ii macrothink !'1-lnstitute 4, no. 1 · , the flipsi back the v at requires haracterist the...
TRANSCRIPT
Sc
Sc
Receive
doi:10.5
Abstra
Cryptocunprececryptocstatisticcryptocin-depthcryptocinvestorand taimainstrthat crydependiobserveimplicacryptoc
Bitco
chool of En
chool of En
ed: Dec. 12,
5296/ifb.v4
ct
currencies edented grocurrencies ecal analysis currencies. Ah univariat
currencies rs—especiail characterream investayptocurrencing on the ed for cryations for currencies)—
oin and
ngineering, Z
Tel: 41-0-5
E-m
ngineering, Z
, 2016 A
i1.10451
became poowth over exist, with
as well as A particularte and muis investig
ally institutiristics are able asset ccies exhibitparticular c
yptocurrencirisk man
—both from
d Crypt
Fain
Zurich Univ
8-934-4594
mail: resear
Zurich Univ
E-mail: m
Accepted: D
URL
opular withthe last feBitcoin stilan extreme
r focus is onultivariate egated (usinional ones—
of utmostclass, studyit strong nocryptocurrenies that sh
nagement, m an investor
56
tocurren
nt-Hear
Jörg Osterr
versity of A
4 E-mail
Julian Lor
rch@algotra
Martin St
versity of A
martin.strika
ec. 22, 2016
L: http://dx.d
h the emerew years. All being thee value analn the tail riextreme vang both e—as well ast importancing these pron-normal ncies and hhare the sfinancial r’s as well a
ncies—
rted
rieder
Applied Scie
: joerg.oster
renz
adingstrateg
trika
Applied Scie
6 Publish
doi.org/10.5
rgence of As of Nove most poplysis of the isk charactealue analysempirical as regulatorsce. For crroperties is characteris
heavy tails. same underengineering
as from a re
International
—Not fo
ences, Winte
rrieder@zha
gies.com
ences, Winte
hed: Jan. 23
5296/ifb.v4i
Bitcoin anvember 201pular one. W
returns of eristics and sis. The taand Gaussi, an unders
ryptocurrencnecessary.
tics, large Statistical
rlying techg (such aegulator’s po
Finance andISSN 2
2017, Vol.
or the
erthur, Swit
aw.ch
erthur, Swit
3, 2017
i1.10451
nd have sh16, more thWe providethe most imwe will proail dependian copulastanding of cies to beOur findintail depensimilarities
hnology. Tas derivatioint of view
d Banking 2374-2089
4, No. 1
tzerland
tzerland
hown an han 720 e both a mportant ovide an ence of
as). For the risk
ecome a gs show
ndencies, s can be
This has ives on
w. To our
II Macrothink !'1-lnstitute ™
knowlecryptoc
Keywoevents
dge, this icurrencies, t
rds: Bitcoi
is the first their correla
in, Cryptoc
detailed sations and ta
currency, E
57
study lookiail dependen
Extreme va
ing at the ncies as we
lue theory,
International
extreme vll as their st
Risk man
Finance andISSN 2
2017, Vol.
value behavtatistical pro
nagement, E
d Banking 2374-2089
4, No. 1
viour of operties.
Extreme
II Macrothink !'1-lnstitute ™
1. Intro
Since cryptocterms o(CoinMRipple, with thDash, A2016).
A crypcryptogcurrenccurrenc
RegulatWe wilinvestm
The main the fo
Table 1
The marates frhowevethe statcharacteinto cr
oduction an
Bitcoin bcurrencies hof total mar
MarketCap, representine top ten o
Augur, Maid
tocurrency graphy to secy. Cryptocucies.
tors need toll show th
ment or curre
arket capitalollowing Ta
. Market ca
Name
Bitcoin
Ethereu
Ripple
Litecoi
Moner
Ethereu
Dash
Augur
Nem
MaidS
ain purposeom a statis
er the actualtistical properistics. Th
ryptocurrenc
nd Motivat
became theave been crrket value, 2016). The
ng 7.6% andof them (BidSafeCoin,
is a digitaecure the traurrencies ar
o ask themsehat the riskency that is
lization of thable 1 (see C
apitalization
n
um
in
ro
um Classic
afeCoin
e of the papstical point l behavior operties of ce analysis ocies and y
ion
e first dereated. Bitcorepresenting
e second and 2.4% of thtcoin, EtheWaves) rep
al asset deansactions are a subset
elves if crypks of cryptavailable to
he top ten cCoinMarket
n of cryptocu
Market c
11350
932
293
185
89
80
67
53
37
36
per is to unof view. Th
of returns is cryptocurrenof extreme you want t
58
ecentralizedoin, as of Ng over 81%nd third larhe market. T
ereum, Ripppresenting ab
esigned to and to contt of alternat
ptocurrencitocurrencieo retail clien
cryptocurrentCap, 2016)
urrencies
cap (bn USD)
nderstand anhe underlyinot yet wel
ncies and wevents is pato understa
d cryptocuNovember 20% of the totrgest cryptoThere are mple, Litecoinbout 95% o
work as atrol the creative currenc
ies can freels are largents.
ncies as of N):
) Mark
83.4
6.8%
2.2%
1.4%
0.7%
0.6%
0.5%
0.4%
0.3%
0.3%
nd characteing technolll understoowe focus inarticularly rand what r
International
rrency in 016, is the ltal market oocurrencies
more than 72n, Ethereum
of the marke
a medium oation of addcies, or spe
ly be used ber than any
November 8
ket cap, relativ
%
%
%
%
%
%
%
%
%
%
erize cryptoogical issue
od. We are thn particularrelevant if yrisks you
Finance andISSN 2
2017, Vol.
2009, nulargest of itsof cryptocuare Ethere
20 cryptocum Classic, Met (CoinMar
of exchangditional unitecifically o
by retail cuy other tra
8, 2016 can
ve
ocurrency ees are welltherefore der on extremyou are an are facing
d Banking 2374-2089
4, No. 1
umerous s kind in urrencies eum and urrencies Monero, rketCap,
ge using ts of the f digital
stomers. aditional
n be seen
xchange -known,
escribing me value
investor . For a
II Macrothink !'1-lnstitute ™
characteOsterriethem toexchang
The papused ancryptocand resextremethat wedifferenindex. Fcomputvalue difindings
1.1 Wha
A cryptand maCryptocmoney/bitcoin’
1.2 A Sh
In our sare: Bitbecausethe retu
Bitcoin all funcout colgovernmensure digitallyalgorithfrom a designerather th
LitecoinprocessInternetefficien
erization ofeder & Loro the G10 cge rate retur
per is organnd the sourccurrencies, cults from ee events wite own crypnt currencieFurthermorted. A closeistribution ws.
at are Cryp
tocurrency ianaged throcurrencies /centralized ’s blockchai
hort Descrip
study, we atcoin, Litece they have urns.
(BTC) is actions such llectively bment manipthat things
y through ahms and crufiat currenc
ed around thhan relying
n (LTC) issing smallert currency
ntly mined
f the paramerenz (2016)currencies. Frns and con
nized as folces from whcomputes vextreme valth a particu
ptocurrenciees and comre, two imper look at thwill be mad
tocurrencie
is a digital aough the us
use decebanking s
in transactio
ption of the
analyze six coin, Ripplexisted on
a decentralias currency
by the netwpulation or s run smooa “mining”unch numbecy, which is he idea of uon central a
s regarded r transactionbased on th
with cons
etric distribu, the authorFor the ana
nsider the qu
llows. In sehich it was volatilities aue theory t
ular focus ones. We use
mpute the taportant risk he generaliz
de in section
es?
asset designe of advan
entralized systems. Thon database
e Individual
out of the tle, Monero,or before J
ized currency issuance, work. Whilinterference
othly or to ” process thers. These cbacked by
sing cryptoauthorities.
as Bitcoin’ns faster. Ithe Bitcoin sumer-grad
59
utions of crrs analyze
alysis fiat cuestion, wha
ection 2, weretrieved. S
and correlatto analyze tn the left tacopula the
ail dependek measures, zed Pareto n 5. The las
ned to worknced encryp
control ashe decentr
e in the role
Cryptocurr
ten largest c, MaidSafeJune 2014, s
cy that usetransaction le this dece, the flipsiback the v
hat requirescharacteristthe full faitgraphy to c
’s leading t was foundprotocol bu
de hardware
ryptocurrencthe Bitcoin
currencies, Nat flexible d
e give an ovSection 3 lotions. In Sethe behavioail of the diseory to looknce coefficvalue-at-ridistributiont section 6 c
k as a mediuption technis opposedalized contof a distribu
rencies
cryptocurreeCoin, Dashso that we h
s peer-to-peprocessing
centralizatioide is that tvalue of a s powerful ics make Bth and credi
control the c
rival at prded in Octout differs fe. Litecoin
International
cies, see Osn exchange Nadarajah edistribution t
verview of ooks at statiection 4 weour of cryptstribution, ik at the de
cients as wesk and expn and the gconcludes a
um of exchaiques knowd to centtrol is relauted ledger.
ncies. The h. We havehave suffici
eer technolog and verificon renders here is no cBitcoin. Bcomputers
Bitcoin fundit of its govcreation and
esent, and ober 2011. from Bitcoinn provides
Finance andISSN 2
2017, Vol.
sterrieder (2rates and c
et al. (2015to use.
f the data wistical prope will use ctocurrenciesimplicitly aependence bell as the epected shorgeneralized and summar
ange that iswn as cryptotralized elated to the.
selected cue chosen thient data to
ogy, which cation to beBitcoin frecentral auth
Bitcoins are s to solve cdamentally dvernment. Bd transfer of
it is desigIt is a peer
in in that itfaster tran
d Banking 2374-2089
4, No. 1
2016). In compare
5) model
which we erties of concepts s during ssuming between extremal rtfall are extreme
rizes our
s created ography. lectronic
use of
urrencies hose six analyze
enables e carried ee from hority to
created complex different
Bitcoin is f money,
gned for r-to-peer t can be nsaction
II Macrothink !'1-lnstitute ™
confirmtarget thwas to used to
Dash (cryptocDash odecentrfor theAlgorith
MoneroprivacyBitcoindifferen
Ripple Chris Lcurrencof the mutilizing
MaidSaNetworan algorcoin wibeing exSafecoiresourcconnectplatform
2. Data
We are from quprovidindifferencryptocdownlochosen cryptoc
The topAugur, cryptoc
mations andhe regular cprovide a mmine bitcoi
(formerly currency thaoperates a alized auto proof-of-whm family)
o (XMR) isy, decentrali, Monero is
nces relating
(XRP) wasLarsen in 20cy componemain contesg collective
afeCoin (Sark. Safecoinrithm. Onlyill have its oxchanged foins are givees to the ntivity, this pm is referred
a
using histouandl.com, ng a weight
nt online excurrencies boaded from
to start onlcurrencies by
p ten of themMaidSafeC
currencies t
d uses a mcomputers amining algoins.
known as at offers indecentraliz
nomous orgwork. Insteor scrypt it
s an open ssation and ss based on g to blockch
s launched b012. Like Bnt is XRP, wstants for th
ely-trusted s
afecoin) is ans are beingy 4.3 billionown uniquefor network en to users etwork. Thprocess of pd to as farm
orical globawhich showted average
xchanges, cobased on act
June 23, 2ly in June y market ca
m (Bitcoin,Coin, NEM)that we hav
emory-hardand GPUs morithm that
Darkcoinnstant transzed governganization. ead of usint uses 11 rou
ource cryptscalability. Uthe Crypto
hain obfusca
by OpenCoBitcoin, Ripwhich has ahe title of Bubnetworks
a digital cryg distributedn MaidSafeCe identity. Fuservices, soin return foese includeproviding y
ming.
al price indws aggrega
e cryptocurrollate it andtivity, tradin2014 until t2014, becau
apitalization
, Ethereum,) represent mve chosen
60
d, scrypt-bamost people
could run
n and XCsactions, prnance and
Dash uses ng the SHunds of diff
tocurrency Unlike man
oNote protoation.
oin, a comppple is botha mathematiBitcoin sucs within the
yptocurrencd entirely aCoins will eurthermore o there will for them proe free storagyour resourc
dices for crated cryptocrency price.d calculate ang volume,the end of use that all
n as of Nove
, Ripple, Limore than 9out of the
ased miningalready havat the same
oin) is anivate transabudgeting a chained A-256 (fro
ferent hashin
created in ny cryptocucol and pos
pany foundeh a currencyical foundatccessor. It u larger netw
cy token thautomated w
ever be in ciSafecoins wbe a new suoviding somge space, pces and earn
ryptocurrenccurrency pr. They regua weighted liquidity aSeptember
lows us to ember 2016
itecoin, Ethe95% of the e top ten a
International
g proof-of-ve. One of the time, on t
n open soactions andsystem, m
hashing algom well-knng functions
April 2014 rrencies thassesses sign
ed by techny and a paytion like Bituses a consework.
at is in the with no humirculation atwill be recyupply of coime of their processing pning Safeco
cies from thrices from mularly collecaverage prind other far 2016. On analyze six
6.
ereum Clasmarket cap
are: Bitcoin
Finance andISSN 2
2017, Vol.
-work algorthe aims of the same h
ource peerd token funmaking it tgorithm calnown Securs.
that is focat are derivanificant algo
nology entreyment systetcoin. Ripplensus algor
heart of thman intervent one time a
ycled when ins for usersunused com
power, and oin in return
the databasemultiple exct informatiice for the d
actors. Dailyn purpose, wx out of the
ssic, Moneropitalization. n, Dash, L
d Banking 2374-2089
4, No. 1
rithm to Litecoin
hardware
r-to-peer ngibility. the first led X11 re Hash
cused on atives of orithmic
epreneur em. The le is one rithm by
he SAFE ntion by and each they are s to earn. mputing Internet
n on the
e BNC2 xchanges ion from different y data is we have e top ten
o, Dash, The six
LiteCoin,
II Macrothink !'1-lnstitute ™
MaidSa30, 201only cathe mar
3. Stati
Before properti
3.1 Dist
We wanDeviatichoice histograshows Bitcoin/with ba
Figu
The nexthe standistribu
afeCoin, Mo5, Ethereum
ame into exirket capitali
istical Prop
we go into ies of crypto
tribution of
nt to showons of the of using exam of daily a normal d/ USD exch
andwidth mu
ure 1. Histo
xt Figure 2 ndard norm
ution both on
onero and Rm Classic, wistence in 2zation as of
perties of C
the extremeocurrencies
f Returns
w the distribcryptocurre
xtreme valureturns of t
distribution hange ratesultiple 0.5. W
ogram of Bi
shows the mal distribun the left an
Ripple. We hwhich only 2015. In totaf November
ryptocurre
e value behs based on th
bution of reency distrib
ue theory tothe Bitcoin
with mean. The blue We see a su
tcoin/USD Gauss
qq-plot of ution. Againd the right
61
have omittestarted tradal, our choir 2016.
encies
haviour of crheir returns
eturns and bution fromo describe th
/ USD exchn and standline is show
ubstantial de
exchange rsian KDE ov
the empiricin, we obstail.
ed Ethereumding in 2016ice of crypt
ryptocurren.
compare thm the normahe tails. In hange rate. dard deviatwing a Gaueviation from
ate with fittverlaid
cal Bitcoin rserve large
International
m, with an in6, Augur andocurrencies
ncies, we an
hem to a nal distributiFigure 1, wThe red lineion taken f
ussian kernem the norm
ted normal d
returns versdeviations
Finance andISSN 2
2017, Vol.
nitial released NEM, whs represents
nalyze the st
normal distrion will juswe are showe which is ofrom the emel-density e
mal distributi
distribution
sus the quans from the
d Banking 2374-2089
4, No. 1
e in July hich also 88% of
tatistical
ribution. stify our wing the overlaid, mpirical stimator ion.
n and
ntiles of normal
II Macrothink !'1-lnstitute ™
,0.2 -t1
CWA_BTC: histogram ot daity r•lums
0.0 01 02
Fi
The folcryptocrestrict
Figure
Figure
igure 2. QQ
llowing fivcurrencies, Dthe x-axis t
3. Histogram
4. Histogra
-Plot of the
ve Figures 3Dash, Litecto returns be
m of Dash/U
am of LTC/U
e Bitcoin/US
3, 4, 5, 6, coin, MaidSetween -0.2
USD exchanK
USD exchanK
62
SD exchang
7 show thSafeCoin, R and 0.2, th
ange rate witKDEoverlai
nge rate witKDE overlai
ge rate versu
he empiricalRipple and
hus omitting
th fitted norid
th fitted norid
International
us the norm
l histogramMonero. W
g occasional
rmal distrib
rmal distribu
Finance andISSN 2
2017, Vol.
mal distributi
m of our fivWe have chl large outlie
bution and G
ution and G
d Banking 2374-2089
4, No. 1
ion
ve other hosen to ers.
Gaussian
Gaussian
II Macrothink !'1-lnstitute ™
i g
i
h 0
;
0 0
;
.,
-<l.2 .. ,
GWA_BTC
~ ooo•o•
GWA_OASH: hl stogra.m cf dillly returns
GWA_L TC: hlstognm of dillly retums
00 0' 02
Fig
Fig
Figure
We see five chaof the distribu
gure 5. Histo
gure 6. Hist
7. Histogra
a substantiarts, 8, 9, 10standard no
ution both on
ogram of M
togram of X
m of XRP/U
ial deviation0, 11, 12 wormal distrin the left an
MAID/USD eGauss
XMR/USD eGauss
USD exchanK
n from the we show the
ibution. Agnd the right
63
exchange rasian KDEov
exchange rasian KDEov
nge rate witKDEoverlai
normal distqq-plot of
gain, we obtail for all c
ate with fitteverlaid
ate with fitteverlaid
th fitted norid
tribution in the empiric
bserve largecryptocurren
International
ed normal d
ed normal d
rmal distribu
all six plotcal returns ve deviationncies.
Finance andISSN 2
2017, Vol.
distribution
distribution a
ution and G
ts. In the foversus the q
ns from the
d Banking 2374-2089
4, No. 1
and
and
Gaussian
ollowing quantiles normal
II Macrothink !'1-lnstitute ™
<12 .. ,
CWA_MAID: histogram of dai ty retums
00 01 02
GWA_XMR: hl stogram of dalty retums
GWA_XRP: hlstogram of dalty returns
~ilyrnffls
Fi
F
Fig
igure 8. QQ
Figure 9. QQ
gure 10. QQ
Q-Plot of the
Q-Plot of th
Q-Plot of th
e DASH/US
he LTC/USD
e MAID/US
64
SD exchang
D exchange
SD exchang
ge rate versu
e rate versus
ge rate versu
International
us the norma
s the normal
us the norm
Finance andISSN 2
2017, Vol.
al distributi
l distributio
mal distribut
d Banking 2374-2089
4, No. 1
on
on
ion
II Macrothink !'1-lnstitute ™
GWA_DAs.H
•• o. N Q ~ _ .. __ _ ---.-.... -.~~---=----------==-
. Q
i 8 .. 1
. q
.,
G\VA_lTC
......... . . 0 ••••• . ..
., nonnOulllllln
~ --,____··_· ----_----~~"'"'"'"'"'--~·-~~~-------------------------. ·······-
.. nonnOuanlln
Fi
F
3.2 Vola
We are Septemthe follo
Table 2
In the fcryptoc
igure 11. QQ
igure 12. Q
atility
recording hmber 2016. T
owing Table
. Annualize
following Ficurrencies v
Q-Plot of th
QQ-Plot of th
here the volThe annualize 2:
ed volatility
Excha
Bitcoi
Dash/U
Liteco
Maid/U
Moner
Ripple
igure 13, wersus the U
he XMR/US
he XRP/US
latilities of ozed standar
ange rate
n/USD
USD
oin/USD
USD
ro/USD
e/USD
we have plotSD.
65
SD exchang
SD exchange
our cryptocrd deviation
Annualized
62%
116%
97%
138%
149%
116%
tted the 90-
ge rate versu
e rate versu
urrencies frn of returns
Volatility
day rolling
International
us the norma
s the norma
rom June 20over this pe
volatility (
Finance andISSN 2
2017, Vol.
al distributi
al distributio
014 until theriod can be
(annualized)
d Banking 2374-2089
4, No. 1
ion
on
he end of e seen in
) for our
II Macrothink !'1-lnstitute ™
1
. 0 . 0
. 0
a
.,
GW"-.)(RP
i 0 0
~ ~--_·_·_··_·_· ___ _ --~ ----------~ q . ... q
., ., nomiOuanlln
We obsany of t
3.3 Cor
An intepropertiPearsoncorrelatsee forcorrelat(2014/0
Table 3
Table 4
serve that crthe establish
rrelations
eresting quies look likn’s correlatition coefficr instance (tion of the 06 to 2016/0
. Kendall co
Bitco
DashLitec
MAI
MonRipp
. Spearman
Bitco
Dash
LitecMAI
Mon
Ripp
Figure 13.
ryptocurrenhed fiat curr
uestion regke, and we on coefficie
cient (for de(Chok, 201
six crypto09).
orrelation fo
Bitco
oin 1
h 0.255coin 0.584
ID 0.192
ero 0.242ple 0.125
n correlation
Bitcooin 1
h 0.362
coin 0.753ID 0.278
ero 0.351
ple 0.183
90-days rol
ncies have erencies or a
arding crypstart by st
ent , Speaefinitions an10) and (Shocurrencies
or cryptocur
oin Dash
0.255
5 1 4 0.214
2 0.142
2 0.1455 0.116
n for cryptoc
oin Dash0.362
2 1
3 0.3118 0.208
1 0.208
3 0.172
66
lling volatil
enormous voany of the st
ptocurrencitudying cry
arman’s ranknd compariheskin, 200(versus U
rrencies
Litecoin
0.584
0.214 1
0.126
0.180 0.130
currencies
Litecoin 0.753
0.311
1 0.186
0.263
0.186
lity (annuali
olatility whtandard fina
ies is howyptocurrenck correlationisons of the07). Tables
USD) in our
Maid M
0.192 0.
0.142 0.0.126 0.
1 0.
0.147 10.083 0.
Maid M0.278 0.
0.208 0.
0.186 0.1 0.
0.211 1
0.122 0.
International
ized, in pct)
ich is substancial assets
their muly correlation coefficienese three cos 3, 4 and r study ove
Monero Rip
242 0.1
145 0.1180 0.1
147 0.0
0.0054 1
Monero Rip351 0.1
208 0.1
263 0.1211 0.1
0.0
079 1
Finance andISSN 2
2017, Vol.
)
tantially lars.
ltivariate ston in the fnt and Kendorrelation s
d 5 summaer our full
pple
125
116 130
083
054
pple 183
172
186 122
079
d Banking 2374-2089
4, No. 1
rger than
tatistical forms of dall’s tau tatistics,
arize the dataset
II Macrothink !'1-lnstitute ™
2015
90-ditys rolling volo1tllity (o1nnui1Uzed, In pct)
2016
Date
Table 5
Overallshowing
Here, thstatistic
Lookingshows tbe seen
(Gandacrypto-cprior towinner-substitupeople positiveif overavarious provide
. Pearson co
Bitco
Dash
Litec
MAI
Mon
Ripp
l, the levels g the highes
he technicalcal propertie
g at the timthe rolling 9
n, correlation
l & Halabcurrency spo ours): rei-take-all raution (crypseek altern
e correlationall sentimen
cryptocurre alternativ
orrelation fo
Bitco
oin 1
h 0.393
coin 0.611
ID 0.337
ero 0.353
ple 0.152
of correlatist correlatio
l similaritiees.
me-evolution90 day correns vary subs
Fi
burda, 201pace to twoinforcemence against
ptocurrencienative invesn as cryptocnt for cryptrencies are es to Bitc
or cryptocur
oin Dash
0.393
3 1
1 0.248
7 0.194
3 0.175
2 0.110
ion are mixeon (Pearson
s between B
n of correlaelation of Bstantially ov
igure 14. 90
16) have o opposing nt (one or
other crypes treated asstments as currencies ctocurrenciesto a large oin by aim
67
rrencies
Litecoin
0.611
0.248
1
0.214
0.212
0.103
ed and some’s correlatio
Bitcoin and
ations, a mBitcoin (BTCver time.
0-days rollin
attributed effects (albmore speci
ptocurrencies financial aone cryptoc
collectively s is losing extent sim
ming to fix
Maid M
0.337 0.
0.194 0.
0.214 0.
1 0.
0.165 1
0.089 0.
ewhat modeon coefficien
Litecoin ar
more compleC) to the oth
ng correlati
this time-beit investigific cryptoces, leading assets—popcurrency gegain in marground). T
milar in techx what the
International
Monero Rip
353 0.1
175 0.1
212 0.1
165 0.0
0.0
047 1
est—with Bnt of approx
re also clear
ex picture eher 5 crypto
on
changing cgating a difcurrencies o
to negativpularized byets too exprket value, ohis is due t
hnology andeir develop
Finance andISSN 2
2017, Vol.
pple
152
110
103
089
047
Bitcoin and ximately 0.6
rly reflected
emerges. Fiocurrencies
correlation fferent timeoutperformive correlatiy Bitcoin-, ipensive—leaor collectivto the fact d functiona
pers though
d Banking 2374-2089
4, No. 1
Litecoin 6).
d in their
igure 14 . As can
in the e period ing in a on) and in which ading to ely drop that the
ality, but ht to be
II Macrothink !'1-lnstitute ™
shortcomincentiv
In our somewhdroppineffects bfour: Lcryptocpointing
4. Extre
Apart frwe alsorisk mamultipleas the dextremebehavio
4.1 Cop
Copulasdependedecompspecificand, forfor all
A copudistribucontinuprobabiThen and in tthe reprtheoremfunction, … ,exponenbivariatallows tdistribu
mings of thve structure
study perihat stable—ng in the sebetween Bit
Litecoin seecurrencies ag towards th
eme Value
from lookino want to ananagement ae cryptocurrdependencee value theour under ta
pulas
s allows onence from posed into itc, let r let den, …
ula of thutions , …uous randomility integra, … ,this case a cresentation
m: given ann define. For exntially distrte distributithe research
utions, leavi
he original of mining n
od, we can—whereas th
econd half, tcoin and Lems to beare less corhese cryptoc
Analysis of
ng at the pronalyze extrand investmrencies, it is
e of returnseory to anail events.
ne to focuthe univarits univar, … ,note the ma, ∈ ,
e random v… , to them variablesal transform, ~ .copula is anin (1) parti
ny collectioed by (1) abxample, oneributed varion. The abher to direcng only the
Bitcoin prnew coins in
n see the Bhe correlatio
especially Litecoin, ande perceivedrrelated to currencies g
f Cryptocu
operties of treme eventsment purposs important in extremealyse the c
s on the diate distriburiate margin be a rand
arginal distr
vector is joint distris, the copu
ms of the or It is well
ny joint distrcularly usef
on of univarove definese might coriable via ability to coctly draw on task of mo
68
rotocol (forn the netwo
Bitcoin/Liteon of Bitcoin 2016. T
d reinforcemd as the m
Bitcoin—agiving way
urrencies
the cryptocs. Large losses. Furtherm
to considere scenarioscorrelation
dependence utions. An nal distributdom vector ribution of
,s thus a funibution , ula is uniqriginal vari
known tharibution funful for appl
ariate distribs a valid joinombine a
a Clayton combine marn the large deling the d
r instance, ork).
ecoin correoin to the oThis points ment in favomost viableand even nto the other
currency retsses or gainmore, whenr their corre. We will uof cryptoc
of random-dimensio
tions and anwith cumu. Then th
… ,nction that and we wr
que. It is ables, whicat if is nction with lied researchbutions ,nt distributinormally
copula, andrginal distribody of res
dependence
International
slow transa
elation beinother four c
to possiblyor of these twe substituteegatively cr two.
urns over thns are particn considerinlation, com
use techniqucurreny retu
m variables onal joint dn -dimensulative distrhere exists
maps the urite ~the joint d
ch are defincontinuous,0,1 mahers is the c… , and ion with madistributed
d obtain an ibutions witsearch on mstructure.
Finance andISSN 2
2017, Vol.
action time
ng rather hcrypto-currey mild subwo versus t
e, while thcorrelated a
he entire spcularly releng an invest
mmon movesues and tooturns as we
by separadistributionsional copulribution funa copula su
univariate m, … ,distributionned as , then ~argins. Whaconverse ofany copula
arginal distrvariable w
n unusual bith a copulamodeling un
d Banking 2374-2089
4, No. 1
s or the
high and encies is stitution
the other he other at times,
pectrum, evant for tment in s as well ols from ell their
ating the n can be la. To be nction uch that,
(1)
marginal . For n of the
. ~ 0,1 at makes f Sklar’s a , the ributions with an
but valid a model nivariate
II Macrothink !'1-lnstitute ™
Definitiif is cube
-dimen
• ,• 1,…• is
-volum
where t
For ins0, ,Copulasdownsidmultiva
In risk/pthat ardownsid
As far widely in the p
4.1.1 Em
We starcopula.
Definitiinvestigfrom a
copula
Howeveconstru
ion 1 (Copua joint cum0,1 witnsional cop, … , , 0,… ,1, , 1, … -non-deme of is
the
stance, in , 00 for
s are usefude regimes
ariate probab
portfolio mre especialde events m
as derivativused in app
pricing of co
mpirical Co
rt our copul
ion 2 (Emgate the un
random ve
observat
er, the marguct pseudo ∑
ulas). In promulative distth uniformpula if , , … ,… ,1 , th
ecreasing, inon-negati
# :the bivari0 , 1all 0
ul in portfoby allowinbility mode
management,ly importa
may occur (e
ves pricing plications oollateralized
opula
la analysis
mpirical copderlying coector (X1,X
tions wou
ginal distribcopula o
i
obabilistic ttribution fun
m margina
0, the
he copula is
i.e., for eave:
∈. iate case, 1, 1olio and risg the model separately
, copulas arant during e.g., the glob
is concernf financial
d debt obliga
by looking
pula). Whenopula. SuppX2,…,Xd) w
uld be (
bution funcobservations
instead. The
69
terms, : 0nction of a ls. In an
copula is ze
equal to u i
ach hyperre
,: 0,1, 1 a
and 0sk managemelling of they.
re used to p“downside
bal financia
ned, dependrisk assessmations (CDO
g at the sam
n studying pose we hawith continu
(U1i ,U2
i ,…,U
ctions ars by using
en, the pseu
0,1 → 0,-dimensio
nalytic term
ero if one o
if one argum
ectangle
10,1 → 0,
and ,1ment and he marginals
perform stree/crisis/panal crisis of 2
dence modement and acOs).
mple data an
multivariaave observauous margi
Udi )= (F1(X
re usually ng the emp
do copula o
International
1 is a -donal randomms, : 0,1f the argum
ment is u an∏ 0,
1 is a b1.
help us anaand depend
ess-tests andic regimes
2007-2008).
lling with cctuarial ana
nd construct
ate data, oations (X1
i ,Xins. The co
1i ),F2(X2
i ),…
not known. pirical dist
observations
Finance andISSN 2
2017, Vol.
dimensionam vector on1 → 0,1
ments is zero
nd all others, ⊆ 0,
bivariate co,alyse the efdence struct
d robustnesss” where .
copula funcalysis—for e
ting the un
one might X2
i ,…,Xdi ),i
orresponding
…,Fd(Xdi ) ,i
Therefore, tribution fu
s are define
d Banking 2374-2089
4, No. 1
l copula the unit
is a
o
s 1, 1 the
opula if ,ffects of ture of a
s checks extreme
ctions is example
derlying
want to =1,…,n, g "true"
=1,…,n.
one can functions
ed as
II Macrothink !'1-lnstitute ™
U1i,U2
i
then de
copula
Xki :Rk
i =
distribu
One of
Theoremfoundatcumulavector copula distribu
theoremis the Cunique : 0,1cumula
Visualizstructurempiricwe showshows mrelationexhibitsMaidSa
In the Litecoin
,…,Udi
= F
efined as C
samples c
=∑ nj=1 1 Xk
j ≤
ution of the
the most im
m 1 (Sklation for thetive distrib, , … ,
. Indeedution has a, … ,m also statesCartesian pr
if the mar→ 0,1 ative distribu
zing the emre of returnscal copula fow MaidSafemuch a highnship betwes very low afeCoin, Mo
following Fn.
F1n(X1
i ),F2n(X
Cn(u1,…,ud)
can also be
≤Xki . Ther
rank transfo
mportant the
ar). Sklar’s e applicatiobution func
can bed: …a density ⋅s that, giveroduct of thrginals
and marginution functio
mpirical copus. In the folor our six cr
feCoin, Monher correlat
een Bitcoin correlation
onero and R
Figure 15,
X2i ),…,Fd
n(X
)=1
n∑ n
i=1 1
e written a
refore, the
formed data
eorems in co
theorem non of copuction e expressed,, and this ⋅ … ⋅n , the co
he ranges ofare contins thon.
ula for crypllowing Figryptocurrennero and Riion than theand Litecoand large
Ripple.
we have p
70
Xdi ) ,i=1,…,n
U1i≤u1,…,U
as
empirical
a.
opula theory
named afteulas. Sklar’s, … ,d in terms of, … ,
is availab, where opula is uniof the margiuous. The
hen
ptocurrenciegures 15 anncies. First, ipple. We ce second on
oin and, to adispersion,
plotted the
n. The corr
Udi≤ud . Th/ , is
copula ca
y is Sklar’s
er Abe Sks theorem sℙf its margin
. Inble, it holds
is the ique on inal cdf’s. Tconverse i, … ,
es will help d 16 we wiwe have Bi
can see that ne. This is aa lesser ext
indicating
empirical c
International
responding
he compone
the rank o
an be seen
theorem.
klar, providstates that , … ,
nals n case thas further thdensity of ⋯
This impliesis also true
defines
us understaill show theitcoin, Dashthe first of
again an indtent, Dash.
low tail de
copula for
Finance andISSN 2
2017, Vol.
empirical c
ents of the
of the obs
n as the em
des the theevery mult
of a ℙat the multhat , …
f the copu⋯s that the ce: given as a -dime
and the depe three-dimh and Litecof those two dication of tThe secondependence b
Bitcoin, D
d Banking 2374-2089
4, No. 1
copula is
pseudo
ervation
mpirical
eoretical tivariate random and a
tivariate … ,ula. The
, which copula is a copula ensional
endence ensional oin, next copulas
the close d copula between
Dash and
II Macrothink !'1-lnstitute ™
In Figur
Overallcorrelat
4.1.2 G
We wansimultan
Definiti0,1 . probabicopula where the joinvector zanalyticand app
re 16, we ha
Figur
l, we can reptions than th
Gaussian Cop
nt to fit a paneously to t
ion 3 (GauIt is cons
ility integrawith param
is the nt cumulativzero and covcal formulaproximated
Figure 15.
ave plotted
re 16. Empir
produce ourhe other fou
pula
arametric cothe six cryp
ussian copultructed fro
al transformmeter matrix
inverse cumve distributvariance ma
a for the cousing nume
Empirical c
the empiric
rical copula
r earlier resur cryptocur
opula to theptocurrencie
la). The Gam a multiv
m. For a givx can bemulative distion functioatrix equal
opula functierical integr
71
copula for B
cal copula fo
a for MaidS
sult, that Bitrrencies.
e data. For tes under con
aussian copvariate norven correlate written as istribution fon of a multo the correion, ration. The
Bitcoin, Das
for MaidSafe
afeCoin, M
tcoin and Li
this, we usensideration.
pula is a drmal distribtion matrix
function of altivariate nelation matr
, it candensity can
International
sh, Litecoin
feCoin, Mon
onero and R
itecoin have
e the Gaussi
istribution bution over∈ 1,1a standard ormal distrrix . Whiln be upper n be written
Finance andISSN 2
2017, Vol.
n
nero and Rip
Ripple
e higher un
ian copula a
over the unr by us1 , the G, … ,normal and
ribution witle there is noor lower bas
d Banking 2374-2089
4, No. 1
pple.
derlying
and fit it
nit cube sing the
Gaussian ,
d is th mean o simple
bounded,
II Macrothink !'1-lnstitute ™
N 0
0 0
N 0
Q 0
00
"
.. CWU,TC
..
.. ..
Emplrtc.111 copul;a
. \ . Oo , O • II . (II
..
Empl ric.iil copul;a
.. "' "
where
In our Table 6used is
Table 6
The stan
Table 7
Bitc
Dash
Litec
MAI
Mon
Ripp
Also, foLitecoin
is the ide
case, this l6, together wmaximum l
. Coefficien
Bitco
Dash
Litec
MAI
Mon
Ripp
ndard errors
. Standard e
Bitc
oin 0
h 0.02
coin 0.0
ID 0.03
nero 0.02
ple 0.03
or visualizatn and Dash
1entity matrix
leads to a swith their stlikelihood a
nts of Gauss
Bitco
oin 1
h 0.391
coin 0.753
ID 0.315
ero 0.380
ple 0.185
s of our Gau
errors of Ga
coin D
0
28 0
12 0
30 0
28 0
33 0
tion purposin Figure 1
x.
six-dimensiotandard erro
and the log-
sian copula
oin Dash
0.391
1 1
3 0.326
5 0.219
0 0.213
5 0.160
usian copul
aussian copu
Dash L
0.028 0
0 0
0.029 0
0.032 0
0.032 0
0.034 0
es, we show7 as well as
72
12 ⋮onal copulaors in Tablelikelihood r
for cryptoc
Litecoin
0.753
0.326
1
0.220
0.280
0.179
la:
ula for cryp
Litecoin
0.012
0.029
0
0.032
0.030
0.033
w the three-s for MaidS
⋅a for whiche 7. The estresulted in 5
currencies
Maid M
0.315 0.
0.219 0.
0.220 0.
1 0.
0.201 1
0.120 0.
ptocurrencie
Maid
0.030
0.032
0.032
0
0.033
0.034
-dimensionaSafeCoin, Ri
International
⋅ ⋮h we recordtimation pro548.
Monero Rip
380 0.1
213 0.1
280 0.1
201 0.1
0.0
078 1
es
Monero
0.028
0.032
0.030
0.033
0
0.035
al Gaussian ipple and M
Finance andISSN 2
2017, Vol.
d the paramocedure wh
pple
185
160
179
120
078
Ripple
0.033
0.034
0.033
0.034
0.035
0
copula for Moneroin Fig
d Banking 2374-2089
4, No. 1
meters in hich was
e
Bitcoin, gure 18.
II Macrothink !'1-lnstitute ™
4.2 Tail
Apart frextremedependecoeffici(Jawors
The res
Figur
l Dependenc
from the core events isence of theients, usingski, Durante
ults for the
Figure 17.
re 18. Gauss
ce Coefficie
rrelation of s of utmose cryptocurrg the nonpae, & Hardle
lower tail d
Gaussian c
sian copula
ent
the cryptoct importanrency returnarametric e, 2010), com
dependence
73
copula for B
a for MaidSa
currency retnce. For thns. First, westimator omputed for a
can be seen
Bitcoin, Das
afeCoin, Mo
turns, lookinis purpose e compute f (Schmid all pairs of r
n in Table 8
International
sh, Litecoin
onero and R
ng at the bewe are co
the matrix & Schmid
returns.
:
Finance andISSN 2
2017, Vol.
Ripple
ehaviour of onsidering of tail-dep
dt, 2007) (s
d Banking 2374-2089
4, No. 1
f them in the tail endence see also
II Macrothink !'1-lnstitute ™
.. 0
" :; ,r_'
0 0
" 0
0 0
00
.. 0 0
0 Cl. CE: .. >< 0
,r_' ;: 0 Cl
:;i 0
00
0 0
0
0 0 <!>
02
0
00 0 0 0 0
0
0 0 Oo
0.2
o.•
GWA_BTC
0 • oo
• 0
o.•
GWA_t.WD
0
06
0
06
Gaussian copula
0 0 O 0
0.8
0
Gausslan copula
0
f0 0 0 o"
ooooo ooo 0~
• O 0 0 0 • • 0.8
'0
'0
0 0 0 0 oO 0
0 0
oO:t',l>OOOo.o 0 Oo
00 0~ O O O O 0 8 0 0
0 00 \Co O 4) oO 0 0 t" i: 00 oO i 0 0 oo
ooo 03'0 0 0 o 0 o O o o O O (9ct> o o
0 0 00 0 ß, 00 0 O 00 0 0 <f>O O q,> 0
0 ..1. 0 0 0 0 09 0 o o O<) 00 0 0 CE: • 4{,,.,~ qpo8 <P O 0 :ll • '0
>< s. 0 ,r_' • • ;:
<f Cl • 0
'0
Table 8
The cumin
Litecoinwith all
We are the datafound in
Table 9
Using afor Bitc
4.3 Valu
We wacryptocshortfal
Definitiportfoli
. Lower tail
Bitcoin
Dash
Litecoin
MAID
Monero
Ripple
ut-off paramn 100/nobsn have the l other crypt
also compua and returnn (Jaworski
. Lower tail
Bitcoin
Dash
Litecoin
MAID
Monero
Ripple
a t-copula, tcoin and Lit
ue-at-risk a
ant to characurrencies. Tll.
ion 4 (at Rio), then
l dependenc
Bitcoin
1
0.300
0.650
0.272
0.368
0.151
meter in s, 0.1 , werhighest tai
tocurrencies
uting the loning the impi, Durante, &
l dependenc
Bitcoin
1
0.140
0.479
0.102
0.053
0.046
there is onlytecoin.
nd Expected
acterize theThe two mo
Risk). Math is t
ce of crypto
Dash L
0.300 0
1 0
0.270 1
0.116 0
0.164 0
0.024 0
0,1 belore nobs stail dependens.
ower tail depplied tail-de& Hardle, 2
ce of crypto
Dash
0.140
1
0.082
0.048
0.100
0.001
y very low t
d Shortfall
e losses thaost importa
hematically,the negative
74
ocurrencies
Litecoin
0.650
0.270
1
0.187
0.318
0.156
ow which tands for thnce, wherea
pendence cependence c2010) and (S
ocurrencies,
Litecoin
0.479
0.082
1
0.033
0.020
0.048
tail depende
at can occuant tail risk
y, if is ae of the -q∈ |
Maid
0.272
0.116
0.187
1
0.177
0.079
the estimatihe number as Ripple ha
coefficient bcoefficient iSchmid & S
fitting a t-c
Maid
0.102
0.048
0.033
1
0.107
0.000
ency betwee
ur in extremmeasures a
a random vquantile, i.e
International
Monero
0.368
0.164
0.318
0.177
1
0.104
ion takes pof observaas the lowe
by fitting pain Table 9.
Schmidt, 200
copula
Monero
0.053
0.100
0.020
0.107
1
0.005
en all crypt
me events are value-at
variable (e.., 1 .
Finance andISSN 2
2017, Vol.
Ripple
0.151
0.024
0.156
0.079
0.104
1
place is chations. Bitcest tail dep
airwise t coDetails of i07).
Ripple
0.046
0.001
0.048
0.000
0.005
1
tocurrencies
when invet-risk and e
.g., the pri
d Banking 2374-2089
4, No. 1
hosen as coin and endence
opulas to it can be
s, except
esting in expected
ice of a
(2)
II Macrothink !'1-lnstitute ™
Definiti. F
of a loss
(Artznevalue-at
In Tablcryptocwhich cGaussiaobtaininexpecte
where
Table 1
We seealmost one daytranslatGaussiahigher tfrom th
In Tableusing twthe expe
Seconddistribuanalytic
ion 5 (shortFor a givens that excee
er, Delbaent-risk, is a c
le 10, we rcurrencies ucorrespondsan value-at-ng the meaed shortfall,
0.950. Value-at-
e that the vthree times
y, about once to losingan VaR, we than the his
he Gaussian
e 11 below, wo methodsected shortf
, the Gausution to thecally the exp
tfall). Expen random vaeds
n, Eber, & coherent risk
record the under consids to taking -risk, whichan and using the fo
5 denotes t
-risk for cry
Exchange
Bitcoin/U
Dash/USD
Litecoin/U
Maid/USD
Monero/U
Ripple/US
value-at-risk as large as
ce every 20 g more tha
can observtorical VaRdistribution
we record s: First, thefall from the
ssian expece data, obtapected shor
cted shortfaariable ,:
Heath, 19k measure.
value-at-risderation, usthe 95% q
h is computstandard d
formula
.the 95% q
yptocurrenci
e rate Hi
USD 0.0
D 0.
USD 0.0
D 0.
USD 0.
SD 0.
k for MaidSs the one fodays, whenn 4.8% on
ve that thoseR. This shown and the hig
the expectee historical e historical
cted shortfaaining the tfall, using
75
fall estimatethe expecte
|999) argue
sk at the 95sing two m
quantile of tted by first deviation
⋅quantile of t
ies, left tail
istorical VaR
048
106
068
125
125
101
SafeCoin anor Bitcoin. Yn you are invnce every 2e two numb
ws the deviagher risks th
ed shortfall expected shsample.
fall, which and stan
the formula
es the potened shortfall
that expec
5% level fomethods: Fir
the negativefitting a no, and then
0.95the standard
Gaussia
0.054
0.101
0.083
0.123
0.132
0.103
nd MoneroYou can expvested in M20 days. Cbers usuallyation of the hat they bea
at the 95% hortfall, wh
is computndard deviaa
International
ntial size of is defined a
cted shortfa
for Bitcoin st, the histoe exchangeormal distrin computin
d normal dis
an VaR
are the hipect to lose
Monero. For omparing h
y differ. Gaucryptocurrear.
level for ouhich corresp
ted by firstation , an
Finance andISSN 2
2017, Vol.
f the loss exas the expec
fall, as opp
and the otorical valuee rates. Secoibution to tng analytic
stribution.
ighest of ale more thanBitcoin, thihistorical Vussian VaR ency exchan
ur cryptocurponds to com
st fitting a nd then com
d Banking 2374-2089
4, No. 1
xceeding cted size
(3)
posed to
ther five e-at-risk, ond, the the data, ally the
ll, being n 12% in is would VaR and is much
nge rates
rrencies, mputing
normal mputing
II Macrothink !'1-lnstitute ™
where distribu
Table 1
We see Bitcoinlose motruly risGaussiaone wovalue-atthe riskrespecti
Table 1
Expecte
0.ution.
1. Expected
that the exp. Provided ore than 12%sky investman ES is lowould expect-risk and t
k of being loively.
2. Expected
ed shortfall
.95 denot
d shortfall fo
Exchan
Bitcoin/
Dash/U
Litecoin
Maid/U
Monero
Ripple/U
pected shoryou find y% in Mone
ment. Compwer than thect from a the expectedong USD an
d shortfall fo
Exchange ra
Bitcoin/USD
Dash/USD
Litecoin/USD
Maid/USD
Monero/USD
Ripple/USD
for the righ
.tes the 95or cryptocur
nge rate H
/USD
USD
n/USD
USD
o/USD
USD
rtfall is highourself on
ero, you wilparing histoe historical normal disd shortfall fnd short the
for cryptocu
te
D
D
D
ht tail, at the
76
⋅5% expect
rrencies, lef
Historical ES
0.077
0.168
0.135
0.194
0.213
0.180
hest for Monone of thosll actually e
orical ES anES, indicatstribution. for the righcryptocurre
urrencies, rig
Historical Va
0.047
0.083
0.070
0.102
0.103
0.079
e 95% level
0.95ted shortfa
ft tail
Gaussia
0.067
0.127
0.104
0.153
0.164
0.128
nero and abse 1-in-20 dend up losinnd Gaussianting a higheFor compl
ht tail (withency. Those
ght tail
aR Gaus
0.053
0.099
0.084
0.113
0.125
0.098
:
International
all of the
an ES
bout three tidays where ng about 21n ES, we caer risk in cryleteness, w
the sign ree values are
ssian VaR
3
9
4
3
5
8
Finance andISSN 2
2017, Vol.
standard
imes larger e you can e1.3% on thaan observe yptocurrenc
we also receversed), ine in Table 12
d Banking 2374-2089
4, No. 1
normal
than for xpect to at day, a that the
cies than cord the ndicating 2 and 13
II Macrothink !'1-lnstitute ™
Table 1
Bitcoin
4.4 Extr
The extthreshoconversThere adeclustein (Ferrof clustcryptocthe runs
Extrema
Table 1
Exch
Bitco
Dash
Litec
Maid
Mone
Rippl
Table 1
3. Expected
is risky and
remal Index
tremal indeld occurs inse does not are many pering (e.g., ro & Segersters. In the
currency retus method.
al index for
4. Extremal
hange rate
oin/USD
h/USD
coin/USD
d/USD
ero/USD
le/USD
5 shows the
d shortfall fo
Exchan
Bitcoin/
Dash/U
Litecoin
Maid/U
Monero
Ripple/U
d other cryp
x
ex is a un the limit hold) and if
possible estColes & C
s, 2003). It ie following urns. This i
r the left tail
l index for n
Runs declu
0.578
0.506
0.542
0.639
0.602
0.554
e extremal i
for cryptocu
nge rate H
/USD
USD
n/USD
USD
o/USD
USD
ptocurrencie
useful indicof the dist
f 1, thetimators of
Coles, 2001;is unbiased table 14 w
is computed
l:
negative ret
ustering
#clusters
48
42
45
53
50
46
ndex for the
77
urrencies, rig
Historical ES
0.082
0.129
0.116
0.133
0.144
0.128
es are even
cator of hotribution. Fen there is s
f the extrem section 5.3in the mea
we show thd for the 90
turns 90% q
s run leng
4
4
4
4
4
4
e right tail:
ght tail
Gaussia
0.066
0.124
0.105
0.144
0.157
0.123
riskier.
ow much clor independsome depenmal index. 3.2) and then and can b
he extremal % quantile,
quantile
Interv
gth
0.728
0.539
0.787
0.946
0.854
0.725
International
an ES
lustering ofdent data,
ndency (clusThe ones ue intervals ebe used to e
index for , using a run
vals estimator
#cluster
8 54
9 42
7 58
6 73
4 64
5 60
Finance andISSN 2
2017, Vol.
f exceedanc1, (thostering) in thused here aestimator d
estimate the the left tain length of
r
rs run len
3
4
2
1
2
2
d Banking 2374-2089
4, No. 1
ces of a ough the he limit. are runs escribed number
il of the four for
ngth
II Macrothink !'1-lnstitute ™
Table 1
E
r
B
D
L
M
M
R
As far ain cryptcurrencwould clusterin
5. Extre
Extremwe are events, play an distribu
5.1 Cryp
The genIt is usescale
Definitifunction
for ∈ . The Pictheory. distribu
5. Extremal
Exchange
rate
Bitcoin/USD
Dash/USD
Litecoin/USD
Maid/USD
Monero/USD
Ripple/USD
as clusteringtocurrencies
cies, Dash hexpect largng effect.
eme Value
e value disinterested iin particul
n important ution.
yptocurrenci
neralized Paed to mode, and shape
ion 6 (Genn of a gener
when
ckands-BalkIt gives th
ution F of X
l index for p
Runs declu
0.482
0.554
0.530
0.614
0.590
0.494
g of extrems, in particuhas the moge losses t
Distributio
tributions ain the risk car very largrole: the ge
ies and the
areto distribel the tails oe . neralized Pralized Pare
, ,0, an
kema-de Hahe asymptoX is unknow
positive retu
ustering
#clusters
40
46
44
51
49
41
me events is ular in the rist clusterino be cluste
ons
are distributcharacteristige negativeeneralized P
Generalized
bution (GPDof a distribu
Pareto distreto distribut
11 end
aan theoremotic tail diswn. Unlike
78
urns 90% qu
run leng
4
4
4
4
4
4
concerned,ight tail. Cong in the leered). In th
tions characics of crypt
e returns. InPareto distri
d Pareto Di
D) is a famiution and sp
ribution). Ttion is given
1exp/
m is often castribution o for the fir
uantile
Interv
gth
0.564
0.589
0.614
0.627
0.771
0.695
there is clemparing the
eft tail (i.e.,he right ta
cterizing thtocurrenciesn extreme vibution and
istribution
ly of continpecified by
The cumulan by
forforwhen
alled the secof a randomst theorem
International
vals estimator
#cluster
4 46
9 46
4 51
7 51
58
5 56
ear evidencee extremal i, when beinil, Bitcoin
e tails of a s, we are fovalue theorythe general
nuous probay three param
ative proba
00
0, where
cond theoremm variable (the Fisher
Finance andISSN 2
2017, Vol.
r
rs run lengt
3
4
3
4
3
2
e for that toindices for dng long Dashows the
distributionocusing on y, two distrlized extrem
ability distriameters: loc
ability dist
∈ ,m in extremX, when
r-Tippett-Gn
d Banking 2374-2089
4, No. 1
th
o happen different ash, you e largest
n. Since extreme
ributions me value
ibutions. ation , tribution
0 and
me value the true nedenko
II Macrothink !'1-lnstitute ™
theorem
The Pictheory. distributheorem
TheoremidenticafunctionHaan shwell app
where
if if special Haan thevents. left tail threshois givenbased o
Table 1
ExchBitcoDashLitecMaidMoneRipp
The crystandardwith thFigure
m) in extrem
ckands-BalkIt gives th
ution F of Xm) in extrem
m 2 (Pickanally-distribun. In (Balkehow that foproximated
0and
0. Here case of the
heorem is sWe are fittof the distr
ld value whn in table 16on (Hosking
6. Maximum
hange rate oin/ USD h/ USD coin/ USD d/ USD ero/ USD le/ USD
yptocurrencyd error of 0
he shape pa19) versus
me value the
kema-de Hahe asymptoX is unknow
me value the
nds-Balkemuted randomema & de Hor a large cld by the gene
0, and e generalizesometimes ting a GPD ribution of rhich is impl6. The estimg & Wallis, 1
m likelihoo
0.272 0.111 0.233 -0.043 -0.042 0.264
y Bitcoin h0.115. For iarameter
the parame
eory, the inte
aan theoremotic tail diswn. Unlike
eory, the inte
ma-de Haan)m variables,Haan, 1974lass of undeeralized Par→
,,0 when
ed Pareto diused to jusdistribution
returns. Theied by thos
mation proce1987).
d fit of gene
0.019 0.035 0.029 0.040 0.044 0.029
as the higheillustrative p0.272 aneters
79
erest here is
m is often castribution o for the firerest here is
) Let , …, and let
4) and (Pickerlying distreto distribu
, ,1 11 exp
n 0 anistribution istify the usn to the 15e parametere 150 retu
edure is usin
eralized Par
s.e.0.1150.0950.1090.0610.0800.113
est shape papurposes, wnd the scal0.042 and
s in the valu
alled the secof a randomst theorem s in the valu… , be a
be their ckands, 1975tribution funution. That as → ∞
p
nd 0is a power-lse of a pow50 largest drs, together urns as wellng the maxi
reto distribu
s.e.0.0030.0040.0040.0040.0050.004
arameter ,we show thele parameted 0.04
International
ues above a
cond theoremm variable (the Fisher
ues above a
a sequence oconditional 5), Pickandnctions , ais:
/ whlaw, the Pic
wer-law for daily negatiwith their sas the nega
mum likelih
ution
thresh0.0180.0360.0220.0500.0500.031
, with a vale density of er 0.014 (Monero
Finance andISSN 2
2017, Vol.
threshold.
m in extremX, when
r-Tippett-Gnthreshold.
of independexcess dist
ds, Balkemaand large
hen 0. ckands-Balk
modeling ive returns, standard errative log-likhood metho
hold n-406-338-345-339-325-34
lue of 0.27f a GPD dist19 (Bitcoino/ USD, Fig
d Banking 2374-2089
4, No. 1
me value the true nedenko
dent and tribution a and de
, is
Since a kema-de extreme i.e., the
rors, the kelihood od and is
nllh6 8 5 9 5 1
72 and a tribution n/ USD, gure 20)
II Macrothink !'1-lnstitute ™
in the fMoneroLitecoinsimilar.
The den
F
The den
F
By comwe see x-valuemuch ri
For sak
following co/ USD becn and Ripp Dash is in
nsity of the
Figure 19. D
nsity of the
Figure 20. D
mparing boththat the G
es than the oiskier than t
ke of com
chart so thacomes obviple have simthe middle
GPD distrib
Density of a
GPD distrib
Density of a
h Figures 1GPD distribone for Monthe Monero/
mpleteness,
at the differous from a
milar paramof those tw
bution fitted
a GPD distr0.bution fitted
GPD distrib0.9 and 20, abution for nero/ USD. /USD excha
we also
80
rence betwea graphical meters. Furthwo groups.
d to the Bitc
ribution with.019(Bitcoi
d to the Mo
bution with.044(Moner
and looking the BitcoinThis is indiange rate du
show the
een the Bitcpoint of vihermore, M
coin/ USD e
h parameterin)
nero/ USD
h parametersro)
at the scalen/ USD excative for thuring extrem
GPD dist
International
coin/ USD ew. We als
Monero and
exchange ra
rs 0.27exchange ra
s 0.0es of the x-achange ratehe Bitcoin eme events.
tribution fo
Finance andISSN 2
2017, Vol.
exchange rso see that
MaidSafeC
ate:
72 and
ate:
042 and
axis and thee has muchexchange ra
for the oth
d Banking 2374-2089
4, No. 1
rate and Bitcoin,
Coin are
e y-axis, h larger
ate being
her four
II Macrothink !'1-lnstitute ™
0.IIO 0.02
0.IIO
Oensity for GPO dlstrfbution
0.04 0.00
Oensity for GPD distribution
00>
008 010 012
0 ,o 0 ,s
cryptoc
Figure
F
Figure 2
currencies, u
21. Density
Figure 22. D
23. Density
using the fit
y of a GPD
Density of a
y of a GPD d
tted paramet
distribution
a GPD distr0.
distribution
81
ters from Ta
n with param
ribution with029(Liteco
n with param
able 16:
meters
h parameteroin)
meters
International
0.111 and
rs 0.23
0.043 an
Finance andISSN 2
2017, Vol.
d 0.035
33 and
nd 0.04
d Banking 2374-2089
4, No. 1
5(Dash)
4(Maid)
II Macrothink !'1-lnstitute ™
000
0.DO
000
Oenslty for GPO dlstribution
005
Oenslty for GPO dlstribution
0.05
Oenslty for GPO dlstribution
005
__ ........__ __ _ 010 015
010 015
010 015
Figure 2
We alsothat cothresho
24. Density
o show the porresponds ld value
y of a GPD d
plots of theto 150 eis given in
Figure 2
Figure
distribution
cumulativextremes oftable 16.
25. Excess d
26. Excess
82
n with param
e excess disf the cryp
distribution
distribution
meters
stribution fuptocurrency
function fo
n function fo
International
0.264 and
unc tions returns. T
r Bitcoin
for Dash
Finance andISSN 2
2017, Vol.
0.029 fo
The corres
d Banking 2374-2089
4, No. 1
9(Ripple)
or the ponding
II Macrothink !'1-lnstitute ™
000 002 00,
002
005
Oensity for GPD d istribution
008
0.05
J.(Oft loOSUilt)
010
x (on 1o0 sc.ale)
008 010 0 12 014
0.10 020
0.20 050
Figure 2
Figure
Figure 2
Figure 3
7. Excess d
28. Excess
9. Excess d
30. Excess d
83
distribution f
distribution
distribution
distribution
function for
n function fo
function for
function fo
International
r Litecoin
or Maid
r Monero
or Ripple
Finance andISSN 2
2017, Vol.
d Banking 2374-2089
4, No. 1
II Macrothink !'1-lnstitute ™
0 o'-,-----------~--------~--------~~----------~------' 002 005 0.10 0.20 050
005 0 10 0.20 050
l (on log sale)
005 0.10 0.20
:r:(on logsc.ile)
005 0 , . 0.20
x (on log sale)
Using thIn Figuwhich, showingdistribuwith thestraight
Figur
An upwpositivetrend sh
For comand UScorresp
he mean exure 31 we hfor a given g a vertica
ution to the e values t line given
re 31. Mean
ward trend ie gradient ahows thin-ta
mparison, wSD in Figureonding para
xcess plot, ohave plottedthreshold
al line at values exc0.016 a
by
n excess plot
in the plot sabove some ailed behavi
we also showe 32. Here,ameters are
one can visud the mean
, plots the0 andceeding , and 0.2⋅ / 1
t for the Bitthr
shows heavthreshold i
iour wherea
w the mean a threshold0.051
84
ually deciden excess ploe mean valud the meanobtaining t239. Once t
(the blue
tcoin/ USD reshold
vy-tailed behs a sign of
as a line wit
n excess plod of 01 and
e on the appot of the neue of all retun excess lithe scale anthose parame straight lin
exchange r0
haviour. In Pareto beha
th zero grad
ot for the ex is chosen 0.087.
International
propriate chegative of turns exceediine given nd shape pameters are one in Figure
rate with a v
particular, aviour in th
dient shows
xchange ratfor the mea
Finance andISSN 2
2017, Vol.
hoice of a ththe Bitcoin
ding . We by fitting
arameters obtained, wee 31).
vertical line
a straight lhe tail. A do
an exponen
te between an excess li
d Banking 2374-2089
4, No. 1
hreshold. returns, are also a GPD and
e draw a
at the
ine with ownward ntial tail.
Monero ine. The
II Macrothink !'1-lnstitute ™
l'l 0 0
0
"' 0
0 0
0 0 <oo 0
"' 9 0
-0.2 -0.1
Mean excess plot
o„ o,
0,0 0.1 0.2
Threshold
Figur
In Figucryptoc
Figure
e 32. Mean
ures 33, 34,currencies.
e 33. Mean
excess plot
, 35, and 3
excess plot
t for the Mothr
6 we are l
t for the Litethresh
85
onero/ USDreshold
ooking at t
ecoin/ USDhold
exchange r0
the mean e
D exchange r0.004
International
rate with a v
xcess plots
rate with a v
Finance andISSN 2
2017, Vol.
vertical line
s for the oth
vertical line
d Banking 2374-2089
4, No. 1
e at the
her four
e at the
II Macrothink !'1-lnstitute ™
"' 0
"' 0
"' ci
N ci
ci
0
0
0
.(),4
1
.05
0
(1)
~<b
.()2
0
Mean excess plot
0,0
Threshold
Mean excess plot
0.0
Threshold
ll
' 0.5
02 0.4
Figu
Figur
ure 34. Mea
re 35. Mean
an excess plo
n excess plo
ot for the Dthre
ot for the Ripthr
86
Dash/ USD eshold
pple/ USD reshold
exchange ra0.01
exchange ra0
International
ate with a ve
ate with a v
Finance andISSN 2
2017, Vol.
ertical line a
vertical line
d Banking 2374-2089
4, No. 1
at the
at the
II Macrothink !'1-lnstitute ™
0 ... ci
"' "' 0
"' ci
"' 0 ci
..,. 0
... ci
N d
d
0 '<>
-0,3
0
0 00
0
0
-0.2
' <t, o„
0
-04 -02
-0.1
Mean excess plot
0.0
Threshold
Mean excess plot
0.0
Threshold
0.1 02 0.3
(X) 0
02 04
Figure
Note thcharacte
5.2 Cryp
In probfamily combinvalue ditheory sample to one oWeibullis given
The roltheoremfrom anstates thone of normaliconditioapproxi
Theoremidentica
of real
where
36. Mean e
he two groeristics. The
yptocurrenci
bability theoof continu
ne the Gumbistributionsregarding aof iid rando
of three posl distribution to (Gneden
le of the exm for averagny distributhat if the da particul
ized maximons on the timation to m
m 3 (Fisherally distribu
numbers
is a non-
excess plot f
oups whiche second gro
ies and the
ory and stauous probabbel, Fréchet. The Fisherasymptotic om variablessible distribon. Credit fonko, 1948).
xtremal typges, except tion with findistribution ar class of
mum does coail of the di
model the m
r-Tippet-Gnuted random, exis
-degenerate
for the Maidthe t
h we can oup consist
Generalized
atistics, the bility distribt and Weibur-Tippett-Gdistribution
es after propbutions, theor the extrem
pes theoremthat the cennite variancof a norma
f distributioonverge. Thistribution. D
maxima of lo
edenko). Lem variables,
sts such tha
e distributio
87
dSafeCoin/ threshold
detect: Mats of Bitcoin
d Extreme V
generalizedbutions devull families
Gnedenko thn of extremper renormae Gumbel dime value th
m for maximntral limit thce, while thalized maximons. It doehe existenceDespite thisong (finite)
et , , …, and
at each
on function,
USD excha0
aidSafeCoinn, Litecoin,
Value Distri
d extreme vveloped wit also know
heorem is a gme order salization canistribution, heorem (or c
ma is similheorem apphe Fisher-Tmum conve
es not statee of a limit ds, the GEV sequences o… , be a,0 and
, then the li
International
ange rate wi
n and MonDash, Ripp
ibution
value (GEVthin extremn as type I,general resutatistics. Thn only convthe Fréchet convergence
ar to that olies to the aippet-Gnederges, then e that the distributiondistributionof random v
a sequence … , . If a
→∞imit distribu
Finance andISSN 2
2017, Vol.
ith a vertica
nero show ple.
V) distributme value th, II and III ult in extrem
The maximuverge in distt distributione to types th
of the centraverage of adenko theore
the limit hdistribution
n requires ren is often usvariables.
of independa sequence
ution be
d Banking 2374-2089
4, No. 1
al line at
similar
tion is a heory to extreme
me value um of a tribution n, or the heorem)
ral limit a sample em only
has to be n of the egularity sed as an
dent and of pairs
elongs to
l\ Macrothink j Institute™ Mean excess plot
" 0 0
0 0
"" 0 0 Q)
q, 0
"! "<>~ 0
' 0
0
-0.4 -0 2 0.0 02 04
Threshold
either tgeneral
By the properlyvariable
Definitiistributi
for 1parametfor
We are exchangestimatelog-likelikeliho
Table 1
Litecoindistribu39) versin the fMonero
the Gumbelized extrem
extreme vay normalizees.
ion 7 (Genion has cum
/ter and ∈0, it is
fitting a GEge rates vered parameelihood is ood method.
7. Maximum
Exchange r
Bitcoin/USD
Dash/USD
Litecoin/US
Maid/USD
Monero/US
Ripple/USD
n has the hiution with psus the parafollowing co/ USD beco
el, the Frécme value dis
alue theoremed maxima o
neralized emulative dist
, ,/ 0, w∈ the sha/EV distribursus the USDeters, togetgiven in T.with a bloc
m likelihoo
rate
D 0
0
SD 0
0
SD 0
D 0
ighest paparameters ameters hart so thatomes obvio
chet or thetribution.
m the GEV of a sequen
extreme valtribution fun
expexp
where ∈ape parame. For
ution to the D. The blocther with Table 17. Tck size of on
d t of gener
.248 0.029
.236 0.034
.326 0.036
.124 0.029
.003 0.041
.231 0.044
arameter of 0.326,0.003,t the differe
ous from a g
88
e Weibull f
distributionnce of indep
lue distribunction
1exp is the
eter. For 0, ∈ . block maxick maxima their stand
The estimane month of
ralized extre
9 0.050
4 0.079
6 0.062
9 0.112
1 0.112
4 0.084
all cryptocu0.0360.041 anence betwe
graphical po
family. The
n is the onlypendent and
ution).The
fo
location p0, the su
ima of the nare taken odard errorsation procef observatio
eme value d
s.e.
0.289 0
0.196 0
0.182 0
0.140 0
0.147 0
0.205 0
urrencies. W and 0nd 0.1en the Lite
oint of view.
International
ese can be
y possible lidentically
generalized
for 0or 0
parameter, upport is
negative of ver periods s, as welldure is usns.
distribution
s.e. s.e
.006 0.00
.006 0.00
.007 0.00
.005 0.00
.006 0.00
.008 0.01
We show the0.062 (Lite112 (Monercoin/ USD .
Finance andISSN 2
2017, Vol.
e grouped
limit distribdistributed
d extreme
0 th/our cryptoc
s of one monl as the nsing the m
e. nllh
07 -51
08 -46
08 -43
06 -53
09 -45
10 -40
e density ofecoin/ USDro/ USD, Fiexchange
d Banking 2374-2089
4, No. 1
into the
bution of random
value d
he scale , while
currency nth. The negative
maximum
f a GEV D, Figure
igure 41) rate and
II Macrothink !'1-lnstitute ™
Fig
Fi
Figu
gure 37. Den
igure 38. De
ure 39. Den
nsity of a G
ensity of a G
nsity of a GE
EV distribu0.0
GEV distrib0.0
EV distribut0.0
89
ution (Bitco029 and
bution (Dash034 and
tion (Liteco036 and
in/USD) wi0.050
h/USD) with0.079
oin/USD) w0.062
International
ith paramete
h parameter
with paramet
Finance andISSN 2
2017, Vol.
ers 0.2
rs 0.23
ters 0.3
d Banking 2374-2089
4, No. 1
248,
36,
326,
II Macrothink !'1-lnstitute ™
000 005
0 -C
O.!IO 005
0
-0.05 0 00 0.05
Density for GEV distribution
0 ,o 015 0.20 0.25
D•nsity for GEV dlstribution
0 , . 0 ,s 020 0.25
Density for GEV distribution
0.10 0.15 0.20 0.25
Fi
Figu
Fig
By comsee that
gure 40. De
ure 41. Den
gure 42. Den
mparing Figt Monero is
ensity of a G
nsity of a GE
nsity of a G
ures 39 ands riskier tha
GEV distrib0.0
EV distribu0.0
EV distribu0.0d 41, and loan Litecoin,
90
bution (Maid029 and
ution (Mone041 and
ution (Rippl044 and
ooking at th, despite ha
d/USD) wit0.112
ero/USD) w0.112
le/ USD) wi0.084
he scales ofaving the lo
International
h parameter
ith paramet
ith paramete
f the x-axis ower par
Finance andISSN 2
2017, Vol.
rs 0.12
ters 0.0
ers 0.2and the y-
rameter. The
d Banking 2374-2089
4, No. 1
24,
003,
231,
axis, we e higher
II Macrothink !'1-lnstitute ™
005 010
e
CO
<D -.. 0 ...
N
0
0.0 0.1
Density for GEV d istribution
015 0.20 025
Density for GEV distribution
0 2 03 0.4
0.nsity for GEV dlstribution
value fo
6. Conc
Cryptoccryptocview sinBitcoincentral have a June 20volatile
To our kbehaviosubstanequitiesover timreachinLitecoinBitcoin/correlatBoth thcoefficiand expmore th20%. Aclasses.returns likely tothe GPDreturns,Using threturns.not all are not cryptocwhile cin geneliquidityabove.
There acryptocmarket
or for M
clusion
currencies currencies hnce at least , Litecoin, banks worlfull unders
014 until See, much risk
knowledge,our of cryntially larges or commome shows lag very highn which a/Litecoin gtion to all the empiricaient is comppected shorthan 10% onAgain, this r. Using the exhibit cleao happen onD and the , respectivehe GEV dis. The investof the invescontrolled b
currencies aconservativeeral can bey and the lo
are variouscurrency ind
places and
Monero leads
became poave shown
t 2013. We hDash, Monld-wide as
standing of eptember 2
kier and exh
, this is the yptocurrencer than anyodities, witharge swingsh levels. Ouare very generally sthe other cual and the puted both tfall, as the
nce every 20risk is mucextremal inar clusteringn consecutivGEV distrib
ely. Bitcoin,stribution, Btment in crystment is veby an entity
are divided e investors re considereowest annua
s ways for dices from md how the p
s to this high
opular in 2an unprecedhave analyznero, MaidSwell as invthe underly016, we coibit heavier
first study lies. The aof the sta
h the annuas and is veryur cryptocurhighly cor
show a courrencies. InGaussian cwith a paramost comm
0 days, withh higher th
ndex as a mg behaviourve days, anbutions to e, Litecoin aBitcoin, Liteyptocurrenciery high. Iny like a gov
between srather see ited safer coal volatility
future resemultiple excprices differ
91
her risk.
2009 with dented growzed the riskSafeCoin, Rvestors, botying risks o
ould show tr tail behavi
looking at thannualized
andard finanal volatility ry unstable, rrencies are rrelated. Eorrelation an order to copula are ametric and
mon risk meh the loss onhan one wou
measure of cr. In other w
nother elemeexceedance
and Ripple ecoin, Rippies is highlyn contrast tovernment orsupporters wt as a specu
ompared toy of 62% co
earch in thchanges. It r from each
the emergewth, and areks of the sixRipple. It isth institutioof those cuthat cryptocour than the
he statisticavolatility
ncial assetsreaching lewith some mildly corr
Except for above 70%study depefitted to th
d a non-parasures, shown such a dayuld expect
clustering ofwords, you ent of a riskes over a thshow simil
ple and Dashy speculativo typical fiar a bank. Opwhich poinulative bubbo other crypmpared to t
hat area. Wis worthwh
h other. Fur
International
ence of Bie widely ava most impos important onal ones anrrencies. By
currency rete traditional
al propertiesof our c
s, such as tevels beyonof return v
related, excthe secon
%. Ripple ndence we he data. Thrametric mew that you cy then reachin one of tf extreme evexpect that ky asset. Lahreshold andlar, and verh show simve and the riat currenciepinions abount toward thble. The invptocurrencithe other fiv
We have beehile understarthermore, w
Finance andISSN 2
2017, Vol.
itcoin. Sincailable in thortant ones ot for regulatnd retail cly taking daturns are exl fiat curren
s and extremcryptocurrenthe fiat curnd 100%. Vvolatilities rcept for Bitcnd half ofshows theare using
he tail depethod. Valucan expect hing levels the traditionvents, our cextreme ev
astly, we havd monthly ry risky, be
milar and theisk of losinges, cryptocuut an invest
the growingvestment inies due theve cryptocu
en using aganding the dwe have no
d Banking 2374-2089
4, No. 1
ce then, he public of them, tors and ients, to ata from xtremely ncies.
me value ncies is rrencies,
Volatility regularly coin and f 2015,
lowest copulas. endence e-at-risk a loss of of up to
nal asset currency vents are ve fitted maxima
ehaviour. e riskiest g a lot if
urrencies tment in g usage,
n Bitcoin e higher urrencies
ggregate different oted that
II Macrothink !'1-lnstitute ™
Bitcoin technicacorrelat
To sumand beymature in crypt
Referen
Alexandrate https://d
ArtznerMathem
BalkemProbab
Bauer, Busines
Chok, Ncontinu
Chu, J.,PLOS O
CoinMahttps://c
Coles, SSpringe
Coppes47(2), 1
Corlu, distribuhttps://d
Ferro, CRoyal http://dx
Gandal,effects?http://dx
and Litecoal foundatition needs to
mmarize we yond any ofand largest tocurrencies
nces
der, C., & modelling
doi.org/10.1
r, P., Delbamatical Fina
ma, A. A., ility, 2(5), 7
C. (2000). ss, 52(5), 45
N. S. (2010uous data. P
, NadarajahONE, 10(7),
arketCap. coinmarketc
S. (2001). Aer-Verlag Ne
s, R. C. (19117-121. htt
C. G., & ution todoi.org/10.1
C. A. T., & Statistical
x.doi.org/10
, N., & Ha?Competitiox.doi.org/10
oin show sons. Howeo be undert
have showf the risks tcryptocurre
s needs to b
Lazar, E. (2. Journa
1002/jae.849
aen, F., Ebance, 9(3), 2
& Haan de792-804. htt
Value at r55-467. http
0). PearsonPh. D. thesis
h, S., & Cha, e0133678.
(2016). Ccap.com/
An introductew York. ht
95). Are extps://doi.org
Corlu, Ao use?1080/14697
Segers, J. l Society: 0.1111/1467
laburda, H.on in 0.3390/g703
imilar stativer, a thoraken.
wn that crypthat we obseency, is still
be aware of
2006). Normal of 9
ber, J. M.,203-228. htt
e, L. (1974tp://dx.doi:1
risk using hps://doi.org/
’s versus Sps, University
an, S. (2015 https://doi.
Crypto-curre
tion to statisttps://doi.org
xchange ratg/10.1016/0
. (2015.) M? Quan688.2014.9
(2003). InfeSeries B
7-9868.0040
(2016). Cacryp
30016
92
istical proprough study
ptocurrencieerve in tradl one of thethose risks.
mal mixturApplied
& Heath,tps://doi.org
4). Residua10.1214/aop
hyperbolic /10.1016/S0
pearman’s ay of Pittsbur
). Statistica.org/10.137
ency mark
stical modeg/10.1007/9
te changes n165-1765(9
Modelling ntitative 942231
ference for cB (Statist01
an we predptocurrency
erties. Thisy of why t
es show riskditional assee less risky c
re garch (1,Econom
D. (1999).g/10.1111/1
al life time p/11769965
distribution0148-6195(0
and Kendalrgh.
al analysis o1/journal.po
et capitali
eling of extr978-1-4471-
normally di94)00571-I
exchange Finance,
clusters of etical Meth
dict the winmark
International
s is partly bthe returns
k characteriet classes. Bcryptocurren
1): Applicetrics, 2
. Coherent 467-9965.0
at great a48
ns. Journal 00)00026-6
ll’s correlat
of the exchaone.013367
zations. O
reme values -3675-0
istributed? E
rate return15(11
extreme valhodology),
nner in a maket. G
Finance andISSN 2
2017, Vol.
based on thhave such
istics that gBitcoin, as tncies. Any
cations to ex21(3), 3
measures 00068
age. The An
of Econom
tion coeffici
ange rate of78
Online [Av
(2nd ed.). L
Economics
ns: which 1), 185
lues. Journa65(2), 5
arket with Games,
d Banking 2374-2089
4, No. 1
he same h a high
go above the most investor
xchange 307-336.
of risk.
nnals of
mics and
ients for
f bitcoin.
vailable]:
London:
Letters,
flexible 51-1864.
al of the 545-556.
network 7(3),
II Macrothink !'1-lnstitute ™
GnedenNauk, 3
Gurrolacompar10(2), 7
Hoskinggeneralhttp://dx
JaworskProceedSpringehttps://d
Kilic, Rmodels
Linden,exchang573-583
Lopez, peso-do19.
NadarajWhich https://d
Nakajimheavy-tEconom
Osterriehttps://s
OsterrieTail Beh
OsterrieFaint-Hhttps://s
PickandStatistic
Pipien, bayesia
nko, B. V. (3(25), 187-1
a, P. (2007rison with t73. https://d
g, J. R. Mized px.doi.org/10
ki, P., Duradings of ter-Verlag doi.org/10.1
R. (2007). Cwith nigdis
, M. (2005ge rate cha3. https://do
H. F., Rodollarexchang
jah, S., Afuflexible
doi.org/10.1
ma, J. (20tailed errormetrics, 17(5
eder, J. (20ssrn.com/ab
eder, J., & Lhaviour. [O
eder, J., LoHearted. Assrn.com/ab
ds, J. (197cs, 3(1), 119
M. (2004)an analysis f
1948). On a194.
). Capturinthe normalmdoi.org/10.2
M., & Wallpareto 0.2307/1269
ante, F., &the worksh
Berlin 1007/978-3-
Conditional stribution. S
). Estimatinanges. Phy
oi.org/10.10
driguez, B.ge rate: The
uecheta, E., distributio
1080/14697
013). Stochrs for ex5), 499-520
016). The Sbstract=2872
Lorenz, J. (2nline] Avail
renz, J., & Advanced Rbstract=2867
5). Statisti9-131. https
). Garch pforpln/usd e
a local limi
ng fat-tail rmixture and1314/JOR.2
lis, J. R. (distributio
9343
Hardle, Whop held i
and -642-12465
volatility anStudies in No
ng the distysica A: St016/j.physa.2
. D., & Ore floating re
& Chan, S.on to use688.2015.1
hastic volatchangerate
0.
Statistics of2158
2016). A Stlable: https:
Strika, M. Risk & Po7671
cal inferens://doi.org/1
processes wexchange rat
93
it theorem o
risk in excd skewed st2007.163
(1987). Paron. Te
W. K. (201in Warsaw
Heidelbe-5
nd distributNonlinear Dy
tribution oftatistical M2004.12.02
rtiz, A. F.egime in Me
. (2015). A e? Quant032997
tility modereturns.
f Bitcoin an
tatistical Ris://ssrn.com/
(2016). Biortfolio M
nce using e0.1214/aos/
with skewedate. Folia Oe
of the theory
change ratetudent distri
rameter andechnometric
0). Copula w, 25-26 Serg Gm
tion of exchynamics & E
f volatility Mechanics a
4
(2011). Thexico. Inves
note on moitative Fin
el with reStudies in
nd Cryptocu
sk Assessm/abstract=28
itcoin and CManagement
extreme ord/117634300
d-t and staeconomica
International
y of probab
e returns uibutions. Th
d quantile cs, 29
theory anSeptember mbH &
hange rates:Econometri
of realizedand its App
he stochastistigacion Ec
odelling excnance, 15
egime-switchn Nonlinea
urrencies. [
ment of Bitco867339
CryptocurrePaper. [O
der statistic03
able conditiCracoviens
Finance andISSN 2
2017, Vol.
bility. Uspek
using su cuhe Journal
estimation 9(3), 3
nd its appli2009. Hei
& Co.
: Garch andics, 11(3).
d stock retupplications,
ic volatilityconomica,
change rate 5(11), 177
ching skewar Dynami
[Online] Av
oin and Its E
encies—NotOnline] Av
cs. The An
ional distrisia, 45, 45-6
d Banking 2374-2089
4, No. 1
khi Mat.
urves: A of Risk,
for the 339-349.
ications: idelberg:
K.
d figarch
urns and 352(2),
y of the 70(276),
returns: 77-1785.
wness in ics and
vailable:
Extreme
t for the vailable:
nnals of
ibutions. 62.
II Macrothink !'1-lnstitute ™
Schmidrelated http://dx
Sheskin(4th ed.
Copyri
Copyrigthe jour
This is Commo
d, F., & Schmeasures o
x.doi.org/10
n, D. J. (20.). UK: Cha
ight Disclai
ght for this rnal.
an open-aons Attribut
hmidt, R. (2of tail depe0.1016/j.jmv
007). Handbapman & Ha
imer
article is re
access articltion license
2007). Multendence. Jova.2006.05.
book of Paall/CRC.
etained by t
le distribute(http://crea
94
tivariate conJournal of M.005
rametric an
the author(s
ed under thativecommo
nditional veMultivariate
nd Nonpara
s), with firs
he terms anons.org/licen
International
ersions of se Analysis,
ametric Sta
st publicatio
nd conditionses/by/3.0/
Finance andISSN 2
2017, Vol.
spearman’s 98(6), 112
atistical Pro
on rights gr
ons of the C/).
d Banking 2374-2089
4, No. 1
rho and 23-1140.
ocedures
ranted to
Creative
II Macrothink !'1-lnstitute ™