic copy available at: https://ssrn.com/abstract=4238213