This action might not be possible to undo. Are you sure you want to continue?
Research Notes
Goldman
Sac hs
March 1999
More
Than You Ever Wanted To Know
*
About
Volatility Swaps
Kresimir Demeterﬁ
Emanuel Derman
Michael Kamal
Joseph Zou
_____________
* But Less Than Can Be Said
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
2
Copyr ight 1999 Goldman, Sachs & Co. All r ight s r eser ved.
This mat er ial is for your pr ivat e infor mat ion, and we ar e not solicit ing any act ion based upon it . This r epor t is not t o be const r ued as
an offer t o sell or t he solicit at ion of an offer t o buy any secur it y in any jur isdict ion wher e such an offer or solicit at ion would be
illegal. Cer t ain t r ansact ions, including t hose involving fut ur es, opt ions and high yield secur it ies, give r ise t o subst ant ial r isk and ar e
not suit able for all invest or s. Opinions expr essed ar e our pr esent opinions only. The mat er ial is based upon infor mat ion t hat we
consider r eliable, but we do not r epr esent t hat it is accur at e or complet e, and it should not be r elied upon as such. We, our afﬁliat es,
or per sons involved in t he pr epar at ion or issuance of t his mat er ial, may fr om t ime t o t ime have long or shor t posit ions and buy or
sell secur it ies, fut ur es or opt ions ident ical wit h or r elat ed t o t hose ment ioned her ein.
This mat er ial has been issued by Goldman, Sachs & Co. and/or one of it s afﬁliat es and has been appr oved by Goldman Sachs
Int er nat ional, r egulat ed by The Secur it ies and Fut ur es Aut hor it y, in connect ion wit h it s dist r ibut ion in t he Unit ed Kingdom and by
Goldman Sachs Canada in connect ion wit h it s dist r ibut ion in Canada. This mat er ial is dist r ibut ed in Hong Kong by Goldman Sachs
(Asia) L.L.C., and in J apan by Goldman Sachs (J apan) Lt d. This mat er ial is not for dist r ibut ion t o pr ivat e cust omer s, as deﬁned by
t he r ules of The Secur it ies and Fut ur es Aut hor it y in t he Unit ed Kingdom, and any invest ment s including any conver t ible bonds or
der ivat ives ment ioned in t his mat er ial will not be made available by us t o any such pr ivat e cust omer. Neit her Goldman, Sachs & Co.
nor it s r epr esent at ive in Seoul, Kor ea is licensed t o engage in secur it ies business in t he Republic of Kor ea. Goldman Sachs
Int er nat ional or it s afﬁliat es may have act ed upon or used t his r esear ch pr ior t o or immediat ely following it s publicat ion. For eign
cur r ency denominat ed secur it ies ar e subject t o ﬂuct uat ions in exchange r at es t hat could have an adver se effect on t he value or pr ice
of or income der ived fr om t he invest ment . Fur t her infor mat ion on any of t he secur it ies ment ioned in t his mat er ial may be obt ained
upon r equest and for t his pur pose per sons in It aly should cont act Goldman Sachs S.I.M. S.p.A. in Milan, or at it s London br anch
ofﬁce at 133 Fleet St r eet , and per sons in Hong Kong should cont act Goldman Sachs Asia L.L.C. at 3 Gar den Road. Unless gover ning
law per mit s ot her wise, you must cont act a Goldman Sachs ent it y in your home jur isdict ion if you want t o use our ser vices in
effect ing a t r ansact ion in t he secur it ies ment ioned in t his mat er ial.
Note: Opti ons are not sui table for all i nvestors. Please ensure that you have read and understood the current opti ons
di sclosure document before enteri ng i nto any opti ons transacti ons.
1
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
SUMMARY
Volat ilit y swaps ar e for war d cont r act s on fut ur e r ealized
st ock volat ilit y. Var iance swaps ar e similar cont r act s on var i
ance, t he squar e of fut ur e volat ilit y. Bot h of t hese inst r ument s
pr ovide an easy way for invest or s t o gain exposur e t o t he
fut ur e level of volat ilit y.
Unlike a st ock opt ion, whose volat ilit y exposur e is cont ami
nat ed by it s st ockpr ice dependence, t hese swaps pr ovide pur e
exposur e t o volat ilit y alone. You can use t hese inst r ument s t o
speculat e on fut ur e volat ilit y levels, t o t r ade t he spr ead
bet ween r ealized and implied volat ilit y, or t o hedge t he vola
t ilit y exposur e of ot her posit ions or businesses.
In t his r epor t we explain t he pr oper t ies and t he t heor y of bot h
var iance and volat ilit y swaps, ﬁr st fr om an int uit ive point of
view and t hen mor e r igor ously. The t heor y of var iance swaps
is mor e st r aight for war d. We show how a var iance swap can be
t heor et ically r eplicat ed by a hedged por t folio of st andar d
opt ions wit h suit ably chosen st r ikes, as long as st ock pr ices
evolve wit hout jumps. The fair value of t he var iance swap is
t he cost of t he r eplicat ing por t folio. We der ive analyt ic for mu
las for t heor et ical fair value in t he pr esence of r ealist ic vola
t ilit y skews. These for mulas can be used t o est imat e swap
values quickly as t he skew changes.
We t hen examine t he modiﬁcat ions t o t hese t heor et ical
r esult s when r ealit y int r udes, for example when some neces
sar y st r ikes ar e unavailable, or when st ock pr ices under go
jumps. Finally, we br ieﬂy r et ur n t o volat ilit y swaps, and show
t hat t hey can be r eplicat ed by dynamically t r ading t he mor e
st r aight for war d var iance swap. As a r esult , t he value of t he
volat ilit y swap depends on t he volat ilit y of volat ilit y it self.
_________________
Kr esimir Demet er ﬁ (212) 3574611
Emanuel Der man (212) 9020129
Michael Kamal (212) 3573722
J oseph Zou (212) 9029794
_________________
Acknowledgments: We t hank Emmanuel Boussar d, Llewel
lyn Connolly, Rust om Khandalavala, Cyr us Pir ast eh, David
Roger s, Emmanuel Roman, Pet er Selman, Richar d Sussman,
Nicholas War r en and sever al of our client s for many discus
sions and insight ful quest ions about volat ilit y swaps.
_________________
Edit or ial: Bar bar a Dunn
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
0
Table of Contents
INTRODUCTION ......................................................................................... 1
Volat ilit y Swaps ........................................................................... 1
Who Can Use Volat ilit y Swaps? ................................................. 2
Var iance Swaps ............................................................................ 3
Out line .......................................................................................... 4
I. REPLICATING VARIANCE SWAPS: FIRST STEPS...................................... 6
The Int uit ive Appr oach ............................................................... 6
Tr ading Realized Volat ilit y wit h a Log Cont r act ..................... 11
The Vega, Gamma and Thet a of a Log Cont r act ...................... 11
Imper fect Hedges ...................................................................... 13
The Limit at ions of t he Int uit ive Appr oach .............................. 13
II. REPLICATING VARIANCE SWAPS: GENERAL RESULTS ....................... 15
Valuing and Pr icing t he Var iance Swap.................................. 17
III. AN EXAMPLE OF A VARIANCE SWAP ................................................ 20
IV. EFFECTS OF THE VOLATILITY SKEW ................................................ 23
Skew Linear in St r ike ............................................................... 23
Skew Linear in Delt a ................................................................ 25
V. PRACTICAL PROBLEMS WITH REPLICATION ....................................... 27
Imper fect Replicat ion Due t o Limit ed St r ike Range ............... 27
The Effect of J umps on a Per fect ly Replicat ed Log Cont r act .. 29
The Effect of J umps When Replicat ing Wit h a
Finit e St r ike Range.............................................................. 32
VI. FROM VARIANCE TO VOLATILITY CONTRACTS ................................. 33
Dynamic Replicat ion of a Volat ilit y Swap ............................... 34
CONCLUSIONS AND FUTURE INNOVATIONS ............................................ 36
APPENDIX A: REPLICATING LOGARITHMIC PAYOFFS .............................. 37
APPENDIX B: SKEW LINEAR IN STRIKE .................................................. 40
APPENDIX C: SKEW LINEAR IN DELTA ................................................... 44
APPENDIX D: STATIC AND DYNAMIC REPLICATION OF A
VOLATILITY SWAP .................................................. 48
REFERENCES .......................................................................................... 50
1
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
INTRO DUC TIO N A st ock’s volat ilit y is t he simplest measur e of it s r iskiness or uncer 
t aint y. For mally, t he volat ilit y σ
R
is t he annualized st andar d deviat ion
of t he st ock’s r et ur ns dur ing t he per iod of int er est , wher e t he subscr ipt
R denot es t he obser ved or “r ealized” volat ilit y. This not e is concer ned
wit h volat ilit y swaps and ot her inst r ument s suit able for t r ading vola
t ilit y
1
.
Why t r ade volat ilit y? J ust as st ock invest or s t hink t hey know some
t hing about t he dir ect ion of t he st ock mar ket , or bond invest or s t hink
t hey can for esee t he pr obable dir ect ion of int er est r at es, so you may
t hink you have insight int o t he level of fut ur e volat ilit y. If you t hink
cur r ent volat ilit y is low, for t he r ight pr ice you might want t o t ake a
posit ion t hat pr oﬁt s if volat ilit y incr eases.
Invest or s who want t o obt ain pur e exposur e t o t he dir ect ion of a st ock
pr ice can buy or sell shor t t he st ock. What do you do if you simply want
exposur e t o a st ock’s volat ilit y?
St ock opt ions ar e impur e: t hey pr ovide exposur e t o bot h t he dir ect ion
of t he st ock pr ice and it s volat ilit y. If you hedge t he opt ions accor ding
t o BlackScholes pr escr ipt ion, you can r emove t he exposur e t o t he st ock
pr ice. But delt ahedging is at best inaccur at e because t he r eal wor ld
violat es many of t he BlackScholes assumpt ions: volat ilit y cannot be
accur at ely est imat ed, st ocks cannot be t r aded cont inuously, t r ansac
t ions cost s cannot be ignor ed, mar ket s somet imes move discont inu
ously and liquidit y is oft en a pr oblem. Never t heless, imper fect as t hey
ar e, unt il r ecent ly opt ions wer e t he only volat ilit y vehicle available.
Vola tility Swa p s The easy way t o t r ade volat ilit y is t o use volatility swaps, somet imes
called realized volatility forward contracts, because t hey pr ovide pur e
exposur e t o volat ilit y (and only t o volat ilit y).
A st ock volat ilit y swap is a for war d cont r act on annualized volat ilit y.
It s payoff at expir at ion is equal t o
( EQ 1)
wher e σ
R
is t he r ealized st ock volat ilit y (quot ed in annual t er ms) over
t he life of t he cont r act , K
vol
is t he annualized volat ilit y deliver y pr ice,
and N is t he not ional amount of t he swap in dollar s per annualized vol
at ilit y point . The holder of a volat ilit y swap at expir at ion r eceives N
dollar s for ever y point by which t he st ock’s r ealized volat ilit y σ
R
has
1. For a discussion of volat ilit y as an asset class, see Der man, Kamal,
Kani, McClur e, Pir ast eh, and Zou (1996).
σ
R
K
v ol
– ( ) N ×
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
2
exceeded t he volat ilit y deliver y pr ice K
vol
. He or she is swapping a ﬁxed
volat ilit y K
vol
for t he act ual (“ﬂoat ing”) fut ur e volat ilit y σ
R.
The deliver y pr ice K
vol
is t ypically quot ed as a volat ilit y, for example
30%. The not ional amount is t ypically quot ed in dollar s per volat ilit y
point , for example, N = $250,000/(volat ilit y point ). As wit h all for war d
cont r act s or swaps, t he fair value of volat ilit y at any t ime is t he deliv
er y pr ice t hat makes t he swap cur r ent ly have zer o value.
The pr ocedur e for calculat ing t he r ealized volat ilit y should be clear ly
speciﬁed wit h r espect t o t he following aspect s:
• The sour ce and obser vat ion fr equency of st ock or index pr ices – for
example, using daily closing pr ices of t he S&P 500 index;
• The annualizat ion fact or in moving fr om daily or hour ly obser va
t ions t o annualized volat ilit ies – for example, using 260 business
days per year as a mult iplicat ive fact or in comput ing annualized
var iances fr om daily r et ur ns; and
• Whet her t he st andar d deviat ion of r et ur ns is calculat ed by subt r act 
ing t he sample mean fr om each r et ur n, or by assuming a zer o mean.
The zer o mean met hod is t heor et ically pr efer able, because it cor r e
sponds most closely t o t he cont r act t hat can be r eplicat ed by opt ions
por t folios. For fr equent ly obser ved pr ices, t he differ ence is usually
negligible.
Who C a n Use Vola tility
Swa p s?
Volat ilit y has sever al char act er ist ics t hat make t r ading at t r act ive. It is
likely t o gr ow when uncer t aint y and r isk incr ease. As wit h int er est
r at es, volat ilit ies appear t o r ever t t o t he mean; high volat ilit ies will
event ually decr ease, low ones will likely r ise. Finally, volat ilit y is oft en
negat ively cor r elat ed wit h st ock or index level, and t ends t o st ay high
aft er lar ge downwar d moves in t he mar ket . Given t hese t endencies,
sever al uses for volat ilit y swaps follow.
Di recti onal Tradi ng of Volati li ty Levels. Client s who want t o spec
ulat e on t he fut ur e levels of st ock or index volat ilit y can go long or
shor t r ealized volat ilit y wit h a swap. This pr ovides a much mor e dir ect
met hod t han t r ading and hedging opt ions. For example, if you for esee a
r apid decline in polit ical and ﬁnancial t ur moil aft er a for t hcoming elec
t ion, a shor t posit ion in volat ilit y might be appr opr iat e.
Tradi ng the Spread between Reali zed and Impli ed Volati li ty
Levels. As we will show lat er, t he fair deliver y pr ice K
vol
of a volat ilit y
swap is a value close t o t he level of cur r ent implied volat ilit ies for
opt ions wit h t he same expir at ion as t he swap. Ther efor e, by unwinding
3
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
t he swap befor e expir at ion, you can t r ade t he spr ead bet ween r ealized
and implied volat ilit y.
Hedgi ng Impli ci t Volati li ty Exposure. Ther e ar e sever al busi
nesses t hat ar e implicit ly shor t volat ilit y:
• Risk ar bit r ageur s or hedge funds oft en t ake posit ions which assume
t hat t he spr ead bet ween st ocks of companies planning mer ger s will
nar r ow. If over all mar ket volat ilit y incr eases, t he mer ger may
become less likely and t he spr ead may widen.
• Invest or s following act ive benchmar king st r at egies may r equir e
mor e fr equent r ebalancing and gr eat er t r ansact ions expenses dur 
ing volat ile per iods.
• Por t folio manager s who ar e judged against a benchmar k have t r ack
ing er r or t hat may incr ease in per iods of higher volat ilit y.
• Equit y funds ar e pr obably shor t volat ilit y because of t he negat ive
cor r elat ion bet ween index level and volat ilit y. As global equit y cor r e
lat ions have incr eased, diver siﬁcat ion acr oss count r ies has become a
less effect ive por t folio hedge. Since volat ilit y is one of t he few
par amet er s t hat t ends t o incr ease dur ing global equit y declines, a
long volat ilit y hedge may be appr opr iat e, especially for ﬁnancial
businesses.
Va ria nc e Swa p s Alt hough opt ions mar ket par t icipant s t alk of volat ilit y, it is variance,
or volat ilit y squar ed, t hat has mor e fundament al t heor et ical signiﬁ
cance. This is so because t he cor r ect way t o value a swap is t o value t he
por t folio t hat r eplicat es it , and t he swap t hat can be r eplicat ed most
r eliably (by por t folios of opt ions of var ying st r ikes, as we show lat er ) is
a variance swap.
A var iance swap is a for war d cont r act on annualized var iance, t he
squar e of t he r ealized volat ilit y. It s payoff at expir at ion is equal t o
( EQ 2)
wher e is t he r ealized st ock var iance (quot ed in annual t er ms) over
t he life of t he cont r act , K
var
is t he deliver y pr ice for var iance, and N is
t he not ional amount of t he swap in dollar s per annualized volat ilit y
point squar ed. The holder of a var iance swap at expir at ion r eceives N
dollar s for ever y point by which t he st ock’s r ealized var iance has
exceeded t he var iance deliver y pr ice K
var
.
σ
R
2
K
v ar
– ( ) N ×
σ
R
2
σ
R
2
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
4
Though t heor et ically simpler, var iance swaps ar e less commonly
t r aded, and so t heir quot ing convent ions var y. The deliver y pr ice K
var
can be quot ed as a volat ilit y squar ed, for example (30%)
2
. Similar ly, for
example, t he not ional amount can be expr essed as $100,000/(one vola
t ilit y point )
2
. The fair value of var iance is t he deliver y pr ice t hat makes
t he swap have zer o value.
O utline Most of t his not e will focus on t he t heor y and pr oper t ies of var iance
swaps, which pr ovide similar volat ilit y exposur e t o st r aight volat ilit y
swaps. Because of it s fundament al r ole, var iance can ser ve as t he basic
building block for const r uct ing ot her volat ilit ydependent inst r ument s.
At t he end, we will r et ur n t o a discussion of t he addit ional r isks
involved in r eplicat ing and valuing volat ilit y swaps.
Sect ion I pr esent s an int uit ive, BlackScholesbased account of t he fun
dament al st r at egy by which a var iance swap can be r eplicat ed and val
ued. Fir st , we show t hat t he hedging of a (slight ly) exot ic st ock opt ion,
t he log contract, pr ovides a payoff equal t o t he var iance of t he st ock’s
r et ur ns under a fair ly wide set of cir cumst ances. Then, we explain how
t his exot ic opt ion it self can be r eplicat ed by a por t folio of st andar d
st ock opt ions wit h a r ange of st r ikes, so t hat t heir mar ket pr ices det er 
mine t he cost of t he var iance swap. We also pr ovide insight int o t he
swap by showing, fr om a var iet y of viewpoint s, how t he appar ent ly
complex hedged log cont r act pr oduces an inst r ument wit h t he simple
const ant exposur e t o t he r ealized var iance of a var iance swap.
Sect ion II der ives t he same r esult s much mor e r igor ously and gener 
ally, wit hout depending on t he full validit y of t he BlackScholes model.
Though mor e difﬁcult , t his pr esent at ion is capable of much gr eat er
gener alizat ion.
In Sect ion III, we pr ovide a det ailed numer ical example of t he valua
t ion of a var iance swap. Some pr act ical issues concer ning t he choice of
st r ikes ar e also discussed.
The fair value of t he var iance swap is det er mined by t he cost of t he
r eplicat ing por t folio of opt ions. This cost , especially for index opt ions, is
signiﬁcant ly affect ed by t he volat ilit y smile or skew. Ther efor e, we
devot e Sect ion IV t o t he effect s of t he skew. In par t icular, for a skew
linear in st r ike or linear in delt a, we der ive t heor et ical for mulas t hat
allow us t o simply det er mine t he appr oximat e effect of t he skew on t he
fair value of index var iance swaps, wit hout det ailed numer ical compu
t at ion. The for mulas and t he int uit ion t hey pr ovide ar e beneﬁcial in
r apidly est imat ing t he effect of changes in t he skew on swap values.
5
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
The fair value of a var iance swap is based on (1) t he abilit y t o r eplicat e
a log cont r act by means of a por t folio of opt ions wit h a (cont inuous)
r ange of st r ikes, and (2) on classical opt ions valuat ion t heor y, which
assumes cont inuous st ock pr ice evolut ion. In pr act ice, not all st r ikes
ar e available, and st ock pr ices can jump. Sect ion V discusses t he effect s
of t hese r eal limit at ions on pr icing.
Finally, Sect ion VI explains t he r isks involved in r eplicat ing a volat ilit y
cont r act . Since var iance can be r eplicat ed r elat ively simply, it is useful
t o r egar d volat ilit y as t he squar e r oot of var iance. Fr om t his point of
view, volat ilit y is it self a squar er oot der ivat ive cont r act on var iance.
Thus, a volat ilit y swap can be dynamically hedged by t r ading t he
under lying var iance swap, and it s value depends on t he volat ilit y of t he
under lying var iance – t hat is, on t he volat ilit y of volat ilit y.
Four appendices cover some mor e advanced mat hemat ics. In Appendix
A, we der ive t he det ails of t he r eplicat ion of a log cont r act by means of
a cont inuum of opt ion st r ikes. It also shows how t he r eplicat ion can be
appr oximat ed in pr act ice when only a discr et e set of st r ikes ar e avail
able. In Appendix B, we der ive t he appr oximat e for mulas for t he value
of an index var iance cont r act in t he pr esence of a volat ilit y skew t hat
var ies linear ly wit h st r ike. In Appendix C, we der ive t he analogous for 
mulas for a skew var ying linear ly wit h t he delt a exposur e of t he
opt ions. Appendix D pr ovides addit ional insight int o t he st at ic and
dynamic hedging of a volat ilit y swap using t he var iance as an under 
lyer.
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
6
REPLICATING
VARIANC E SWAPS:
FIRST STEPS
In t his sect ion, we explain t he r eplicat ing st r at egy t hat capt ur es r eal
ized var iance. The cost of implement ing t hat st r at egy is t he fair value
of fut ur e r ealized var iance.
The Intuitive
Ap p roa c h: C re a ting a
Portfolio of O p tions
Whose Va ria nc e Se n
sitivity is Ind e p e nd e nt
of Stoc k Pric e
We appr oach var iance r eplicat ion by building on t he r eader ’s assumed
familiar it y wit h t he st andar d BlackScholes model. In t he next sect ion,
we shall pr ovide a mor e gener al pr oof t hat you can r eplicat e var iance,
even when some of t he BlackScholes assumpt ions fail, as long as t he
st ock pr ice evolves cont inuously – t hat is, wit hout jumps.
We ease t he development of int uit ion by assuming her e t hat t he r isk
less int er est r at e is zer o. Suppose at t ime t you own a st andar d call
opt ion of st r ike K and expir at ion T, whose value is given by t he Black
Scholes for mula , wher e S is t he cur r ent st ock pr ice, τ
is t he t ime t o expir at ion (T − t), is t he r et ur n volat ilit y of t he st ock,
is t he st ock’s var iance, and is t he t ot al var iance of t he
st ock t o expir at ion. (We have wr it t en t he opt ion value as a funct ion of
in or der t o make clear t hat all it s dependence on bot h volat ilit y
and t ime t o expir at ion is expr essed in t he combined var iable .)
We will call t he exposur e of an opt ion t o a st ock’s var iance V ; it mea
sur es t he change in value of t he posit ion r esult ing fr om a change in
var iance
2
. Figur e 1a shows a gr aph of how V var ies wit h st ock pr ice S,
for each of t hr ee differ ent opt ions wit h st r ikes 80, 100 and 120. For
each st r ike, t he var iance exposur e V is lar gest when t he opt ion is at
t he money, and falls off r apidly as t he st ock pr ice moves in or out of t he
money. V is closely r elat ed t o t he t ime sensit ivit y or t ime decay of t he
opt ion, because, in t he BlackScholes for mula wit h zer o int er est r at e,
opt ions values depend on t he t ot al var iance .
If you want a long posit ion in fut ur e r ealized var iance, a single opt ion
is an imper fect vehicle: as soon as t he st ock pr ice moves, your sensit iv
it y t o fur t her changes in var iance is alt er ed. What you want is a por t fo
2. Her e, we deﬁne t he sensit ivit y , wher e
. We will somet imes r efer t o V as “var iance
vega”. Not e t hat d
1
depends only on t he t wo combinat ions S / K and
. V decr eases ext r emely r apidly as S leaves t he vicinit y of t he
st r ike K.
C
BS
S K σ τ , , ( )
σ
σ
2
v σ
2
τ =
σ τ
σ τ
V
σ
2
∂
∂C
B S S τ
2σ

d
1
2
2 ⁄ – ( ) exp
2π
 = =
d
1
S K ⁄ ( ) log σ
2
τ ( ) 2 ⁄ +
σ τ
 =
σ τ
σ
2
τ
7
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
lio whose sensit ivit y t o r ealized var iance is independent of t he st ock
pr ice S.
To obt ain a por t folio t hat r esponds t o volat ilit y or var iance indepen
dent of moves in t he st ock pr ice, you need t o combine opt ions of many
st r ikes. What combinat ion of st r ikes will give you such undilut ed var i
ance exposur e?
Figur e 1b shows t he var iance exposur e for t he por t folio consist ing of all
t hr ee opt ion st r ikes in Figur e 1a. The dot t ed line r epr esent s t he sum of
equally weight ed st r ikes; t he solid line r epr esent s t he sum wit h
weight s in inver se pr opor t ion t o t he squar e of t heir st r ike. Figur es 1c, e
and g show t he individual sensit ivit ies t o var iance of incr easing num
ber s of opt ions, each panel having t he opt ions mor e closely spaced. Fig
ur es 1d, f and h show t he sensit ivit y for t he equallyweight ed and
st r ikeweight ed por t folios. Clear ly, t he por t folio wit h weight s inver sely
pr opor t ional t o pr oduces a V t hat is vir t ually independent of st ock
pr ice S, as long as S lies inside t he r ange of available st r ikes and far
fr om t he edge of t he r ange, and pr ovided t he st r ikes ar e dist r ibut ed
evenly and closely.
Appendix A pr ovides a mat hemat ical der ivat ion of t he r equir ement
t hat opt ions be weight ed inver sely pr opor t ional t o in or der t o
achieve const ant V . You can also under st and t his int uit ively. As t he
st ock pr ice moves up t o higher values, each addit ional opt ion of higher
st r ike in t he por t folio will pr ovide an addit ional cont r ibut ion t o V pr o
por t ional t o t hat st r ike. This follows fr om t he for mula in foot not e 2,
and you can obser ve it in t he incr easing height of t he V peaks in Fig
ur e 1a. An opt ion wit h higher st r ike will t her efor e pr oduce a V cont r i
but ion t hat incr eases wit h S. In addit ion, t he cont r ibut ions of all
opt ions over lap at any deﬁnit e S. Ther efor e, t o offset t his accumulat ion
of S dependence, one needs diminishing amount s of higherst r ike
opt ions, wit h weight s inver sely pr opor t ional t o .
If you own a por t folio of opt ions of all st r ikes, weight ed in inver se pr o
por t ion t o t he squar e of t he st r ike level, you will obt ain an exposur e t o
var iance t hat is independent of st ock pr ice, just what is needed t o t r ade
var iance. What does t his por t folio of opt ions look like, and how does
t r ading it capt ur e var iance?
Consider t he por t folio of opt ions of all st r ikes K and a single
expir at ion τ, weight ed inver sely pr opor t ional t o . Because out of
K
2
K
2
K
2
Π S σ τ , ( )
K
2
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
8
20 60 100 140 180 20 60 100 140 180
20 60 100 140 180 20 60 100 140 180
20 60 100 140 180 20 60 100 140 180
20 60 100 140 180 20 60 100 140 180
strikes: 80,100,120
equally
weighted
strikes 60 to 140
strikes 20 to 180
spaced 1 apart
FIG URE 1. The va ria nc e e xp osure , V , of p ortfolios of c a ll op tions of d iffe re nt
strike s a s a func tion of stoc k p ric e S. Ea c h ﬁg ure on the le ft shows the
ind ivid ua l V
i
c ontrib utions for e a c h op tion of strike K
i
. The c orre sp ond ing
ﬁg ure on the rig ht shows the sum of the c ontrib utions, we ig hte d two
d iffe re nt wa ys; the d otte d line c orre sp ond s to a n e q ua lly we ig hte d sum of
op tions; the solid line c orre sp ond s to we ig hts inve rse ly p rop ortiona l to K
2
,
a nd b e c ome s tota lly ind e p e nd e nt of stoc k p ric e S insid e the strike ra ng e
spaced 20 apart
strikes 60 to 140
spaced 10 apart
(a)
(c)
(e)
(g)
(b)
(d)
(f)
(h)
weighted
inversely
proportional
to square
of strike
9
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
t hemoney opt ions ar e gener ally mor e liquid, we employ put opt ions
for st r ikes K var ying cont inuously fr om zer o up t o some
r efer ence pr ice S
*,
and call opt ions for st r ikes var ying
cont inuously fr om S
*
t o inﬁnit y
3
. You can t hink of S
*
as t he appr oxi
mat e at t hemoney for war d st ock level t hat mar ks t he boundar y
bet ween liquid put s and liquid calls.
At expir at ion, when t · T, one can show t hat t he sum of all t he payoff
values of t he opt ions in t he por t folio is simply
( EQ 3)
wher e log( ) denot es t he nat ur al logar it hm funct ion, and S
T
is t he t er 
minal st ock pr ice.
Similar ly, at t ime t you can sum all t he BlackScholes opt ions values t o
show t hat t he t ot al por t folio value is
( EQ 4)
wher e S is t he st ock pr ice at t ime t. Not e how lit t le t he value of t he
por t folio befor e expir at ion differ s fr om it s value at expir at ion at t he
same st ock pr ice. The only differ ence is t he addit ional value due t o half
t he t ot al var iance .
Clear ly, t he var iance exposur e of Π is
( EQ 5)
To obt ain an init ial exposur e of $1 per volat ilit y point squar ed, you
need t o hold (2/T) unit s of t he por t folio Π. Fr om now on, we will use Π t o
r efer t o t he value of t his new por t folio, namely
3. For mally, t he expr ession for t he por t folio is given by
P S K σ τ , , ( )
C S K σ τ , , ( )
Π S σ τ , ( )
1
K
2
C S K σ τ , , ( )
K S
*
>
∑
1
K
2
 P S K σ τ , , ( )
K S
*
<
∑
+ =
Π S
T
0 , ( )
S
T
S
*
–
S
*

S
T
S
*

. ,
 `
log – =
Π S σ τ , ( )
S S
*
–
S
*

S
S
*

. ,
 `
log –
σ
2
τ
2
 + =
σ
2
τ
V
τ
2
 =
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
10
( EQ 6)
The ﬁr st t er m in t he payoff in Equat ion 6, (S − S
*
)/S
*
, descr ibes 1/S
*
for war d cont r act s on t he st ock wit h deliver y pr ice S
*
. It is not r eally an
opt ion; it s value r epr esent s a long posit ion in t he st ock (value S ) and a
shor t posit ion in a bond (value S
*
), which can be st at ically r eplicat ed,
once and for all, wit hout any knowledge of t he st ock’s volat ilit y. The
second t er m, , descr ibes a shor t posit ion in a log contract
4
wit h r efer ence value S
*
, a socalled exot ic opt ion whose payoff is pr o
por t ional t o t he log of t he st ock at expir at ion, and whose cor r ect hedg
ing depends on t he volat ilit y of t he st ock. All of t he volat ilit y sensit ivit y
of t he weight ed por t folio of opt ions we have cr eat ed is cont ained in t he
log cont r act .
Figur e 2 gr aphically illust r at es t he equivalence bet ween (1) t he
summed, weight ed payoffs of put s and calls and (2) a long posit ion in a
for war d cont r act and a shor t posit ion in a log cont r act .
4. The log cont r act was ﬁr st discussed in Neuber ger (1994). See also Neu
ber ger (1996).
Π S σ τ , ( )
2
T

S S
*
–
S
*

S
S
*

. ,
 `
log –
τ
T
σ
2
+ =
S S
*
⁄ ( ) log –
20 60 100 140 180 20 60 100 140 180
FIG URE 2. An e xa mp le tha t illustra te s the e q uiva le nc e a t e xp ira tion
b e twe e n ( 1) Π( S
T
,0) the we ig hte d sum of p uts a nd c a lls, with we ig ht
inve rse ly p rop ortiona l to the sq ua re of the strike , a nd ( 2) the p a yoff of a
long p osition in a forwa rd c ontra c t a nd a short p osition in a log c ontra c t.
( a ) Ind ivid ua l c ontrib utions to the p a yoff from p ut op tions with a ll inte g e r
strike s from 20 to 99, a nd c a ll op tions with a ll inte g e r strike s from 100 to 180.
( b ) The p a yoff of 1/ 100 of a long p osition in a forwa rd c ontra c t with
d e live ry p ric e 100 a nd one short p osition in a log c ontra c t with re fe re nc e
va lue 100.
S
T
S
*
–
S
*

S
T
S
*
 log –
Π S
T
0 , ( )
( a ) ( b )
11
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Tra d ing Re a lize d
Vola tility with a Log
C ontra c t
For now, assume t hat we ar e in a BlackScholes wor ld wher e t he
implied volat ilit y is t he est imat e of fut ur e r ealized volat ilit y. If you
t ake a posit ion in t he por t folio Π, t he fair value you should pay at t ime
when t he st ock pr ice is S
0
is
At expir at ion, if t he r ealized volat ilit y t ur ns out t o have been , t he
init ial fair value of t he posit ion capt ur ed by delt ahedging would have
been
The net payoff on t he posit ion, hedged t o expir at ion, will be
( EQ 7)
Looking back at Equat ion 2, you will see t hat by rehedging the position
in log contracts, you have, in effect, been the owner of a position in a
variance swap with fair strike K
var
= and face value $1. You will
have pr oﬁt ed (or lost ) if r ealized volat ilit y has exceeded (or been
exceeded by) implied volat ilit y.
The Ve g a , G a mma
a nd The ta of a Log
C ontra c t
In Equat ion 6 we showed t hat , in a BlackScholes wor ld wit h zer o
int er est r at es and zer o dividend yield, t he por t folio of opt ions whose
var iance vega is independent of t he st ock pr ice S can be wr it t en
The t er m r epr esent s a long posit ion in t he st ock and a shor t
posit ion in a bond, bot h of which can be st at ically hedged wit h no
dependence on volat ilit y. In cont r ast , t he log( ) t er m needs cont inual
dynamic r ehedging. Ther efor e, let us concent r at e on t he log cont r act
t er m alone, whose value at t ime t for a logar it hmic payoff at t ime T is
( EQ 8)
σ
I
t 0 =
Π
0
2
T

S
0
S
*
–
S
*

S
0
S
*

. ,
 `
log – σ
I
2
+ =
σ
R
Π
0
2
T

S
0
S
*
–
S
*

S
0
S
*

. ,
 `
log – σ
R
2
+ =
payoff σ
R
2
σ
I
2
– ( ) =
σ
I
2
Π S σ t T , , , ( )
2
T

S S
*
–
S
*

S
S
*

. ,
 `
log –
T t – ( )
T
σ
2
+ =
S S
*
– ( )
L S σ t T , , , ( )
2
T

S
S
*

. ,
 `
log –
T t – ( )
T
σ
2
+ =
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
12
The sensit ivit ies of t he value of t his por t folio ar e pr ecisely appr opr iat e
for t r ading var iance, as we now show.
The var iance vega of t he por t folio in Equat ion 8 is
( EQ 9)
The exposur e t o var iance is equal t o 1 at t = 0, and decr eases linear ly t o
zer o as t he cont r act appr oaches expir at ion.
The t ime decay of t he log cont r act , t he r at e at which it s value changes
if t he st ock pr ice r emains unchanged as t ime passes, is
( EQ 10)
The cont r act loses t ime value at a const ant r at e pr opor t ional t o it s var i
ance, so t hat at expir at ion, all t he init ial var iance has been lost .
The log cont r act ’s exposur e t o st ock pr ice is
shar es of st ock. That is, since each shar e of st ock is wor t h S, you need a
const ant long posit ion in $(2/T) wor t h of st ock t o be hedged at any t ime.
The gamma of t he por t folio, t he r at e at which t he exposur e changes as
t he st ock pr ice moves, is
( EQ 11)
Gamma is a measur e of t he r isk of hedging an opt ion. The log con
t r act ’s gamma, being t he sum of t he gammas of a por t folio of put s and
calls, is a smoot her funct ion of S t han t he shar ply peaked gamma of a
single opt ion.
Equat ions 10 and 11 can be combined t o show t hat
( EQ 12)
Equat ion 12 is t he essence of t he BlackScholes opt ions pr icing t heor y.
It st at es t hat t he disadvant age of negat ive t het a (t he decr ease in value
wit h t ime t o expir at ion) is offset by t he beneﬁt of posit ive gamma (t he
cur vat ur e of t he payoff).
V
T t –
T

. ,
 `
=
θ
1
T
σ
2
– =
∆
2
T

1
S
 – =
Γ
2
T

1
S
2
 =
θ
1
2
ΓS
2
σ
2
+ 0 =
13
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Imp e rfe c t He d g e s It t akes an inﬁnit e number of st r ikes, appr opr iat ely weight ed, t o r epli
cat e a var iance swap. In pr act ice, t his isn’t possible, even when t he
st ock and opt ions mar ket sat isfy all t he BlackScholes assumpt ions:
t her e ar e only a ﬁnit e number of opt ions available at any mat ur it y. Fig
ur e 1 illust r at es t hat a ﬁnit e number of st r ikes fails t o pr oduce a uni
for m V as t he st ock pr ice moves out side t he r ange of t he available
st r ikes. As long as t he st ock pr ice r emains wit hin t he st r ike r ange,
t r ading t he imper fect ly r eplicat ed log cont r act will allow var iance t o
accr ue at t he cor r ect r at e. Whenever t he st ock pr ice moves out side, t he
r educed vega of t he imper fect ly r eplicat ed log cont r act will make it less
r esponsive t han a t r ue var iance swap.
Figur e 3 shows how t he var iance vega of a t hr eemont h var iance swap
is affect ed by imper fect r eplicat ion. Figur e 3a shows t he ideal var iance
vega t hat r esult s fr om a por t folio of put s and calls of all st r ikes fr om
zer o t o inﬁnit y, weight ed in inver se pr opor t ion t o t he st r ike squar ed.
Her e t he var iance vega is independent of st ock pr ice level, and
decr eases linear ly wit h t ime t o expir at ion, as expect ed for a swap
whose value is pr opor t ional t o t he r emaining var iance at any t ime.
Figur e 3b shows st r ikes fr om $75 t o $125, unifor mly spaced $1 apar t .
Her e, deviat ion fr om const ant var iance vega develops at t he t ail of t he
st r ike r ange, and t he deviat ion is gr eat er at ear lier t imes. Finally, Fig
ur e 3c shows t he vega for st r ikes fr om $20 t o $200, spaced $10 apar t .
Now, alt hough t he r ange of st r ikes is gr eat er, t he coar ser spacing
causes t he vega sur face t o develop cor r ugat ions bet ween st r ike values
t hat gr ow mor e pr onounced closer t o expir at ion.
The Limita tions of the
Intuitive Ap p roa c h
A var iance swap has a payoff pr opor t ional t o r ealized var iance. In t his
sect ion, assuming t he BlackScholes wor ld for st ock and opt ions mar 
ket s, we have shown t hat t he dynamic, cont inuous hedging of a log con
t r act pr oduces a payoff whose value is pr opor t ional t o fut ur e r ealized
var iance. We have also shown t hat you can use a por t folio of appr opr i
at ely weight ed put s and calls t o appr oximat e a log cont r act .
The somewhat int uit ive der ivat ions in t his sect ion have assumed t hat
int er est r at es and dividend yields ar e zer o, but it is not har d t o gener 
alize t hem. We have also assumed t hat all t he BlackScholes assump
t ions hold. In pr act ice, in t he pr esence of an implied volat ilit y skew, it
is difﬁcult t o ext end t hese ar gument clear ly. Ther efor e, we move on t o a
mor e gener al and r igor ous der ivat ion of t he value of var iance swaps
based on r eplicat ion. Many of t he r esult s will be similar, but t he condi
t ions under which t hey hold, and t he cor r ect answer s when t hey do not
hold, will be mor e easily under st andable.
σ
2
τ
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
14
FIG URE 3. The va ria nc e ve g a , V , of a p ortfolio of p uts a nd c a lls, we ig hte d
inve rse ly p rop ortiona l to the sq ua re of the strike le ve l, a nd c hose n to
re p lic a te a thre e  month va ria nc e swa p . ( a ) An inﬁnite numb e r of strike s.
( b ) Strike s from $75 to $125, uniformly sp a c e d $1 a p a rt. ( c ) Strike s from $20
to $200, uniformly sp a c e d $10 a p a rt.
v
60
80
100
120
140
S
0
0.1
0.2
τ
v
60
80
100
120
140
S
0
0.1
0.2
τ
v
60
80
100
120
140
S
0
0.1
0.2
τ
V
V
V
( a )
( b )
( c )
15
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
REPLICATING
VARIANC E SWAPS:
G ENERAL RESULTS
In t he pr evious sect ion, we explained how t o r eplicat e a var iance swap
by means of a por t folio of opt ions whose payoffs appr oximat e a log con
t r act . Alt hough our explanat ion depended on t he validit y of t he Black
Scholes model, t he r esult – t hat t he dynamic hedging of a log cont r act
capt ur es r ealized volat ilit y – holds t r ue mor e gener ally. The only
assumpt ion we will make about t he fut ur e under lyer evolut ion is t hat
it is diffusive, or cont inuous – t his means t hat no jumps ar e allowed.
(In a lat er sect ion, we will consider t he effect s of discont inuous st ock
pr ice movement s on t he success of t he r eplicat ion.) Ther efor e, we
assume t hat t he st ock pr ice evolut ion is given by
( EQ 13)
Her e, we assume t hat t he dr ift µ and t he cont inuouslysampled volat il
it y σ ar e ar bit r ar y funct ions of t ime and ot her par amet er s. These
assumpt ions include, but ar e not r est r ict ed t o, implied t r ee models in
which t he volat ilit y is a funct ion of st ock pr ice and t ime only.
For simplicit y of pr esent at ion, we assume t he st ock pays no dividends;
allowing for dividends does not signiﬁcant ly alt er t he der ivat ion.
The t heor et ical deﬁnit ion of r ealized var iance for a given pr ice hist or y
is t he cont inuous int egr al
( EQ 14)
This is a good appr oximat ion t o t he var iance of daily r et ur ns used in
t he cont r act t er ms of most var iance swaps.
Concept ually, valuing a var iance for war d cont r act or “swap” is no dif
fer ent t han valuing any ot her der ivat ive secur it y. The value of a for 
war d cont r act F on fut ur e r ealized var iance wit h st r ike K is t he
expect ed pr esent value of t he fut ur e payoff in t he r iskneut r al wor ld:
( EQ 15)
Her e r is t he r iskfr ee discount r at e cor r esponding t o t he expir at ion
dat e T, and E[ ] denot es t he expect at ion.
The fair deliver y value of fut ur e r ealized var iance is t he st r ike
for which t he cont r act has zer o pr esent value:
( EQ 16)
d S
t
S
t
 µ t ... , ( )d t σ t ... , ( )d Z
t
+ =
σ S t , ( )
V
1
T
 σ
2
t … , ( ) t d
0
T
∫
=
F E e
rT –
V K – ( ) [ ] =
K
v ar
K
v ar
E V [ ] =
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
16
If t he fut ur e volat ilit y in Equat ion 13 is speciﬁed, t hen one appr oach
for calculat ing t he fair pr ice of var iance is t o dir ect ly calculat e t he r isk
neut r al expect at ion
( EQ 17)
No one knows wit h cer t aint y t he value of fut ur e volat ilit y. In implied
t r ee models
5
, t he socalled local volat ilit y consist ent wit h all
cur r ent opt ions pr ices is ext r act ed fr om t he mar ket pr ices of t r aded
st ock opt ions. You can t hen use simulat ion t o calculat e t he fair var i
ance K
var
as t he aver age of t he exper ienced var iance along each simu
lat ed pat h consist ent wit h t he r iskneut r al st ock pr ice evolut ion given
of Equat ion 13, wher e t he dr ift µ is set equal t o t he r iskless r at e.
The above appr oach is good for valuing t he cont r act , but it does not
pr ovide insight int o how t o r eplicat e it . The essence of t he r eplicat ion
st r at egy is t o devise a posit ion t hat , over t he next inst ant of t ime, gen
er at es a payoff pr opor t ional t o t he incr ement al var iance of t he st ock
dur ing t hat t ime
6
.
By applying It o’s lemma t o , we ﬁnd
( EQ 18)
Subt r act ing Equat ion 18 fr om Equat ion 13, we obt ain
( EQ 19)
in which all dependence on t he dr ift has cancelled. Summing Equa
t ion 19 over all t imes fr om 0 t o T, we obt ain t he cont inuouslysampled
var iance
5. See, for example, Der man and Kani (1994), Dupir e (1994) and Der man,
Kani and Zou (1996).
6. This appr oach was ﬁr st out lined in Der man, Kamal, Kani, and Zou
(1996). For an alt er nat ive discussion, see Car r and Madan (1998).
K
v ar
1
T
 E σ
2
t … , ( ) t d
0
T
∫
=
σ S t , ( )
S
t
log
d S
t
log ( ) µ
1
2
σ
2
–
. ,
 `
d t σd Z
t
+ =
d S
t
S
t
 d S
t
log ( ) –
1
2
σ
2
d t =
µ
17
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
( EQ 20)
This mat hemat ical ident it y dict at es t he r eplicat ion st r at egy for var i
ance. The ﬁr st t er m in t he br acket s can be t hought of as t he net out 
come of cont inuous r ebalancing a st ock posit ion so t hat it is always
inst ant aneously long shar es of st ock wor t h $1. The second t er m
r epr esent s a static shor t posit ion in a cont r act which, at expir at ion,
pays t he logar it hm of t he t ot al r et ur n. Following t his cont inuous r ebal
ancing st r at egy capt ur es t he r ealized var iance of t he st ock fr om incep
t ion t o expir at ion at t ime T. Not e t hat no expect at ions or aver ages have
been t aken – Equat ion 20 guar ant ees t hat var iance can be capt ur ed no
mat t er which pat h t he st ock pr ice t akes, as long as it moves cont inu
ously.
Va luing a nd Pric ing
the Va ria nc e Swa p
Equat ion 20 pr ovides anot her met hod for calculat ing t he fair var iance.
Inst ead of aver aging over fut ur e var iances, as in Equat ion 17, one can
t ake t he expect ed r iskneut r al value of t he r ight hand side of Equat ion
20 t o obt ain t he cost of r eplicat ion dir ect ly, so t hat
( EQ 21)
The expect ed value of t he ﬁr st t er m in Equat ion 21 account s for t he
cost of r ebalancing. In a r iskneut r al wor ld wit h a const ant r iskfr ee
r at e r, t he under lyer evolves accor ding t o:
( EQ 22)
so t hat t he r iskneut r al pr ice of t he r ebalancing component of t he hedg
ing st r at egy is given by
( EQ 23)
This equat ion r epr esent s t he fact t hat a shar es posit ion, cont inuously
r ebalanced t o be wor t h $1, has a for war d pr ice t hat gr ows at t he r isk
less r at e.
V
1
T
 σ
2
t d
0
T
∫
≡
2
T

S
t
d
S
t

0
T
∫
S
T
S
0
 log – =
1/S
t
K
v ar
2
T
 E
S
t
d
S
t

0
T
∫
S
T
S
0
 log – =
S
t
d
S
t
 rd t σ t … , ( )d Z + =
E
S
t
d
S
t

0
T
∫
rT =
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
18
As t her e ar e no act ively t r aded log cont r act s for t he second t er m in
Equat ion 21, one must duplicat e t he log payoff, at all st ock pr ice levels
at expir at ion, by decomposing it s shape int o linear and cur ved compo
nent s, and t hen duplicat ing each of t hese separ at ely. The linear compo
nent can be duplicat ed wit h a for war d cont r act on t he st ock wit h
deliver y t ime T; t he r emaining cur ved component , r epr esent ing t he
quadr at ic and higher or der cont r ibut ions, can be duplicat ed using st an
dar d opt ions wit h all possible st r ike levels and t he same expir at ion
t ime T.
For pr act ical r easons we want t o duplicat e t he log payoff wit h liquid
opt ions – t hat is, wit h a combinat ion of out oft hemoney calls for high
st ock values and out oft hemoney put s for low st ock values. We int r o
duce a new ar bit r ar y par amet er S
*
t o deﬁne t he boundar y bet ween
calls and put s. The log payoff can t hen be r ewr it t en as
( EQ 24)
The second t er m is const ant , independent of t he ﬁnal st ock
pr ice S
T
, so only t he ﬁr st t er m has t o be r eplicat ed.
The following mat hemat ical ident it y, which holds for all fut ur e values
of S
T
, suggest s t he decomposit ion of t he logpayoff:
( EQ 25)
Equat ion 25 r epr esent s t he decomposit ion of a log payoff int o a por t fo
lio consist ing of:
• a shor t posit ion in for war d cont r act s st r uck at S
*
;
• a long posit ion in put opt ions st r uck at K, for all st r ikes
fr om 0 t o S
*
; and
• a similar long posit ion in call opt ions st r uck at K, for all
st r ikes fr om S
*
t o .
All cont r act s expir e at t ime T. Figur e 4 shows t his decomposit ion sche
mat ically.
S
T
S
0
 log
S
T
S
*
 log
S
*
S
0
 log + =
S
*
S
0
⁄ ( ) log
S
T
S
*
 log –
S
T
S
*
–
S
*
 (for war d cont r act ) – =
1
K
2
 Max K S
T
– 0 , ( ) K d
0
S
*
∫
+ (put opt ions)
1
K
2
 Max S
T
K – 0 , ( ) K (call opt ions) d
S
*
∞
∫
+
1 S
*
⁄ ( )
1 K
2
⁄ ( )
1 K
2
⁄ ( )
∞
19
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
The fair value of fut ur e var iance can be r elat ed t o t he init ial fair value
of each t er m on t he r ight hand side of Equat ion 21. By using t he ident i
t ies in Equat ions 23 and 25, we obt ain
( EQ 26)
wher e P(K) and (C(K)), r espect ively, denot e t he cur r ent fair value of a
put and call opt ion of st r ike K. If you use t he mar ket pr ices of t hese
opt ions, you obt ain an est imat e of t he cur r ent mar ket pr ice of fut ur e
var iance.
This appr oach t o t he fair value of fut ur e var iance is t he most r igor ous
fr om a t heor et ical point of view, and makes less assumpt ions t han our
int uit ive t r eat ment in t he sect ion on page 6. Equat ion 26 makes pr e
cise t he int uit ive not ion t hat implied volat ilit ies can be r egar ded as t he
mar ket ’s expect at ion of fut ur e r ealized volat ilit ies. It pr ovides a dir ect
connect ion bet ween t he mar ket cost of opt ions and t he st r at egy for cap
t ur ing fut ur e r ealized volat ilit y, even when there is an implied volatility
skew and t he simple BlackScholes for mula is invalid.
K
v ar
2
T
 rT
S
0
S
*
e
rT
1 –
. ,
 `
S
*
S
0
 log – –
.

=
e
rT
1
K
2
 P K ( ) K d
0
S
*
∫
+
e
rT
1
K
2
C K ( ) K d
S
*
∞
∫ ,
`
+
FIG URE 4. Re p lic a tion of the log p a yoff. ( a ) The p a yoff of a short p osition in a
log c ontra c t a t e xp ira tion. ( b ) Da she d line : the line a r p a yoff a t e xp ira tion of
a forwa rd c ontra c t with d e live ry p ric e S
*
; Solid line : the c urve d p a yoff of
c a lls struc k a b ove S
*
a nd p uts struc k b e low S
*
. Ea c h op tion is we ig hte d by
the inve rse sq ua re of its strike . The sum of the p a yoffs for the d a she d a nd
solid line s p rovid e the sa me p a yoff a s the log c ontra c t.
S
*
S
*
S
T
S
*
 log –
S
T
S
*
–
S
*
 –
por t folio of opt ions
( a ) ( b )
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
20
AN EXAM PLE O F A
VARIANC E SWAP
We now pr esent a det ailed pr act ical example. Suppose you want t o
pr ice a swap on t he r ealized var iance of t he daily r et ur ns of some hypo
t het ical equit y index. The fair deliver y var iance is det er mined by t he
cost of t he r eplicat ing st r at egy discussed in t he pr evious sect ion. If you
could buy opt ions of all st r ikes bet ween zer o and inﬁnit y, t he fair var i
ance would be given by Equat ion 26 wit h some choice of S
*
, say S
*
= S
0
.
In pr act ice, however, only a small set of discr et e opt ion st r ikes is avail
able, and using Equat ion 26 wit h only a few st r ikes leads t o appr ecia
ble er r or s. Her e we suggest a bet t er appr oximat ion.
We st ar t wit h t he deﬁnit ion of fair var iance given by Equat ion 21,
which can be wr it t en as
Taking expect at ions, t his becomes
( EQ 27)
wher e is t he pr esent value of t he por t folio of opt ions wit h payoff
at expir at ion given by
( EQ 28)
Suppose t hat you can t r ade call opt ions wit h st r ikes K
ic
such t hat
and put opt ions wit h st r ikes K
ip
such
t hat
In Appendix A we der ive t he for mula t hat det er mines how many
opt ions of each st r ike you need in or der t o appr oximat e t he payoff f(S
T
)
by piecewise linear opt ions payoffs. The pr ocedur e in Appendix A
guar ant ees t hat t hese payoffs will always exceed or mat ch t he value of
t he log cont r act , but never be wor t h less. Once t hese weight s ar e calcu
lat ed, is obt ained fr om
( EQ 29)
We now illust r at e t his pr ocedur e wit h a concr et e numer ical example.
K
v ar
2
T
 E
S
t
d
S
t

0
T
∫
S
T
S
*
–
S
*
 –
S
*
S
0

S
T
S
*
–
S
*

S
T
S
*
 log – + log – ≡
K
v ar
2
T
 rT
S
0
S
*
e
rT
1 –
. ,
 `
S
*
S
0
 log – – e
rT
Π
CP
+ =
Π
CP
f S
T
( )
2
T

S
T
S
*
–
S
*

S
T
S
*
 log –
. ,
 `
=
K
0
S
*
K
1c
K
2c
K
3c
... < < < < =
K
0
S
*
K
1 p
K
2 p
K
3 p
... > > > > =
Π
CP
Π
CP
w K
i p
( )P S K
i p
, ( ) w K
i c
( )C S K
i c
, ( )
i
∑
+
i
∑
=
21
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
TABLE 1. The p ortfolio of Europ e a n style p ut a nd c a ll op tions use d for
c a lc ula ting the c ost of c a p turing re a lize d va ria nc e in the p re se nc e of the
imp lie d vola tility ske w with a d isc re te se t of op tions strike s.
Assume t hat t he index level S
0
is 100, t he cont inuously compounded
annual r iskless int er est r at e r is 5%, t he dividend yield is zer o, and t he
mat ur it y of t he var iance swap is t hr ee mont hs (T = 0.25. Suppose t hat
Strike Vola tility We ig ht
Va lue
p e r
O p tion
C ontrib ution
P
U
T
S
50 30 163.04 0.000002 0.0004
55 29 134.63 0.00003 0.0035
60 28 113.05 0.0002 0.0241
65 27 96.27 0.0013 0.1289
70 26 82.98 0.0067 0.5560
75 25 72.26 0.0276 1.9939
80 24 63.49 0.0958 6.0829
85 23 56.23 0.2854 16.0459
90 22 50.15 0.7384 37.0260
95 21 45.00 1.6747 75.3616
100 20 20.98 3.3537 70.3615
C
A
L
L
S
100 20 19.63 4.5790 89.8691
105 19 36.83 2.2581 83.1580
110 18 33.55 0.8874 29.7752
115 17 30.69 0.2578 7.9130
120 16 28.19 0.0501 1.4119
125 15 25.98 0.0057 0.1476
130 14 24.02 0.0003 0.0075
135 13 22.27 0.000006 0.0001
TOTAL C O ST 419.8671
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
22
you can buy opt ions wit h st r ikes in t he r ange fr om 50 t o 150, unifor mly
spaced 5 point s apar t . We assume t hat at t hemoney implied volat ilit y
is 20%, wit h a skew such t hat t he implied volat ilit y incr eases by 1 vola
t ilit y point for ever y 5 point decr ease in t he st r ike level. In Table 1 we
pr ovide t he list of st r ikes and t heir cor r esponding implied volat ilit ies.
We t hen show t he weight s, t he value of each individual opt ion and t he
cont r ibut ion of each st r ike level t o t he t ot al cost of t he por t folio. At t he
bot t om of t he t able we show t he t ot al cost of t he opt ions por t folio,
. It is clear fr om Table 1 t hat most of t he cost comes
fr om opt ions wit h st r ikes near t he spot value. Alt hough t he number of
opt ions which ar e far out of t he money is lar ge, t heir value is small and
cont r ibut es lit t le t o t he t ot al cost .
The cost of capt ur ing var iance is now simply calculat ed using Equat ion
27 wit h t he r esult . This is not st r ict ly t he fair var i
ance; because t he pr ocedur e of appr oximat ing t he log cont r act in
Appendix A always overest imat es t he value of t he log cont r act , t his
value is higher t han t he t r ue t heor et ical value for t he fair var iance
obt ained by appr oximat ing t he log cont r act wit h a cont inuum of
st r ikes. In Figur e 5 we illust r at e t he cost of var iance as funct ion of t he
spacing bet ween st r ikes, for t wo cases, wit h and wit hout a volat ilit y
skew. You can see t hat as t he spacing bet ween st r ikes appr oaches zer o,
t he cost of capt ur ing var iance appr oaches t he t heor et ically fair var i
ance.
0 1 2 3 4 5
DK
400
405
410
415
420
K
v
a
r
FIG URE 5. C onve rg e nc e of K
va r,
the c ost of c a p turing va ria nc e with a
d isc re te se t of strike s, towa rd s the fa ir va lue of va ria nc e a s a func tion of
∆K, the sp a c ing b e twe e n strike s. The line with sq ua re symb ols shows the
c onve rg e nc e for no ske w, with a ll imp lie d vola tilitie s a t the sa me va lue of
20%. The the ore tic a l fa ir va ria nc e for ∆K = 0 is the n ( 20)
2
= 400. The line
with d ia mond symb ols shows simila r c onve rg e nc e to a hig he r fa ir
va ria nc e of a b out 402, the e xtra c ontrib ution c oming from the e ffe c t of
the ske w.
Π
CP
419.8671 =
K
v ar
20.467 ( )
2
=
23
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
EFFEC TS O F THE
VO LATILITY SKEW
The gener al st r at egy discussed in t he pr evious sect ion can be used t o
det er mine t he fair var iance and t he hedging por t folio fr om t he set of
available opt ions and t heir implied volat ilit ies. Her e we discuss t he
effect s of a volat ilit y skew on t he fair var iance. We assume t hat t her e is
no t er m st r uct ur e and consider t wo differ ent skew par amet er izat ions,
bot h of which r esemble t ypical index skews. The ﬁr st is a skew t hat
var ies linear ly wit h t he st r ike of t he opt ion, t he second a skew t hat
var ies linear ly wit h t he BlackScholes delt a. In bot h cases we will com
par e t he numer ically cor r ect value of fair var iance, comput ed fr om
Equat ion 26, wit h an appr oximat e analyt ic for mula t hat we der ive.
This for mula pr ovides a good r ule of t humb for a quick est imat e of t he
impact of t he volat ilit y skew on t he fair var iance.
Ske w Line a r in Strike We ﬁr st consider a skew for which t he implied volat ilit y var ies linear ly
wit h st r ike, so t hat
( EQ 30)
Her e Σ
0
is t he implied volat ilit y of an opt ion st r uck at t he for war d. The
st eepness of t he skew is det er mined by t he slope b, wit h a posit ive
value indicat ing a higher volat ilit y for st r ikes below t he for war d. Not e
t hat t his par amet r izat ion cannot hold for all st r ikes, because, for a
lar ge enough value of K, t he implied volat ilit y would become negat ive
7
.
A value of b = 0.2 means t hat t he implied volat ilit y cor r esponding t o a
st r ike 10% below t he for war d, for example, is 2 volat ilit y point s higher
t han Σ
0
. In Appendix B we der ive t he following appr oximat e for mula
for t he fair var iance of t he cont r act wit h t ime t o expir at ion T:
( EQ 31)
The skew incr eases t he value of t he fair var iance above t he at t he
moneyfor war d level of volat ilit y, and t he size of t he incr ease is pr opor 
t ional t o t ime t o mat ur it y and t he squar e of t he skew slope. (Not e t hat
b in Equat ion 30 has t he same dimension as volat ilit y, so t hat b
2
T is a
dimensionless par amet er, and t her efor e a nat ur al candidat e for t he
or der of magnit ude of t he per cent age cor r ect ion t o K
var
. Not e also t hat
t her e is no t er m in Equat ion 31. This appr oximat ion wor ks best
for shor t mat ur it ies and skews t hat ar e not t oo st eep.
7. Not e t hat for lar ge values of K, wher e t his par amet er izat ion is invalid,
t he opt ions pr ices in Equat ion 26 ar e negligible and t her efor e do not
affect t he value of t he fair var iance.
Σ K ( ) Σ
0
b
K S
F
–
S
F
 – =
K
v ar
Σ
0
2
1 3T b
2
.... + + ( ) ≈
b T
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
24
In Table 2 we compar e t he exact r esult s for fair var iance, comput ed
numer ically, wit h t he appr oximat e values given by t he analyt ic for 
mula in Equat ion 31.
TABLE 2. C omp a rison of the e xa c t fa ir va ria nc e , c omp ute d nume ric a lly, with
the a p p roxima te a na lytic formula of Eq ua tion 31. We a ssume Σ
0
= 30%, S =
100, the c ontinuously c omp ound e d a nnua l d isc ount ra te r = 5%, ze ro
d ivid e nd yie ld , a nd use strike s e ve nly sp a c e d one p oint a p a rt from K = 10
to K = 200 to re p lic a te the log p a yoff.
Figur e 6 cont ains a gr aph of t hese r esult s. We see excellent agr eement
in t he case of t he t hr eemont h var iance swap, and r easonable agr ee
ment for one year.
Ske w Slop e
b
T = 3 months T = 1 ye a r
Exa c t
Va lue
Ana lytic
Ap p roxima tion
Exa c t
Va lue
Ana lytic
Ap p roxima tion
0.0
0.1
0.2
0.3
30.01 ( )
2
30.00 ( )
2
29.97 ( )
2
30.00 ( )
2
30.01 ( )
2
30.11 ( )
2
30.05 ( )
2
30.45 ( )
2
30.22 ( )
2
30.44 ( )
2
30.82 ( )
2
31.75 ( )
2
30.65 ( )
2
30.99 ( )
2
32.33 ( )
2
33.81 ( )
2
0.1 0.2 0.3
b
900
950
1000
1050
1100
1150
K
v
a
r
0.1 0.2 0.3
b
900
950
1000
1050
1100
1150
K
v
a
r
FIG URE 6. C omp a rison of the e xa c t va lue of fa ir va ria nc e , K
va r
, with the
a p p roxima te va lue from the formula of Eq ua tion 31, a s a func tion of the
ske w slop e b . The thin line with sq ua re s shows the e xa c t va lue s ob ta ine d
by re p lic a ting the log  p a yoff. The thic k line d e p ic ts the a p p roxima te
va lue g ive n by Eq ua tion 31. ( a ) thre e  month va ria nc e swa p . ( b ) one  ye a r
va ria nc e swa p .
( a ) ( b )
25
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Ske w Line a r in De lta Next we consider a skew t hat var ies linear ly wit h t he BlackScholes
delt a of t he opt ion, so t hat :
( EQ 32)
Her e ∆
p
is t he BlackScholes exposur e of a put opt ion, given by
, wher e d
1
is deﬁned in Foot not e 2, is t he implied
volat ilit y of a “50delt a” put opt ion and b is t he slope of t he skew – t hat
is, t he change in t he skew per unit delt a. This par amet er b is not t he
same as t he b in t he pr evious sect ion. In par t icular, t her e is an implicit
dependence on t he t ime t o expir at ion in t he for mula of Equat ion 32,
because of t he ∆
p
t er m, which was absent fr om Equat ion 30. Since ∆
p
is
bounded, t he implied volat ilit y is always posit ive pr ovided b < 2Σ
0
.
This r est r ict ion is ir r elevant , since Equat ion 32 leads t o ar bit r age vio
lat ion befor e b r eaches t his limit . In pr act ice, t his par amet er izat ion
leads t o mor e r ealist ic skews t han t hose pr oduced by t he linearst r ike
for mula.
Appendix C pr esent s a det ailed der ivat ion of t he following appr oximat e
for mula for t he fair var iance of t he cont r act wit h t ime t o expir at ion T:
( EQ 33)
Her e, in cont r ast t o t he skew linear in st r ike, t he ﬁr st or der cor r ect ion
is of magnit ude , because a var iat ion linear in delt a about t he at 
t hemoneyfor war d st r ike is not equivalent t o a var iat ion linear in
st r ike.
Σ ∆
p
( ) Σ
0
b ∆
p
1
2
 +
. ,
 `
+ =
∆
p
N d
1
– ( ) – = Σ
0
K
v ar
Σ
0
2
1
1
π
b T
1
12

b
2
Σ
0
2
 .... + + +
. ,
 `
≈
b T
1 0.75 0.5 0.25 0
∆
p
20
25
30
35
40
Σ
60 80 100 120 140
K
20
25
30
35
40
Σ
FIG URE 7. ( a ) A vola tility ske w tha t va rie s line a rly in d e lta . ( b ) The
c orre sp ond ing ske w p lotte d a s a func tion of strike . We ha ve a ssume d tha t
the stoc k p ric e S is 100, the c ontinuously c omp ound e d a nnua l d isc ount
ra te r is 5%, the te rm to ma turity is thre e months, a nd the ske w slop e is 0.2.
( a ) ( b )
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
26
Fir st , we conver t t he skew by delt a in Equat ion 32 int o a skew by
st r ike, as displayed in Figur e 7. Again, we compar e t he exact r esult s
comput ed accor ding t o Appendix A wit h t he appr oximat e values given
by Equat ion 33. In Table 3 we compar e t he r esult s for fair var iance,
comput ed numer ically, wit h t he appr oximat e values given by t he ana
lyt ic for mula in Equat ion 33. The analyt ic for mula wor ks ver y well for
t he t hr eemont h var iance swap, and t r uly impr essively for t he one
year swap, as displayed in Figur e 8.
TABLE 3. C omp a rison of the fa ir va ria nc e , c omp ute d nume ric a lly, with the
a p p roxima te a na lytic formula of Eq ua tion 33. We a ssume Σ
0
= 30%, S = 100,
the c ontinuously c omp ound e d a nnua l d isc ount ra te r = 5%, ze ro d ivid e nd
yie ld , a nd use strike s e ve nly sp a c e d one p oint a p a rt from K = 10 to K = 200
to re p lic a te the log p a yoff.
Ske w Slop e
b
T = 3 months T = 1 ye a r
Exa c t
Va lue
Ana lytic
Ap p roxima tion
Exa c t
Va lue
Ana lytic
Ap p roxima tion
0.0
0.1
0.2
0.3
30.01 ( )
2
30.00 ( )
2
29.97 ( )
2
30.00 ( )
2
30.61 ( )
2
30.62 ( )
2
31.06 ( )
2
31.03 ( )
2
31.49 ( )
2
31.60 ( )
2
32.42 ( )
2
32.40 ( )
2
32.64 ( )
2
32.93 ( )
2
34.06 ( )
2
34.06 ( )
2
0.1 0.2 0.3
b
900
950
1000
1050
1100
1150
K
v
a
r
0.1 0.2 0.3
b
900
950
1000
1050
1100
1150
K
v
a
r
FIG URE 8. C omp a rison of the e xa c t va lue of fa ir va ria nc e , K
va r
, with the
a p p roxima te va lue from the formula of Eq ua tion 33, a s a func tion of the
ske w slop e b . The thin line with sq ua re s shows the e xa c t va lue s ob ta ine d
by re p lic a ting the log  p a yoff. The thic k line d e p ic ts the a p p roxima te
va lue g ive n by Eq ua tion 33. ( a ) Thre e  month va ria nc e swa p . ( b ) O ne 
ye a r va ria nc e swa p .
( a ) ( b )
27
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
PRAC TICAL PRO BLEM S
WITH REPLICATIO N
We have shown in Equat ion 20 t hat a var iance swap is t heor et ically
equivalent t o a dynamically adjust ed, const ant dollar exposur e t o t he
st ock, t oget her wit h a st at ic long posit ion in a por t folio of opt ions and a
for war d t hat t oget her r eplicat e t he payoff of a log cont r act . This por t fo
lio st r at egy capt ur es var iance exact ly, pr ovided t he por t folio of opt ions
cont ains all st r ikes bet ween zer o and inﬁnit y in t he appr opr iat e weight
t o mat ch t he log payoff, and pr ovided t he st ock pr ice evolves cont inu
ously.
Two obvious t hings can go wr ong. Fir st , you may be able t o t r ade only a
limit ed r ange of opt ions st r ikes, insufﬁcient t o accur at ely r eplicat e t he
log payoff. Second, t he st ock pr ice may jump. Bot h of t hese effect s
cause t he st r at egy t o capt ur e a quant it y t hat is not t he t r ue r ealized
var iance. We will focus on t he effect s of t hese t wo limit at ions below,
t hough ot her pr act ical issues, like liquidit y, may also cor r upt t he ideal
st r at egy.
Imp e rfe c t Re p lic a tion
Due to Limite d Strike
Ra ng e
Var iance r eplicat ion r equir es a log cont r act . Since log cont r act s ar e not
t r aded in pr act ice, we r eplicat e t he payoff wit h t r aded st andar d opt ions
in a limit ed st r ike r ange. Because t hese st r ikes fail t o duplicat e t he log
cont r act exact ly, t hey will capt ur e less t han t he t r ue r ealized var iance.
Ther efor e, t hey have lower value t han t hat of a t r ue log cont r act , and
so pr oduce an inaccur at e, lower est imat e of t he fair var iance.
In Table 4 below we show how t he est imat ed value of fair var iance is
affect ed by t he r ange of st r ikes t hat make up t he r eplicat ing por t folio.
The fair var iances ar e est imat ed fr om (1) a r eplicat ing por t folio wit h a
nar r ow r ange of st r ikes, r anging fr om 75% t o 125% of t he init ial spot
level, and (2) a por t folio wit h a wide r ange of st r ikes, fr om 50% t o 200%
of t he init ial spot level. In bot h cases t he st r ikes ar e unifor mly spaced,
one point apar t . (The fair var iance is calculat ed accor ding t o Equat ion
26, except t hat t he int egr als ar e r eplaced by sums over t he available
opt ion st r ikes whose weight s ar e chosen accor ding t o t he pr ocedur e of
Appendix A). We assume her e t hat implied volat ilit y is 25% per year
for all st r ikes, wit h no volat ilit y skew, so t hat all opt ions ar e valued at
t he same implied volat ilit y. We also assume a cont inuously com
pounded annual int er est r at e of 5%.
For bot h expir at ions, t he wide st r ike r ange accur at ely appr oximat es
t he act ual squar e of t he implied volat ilit y. However, t he nar r ow st r ike
r ange under est imat es t he fair var iance, mor e dr amat ically so for
longer expir at ions.
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
28
TABLE 4. The e ffe c t of strike ra ng e on e stima te d fa ir va ria nc e .
In t he sect ion ent it led Replicating Variance Swaps: First Stepson page 6, we have
alr eady discussed one appr oach t o under st anding why t he nar r ow
st r ike r ange fails t o capt ur e var iance. As shown in Figur e 3, t he vega
and gamma of a limit ed st r ike r ange bot h fall t o zer o when t he index
moves out side t he st r ike r ange, and t he st r at egy t hen fails t o accr ue
r ealized var iance as t he st ock pr ice moves. Consequent ly, t he est i
mat ed var iance is lower t han t he t r ue fair value for bot h expir at ions
above, and t he r educt ion in value is gr eat er for t he oneyear case. Over
a longer t ime per iod it is mor e likely t hat t he st ock pr ice will evolve
out side t he st r ike r ange.
In essence, capt ur ing var iance r equir es owning t he full log cont r act ,
whose duplicat ion demands an inﬁnit e r ange of st r ikes. If you own a
limit ed number of st r ikes, st ill appr opr iat ely weight ed, you pay less
t han t he full value, and, when t he st ock pr ice evolves int o r egions
wher e t he cur vat ur e of t he por t folio is insufﬁcient ly lar ge, you capt ur e
less t han t he full r ealized var iance, even if no jumps occur and t he
st ock always moves cont inuously. In or der t o keep capt ur ing var iance,
you need t o maint ain t he cur vat ur e of t he log cont r act at t he cur r ent
st ock pr ice, what ever value it t akes.
A simpler way of under st anding why a nar r ow st r ike r ange leads t o a
lower fair var iance is t o compar e t he payoff of t he nar r owst r ike r epli
cat ing por t folio at expir at ion t o t he t er minal payoff t hat t he por t folio is
at t empt ing t o r eplicat e, t hat is, t he nonlinear par t of t he log payoff:
( EQ 34)
Figur e 9 displays t he mismat ch bet ween t he t wo payoffs. The nar r ow
st r ike opt ion por t folio mat ches t he cur ved par t of t he log payoff well at
st ock pr ice levels bet ween t he r ange of st r ikes, t hat is, fr om 75 t o 125.
Beyond t his r ange, t he opt ion por t folio payoff r emains linear, always
gr owing less r apidly t han t he nonlinear par t of t he log cont r act . The
lack of cur vat ur e (or gamma, or vega) in t he opt ions por t folio out side
Exp ira tion Wid e strike ra ng e
( 50%  200%)
Na rrow strike ra ng e
( 75%  125%)
Thr eemont h
Oneyear
25.0 ( )
2
24.9 ( )
2
25.0 ( )
2
23.0 ( )
2
S
T
S
0
–
S
0

S
T
S
0
 log –
29
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
t he nar r ow st r ike r ange is r esponsible for t he inabilit y t o capt ur e var i
ance.
The Effe c t of Jump s
on a Pe rfe c tly Re p li
c a te d Log C ontra c t
When t he st ock pr ice jumps, t he log cont r act may no longer capt ur e
r ealized volat ilit y, for t wo r easons. Fir st , if t he log cont r act has been
appr oximat ely r eplicat ed by only a ﬁnit e r ange of st r ikes, a lar ge jump
may t ake t he st ock pr ice int o a r egion in which var iance does not
accr ue at t he r ight r at e. Second, even wit h per fect r eplicat ion, a discon
t inuous st ockpr ice jump causes t he var iancecapt ur e st r at egy of Equa
t ion 20 t o capt ur e an amount not equal t o t he t r ue r ealized var iance. In
r ealit y, bot h t hese effect s cont r ibut e t o t he r eplicat ion er r or. In t his sec
t ion, we focus only on t he second effect and examine t he effect s of
jumps assuming t hat t he logpayoff can be r eplicat ed per fect ly wit h
opt ions.
For t he sake of discussion, fr om now on we will assume t hat we ar e
shor t t he var iance swap, which we will hedge by following a discr et e
ver sion of t he var iancecapt ur e st r at egy
( EQ 35)
wher e is t he change in st ock pr ice bet ween successive
obser vat ions. Rat her t han cont inuously r ebalance as t he st ock pr ice
moves, we inst ead adjust t he exposur e t o (2/T) dollar s wor t h of st ock
only when a new st ock pr ice is r ecor ded for updat ing t he r ealized var i
ance.
Because of t he addit ive pr oper t ies of t he logar it hm funct ion, t he t er mi
nal log payoff is equivalent t o a daily accumulat ion of log payoffs:
50 100 150 200
S
T
T
e
r
m
i
n
a
l
p
a
y
o
f
f
FIG URE 9. C omp a rison of the te rmina l p a yoff of the na rrow strike re p lic a ting
p ortfolio ( d a she d line ) a nd the nonline a r p a rt of the log  p a yoff ( solid line ) .
V
2
T

∆S
i
S
i 1 –

i 1 =
N
∑
S
T
S
0
 log – =
∆S
i
S
i
S
i 1 –
– =
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
30
( EQ 36)
Suppose t hat all but one of t he daily pr ice changes ar e wellbehaved –
t hat is, all changes ar e diffusive, except for a single jump event . We
char act er ize t he jump by t he par amet er , t he per cent age jump down
wards, fr om ; a jump downwar ds of 10% cor r esponds t o J
= 0.1. A jump up cor r esponds t o a value J < 0.
The cont r ibut ion of t his one jump t o t he var iance is easy t o isolat e,
because var iance is addit ive; t he t ot al (unannualized) r ealized var i
ance for a zer omean cont r act is t he sum
( EQ 37)
The cont r ibut ion of t he jump t o t he r ealized t ot al var iance is given by:
( EQ 38)
On t he ot her hand, t he impact of t he jump on t he quant it y capt ur ed by
our var iance r eplicat ion st r at egy in Equat ion 36 is
( EQ 39)
In t he limit t hat t he jump size J is small enough t o be r egar ded as par t
of a cont inuous st ock evolut ion pr ocess, t he r ight hand side of Equat ion
39 does r educe t o t he cont r ibut ion of t his (now small) move t o t he t r ue
r ealized var iance. It is only because J is not small t hat t he var iance
capt ur e st r at egy is inaccur at e. Ther efor e, t he r eplicat ion er r or, or t he
P&L (pr oﬁt /loss) due t o t he jump for a shor t posit ion in a var iance
swap hedged by a long posit ion in a var iancecapt ur e st r at egy is
( EQ 40)
To under st and t his r esult bet t er, it is helpful t o expand t he log funct ion
as a ser ies in J :
( EQ 41)
V
2
T

∆S
i
S
i 1 –

S
i
S
i 1 –
 log –
i 1 =
N
∑
=
J
S S 1 J – ( ) →
V
1
T

∆S
i
S
i 1 –

. ,
 `
2
∑
1
T

∆S
i
S
i 1 –

. ,
 `
2
1
T

∆S
S

. ,
 `
2
jump
+
no jumps
∑
= =
1
T

∆S
S

. ,
 `
2
jump
J
2
T
 =
2
T

∆S
i
S
i 1 –

S
i
S
i 1 –
 log –
. ,
 `
jump
2
T
 J – 1 J – ( ) log – [ ] =
P&L due t o jump =
2
T
 J – 1 J – ( ) log – [ ]
J
2
T
 –
1 J – ( ) log – J
J
2
2

J
3
3
 … + + + =
31
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
The leading cont r ibut ion t o t he r eplicat ion er r or is t hen
( EQ 42)
The quadr at ic cont r ibut ion of t he jump is t he same for t he var iance
swap as it is for t he var iancecapt ur e st r at egy, and has no impact on
t he hedging mismat ch. The leading cor r ect ion is cubic in t he jump size
J and has a differ ent sign for upwar ds or downwar ds jumps. A lar ge
move downwar ds (J > 0) leads t o a pr oﬁt for t he (shor t var iance swap)
(long var iancecapt ur e st r at egy), while a lar ge move upwar ds (J < 0)
leads t o a loss. Fur t her mor e, a lar ge move one day, followed by a lar ge
move in t he opposit e dir ect ion t he next day would t end t o offset each
ot her. Figur e 10 shows t he impact of t he jump on t he st r at egy for a
r ange of jump values. Not e t hat t he simple cubic appr oximat ion of
Equat ion 42 cor r ect ly pr edict s t he sign of t he P&L for all values of t he
jump size.
FIG URE 10. he imp a c t of a sing le jump on the p roﬁt or loss of a short p osition
in a va ria nc e swa p a nd a long p osition in the va ria nc e re p lic a tion stra te g y,
a s g ive n by Eq ua tion 40 a s a func tion of ( d ownwa rd ) jump size for T=1 ye a r.
Ther e is an analogy bet ween t he cancellat ion of t he quadr at ic jump
t er m in var iance r eplicat ion and t he linear jump t er m in opt ions r epli
cat ion. When you ar e long an opt ion you ar e long linear, quadr at ic and
higheror der dependence on t he st ock pr ice. If you ar e also shor t t he
opt ion’s delt ahedge, t hen t he linear dependence of t he net posit ion
cancels, leaving only t he quadr at ic and higheror der dependencies.
Because t he leadingor der t er m is quadr at ic, lar ge moves in eit her
dir ect ion beneﬁt t he posit ion; t his is pr ecisely why hedged long opt ions
posit ions capt ur e var iance. In cont r ast , in t he case of var iance r eplica
t ion consider ed her e, t he var iance r eplicat ion st r at egy is long qua
dr at ic, cubic and higheror der t er ms in t he st ock pr ice, while t he
posit ion in t he var iance swap is shor t only t he quadr at ic dependence.
Now t he quadr at ic t er m in t he net posit ion cancels, leaving only cubic
and higheror der dependencies on t he jump size. Since t he leading
P&L due t o jump =
2
3

J
3
T
 … +
20 10 0 10 20
Jump size (downward)
0.004
0.002
0
0.002
0.004
0.006
I
m
p
a
c
t
o
f
j
u
m
p
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
32
t er m is cubic, t he dir ect ion of t he jump det er mines whet her t her e is a
net pr oﬁt or loss.
Table 5 displays t he pr oﬁt or loss due t o jumps of var ying sizes for
t hr eemont h and oneyear var iance swaps.
TABLE 5. The p roﬁt/ loss d ue to a sing le jump for a short va ria nc e swa p with a
notiona l va lue of $1 p e r sq ua re d va ria nc e p oint, tha t is he d g e d with the
va ria nc e re p lic a tion stra te g y of Eq ua tion 36 for T=1 ye a r.
The Effe c t of Jump s
Whe n Re p lic a ting With
a Finite Strike Ra ng e
In pr act ice, bot h t he effect s of jumps and t he r isks of log r eplicat ion
wit h only a limit ed st r ike r ange cause t he st r at egy t o capt ur e a quan
t it y differ ent fr om t he t r ue r ealized var iance of t he st ock pr ice. The
combined effect of bot h t hese r isks is har der t o char act er ize because
t hey int er act wit h one anot her in a complicat ed manner.
Consider again a shor t posit ion in a var iance cont r act t hat is being
hedged by t he var iancecapt ur e st r at egy. Suppose t hat a downwar d
jump occur s, lar ge enough t o move t he st ock pr ice out side t he r ange of
opt ion st r ikes. If t he logpayoff wer e r eplicat ed per fect ly, t he const ant 
dollar exposur e would cancel t he linear par t of t he st ock pr ice change,
and lead t o a convexit y gain. Alt hough t he logpayoff is not being r epli
cat ed per fect ly, t her e is st ill a convexit y gain fr om t he jump, but it is
smaller in size. However, aft er t his jump, wit h t he st ock pr ice now out 
side t he st r ike r ange, t he vega and gamma of t he r eplicat ing por t folio
ar e now t oo low t o accr ue sufﬁcient var iance, even if no fur t her jumps
occur. In t his scenar io, t he gain fr om t he jump has t o be balanced
against t he subsequent failur e of t he hedge t o capt ur e t he smoot h var i
ance. The net r esult s will depend on t he det ails of t he scenar io.
In cont r ast , a lar ge move upwar ds will be doubly damaging: t her e will
be convexit y loss due t o t he jump and t he hedge will not capt ur e var i
ance if t he jump t akes t he index out side t he st r ike r ange.
Jump size a nd
d ire c tion
Thre e  month O ne  ye a r
J = 15% (down) 101.5 25.4
J = 10% (down) 28.8 7.2
J = 5% (down) 3.5 0.9
J · −5% (up) −3.2 −0.8
J = −10% (up) −24.8 −6.2
J = −20% (up) −80.9 −20.2
33
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
FRO M VARIANC E TO
VO LATILITY
C O NTRAC TS
For most of t his not e we have focused on valuing and r eplicat ing var i
ance swaps. But most mar ket par t icipant s pr efer t o quot e levels of vol
at ilit y r at her t han var iance, and so we now consider volat ilit y swaps.
Ther e is no simple r eplicat ion st r at egy for synt hesizing a volat ilit y
swap; it is var iance t hat emer ges nat ur ally fr om hedged opt ions t r ad
ing. The r eplicat ion st r at egy for t he var iance swap makes no assump
t ions about t he level of fut ur e volat ilit y, ot her t han assuming t hat t he
st ock pr ice evolves cont inuously (wit hout jumps). Changes in volat ilit y
have no effect on t he st r at egy, which st ill capt ur es t he t ot al var iance
over t he life of t he log cont r act . In cont r ast , as we will show, t he r epli
cat ion st r at egy for a volat ilit y swap is fundament ally differ ent ; it is
affect ed by changes in volat ilit y and it s value depends on t he volat ilit y
of fut ur e r ealized volat ilit y. In essence, fr om a cont ingent claims or
der ivat ives point of view, var iance is t he pr imar y under lyer and all
ot her volat ilit y payoffs, such as volat ilit y swaps, ar e best r egar ded as
der ivat ive secur it ies on t he var iance as under lyer. Fr om t his per spec
t ive, volat ilit y it self is a nonlinear funct ion (t he squar e r oot ) of var i
ance and is t her efor e mor e difﬁcult , bot h t heor et ically and pr act ically,
t o value and hedge.
To illust r at e t he issues involved, let ’s consider a naive st r at egy:
appr oximat e a volat ilit y swap by st at ically holding a suit ably chosen
var iance cont r act . In or der t o appr oximat e a volat ilit y swap st r uck at
, which has payoff , we can use t he appr oximat ion
( EQ 43)
This means t hat var iance cont r act s wit h st r ike can
appr oximat e a volat ilit y swap wit h a not ional $1/(vol point ), for r eal
ized volat ilit ies near . Wit h t his choice, t he var iance and volat ilit y
payoffs agr ee in value and volat ilit y sensit ivit y (t he ﬁr st der ivat ive
wit h r espect t o ) when . Naively, t his would also imply
t hat t he fair pr ice of fut ur e volat ilit y (t he st r ike for which t he volat ilit y
swap has zer o value) is simply t he squar e r oot of fair var iance :
(naive est imat e) ( EQ 44)
In Figur e 11 we compar e t he t wo sides of Equat ion 43 for K
vol
= 30%
for differ ent values of t he r ealized volat ilit y. We see t hat t he act ual vol
at ilit y swap and t he appr oximat ing var iance swap differ appr eciably
K
v ol
σ
R
K
v ol
–
σ
R
K
v ol
–
1
2K
v ol
 σ
R
2
K
v ol
2
– ( ) ≈
1 2K
v ol
( ) ⁄ K
v ol
2
K
v ol
σ
R
σ
R
K
v ol
=
K
v ar
K
v ol
K
v ar
=
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
34
only when t he fut ur e r ealized volat ilit y moves away fr om ; you
cannot ﬁt a line ever ywher e wit h a par abola.
The naive est imat e of Equat ions 43 and 44 is not quit e cor r ect . Wit h
t his choice, t he var iance swap payoff is always gr eat er t han t he volat il
it y swap payoff. The mismat ch bet ween t he var iance and volat ilit y
swap payoffs in Equat ion 43, is t he
This squar e is always posit ive, so t hat wit h t his choice of t he fair deliv
er y pr ice for volat ilit y, t he var iance swap always out per for ms t he vola
t ilit y swap. To avoid t his ar bit r age, we should cor r ect our naive
est imat e t o make t he fair st r ike for t he volat ilit y cont r act lower t han
t he squar e r oot of t he fair st r ike for a var iance cont r act , so t hat
. In t his way, t he st r aight line in Figur e 11 will shift t o
t he left and will not always lie below t he par abola.
In or der t o est imat e t he size of t he convexit y bias, and t her efor e t he
fair st r ike for t he volat ilit y swap, it is necessar y t o make an assump
t ion about bot h t he level and volat ilit y of fut ur e r ealized volat ilit y. In
Appendix D we est imat e t he expect ed hedging mismat ch and st at ic
hedging par amet er s under t he assumpt ion t hat fut ur e r ealized volat il
it y is nor mally dist r ibut ed.
Dyna mic Re p lic a tion
of a Vola tility Swa p
In pr inciple, some of t he r isks inher ent in t he st at ic appr oximat ion of a
volat ilit y swap by a var iance swap could be r educed by dynamically
t r ading new var iance cont r act s t hr oughout t he life of t he volat ilit y
swap. This dynamic r eplicat ion of a volat ilit y swap by means of var i
20 30 40
σ
R
15
10
5
0
5
10
p
a
y
o
f
f
FIG URE 11. Pa yoff of a vola tility swa p ( stra ig ht line ) a nd va ria nc e swa p
( c urve d line ) a s a func tion of re a lize d vola tility, for . K
v ol
30% =
K
v ol
convexit y bias
1
2K
v ol
 σ
R
K
v ol
– ( )
2
=
K
v ol
K
v ar
<
35
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
ance swaps would (in pr inciple) pr oduce t he payoff of a volat ilit y swap
independent of t he moves in fut ur e volat ilit y. This is closely analogous
t o r eplicat ing a cur ved st ock opt ion payoff by means of delt ahedging
using t he linear under lying st ock pr ice. In pr act ice, of cour se, t her e is
no mar ket in var iance swaps liquid enough t o pr ovide a usable under 
lyer.
In t he same way t hat t he appr opr iat e opt ion hedge r at io depends on
t he assumed fut ur e volat ilit y of t he st ock, t he dynamic r eplicat ion of a
volat ilit y swap r equir es a model for t he volat ilit y of volat ilit y. Taking
t he analogy fur t her, one could imagine t hat t he st r at egy would call for
holding at ever y inst ant a “var iancedelt a” equivalent of var iance con
t r act s t o hedge a volat ilit y der ivat ive.
The pr act ical implement at ion of t hese ideas r equir es an ar bit r agefr ee
model for t he st ochast ic evolut ion of t he volat ilit y sur face. Due t o t he
complexit y of t he mat hemat ics involved, it is only ver y r ecent ly t hat
such models have been developed [see for example Der man and Kani
(1998) and Ledoit (1998)]. When t her e is a liquid mar ket in var iance
swaps, t hese models may be useful in hedging volat ilit y swaps and
ot her var iance der ivat ives.
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
36
C O NC LUSIO NS AND
FUTURE INNOVATIO NS
We have t r ied t o pr esent a compr ehensive and didact ic account of bot h
t he pr inciples and met hods used t o value and hedge var iance swaps.
We have explained bot h t he int uit ive and t he r igor ous appr oach t o r ep
licat ion. In mar ket s wit h a volat ilit y skew (t he r eal wor ld for most
swaps of int er est ), t he int uit ive appr oach loses it s foot ing. Her e, using
t he r igor ous appr oach, one can st ill value var iance swaps by r eplica
t ion. Remar kably, we have succeeded in der iving analyt ic appr oxima
t ions t hat wor k well for t he swap value under commonly used skew
par amet er izat ions. These for mulas enable t r ader s t o updat e pr ice
quot es quickly as t he mar ket skew changes.
Ther e ar e at least t wo ar eas wher e fur t her development is welcome.
Fir st , our abilit y t o effect ively pr ice and hedge volat ilit y swaps is st ill
limit ed. To fully implement a r eplicat ion st r at egy for volat ilit y swaps,
we need a consist ent st ochast ic volat ilit y model for opt ions. Much wor k
r emains t o be done in t his ar ea.
Second, some mar ket par t icipant s pr efer t o ent er a capped var iance
swap or volat ilit y swap t hat limit s t he possible loss on t he posit ion. The
capped var iance swap has embedded in it an opt ion on r ealized var i
ance. The development of a t r uly liquid mar ket in volat ilit y swaps, for 
war ds or fut ur es would lead t o t he possibilit y of t r ading and hedging
volat ilit y opt ions. Once again, t his r equir es a consist ent model for st o
chast ic volat ilit y.
37
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
APPENDIX A:
REPLICATING
LO G ARITHM IC
PAYO FFS
In t his Appendix we der ive sever al r esult s concer ning t he r eplicat ion of
a logar it hmic payoff wit h por t folios of st andar d opt ions.
C onsta nt Ve g a
Re q uire s O p tions
We ig hte d Inve rse ly
Prop ortiona l to the
Sq ua re of the Strike
Consider a por t folio of st andar d opt ions
( A 1)
wher e O(S,K,v) r epr esent s a st andar d BlackScholes opt ion of st r ike K
and t ot al var iance when t he st ock pr ice is S.
Vega, t he sensit ivit y t o t he t ot al var iance of an individual opt ion O in
t his por t folio, is given by
wher e
and
.
The var iance sensit ivit y of t he whole por t folio is t her efor e
( A 2)
The sensit ivit y of vega t o S is
Π S ( ) ρ K ( )O S K v , , ( ) K d
0
∞
∫
=
v σ
2
τ =
V
O
τ
v ∂
∂
O ( ) τS f
K
S
 v ,
. ,
 `
= =
f S K v , , ( )
1
2 v

d
1
2
2 ⁄ – ( ) exp
2π
 =
d
1
S K ⁄ ( ) ln v 2 ⁄ +
v
 =
V
Π
S ( ) τ ρ K ( )S f
K
S
 v ,
. ,
 `
K d
0
∞
∫
=
S ∂
∂V
Π
τ
S ∂
∂
S
2
ρ xS ( ) [ ] f x v , ( ) x d
0
∞
∫
=
τ S 2ρ xS ( ) xS ρ' xS ( ) + [ ] f x v , ( ) x d
0
∞
∫
=
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
38
wher e, in t he second line of t he above equat ion, we changed t he int e
gr at ion var iable t o .
We want vega t o be independent of S, t hat is , which implies
t hat
The solut ion t o t his equat ion is
( A 3)
Log Pa yoff Re p lic a tion
with a Disc re te Se t of
O p tions
It was shown in t he main t ext t hat t he r ealized var iance is r elat ed t o
t r ading a log cont r act . Since t her e is no logcont r act t r aded, we want t o
r epr esent it in t er ms of st andar d opt ions. It is useful t o subt r act t he
linear par t (cor r esponding t o t he for war d cont r act ) and look at t he
funct ion
( A 4)
wher e S
*
is some r efer ence pr ice. In pr act ice, only a discr et e set of
opt ion st r ikes is available for r eplicat ing , and we need t o det er 
mine t he number of opt ions for each st r ike. Assume t hat you can t r ade
call opt ions wit h st r ikes
and put opt ions wit h st r ikes
We can appr oximat e wit h a piecewise linear funct ion as in Fig
ur e 11. The ﬁr st segment t o t he r ight of S
*
is equivalent t o t he payoff of
a call opt ion wit h st r ike K
0
. The number of opt ions is det er mined by
t he slope of t his segment :
( A 5)
x K S ⁄ =
S ∂
∂V
Π
0 =
2ρ K
K ∂
∂ρ
+ 0 =
ρ
const
K
2
 =
f S
T
( )
2
T

S
T
S
*
–
S
*

S
T
S
*
 log – =
f S
T
( )
K
0
S
*
K
1c
K
2c
K
3c
... < < < < =
K
0
S
*
K
1 p
K
2 p
K
3 p
... > > > > =
f S
T
( )
w
c
K
0
( )
f K
1c
( ) f K
0
( ) –
K
1c
K
0
–
 =
39
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
FIG URE 12. Log  p a yoff a nd op tions p ortfolio a t ma turity.
Similar ly, t he second segment looks like a combinat ion of calls wit h
st r ikes and . Given t hat we alr eady hold opt ions wit h
st r ike we need t o hold calls wit h st r ike wher e
( A 6)
Cont inuing in t his way we can build t he ent ir e payoff cur ve one st ep at
t he t ime. In gener al, t he number of call opt ions of st r ike is given
by
( A 7)
The ot her side of t he cur ve can be built using put opt ions:
( A 8)
K
2p
K
1p
K
0
K
1c
K
2c
K
0
K
1c
w
c
K
0
( )
K
0
w
c
K
1
( ) K
1
w
c
K
1
( )
f K
2c
( ) f K
1c
( ) –
K
2c
K
1c
–
 w
c
K
0
( ) – =
K
n c ,
w
c
K
n c ,
( )
f K
n 1 c , +
( ) f K
n c ,
( ) –
K
n 1 c , +
K
n c ,
–
 w
c
K
i c ,
( )
i 0 =
n 1 –
∑
– =
w
p
K
n p ,
( )
f K
n 1 p , +
( ) f K
n p ,
( ) –
K
n p ,
K
n 1 + p ,
–
 w
c
K
i p ,
( )
i 0 =
n 1 –
∑
– =
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
40
APPENDIX B:
SKEW LINEAR IN STRIKE
Her e, we der ive a for mula which gives t he appr oximat e value of t he
var iance swap when t he skew is linear in st r ike. We par amet er ize t he
implied volat ilit y by
( B 1)
wher e is t he for war d value cor r esponding t o t he cur r ent
spot , is at t hemoney for war d implied volat ilit y and is t he slope of
t he skew.
We st ar t wit h t he gener al expr ession for t he fair var iance discussed in
t he main t ext :
( B 2)
We now expand opt ion pr ices as a power ser ies in ar ound a ﬂat
implied volat ilit y ( ),
( B 3)
Using t his expansion we can for mally wr it e an expansion of fair var i
ance in power s of as follows:
( B 4)
Σ K ( ) Σ
0
b
K S
F
–
S
F
 – =
S
F
S
0
e
rT
=
Σ
0
b
K
v ar
2
T
 rT
S
0
S
*
e
rT
1 –
. ,
 `
S
*
S
0
 log – + –
.

=
e
rT 1
K
2
 P K Σ b ( ) , ( ) K d
0
S
*
∫
e
rT
+
1
K
2
C K Σ b ( ) , ( ) K d
S
*
∞
∫ ,
`
b
b 0 =
C K Σ b ( ) , ( ) C K Σ
0
, ( ) b
b ∂
∂C
b 0 =
1
2
b
2
b
2
2
∂
∂ C
b 0 =
... + + + =
P K Σ b ( ) , ( ) P K Σ
0
, ( ) b
b ∂
∂P
b 0 =
1
2
b
2
b
2
2
∂
∂ P
b 0 =
... + + + =
b
K
v ar
Σ
0
2
b
2
T
e
rT
. ,
 `
1
K
2

b ∂
∂P
b 0 =
K d
0
S
*
∫
1
K
2

b ∂
∂C
b 0 =
K d
S
*
∞
∫
+
¹ ¹
' '
¹ ¹
+ + =
1
2
b
2 2
T
e
rT
. ,
 `
1
K
2

b
2
2
∂
∂ P
b 0 =
K d
0
S
*
∫
1
K
2

b
2
2
∂
∂ C
b 0 =
K d
S
*
∞
∫
+
¹ ¹
¹ ¹
' '
¹ ¹
¹ ¹
... +
41
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Her e is t he fair var iance in t he “ﬂat wor ld” wher e volat ilit y is con
st ant and is given by Equat ion B2 wit h r eplaced by .
The der ivat ives which ent er Equat ion B4 ar e given by
The der ivat ives wit h r espect t o volat ilit y ar e easily calculat ed using
t he BlackScholes for mula
( B 5)
wher e, for t he model we ar e consider ing her e
( B 6)
The fact t hat call and put opt ions have t he same vega in t he Black
Scholes fr amewor k makes it possible t o combine t he int egr als in Equa
t ion B4 int o one int egr al fr om 0 t o :
( B 7)
Σ
0
2
Σ b ( ) Σ
0
b ∂
∂P
b 0 =
Σ ∂
∂P
Σ
0
b ∂
∂Σ
b 0 =
, =
b
2
2
∂
∂ P
b 0 =
Σ
2
2
∂
∂ P
Σ
0
b ∂
∂Σ
. ,
 `
2
b 0 =
=
b ∂
∂C
b 0 =
Σ ∂
∂C
Σ
0
b ∂
∂Σ
b 0 =
, =
b
2
2
∂
∂ C
b 0 =
Σ
2
2
∂
∂ C
Σ
0
b ∂
∂Σ
. ,
 `
2
b 0 =
=
Σ ∂
∂P
Σ
0
Σ ∂
∂C
Σ
0
S T
2π
e
d
1
2
2 ⁄ –
= =
Σ
2
2
∂
∂ P
Σ
0
Σ
2
2
∂
∂ C
Σ
0
S T
2π
d
1
Σ
0
∂
∂d
1
e
d
1
2
2 ⁄ –
– = =
d
1
S
F
K

. ,
 `
1
2
Σ
0
2
T + log
Σ
0
T
 =
b ∂
∂Σ
b 0 =
K
S
F
 1 –
. ,
 `
– =
∞
K
v ar
Σ
0
2
b
2
T
e
rT
. ,
 `
S T
2π

1
K
2

K
S
F
 1 –
. ,
 `
e
d
1
2
2 ⁄ –
K d
0
∞
∫
– – =
1
2
b
2 2
T
e
rT
. ,
 `
S T
2π

1
K
2

K
S
F
 1 –
. ,
 `
2
d
1
Σ
0
∂
∂d
1
e
d
1
2
2 ⁄ –
K d
0
∞
∫
... +
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
42
To evaluat e t hese int egr als, one can, for example, change t he int egr a
t ion var iable t o , wher e , and
t hen wr it e Equat ion B7 as
The t er m linear in vanishes and t he t er m quadr at ic in has coefﬁ
cient , so t hat
( B 8)
We now pr esent an alt er nat ive der ivat ion of t his r esult . We st ar t wit h
t he fundament al deﬁnit ion of t he fair deliver y var iance as t he expect ed
value of fut ur e r ealized var iance, i.e.
( B 9)
This can be evaluat ed appr oximat ely as follows. Fir st , we use t he r ela
t ion bet ween implied and local volat ilit y:
( B 10)
Denot e . Equat ion B10 can be wr it t en as
( B 11)
z
S
F
K
 log
1
2
v
0
+
. ,
 `
v
0
⁄ d
1
≡ = v
0
Σ
0
2
T =
K
v ar
Σ
0
2
b 2Σ
0
1 e
v
0
z v
0
2 ⁄ –
–
. ,
 `
e
z
2
2 ⁄ – d z
2π

∞ –
∞
∫
¹ ¹
' '
¹ ¹
– + =
b
2
e
v
0
z v
0
2 ⁄ –
e
v
0
– z v
0
2 ⁄ +
2 – +
. ,
 `
z
2
v
0
z – ( )e
z
2
2 ⁄ – d z
2π

∞ –
∞
∫
¹ ¹
' '
¹ ¹
b b
3Σ
0
2
T
K
v ar
Σ
0
2
1 3T b
2
.... + + ( ) =
K
v ar
E
1
T
 σ
2
S t , ( ) t d
0
T
∫
=
σ
2
S t , ( )
Σ
T
 2
T ∂
∂Σ
2rK
K ∂
∂Σ
+ +
K
2
K
2
2
∂
∂ Σ
d
1
T
K ∂
∂Σ
. ,
 `
2
1
Σ

1
K T
 d
1
K ∂
∂Σ
+
. ,
 `
2
+ –
¹ ¹
' '
¹ ¹

K S =
T t =
=
x
S
S
F
 1 – =
σ
2
S t , ( )
Σ
0
bx – 2br 1 x + ( )t –
1 x + ( )
2
t b
2
t d
1
–
1
Σ
0
bx –

1
1 x + ( ) t
 bd
1
–
2
+
¹ ¹
' '
¹ ¹
 =
43
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
wher e
( B 12)
We expand in power s of and calculat e t he expect ed value in
a lognor mal wor ld wit h volat ilit y using
Expect ed values of higher power s of ar e easily calculat ed using
Aft er aver aging over t he st ock pr ice dist r ibut ion, we aver age over t ime
and, ﬁnally, expand t he r esult in power s of t he skew slope . Tedious
calculat ion leads t o t he r elat ion
It is r eassur ing t hat t hese t wo ver y differ ent met hods lead t o t he same
appr oximat ion for mula.
d
1
1 x + ( ) log –
Σ
0
bx – ( ) t

1
2
 Σ
0
bx – ( ) t + =
σ
2
S t , ( ) x
Σ
0
E x [ ] 0 =
E x
2
[ ] e
Σ
0
2
t
1 – =
E x
3
[ ] e
3Σ
0
2
t
3e
Σ
0
2
t
– 2 + =
...
x
E
S
S
F

. ,
 `
n
e
n
2
n – ( )Σ
0
2
t 2 ⁄
=
b
K
v ar
Σ
0
2
1 3T b
2
.... + + ( ) =
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
44
APPENDIX C :
SKEW LINEAR IN DELTA
Her e we consider t he case wher e implied volat ilit y var ies linear ly wit h
delt a. Such a skew can be par amet er ized in t er ms of , t he delt a of a
Eur opeanst yle put , as
( C 1)
wher e is t he implied volat ilit y of opt ions wit h (t he “50
delt a volat ilit y”). (We could also par amet r ize t he skew in t er ms of t he
call delt a as .)
To der ive t he for mula for t he fair var iance we follow t he same pr oce
dur e as in Appendix B, st ar t ing wit h Equat ion B2. One impor t ant dif
fer ence is t hat now implied volat ilit y is nonlinear in (since
depends implicit ly on ) so t hat second der ivat ives have an addit ional
t er m:
( C 2)
Ot her der ivat ives we need ar e easily calculat ed:
( C 3)
wher e
and
∆
p
Σ ∆
p
( ) Σ
0
b ∆
p
1
2
 +
. ,
 `
+ =
Σ
0
∆
p
1 2 ⁄ – =
Σ ∆
c
( ) Σ
0
b ∆
c
1
2
 –
. ,
 `
+ =
b ∆
p
b
b
2
2
∂
∂ P
b 0 =
Σ
2
2
∂
∂ P
Σ
0
b ∂
∂Σ
. ,
 `
2
b 0 =
Σ ∂
∂P
Σ
0
b
2
2
∂
∂ Σ
b 0 =
+ =
b
2
2
∂
∂ C
b 0 =
Σ
2
2
∂
∂ C
Σ
0
b ∂
∂Σ
. ,
 `
2
b 0 =
Σ ∂
∂C
Σ
0
b
2
2
∂
∂ Σ
b 0 =
+ =
b ∂
∂Σ
b 0 =
∆
p
1
2
 + =
b
2
2
∂
∂ Σ
b 0 =
2 ∆
p
1
2
 +
. ,
 `
Σ
0
∂
∂∆
p
=
∆
p
N d
1
– ( ) – =
45
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Combining t hese r elat ions, t he fair var iance can be wr it t en as
( C 4)
Again, int egr als can be evaluat ed by changing t he int egr at ion var iable
t o , wher e , so t hat
All t hese int egr als can be evaluat ed exact ly. Since we ar e event ually
int er est ed in expanding t he r esult in power s of , one can ﬁr st
expand in power s of and int egr at e t er m by t er m. It is also
useful t o not e t hat is ant isymmet r ic in t o simplify calcula
t ions. In addit ion t he following r esult s ar e useful:
d
1
S
F
K
 log
Σ
0
T

1
2
Σ
0
T + =
K
v ar
Σ
0
2
b
2
T
e
rT
. ,
 `
S T
2π

1
K
2

0
∞
∫
∆
p
1
2
 +
. ,
 `
e
d
1
2
2 ⁄ –
K d – – =
1
2
b
2 2
T
e
rT
. ,
 `
S T
2π

1
K
2
 ∆
p
1
2
 +
. ,
 `
2
d
1
Σ
0
∂
∂d
1
e
d
1
2
2 ⁄ –
K d
0
∞
∫
.

2
1
K
2
 ∆
p
1
2
 +
. ,
 `
Σ
0
∂
∂∆
p
e
d
1
2
2 ⁄ –
K d
0
∞
∫
–
,
`
z
S
F
K
 log
1
2
v
0
+
. ,
 `
v
0
⁄ d
1
≡ = v
0
Σ
0
2
T =
K
v ar
Σ
0
2
b 2Σ
0
N z ( )
1
2
 –
∞ –
∞
∫
e
v
0
z v
0
2 ⁄ –
e
z
2
2 ⁄ –
d z
2π

¹ ¹
¹ ¹
' '
¹ ¹
¹ ¹
– + =
b
2
N z ( )
1
2
 –
2
z
2
v
0
z – ( )e
v
0
z v
0
2 ⁄ –
e
z
2
2 ⁄ –
d z
2π

∞ –
∞
∫
.

2 N z ( )
1
2
 – z v
0
– ( )e
v
0
z v
0
2 ⁄ –
e
z
2
– d z
2π

∞ –
∞
∫
–
,
`
v
0
Σ
0
2
T =
e
v
0
z v
0
2 ⁄ –
z
N z ( )
1
2
 – z
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
46
( C 5)
Aft er evaluat ing all int egr als we ﬁnd t he ﬁnal answer t o be
( C 6)
Two Slop e M od e l Our calculat ions can easily be gener alized t o t he model wher e t he slope
of t he skew is differ ent for put and call opt ions, i.e.
( C 7)
We now br ieﬂy sket ch t he der ivat ion emphasizing only t he differ ences
wit h t he above det ailed calculat ions. We st ar t wit h t he same funda
ment al expr ession:
( C 8)
z
2n
e
az
2
2 ⁄ –
0
∞
∫
N bz ( )
1
2
 –
d z
2π

2 – ( )
n
2π

a
n
n
∂
∂ 1
a
 ar ct an
b
a
 =
z
2n 1 +
e
az
2
2 ⁄ –
0
∞
∫
N bz ( )
1
2
 –
d z
2π

2 – ( )
n
b
2 2π

a
n
n
∂
∂ 1
a a b
2
+
 =
z
2n
e
az
2
2 ⁄ –
0
∞
∫
N bz ( )
1
2
 –
2
d z
2π

2 – ( )
n
2π

a
n
n
∂
∂ 1
a
 ar ct an
a 2b
2
+
a

π
4
 –
. ,
 `
=
z
2n 1 +
e
az
2
2 ⁄ –
0
∞
∫
N bz ( )
1
2
 –
2
d z
2π

2 – ( )
n
b
2π ( )
3 2 ⁄

a
n
n
∂
∂ 1
a a b
2
+
 ar ct an
b
a b
2
+

. ,
 `
=
K
v ar
Σ
0
2
bΣ
0
2 T
π

1
12
b
2
.... + + + =
Σ
p
∆
p
( ) Σ
0
b
p
∆
p
1
2
 +
. ,
 `
+ = for
1
2
 – ∆
p
0 ≤ ≤
Σ
c
∆
c
( ) Σ
0
b
c
∆
c
1
2
 –
. ,
 `
+ = for 0 ∆
c
1
2
 ≤ ≤
K
v ar
2
T
 rT
S
0
S
*
e
rT
1 –
. ,
 `
S
*
S
0
 log – + –
.

=
e
rT 1
K
2
 P K Σ
p
b
p
( ) , ( ) K d
0
S
*
∫
e
rT
+
1
K
2
C K Σ
c
b
c
( ) , ( ) K d
S
*
∞
∫ ,
`
47
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
We use differ ent implied volat ilit y par amet er izat ions for put and call
opt ions, as given by Equat ion C7. Not e t hat we should choose S
*
so
t hat
This ensur es t hat we use t he put (call) par amet er izat ion in Equat ion
C7 for st r ikes below (above) S
*.
We expand put opt ion pr ices in power s
of and call opt ion pr ices in power s of . Evaluat ing all int egr als as
above we ﬁnd
( C 9)
Obviously, for t his r educes t o t he r esult for single slope given
in Equat ion C6. Not e t hat by changing t he sign of we t ur n t he
implied skew int o a smile.
S
*
S
F
e
Σ
0
2
T 2 ⁄ –
=
b
p
b
c
K
v ar
Σ
0
2 1
4
Σ
0
b
p
b
c
– ( ) Σ
0
b
p
b
c
+
2
 Σ
0
T
π

. ,
 `
+
1
12

b
p
2
b
c
2
+
2
 .... + + + =
b
p
b
c
=
b
c
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
48
APPENDIX D:
STATIC AND DYNAM IC
REPLICATIO N O F A
VO LATILITY SWAP
We have ar gued t hat volat ilit y swaps ar e fundament ally differ ent fr om
var iance swaps and t hat , unlike t he var iance swap, t her e is no simple
r eplicat ing st r at egy t o synt het ically cr eat e a volat ilit y swap.
In t he sect ion From Variance to Volatility Contracts on page 33, we showed t hat
at t empt ing t o cr eat e a volat ilit y swap fr om a var iance swap by means
of a “buyandhold” st r at egy invar iably leads t o misr eplicat ion, since
t his amount s t o t r ying t o ﬁt a linear payoff (t he volat ilit y payoff) wit h a
quadr at ic payoff (t he var iance swap).
Given a view on bot h t he dir ect ion and volat ilit y of fut ur e volat ilit y, we
will show t hat it is possible t o pick t he st r ike and not ional size of a
var iance cont r act t o mat ch t he payoff of a volat ilit y cont r act , on aver 
age, as closely as possible. The ext ent of t he r eplicat ion mismat ch will
depend on how close t he r ealized volat ilit y is t o it s expect ed value.
The hedging inst r ument is t he r ealized var iance ( ), while t he t ar get
of t he r eplicat ion is t he r ealized volat ilit y ( ). We want t o appr oxi
mat e t he volat ilit y as a funct ion of t he var iance by wr it ing
( D 1)
and choose a and b t o minimize t he expect ed squar ed deviat ion of t he
t wo sides of Equat ion D1:
( D 2)
Differ ent iat ion leads t o t he following equat ions for t he coefﬁcient s
and :
( D 3)
The dist r ibut ion of fut ur e volat ilit y could be assumed t o be nor mal,
wit h mean and st andar d deviat ion :
( D 4)
This model only makes sense if t he pr obabilit y of negat ive volat ilit ies
is negligible. This st r at egy will r eplicat e only on aver age; t he expect ed
squar ed r eplicat ion er r or is given by:
Σ
T
2
Σ
T
Σ
T
aΣ
T
2
b + ≈
min E Σ
T
aΣ
T
2
– b – ( )
2
[ ]
a
b
E Σ
T
[ ] aE Σ
T
2
[ ] b
E Σ
T
3
[ ]
+
aE Σ
T
4
[ ] bE Σ
T
2
[ ] +
=
=
Σ σ
Σ
Σ
T
N Σ σ
Σ
, ( ) ∼
49
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
( D 5)
For r ealized volat ilit ies dist r ibut ed nor mally as in Equat ion D4, t he
hedging coefﬁcient s ar e
( D 6)
and t he expect ed squar ed r eplicat ion er r or is:
( D 7)
min E Σ
T
aΣ
T
2
– b – ( )
2
[ ] Var Σ
T
( ) 1 cor r Σ
T
Σ
T
2
, ( ) ( )
2
– [ ] =
a
1
2Σ
σ
Σ
2
Σ
 +

b
Σ
2
σ
Σ
2
Σ
2
 +

=
=
min E Σ
T
aΣ
T
2
– b – ( )
2
[ ]
σ
Σ
2
1
2Σ
2
σ
Σ
2
 +
 =
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
50
REFERENC ES Car r, P. and D. Madan (1998). Towar ds a Theor y of Volat ilit y Tr ading,
in Volat ilit y: New Est imat ion Techniques for Pr icing Der ivat ives,
edit ed by R. J ar r ow, 417427.
Der man, E. and I. Kani (1994). Riding on a Smile, RIS K 7, No. 2, 3239.
Der man, E., I. Kani and J. Zou (1996). The Local Volat ilit y Sur face,
Financial Analyst Journal, J uly/ August, 2536.
Der man, E., M. Kamal, I. Kani, and J. Zou (1996). Valuing Contracts
with Payoffs Based on Realized Volatility, Global Der ivat ives (J uly)
Quar t er ly Review, Equit y Der ivat ives Resear ch, Goldman, Sachs & Co.
Der man, E., M. Kamal, I. Kani, J. McClur e, C. Pir ast eh, and J. Zou
(1996). Investing in Volatility, Quant it at ive St r at egies Resear ch Not es,
Oct ober, Goldman, Sachs & Co.
Der man, E. and I. Kani (1998). St ochast ic Implied Tr ees: Ar bit r age
Pr icing wit h St ochast ic Ter m and St r ike St r uct ur e of Volat ilit y, Inter
national Journal of Theoretical and Applied Finance, Vol. 1, No. 1, 61
110.
Dupir e, B. (1994). Pr icing wit h a Smile, RIS K 7, No. 1, 1820.
Ledoit , O. and P. Sant aClar a (1998). Relat ive Pr ice of Opt ions wit h
St ochast ic Volat ilit y. Wor king Paper, The Ander son Gr aduat e Schools
of Management , Univer sit y of Califor nia, Los Angeles.
Neuber ger, A. (1994). The Log Cont r act : A new inst r ument t o hedge
volat ilit y, Journal of Portfolio Management, Winter, 7480.
Neuber ger, A. (1996). The Log Cont r act and Ot her Power Cont r act s, in
The Handbook of Exot ic Opt ions, edit ed by I. Nelken, 200212.
51
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
SELECTED QUANTITATIVE STRATEGIES PUBLICATIONS
J une 1990 Understanding Guaranteed ExchangeRate
Contracts In Foreign S tock Investments
Emanuel Der man, Piot r Kar asinski and J effr ey Wecker
J an. 1992 Valuing and Hedging Outperformance Options
Emanuel Der man
Mar. 1992 PayOnExercise Options
Emanuel Der man and Ir aj Kani
J une 1993 The Ins and Outs of Barrier Options
Emanuel Der man and Ir aj Kani
J an. 1994 The Volatility S mile and Its Implied Tree
Emanuel Der man and Ir aj Kani
May 1994 S tatic Options Replication
Emanuel Der man, Deniz Er gener and Ir aj Kani
May 1995 Enhanced Numerical Methods for Options
with Barriers
Emanuel Der man, Ir aj Kani, Deniz Er gener
and Indr ajit Bar dhan
Dec. 1995 The Local Volatility S urface: Unlocking the
Information in Index Option Prices
Emanuel Der man, Ir aj Kani and J oseph Z. Zou
Feb. 1996 Implied Trinomial Trees of the Volatility S mile
Emanuel Der man, Ir aj Kani and Neil Chr iss
Apr. 1996 Model Risk
Emanuel Der man,
Aug. 1996 Trading and Hedging Local Volatility
Ir aj Kani, Emanuel Der man and Michael Kamal
Oct . 1996 Investing in Volatility
Emanuel Der man, Michael Kamal, Ir aj Kani,
J ohn McClur e, Cyr us Pir ast eh and J oseph Zou
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
52
Apr. 1997 Is the Volatility S kew Fair?
Emanuel Der man, Michael Kamal, Ir aj Kani
and J oseph Zou
Apr. 1997 S tochastic Implied Trees: Arbitrage Pricing with
S tochastic Term and S trike S tructure of Volatility
Emanuel Der man and Ir aj Kani
Sept . 1997 The Patterns of Change in Implied Index Volatilities
Michael Kamal and Emanuel Der man
Nov. 1997 Predicting the Response of Implied Volatility to Large
Index Moves: An October 1997 S &P Case S tudy
Emanuel Der man and J oe Zou
Sept . 1998 How to Value and Hedge Options on Foreign Indexes
Kr esimir Demet er ﬁ
J an. 1999 Regimes of Volatility: S ome Observations on the
Variation of S &P 500 Implied Volatilities
Emanuel Der man
Goldman Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
Copyright 1999 Goldman, Sachs & Co. All rights reserved. This material is for your private information, and we are not soliciting any action based upon it. This report is not to be construed as an offer to sell or the solicitation of an offer to buy any security in any jurisdiction where such an offer or solicitation would be illegal. Certain transactions, including those involving futures, options and high yield securities, give rise to substantial risk and are not suitable for all investors. Opinions expressed are our present opinions only. The material is based upon information that we consider reliable, but we do not represent that it is accurate or complete, and it should not be relied upon as such. We, our afﬁliates, or persons involved in the preparation or issuance of this material, may from time to time have long or short positions and buy or sell securities, futures or options identical with or related to those mentioned herein. This material has been issued by Goldman, Sachs & Co. and/or one of its afﬁliates and has been approved by Goldman Sachs International, regulated by The Securities and Futures Authority, in connection with its distribution in the United Kingdom and by Goldman Sachs Canada in connection with its distribution in Canada. This material is distributed in Hong Kong by Goldman Sachs (Asia) L.L.C., and in Japan by Goldman Sachs (Japan) Ltd. This material is not for distribution to private customers, as deﬁned by the rules of The Securities and Futures Authority in the United Kingdom, and any investments including any convertible bonds or derivatives mentioned in this material will not be made available by us to any such private customer. Neither Goldman, Sachs & Co. nor its representative in Seoul, Korea is licensed to engage in securities business in the Republic of Korea. Goldman Sachs International or its afﬁliates may have acted upon or used this research prior to or immediately following its publication. Foreign currency denominated securities are subject to ﬂuctuations in exchange rates that could have an adverse effect on the value or price of or income derived from the investment. Further information on any of the securities mentioned in this material may be obtained upon request and for this purpose persons in Italy should contact Goldman Sachs S.I.M. S.p.A. in Milan, or at its London branch ofﬁce at 133 Fleet Street, and persons in Hong Kong should contact Goldman Sachs Asia L.L.C. at 3 Garden Road. Unless governing law permits otherwise, you must contact a Goldman Sachs entity in your home jurisdiction if you want to use our services in effecting a transaction in the securities mentioned in this material. Note: Options are not suitable for all investors. Please ensure that you have read and understood the current options disclosure document before entering into any options transactions.
2
Goldman Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
SUMMARY Volatility swaps are forward contracts on future realized stock volatility. Variance swaps are similar contracts on variance, the square of future volatility. Both of these instruments provide an easy way for investors to gain exposure to the future level of volatility. Unlike a stock option, whose volatility exposure is contaminated by its stockprice dependence, these swaps provide pure exposure to volatility alone. You can use these instruments to speculate on future volatility levels, to trade the spread between realized and implied volatility, or to hedge the volatility exposure of other positions or businesses. In this report we explain the properties and the theory of both variance and volatility swaps, ﬁrst from an intuitive point of view and then more rigorously. The theory of variance swaps is more straightforward. We show how a variance swap can be theoretically replicated by a hedged portfolio of standard options with suitably chosen strikes, as long as stock prices evolve without jumps. The fair value of the variance swap is the cost of the replicating portfolio. We derive analytic formulas for theoretical fair value in the presence of realistic volatility skews. These formulas can be used to estimate swap values quickly as the skew changes. We then examine the modiﬁcations to these theoretical results when reality intrudes, for example when some necessary strikes are unavailable, or when stock prices undergo jumps. Finally, we brieﬂy return to volatility swaps, and show that they can be replicated by dynamically trading the more straightforward variance swap. As a result, the value of the volatility swap depends on the volatility of volatility itself. _________________ Kresimir Demeterﬁ Emanuel Derman Michael Kamal Joseph Zou _________________ Acknowledgments: We thank Emmanuel Boussard, Llewellyn Connolly, Rustom Khandalavala, Cyrus Pirasteh, David Rogers, Emmanuel Roman, Peter Selman, Richard Sussman, Nicholas Warren and several of our clients for many discussions and insightful questions about volatility swaps. _________________ Editorial: Barbara Dunn (212) 3574611 (212) 9020129 (212) 3573722 (212) 9029794
1
Goldman Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
Table of Contents
INTRODUCTION ......................................................................................... 1 Volatility Swaps ........................................................................... 1 Who Can Use Volatility Swaps? ................................................. 2 Variance Swaps ............................................................................ 3 Outline .......................................................................................... 4 I. REPLICATING VARIANCE SWAPS: FIRST STEPS ...................................... 6 The Intuitive Approach ............................................................... 6 Trading Realized Volatility with a Log Contract ..................... 11 The Vega, Gamma and Theta of a Log Contract ...................... 11 Imperfect Hedges ...................................................................... 13 The Limitations of the Intuitive Approach .............................. 13 II. REPLICATING VARIANCE SWAPS: GENERAL RESULTS ....................... 15 Valuing and Pricing the Variance Swap.................................. 17 III. AN EXAMPLE OF A VARIANCE SWAP ................................................ 20 IV. EFFECTS OF THE VOLATILITY SKEW ................................................ 23 Skew Linear in Strike ............................................................... 23 Skew Linear in Delta ................................................................ 25 V. PRACTICAL PROBLEMS WITH REPLICATION ....................................... 27 Imperfect Replication Due to Limited Strike Range ............... 27 The Effect of Jumps on a Perfectly Replicated Log Contract .. 29 The Effect of Jumps When Replicating With a Finite Strike Range.............................................................. 32 VI. FROM VARIANCE TO VOLATILITY CONTRACTS ................................. 33 Dynamic Replication of a Volatility Swap ............................... 34 CONCLUSIONS AND FUTURE INNOVATIONS ............................................ 36 APPENDIX A: REPLICATING LOGARITHMIC PAYOFFS .............................. 37 APPENDIX B: SKEW LINEAR IN STRIKE .................................................. 40 APPENDIX C: SKEW LINEAR IN DELTA ................................................... 44 APPENDIX D: STATIC AND DYNAMIC REPLICATION OF A VOLATILITY SWAP .................................................. 48 REFERENCES .......................................................................................... 50
0
Goldman Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
INTRODUCTION
A stock’s volatility is the simplest measure of its riskiness or uncertainty. Formally, the volatility σR is the annualized standard deviation of the stock’s returns during the period of interest, where the subscript R denotes the observed or “realized” volatility. This note is concerned with volatility swaps and other instruments suitable for trading volatility1. Why trade volatility? Just as stock investors think they know something about the direction of the stock market, or bond investors think they can foresee the probable direction of interest rates, so you may think you have insight into the level of future volatility. If you think current volatility is low, for the right price you might want to take a position that proﬁts if volatility increases. Investors who want to obtain pure exposure to the direction of a stock price can buy or sell short the stock. What do you do if you simply want exposure to a stock’s volatility? Stock options are impure: they provide exposure to both the direction of the stock price and its volatility. If you hedge the options according to BlackScholes prescription, you can remove the exposure to the stock price. But deltahedging is at best inaccurate because the real world violates many of the BlackScholes assumptions: volatility cannot be accurately estimated, stocks cannot be traded continuously, transactions costs cannot be ignored, markets sometimes move discontinuously and liquidity is often a problem. Nevertheless, imperfect as they are, until recently options were the only volatility vehicle available.
Volatility Swaps
The easy way to trade volatility is to use volatility swaps, sometimes called realized volatility forward contracts, because they provide pure exposure to volatility (and only to volatility). A stock volatility swap is a forward contract on annualized volatility. Its payoff at expiration is equal to ( σ R – K vol ) × N
(EQ 1)
where σR is the realized stock volatility (quoted in annual terms) over the life of the contract, Kvol is the annualized volatility delivery price, and N is the notional amount of the swap in dollars per annualized volatility point. The holder of a volatility swap at expiration receives N dollars for every point by which the stock’s realized volatility σR has
1. For a discussion of volatility as an asset class, see Derman, Kamal, Kani, McClure, Pirasteh, and Zou (1996).
1
low ones will likely rise.000/(volatility point). • The annualization factor in moving from daily or hourly observations to annualized volatilities – for example. volatility is often negatively correlated with stock or index level. using 260 business days per year as a multiplicative factor in computing annualized variances from daily returns. This provides a much more direct method than trading and hedging options. It is likely to grow when uncertainty and risk increase. As with interest rates. As we will show later. and tends to stay high after large downward moves in the market. Therefore. for example 30%. The procedure for calculating the realized volatility should be clearly speciﬁed with respect to the following aspects: • The source and observation frequency of stock or index prices – for example. high volatilities will eventually decrease.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES exceeded the volatility delivery price Kvol. Given these tendencies. by unwinding 2 . and • Whether the standard deviation of returns is calculated by subtracting the sample mean from each return. N = $250. if you foresee a rapid decline in political and ﬁnancial turmoil after a forthcoming election. a short position in volatility might be appropriate. As with all forward contracts or swaps. For frequently observed prices. several uses for volatility swaps follow. Who Can Use Volatility Swaps? Volatility has several characteristics that make trading attractive. Clients who want to speculate on the future levels of stock or index volatility can go long or short realized volatility with a swap. The notional amount is typically quoted in dollars per volatility point. or by assuming a zero mean. volatilities appear to revert to the mean. Finally. Trading the Spread between Realized and Implied Volatility Levels. because it corresponds most closely to the contract that can be replicated by options portfolios. for example. The zero mean method is theoretically preferable. For example. the difference is usually negligible. using daily closing prices of the S&P 500 index. Directional Trading of Volatility Levels. the fair delivery price Kvol of a volatility swap is a value close to the level of current implied volatilities for options with the same expiration as the swap. The delivery price Kvol is typically quoted as a volatility. He or she is swapping a ﬁxed volatility Kvol for the actual (“ﬂoating”) future volatility σR. the fair value of volatility at any time is the delivery price that makes the swap currently have zero value.
that has more fundamental theoretical signiﬁcance. or volatility squared. Variance Swaps Although options market participants talk of volatility. There nesses that are implicitly short volatility: are several busi • Risk arbitrageurs or hedge funds often take positions which assume that the spread between stocks of companies planning mergers will narrow. Since volatility is one of the few parameters that tends to increase during global equity declines. it is variance. This is so because the correct way to value a swap is to value the portfolio that replicates it. the merger may become less likely and the spread may widen. the square of the realized volatility. The holder of a variance swap at expiration receives N dollars for every point by which the stock’s realized variance σ R has exceeded the variance delivery price Kvar. • Equity funds are probably short volatility because of the negative correlation between index level and volatility. A variance swap is a forward contract on annualized variance. • Investors following active benchmarking strategies may require more frequent rebalancing and greater transactions expenses during volatile periods. and the swap that can be replicated most reliably (by portfolios of options of varying strikes. and N is the notional amount of the swap in dollars per annualized volatility point squared.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES the swap before expiration. • Portfolio managers who are judged against a benchmark have tracking error that may increase in periods of higher volatility. especially for ﬁnancial businesses. Kvar is the delivery price for variance. If overall market volatility increases. as we show later) is a variance swap. Its payoff at expiration is equal to ( σ R – K var ) × N 2 2 (EQ 2) where σ R is the realized stock variance (quoted in annual terms) over the life of the contract. a long volatility hedge may be appropriate. diversiﬁcation across countries has become a less effective portfolio hedge. As global equity correlations have increased. 2 3 . you can trade the spread between realized and implied volatility. Hedging Implicit Volatility Exposure.
At the end.000/(one volatility point)2. we provide a detailed numerical example of the valuation of a variance swap. this presentation is capable of much greater generalization. Some practical issues concerning the choice of strikes are also discussed. provides a payoff equal to the variance of the stock’s returns under a fairly wide set of circumstances. for a skew linear in strike or linear in delta. especially for index options. is signiﬁcantly affected by the volatility smile or skew. Therefore. from a variety of viewpoints. The fair value of variance is the delivery price that makes the swap have zero value. The formulas and the intuition they provide are beneﬁcial in rapidly estimating the effect of changes in the skew on swap values. variance swaps are less commonly traded. so that their market prices determine the cost of the variance swap. the log contract. 4 . First. we will return to a discussion of the additional risks involved in replicating and valuing volatility swaps. We also provide insight into the swap by showing. for example (30%)2. without depending on the full validity of the BlackScholes model. for example. we explain how this exotic option itself can be replicated by a portfolio of standard stock options with a range of strikes. The delivery price Kvar can be quoted as a volatility squared. Section II derives the same results much more rigorously and generally. In Section III. we devote Section IV to the effects of the skew. Outline Most of this note will focus on the theory and properties of variance swaps. This cost. without detailed numerical computation. Because of its fundamental role. which provide similar volatility exposure to straight volatility swaps. In particular. The fair value of the variance swap is determined by the cost of the replicating portfolio of options. we derive theoretical formulas that allow us to simply determine the approximate effect of the skew on the fair value of index variance swaps. and so their quoting conventions vary. Section I presents an intuitive. variance can serve as the basic building block for constructing other volatilitydependent instruments.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES Though theoretically simpler. the notional amount can be expressed as $100. how the apparently complex hedged log contract produces an instrument with the simple constant exposure to the realized variance of a variance swap. we show that the hedging of a (slightly) exotic stock option. Though more difﬁcult. Similarly. BlackScholesbased account of the fundamental strategy by which a variance swap can be replicated and valued. Then.
it is useful to regard volatility as the square root of variance. Finally. we derive the analogous formulas for a skew varying linearly with the delta exposure of the options. In Appendix B. on the volatility of volatility. volatility is itself a squareroot derivative contract on variance. which assumes continuous stock price evolution. Appendix D provides additional insight into the static and dynamic hedging of a volatility swap using the variance as an underlyer. a volatility swap can be dynamically hedged by trading the underlying variance swap. not all strikes are available. and its value depends on the volatility of the underlying variance – that is. Since variance can be replicated relatively simply.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES The fair value of a variance swap is based on (1) the ability to replicate a log contract by means of a portfolio of options with a (continuous) range of strikes. In practice. In Appendix C. In Appendix A. Four appendices cover some more advanced mathematics. we derive the details of the replication of a log contract by means of a continuum of option strikes. and (2) on classical options valuation theory. and stock prices can jump. It also shows how the replication can be approximated in practice when only a discrete set of strikes are available. Section V discusses the effects of these real limitations on pricing. From this point of view. Thus. Section VI explains the risks involved in replicating a volatility contract. 5 . we derive the approximate formulas for the value of an index variance contract in the presence of a volatility skew that varies linearly with strike.
In the next section. a single option is an imperfect vehicle: as soon as the stock price moves.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES REPLICATING VARIANCE SWAPS: FIRST STEPS The Intuitive Approach: Creating a Portfolio of Options Whose Variance Sensitivity is Independent of Stock Price In this section. Figure 1a shows a graph of how V varies with stock price S. (We have written the option value as a function of σ τ in order to make clear that all its dependence on both volatility and time to expiration is expressed in the combined variable σ τ . 6 . The cost of implementing that strategy is the fair value of future realized variance. where 2σ 2π 2 log ( S ⁄ K ) + ( σ τ ) ⁄ 2 d 1 = . because. in the BlackScholes formula with zero interest rate. we explain the replicating strategy that captures realized variance. your sensitivity to further changes in variance is altered. We approach variance replication by building on the reader’s assumed familiarity with the standard BlackScholes model. What you want is a portfo2 2 2 2. σ is the return volatility of the stock. we deﬁne the sensitivity V = 2 ∂C BS ∂σ 2 S τ exp ( – d 1 ⁄ 2 ) = .. as long as the stock price evolves continuously – that is. we shall provide a more general proof that you can replicate variance. for each of three different options with strikes 80. 100 and 120.. σ is the stock’s variance. Note that d1 depends only on the two combinations S/K and σ τ . and falls off rapidly as the stock price moves in or out of the money. We will sometimes refer to V as “variance σ τ vega”. We ease the development of intuition by assuming here that the riskless interest rate is zero. it measures the change in value of the position resulting from a change in variance2. Here. whose value is given by the BlackScholes formula C BS ( S. Suppose at time t you own a standard call option of strike K and expiration T. the variance exposure V is largest when the option is at the money. V decreases extremely rapidly as S leaves the vicinity of the strike K. and v = σ τ is the total variance of the stock to expiration. without jumps. For each strike. options values depend on the total variance σ τ .. K . σ τ ) . even when some of the BlackScholes assumptions fail. If you want a long position in future realized variance. τ is the time to expiration (T − t).) We will call the exposure of an option to a stock’s variance V . V is closely related to the time sensitivity or time decay of the option. where S is the current stock price.
An option with higher strike will therefore produce a V contribution that increases with S. Figures 1c. the portfolio with weights inversely proportional to K produces a V that is virtually independent of stock price S. one needs diminishing amounts of higherstrike options.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES lio whose sensitivity to realized variance is independent of the stock price S. As the stock price moves up to higher values. Therefore. weighted inversely proportional to K . to offset this accumulation of Sdependence. the solid line represents the sum with weights in inverse proportion to the square of their strike. Because outof2 2 2 2 7 . Figures 1d. the contributions of all options overlap at any deﬁnite S. f and h show the sensitivity for the equallyweighted and strikeweighted portfolios. In addition. Clearly. Appendix A provides a mathematical derivation of the requirement that options be weighted inversely proportional to K in order to achieve constant V. just what is needed to trade variance. each additional option of higher strike in the portfolio will provide an additional contribution to V proportional to that strike. with weights inversely proportional to K . What does this portfolio of options look like. e and g show the individual sensitivities to variance of increasing numbers of options. and how does trading it capture variance? Consider the portfolio Π ( S. σ τ ) of options of all strikes K and a single expiration τ. you need to combine options of many strikes. weighted in inverse proportion to the square of the strike level. The dotted line represents the sum of equally weighted strikes. This follows from the formula in footnote 2. each panel having the options more closely spaced. You can also understand this intuitively. as long as S lies inside the range of available strikes and far from the edge of the range. and you can observe it in the increasing height of the V peaks in Figure 1a. What combination of strikes will give you such undiluted variance exposure? Figure 1b shows the variance exposure for the portfolio consisting of all three option strikes in Figure 1a. you will obtain an exposure to variance that is independent of stock price. and provided the strikes are distributed evenly and closely. If you own a portfolio of options of all strikes. To obtain a portfolio that responds to volatility or variance independent of moves in the stock price.
V. Each ﬁgure on the left shows the individual V i contributions for each option of strike Ki.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES FIGURE 1. and becomes totally independent of stock price S inside the strike range strikes: 80. weighted two different ways.100. of portfolios of call options of different strikes as a function of stock price S. The corresponding ﬁgure on the right shows the sum of the contributions. the solid line corresponds to weights inversely proportional to K2. the dotted line corresponds to an equallyweighted sum of options.120 equally weighted weighted inversely proportional to square of strike (a) (b) 20 60 100 140 180 20 60 100 140 180 strikes 60 to 140 spaced 20 apart (c) (d) 20 60 100 140 180 20 60 100 140 180 strikes 60 to 140 spaced 10 apart (e) (f) 20 60 100 140 180 20 60 100 140 180 strikes 20 to 180 spaced 1 apart (g) (h) 20 60 100 140 180 20 60 100 140 180 8 . The variance exposure.
0 ) = . σ τ ) for strikes K varying continuously from zero up to some reference price S*. the expression for the portfolio is given by Π ( S. at time t you can sum all the BlackScholes options values to show that the total portfolio value is 2 S – S* S σ τ Π ( S. namely 3. one can show that the sum of all the payoff values of the options in the portfolio is simply ST – S* S T Π ( S T . the variance exposure of Π is τ V = 2 (EQ 5) 2 To obtain an initial exposure of $1 per volatility point squared. we employ put options P ( S. σ τ ) = 1 1 C ( S.– log  S* S* (EQ 3) where log( ) denotes the natural logarithm function. σ τ ) = . and call options C ( S. σ τ ) + . we will use Π to refer to the value of this new portfolio. when t = T. Note how little the value of the portfolio before expiration differs from its value at expiration at the same stock price. From now on. You can think of S* as the approximate atthemoney forward stock level that marks the boundary between liquid puts and liquid calls. K . K . At expiration.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES themoney options are generally more liquid. Similarly. and ST is the terminal stock price. Formally. σ τ ) ∑ K2 2 K K > S* K < S* ∑ 9 . +  S * S* 2 (EQ 4) where S is the stock price at time t. σ τ ) for strikes varying continuously from S* to inﬁnity3.P ( S.– log . K . you need to hold (2/T) units of the portfolio Π. The only difference is the additional value due to half the total variance σ τ . K . Clearly.
0) the weighted sum of puts and calls. (a) Individual contributions to the payoff from put options with all integer strikes from 20 to 99.– log S* S* Π ( ST.σ S * T T S* (EQ 6) The ﬁrst term in the payoff in Equation 6. describes 1/S* forward contracts on the stock with delivery price S*. All of the volatility sensitivity of the weighted portfolio of options we have created is contained in the log contract. 0 ) 20 60 100 140 180 20 60 100 140 180 (a) (b) 4. 10 . It is not really an option. – log ( S ⁄ S * ) . The second term. once and for all. a socalled exotic option whose payoff is proportional to the log of the stock at expiration. σ τ ) = .. The log contract was ﬁrst discussed in Neuberger (1994). See also Neuberger (1996). FIGURE 2. weighted payoffs of puts and calls and (2) a long position in a forward contract and a short position in a log contract. its value represents a long position in the stock (value S) and a short position in a bond (value S*). and (2) the payoff of a long position in a forward contract and a short position in a log contract. with weight inversely proportional to the square of the strike. Figure 2 graphically illustrates the equivalence between (1) the summed. and whose correct hedging depends on the volatility of the stock. describes a short position in a log contract4 with reference value S*. (b) The payoff of 1/100 of a long position in a forward contract with delivery price 100 and one short position in a log contract with reference value 100. (S − S*)/S*.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES τ 2 2 S – S* S Π ( S. which can be statically replicated. ST ST – S* . + .– log . An example that illustrates the equivalence at expiration between (1) Π(ST. and call options with all integer strikes from 100 to 180. without any knowledge of the stock’s volatility.
+ . If you take a position in the portfolio Π. will be payoff = ( σ R – σ I ) 2 2 (EQ 7) Looking back at Equation 2. if the realized volatility turns out to have been σ R . in a BlackScholes world with zero interest rates and zero dividend yield. the initial fair value of the position captured by deltahedging would have been S 0 2 2 S0 – S* Π 0 = . both of which can be statically hedged with no dependence on volatility. σ..– log  + σ R T S* S * The net payoff on the position. + . assume that we are in a BlackScholes world where the implied volatility σ I is the estimate of future realized volatility. the portfolio of options whose variance vega is independent of the stock price S can be written (T – t) 2 2 S – S* S Π ( S. you will see that by rehedging the position in log contracts. You will have proﬁted (or lost) if realized volatility has exceeded (or been exceeded by) implied volatility. T ) = . The Vega. been the owner of a position in a variance swap with fair strike Kvar = σ I and face value $1.– log  + σ I T S* S * At expiration. t. the fair value you should pay at time t = 0 when the stock price is S0 is S 0 2 2 S0 – S* Π 0 = . in effect.. In contrast.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES Trading Realized Volatility with a Log Contract For now. Therefore. hedged to expiration.σ S * T T S* The ( S – S * ) term represents a long position in the stock and a short position in a bond. Gamma and Theta of a Log Contract 2 In Equation 6 we showed that. σ.σ S * T T (EQ 8) 11 . you have.log . the log( ) term needs continual dynamic rehedging.– log . T ) = – . t. let us concentrate on the log contract term alone.. whose value at time t for a logarithmic payoff at time T is (T – t) 2 2 S L ( S.
as we now show. The log contract’s exposure to stock price is 21 . all the initial variance has been lost. is a smoother function of S than the sharply peaked gamma of a single option. The time decay of the log contract. It states that the disadvantage of negative theta (the decrease in value with time to expiration) is offset by the beneﬁt of positive gamma (the curvature of the payoff). and decreases linearly to zero as the contract approaches expiration. the rate at which the exposure changes as the stock price moves.Γ = . the rate at which its value changes if the stock price remains unchanged as time passes. The variance vega of the portfolio in Equation 8 is T–t V = . you need a constant long position in $(2/T) worth of stock to be hedged at any time. The gamma of the portfolio. 12 .σ T (EQ 10) The contract loses time value at a constant rate proportional to its variance. T (EQ 9) The exposure to variance is equal to 1 at t = 0.∆ = – . since each share of stock is worth S. Equations 10 and 11 can be combined to show that 2 1 θ + . so that at expiration. is 2 1 .ΓS 2 σ = 0 2 (EQ 12) Equation 12 is the essence of the BlackScholes options pricing theory. The log contract’s gamma. That is. is 1 2 θ = – .TS shares of stock.T S2 (EQ 11) Gamma is a measure of the risk of hedging an option.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES The sensitivities of the value of this portfolio are precisely appropriate for trading variance. being the sum of the gammas of a portfolio of puts and calls.
it is difﬁcult to extend these argument clearly. deviation from constant variance vega develops at the tail of the strike range. weighted in inverse proportion to the strike squared. we have shown that the dynamic. Now. 2 The Limitations of the Intuitive Approach A variance swap has a payoff proportional to realized variance. we move on to a more general and rigorous derivation of the value of variance swaps based on replication. and the correct answers when they do not hold. will be more easily understandable. although the range of strikes is greater. Therefore. trading the imperfectly replicated log contract will allow variance to accrue at the correct rate. Figure 3b shows strikes from $75 to $125. Whenever the stock price moves outside. In practice. As long as the stock price remains within the strike range. as expected for a swap whose value is proportional to the remaining variance σ τ at any time. and the deviation is greater at earlier times. even when the stock and options market satisfy all the BlackScholes assumptions: there are only a ﬁnite number of options available at any maturity. Figure 3a shows the ideal variance vega that results from a portfolio of puts and calls of all strikes from zero to inﬁnity. Here. In this section.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES Imperfect Hedges It takes an inﬁnite number of strikes. the coarser spacing causes the vega surface to develop corrugations between strike values that grow more pronounced closer to expiration. continuous hedging of a log contract produces a payoff whose value is proportional to future realized variance. the reduced vega of the imperfectly replicated log contract will make it less responsive than a true variance swap. The somewhat intuitive derivations in this section have assumed that interest rates and dividend yields are zero. in the presence of an implied volatility skew. this isn’t possible. Figure 3 shows how the variance vega of a threemonth variance swap is affected by imperfect replication. Finally. appropriately weighted. and decreases linearly with time to expiration. but the conditions under which they hold. Many of the results will be similar. assuming the BlackScholes world for stock and options markets. In practice. spaced $10 apart. We have also assumed that all the BlackScholes assumptions hold. We have also shown that you can use a portfolio of appropriately weighted puts and calls to approximate a log contract. Here the variance vega is independent of stock price level. to replicate a variance swap. Figure 3c shows the vega for strikes from $20 to $200. but it is not hard to generalize them. 13 . Figure 1 illustrates that a ﬁnite number of strikes fails to produce a uniform V as the stock price moves outside the range of the available strikes. uniformly spaced $1 apart.
(a) An inﬁnite number of strikes.1 0 60 (c) V v 140 120 100 0.2 80 S τ 0. (c) Strikes from $20 to $200. V (a) v 120 100 0.1 0 60 14 .2 80 140 S τ 0. of a portfolio of puts and calls. and chosen to replicate a threemonth variance swap. uniformly spaced $10 apart.1 0 60 (b) V v 120 100 0.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES FIGURE 3.2 80 140 S τ 0. weighted inversely proportional to the square of the strike level. uniformly spaced $1 apart. The variance vega. (b) Strikes from $75 to $125. V .
or continuous – this means that no jumps are allowed.) Therefore. allowing for dividends does not signiﬁcantly alter the derivation. and E[ ] denotes the expectation. Conceptually. Although our explanation depended on the validity of the BlackScholes model... but are not restricted to. . . we explained how to replicate a variance swap by means of a portfolio of options whose payoffs approximate a log contract. )dZ t St (EQ 13) Here.. These assumptions include. (In a later section. the result – that the dynamic hedging of a log contract captures realized volatility – holds true more generally. we assume that the stock price evolution is given by dS t .. valuing a variance forward contract or “swap” is no different than valuing any other derivative security.= µ ( t. we assume the stock pays no dividends. The only assumption we will make about the future underlyer evolution is that it is diffusive. )dt + σ ( t. we assume that the drift µ and the continuouslysampled volatility σ are arbitrary functions of time and other parameters. The fair delivery value of future realized variance is the strike K var for which the contract has zero present value: K var = E [ V ] (EQ 16) 15 . The value of a forward contract F on future realized variance with strike K is the expected present value of the future payoff in the riskneutral world: F = E [ e – rT ( V – K ) ] (EQ 15) Here r is the riskfree discount rate corresponding to the expiration date T. The theoretical deﬁnition of realized variance for a given price history is the continuous integral 1 T V = .σ 2 ( t. we will consider the effects of discontinuous stock price movements on the success of the replication. implied tree models in which the volatility is a function σ ( S. For simplicity of presentation.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES REPLICATING VARIANCE SWAPS: GENERAL RESULTS In the previous section. … ) dt T 0 ∫ (EQ 14) This is a good approximation to the variance of daily returns used in the contract terms of most variance swaps. t ) of stock price and time only.
You can then use simulation to calculate the fair variance Kvar as the average of the experienced variance along each simulated path consistent with the riskneutral stock price evolution given of Equation 13. where the drift µ is set equal to the riskless rate. see Carr and Madan (1998). This approach was ﬁrst outlined in Derman. the socalled local volatility σ ( S. then one approach for calculating the fair price of variance is to directly calculate the riskneutral expectation T 1 K var = . See. For an alternative discussion. over the next instant of time. we ﬁnd 1 d ( log S t ) = µ – . … ) dt T 0 ∫ (EQ 17) No one knows with certainty the value of future volatility.σ dt 2 St (EQ 19) (EQ 18) in which all dependence on the drift µ has cancelled.E σ 2 ( t. we obtain dS t 1 2 .– d ( log S t ) = . By applying Ito’s lemma to log S t . generates a payoff proportional to the incremental variance of the stock during that time6. Dupire (1994) and Derman. t ) consistent with all current options prices is extracted from the market prices of traded stock options. Derman and Kani (1994). Summing Equation 19 over all times from 0 to T. but it does not provide insight into how to replicate it. Kamal. The essence of the replication strategy is to devise a position that.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES If the future volatility in Equation 13 is speciﬁed. In implied tree models5. 6. Kani and Zou (1996). for example. we obtain the continuouslysampled variance 5. The above approach is good for valuing the contract. Kani.σ 2 dt + σdZ t 2 Subtracting Equation 18 from Equation 13. 16 . and Zou (1996).
so that ST T dS t 2 K var = .σ dt T ∫ 0 (EQ 20) ST 2 = . Following this continuous rebalancing strategy captures the realized variance of the stock from inception to expiration at time T.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES T 1 2 V ≡ . The second term represents a static short position in a contract which. … )dZ St (EQ 22) so that the riskneutral price of the rebalancing component of the hedging strategy is given by E ∫0 St T dS t = rT (EQ 23) This equation represents the fact that a shares position. has a forward price that grows at the riskless rate. continuously rebalanced to be worth $1. the underlyer evolves according to: dS t . pays the logarithm of the total return. The ﬁrst term in the brackets can be thought of as the net outcome of continuous rebalancing a stock position so that it is always instantaneously long 1/S t shares of stock worth $1. In a riskneutral world with a constant riskfree rate r. Instead of averaging over future variances.– log T S0 0 St ∫ (EQ 21) The expected value of the ﬁrst term in Equation 21 accounts for the cost of rebalancing. as long as it moves continuously.– log T 0 St S0 ∫ T dS t This mathematical identity dictates the replication strategy for variance. Note that no expectations or averages have been taken – Equation 20 guarantees that variance can be captured no matter which path the stock price takes. 17 . as in Equation 17.E . Valuing and Pricing the Variance Swap Equation 20 provides another method for calculating the fair variance. one can take the expected riskneutral value of the righthand side of Equation 20 to obtain the cost of replication directly.= rdt + σ ( t. at expiration.
+ log S0 S* S0 (EQ 24) The second term log ( S * ⁄ S 0 ) is constant. by decomposing its shape into linear and curved components. 18 . and then duplicating each of these separately. The following mathematical identity. so only the ﬁrst term has to be replicated.Max ( ST – K .Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES As there are no actively traded log contracts for the second term in Equation 21. with a combination of outofthemoney calls for high stock values and outofthemoney puts for low stock values. and • a similar long position in ( 1 ⁄ K 2 ) call options struck at K. suggests the decomposition of the logpayoff: ST – S* ST – log . can be duplicated using standard options with all possible strike levels and the same expiration time T. 0 ) d K K2 1 ∫S . representing the quadratic and higher order contributions. one must duplicate the log payoff. • a long position in ( 1 ⁄ K 2 ) put options struck at K. at all stock price levels at expiration. The linear component can be duplicated with a forward contract on the stock with delivery time T. 0 ) dK K2 * Equation 25 represents the decomposition of a log payoff into a portfolio consisting of: • a short position in ( 1 ⁄ S * ) forward contracts struck at S*. which holds for all future values of ST. For practical reasons we want to duplicate the log payoff with liquid options – that is. the remaining curved component. for all strikes from S* to ∞ . for all strikes from 0 to S*. The log payoff can then be rewritten as ST ST S* log .Max ( K – S T . Figure 4 shows this decomposition schematically. We introduce a new arbitrary parameter S* to deﬁne the boundary between calls and puts.= log . independent of the ﬁnal stock price ST.= – S* S* + + (forward contract) (put options) (call options) (EQ 25) ∫0 S* ∞ 1 . All contracts expire at time T.
respectively. FIGURE 4. By using the identities in Equations 23 and 25. Replication of the log payoff. The sum of the payoffs for the dashed and solid lines provide the same payoff as the log contract. even when there is an implied volatility skew and the simple BlackScholes formula is invalid. we obtain S* S 0 rT 2 K var = . (a) The payoff of a short position in a log contract at expiration. rT – . ST – log S* portfolio of options ST – S* – S* S* S* (a) (b) 19 . Equation 26 makes precise the intuitive notion that implied volatilities can be regarded as the market’s expectation of future realized volatilities. denote the current fair value of a put and call option of strike K. you obtain an estimate of the current market price of future variance. Each option is weighted by the inverse square of its strike.C ( K ) dK K2 * where P(K) and (C(K)).P ( K ) d K K2 1 (EQ 26) ∫S . Solid line: the curved payoff of calls struck above S* and puts struck below S* . and makes less assumptions than our intuitive treatment in the section on page 6. This approach to the fair value of future variance is the most rigorous from a theoretical point of view. (b) Dashed line: the linear payoff at expiration of a forward contract with delivery price S*.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES The fair value of future variance can be related to the initial fair value of each term on the right hand side of Equation 21. It provides a direct connection between the market cost of options and the strategy for capturing future realized volatility.e – 1 – log T S0 S* + e rT + e rT ∫0 ∞ S* 1 . If you use the market prices of these options.
K ip ) + ∑ w ( K ic )C ( S.– log .e – 1 – log . K ic ) i i (EQ 29) We now illustrate this procedure with a concrete numerical example.– . say S* = S0. and using Equation 26 with only a few strikes leads to appreciable errors. The fair delivery variance is determined by the cost of the replicating strategy discussed in the previous section... The procedure in Appendix A guarantees that these payoffs will always exceed or match the value of the log contract. the fair variance would be given by Equation 26 with some choice of S*. but never be worth less.+ . If you could buy options of all strikes between zero and inﬁnity.E . Once these weights are calculated. 20 . Π CP is obtained from Π CP = ∑ w ( K ip )P ( S.– log T S* S* S* S0 0 St ∫ Taking expectations. however.. this becomes S* S 0 rT 2 K var = . only a small set of discrete option strikes is available. ..Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES AN EXAMPLE OF A VARIANCE SWAP We now present a detailed practical example. and put options with strikes Kip such that K 0 = S * > K 1 p > K 2 p > K 3 p > . In Appendix A we derive the formula that determines how many options of each strike you need in order to approximate the payoff f(ST) by piecewise linear options payoffs. which can be written as ST S* ST – S* T dS t S T – S * 2 K var ≡ . Suppose you want to price a swap on the realized variance of the daily returns of some hypothetical equity index.rT – .+ e rT Π CP T S0 S* (EQ 27) where Π CP is the present value of the portfolio of options with payoff at expiration given by S T 2 ST – S* f ( S T ) = . In practice.– log  T S* S* (EQ 28) Suppose that you can trade call options with strikes Kic such that K 0 = S * < K 1c < K 2c < K 3c < . We start with the deﬁnition of fair variance given by Equation 21. Here we suggest a better approximation.
7384 1.55 30.0002 0.69 28.3615 89.05 96.8671 Assume that the index level S0 is 100.0829 16.8874 0.4119 0.27 0.8691 83. The portfolio of Europeanstyle put and call options used for calculating the cost of capturing realized variance in the presence of the implied volatility skew with a discrete set of options strikes.0459 37.000002 0.63 36.27 82.0958 0.0075 0.0057 0.00 20.5790 2.6747 3.1580 29.9130 1.04 134.0501 0.0013 0.26 63.2578 0.49 56.1289 0.7752 7.23 50. Strike Volatility Weight Value per Option Contribution 50 55 60 65 70 PUTS 75 80 85 90 95 100 100 105 110 CALLS 115 120 125 130 135 30 29 28 27 26 25 24 23 22 21 20 20 19 18 17 16 15 14 13 163.0241 0. the continuously compounded annual riskless interest rate r is 5%.0004 0.63 113.02 22.00003 0. and the maturity of the variance swap is three months (T = 0.0067 0.83 33.2581 0.9939 6. Suppose that 21 .0001 419.0260 75.2854 0.5560 1.3537 4.3616 70.98 24. the dividend yield is zero.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES TABLE 1.98 19.98 72.0003 0.0035 0.15 45.1476 0.0276 0.25.000006 TOTAL COST 0.19 25.
towards the fair value of variance as a function of ∆K. their value is small and contributes little to the total cost. The cost of capturing variance is now simply calculated using Equation 27 with the result K var = ( 20.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES you can buy options with strikes in the range from 50 to 150. with and without a volatility skew. You can see that as the spacing between strikes approaches zero. the spacing between strikes. In Table 1 we provide the list of strikes and their corresponding implied volatilities. At the bottom of the table we show the total cost of the options portfolio. The line with diamond symbols shows similar convergence to a higher fair variance of about 402. 2 420 415 K var 410 405 400 0 1 2 DK 3 4 5 22 . for two cases. Π CP = 419. The theoretical fair variance for ∆K = 0 is then (20)2 = 400. We then show the weights. Although the number of options which are far out of the money is large. It is clear from Table 1 that most of the cost comes from options with strikes near the spot value. the cost of capturing variance with a discrete set of strikes. the extra contribution coming from the effect of the skew. the value of each individual option and the contribution of each strike level to the total cost of the portfolio. Convergence of Kvar. FIGURE 5. We assume that atthemoney implied volatility is 20%. because the procedure of approximating the log contract in Appendix A always overestimates the value of the log contract. The line with square symbols shows the convergence for no skew.8671 . with a skew such that the implied volatility increases by 1 volatility point for every 5 point decrease in the strike level. this value is higher than the true theoretical value for the fair variance obtained by approximating the log contract with a continuum of strikes. In Figure 5 we illustrate the cost of variance as function of the spacing between strikes. with all implied volatilities at the same value of 20%. the cost of capturing variance approaches the theoretically fair variance. This is not strictly the fair variance.467 ) . uniformly spaced 5 points apart.
. In Appendix B we derive the following approximate formula for the fair variance of the contract with time to expiration T: K var ≈ Σ 0 ( 1 + 3Tb 2 + . Note also that there is no term b T in Equation 31. both of which resemble typical index skews. The ﬁrst is a skew that varies linearly with the strike of the option... so that b2T is a dimensionless parameter. A value of b = 0.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES EFFECTS OF THE VOLATILITY SKEW The general strategy discussed in the previous section can be used to determine the fair variance and the hedging portfolio from the set of available options and their implied volatilities. the second a skew that varies linearly with the BlackScholes delta. and the size of the increase is proportional to time to maturity and the square of the skew slope. and therefore a natural candidate for the order of magnitude of the percentage correction to Kvar. Here we discuss the effects of a volatility skew on the fair variance. 7. This formula provides a good rule of thumb for a quick estimate of the impact of the volatility skew on the fair variance. We ﬁrst consider a skew for which the implied volatility varies linearly with strike. ) 2 (EQ 31) The skew increases the value of the fair variance above the atthemoneyforward level of volatility. with an approximate analytic formula that we derive. 23 . The steepness of the skew is determined by the slope b. Note that this parametrization cannot hold for all strikes. for example. the options prices in Equation 26 are negligible and therefore do not affect the value of the fair variance. is 2 volatility points higher than Σ0. with a positive value indicating a higher volatility for strikes below the forward. for a large enough value of K. because. computed from Equation 26. so that K – SF Σ ( K ) = Σ 0 – b S F (EQ 30) Skew Linear in Strike Here Σ0 is the implied volatility of an option struck at the forward. the implied volatility would become negative7.2 means that the implied volatility corresponding to a strike 10% below the forward. Note that for large values of K. This approximation works best for short maturities and skews that are not too steep. (Note that b in Equation 30 has the same dimension as volatility. where this parameterization is invalid. In both cases we will compare the numerically correct value of fair variance. We assume that there is no term structure and consider two different skew parameterizations.
0 0.01 ) ( 30.45 ) ( 31. the continuously compounded annual discount rate r = 5%.3 b (a) b (b) 24 .97 ) ( 30.2 0.2 0. with the approximate value from the formula of Equation 31.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES In Table 2 we compare the exact results for fair variance. We assume Σ0 = 30%. with the approximate analytic formula of Equation 31.1 0. FIGURE 6.99 ) 2 2 2 2 ( 29. as a function of the skew slope b. computed numerically. with the approximate values given by the analytic formula in Equation 31.00 ) ( 30.2 0.33 ) ( 30.3 ( 30.05 ) ( 30. computed numerically. T = 3 months Skew Slope b Exact Value Analytic Approximation 2 2 2 2 T = 1 year Exact Value Analytic Approximation 2 2 2 2 0. 1150 1100 1150 1100 Kvar 1000 950 900 0.75 ) ( 33. and reasonable agreement for one year.81 ) 2 2 2 2 Figure 6 contains a graph of these results.44 ) ( 30. Kvar. and use strikes evenly spaced one point apart from K = 10 to K = 200 to replicate the log payoff.82 ) ( 32.22 ) ( 30. TABLE 2. We see excellent agreement in the case of the threemonth variance swap.3 Kvar 1050 1050 1000 950 900 0. (a) threemonth variance swap. zero dividend yield. (b) oneyear variance swap.65 ) ( 30.1 0. The thin line with squares shows the exact values obtained by replicating the logpayoff.00 ) ( 30.11 ) ( 30. S = 100. Comparison of the exact value of fair variance. Comparison of the exact fair variance. The thick line depicts the approximate value given by Equation 31.1 0.01 ) ( 30.
so that: 1 Σ ( ∆ p ) = Σ 0 + b ∆ p + .2. (b) The corresponding skew plotted as a function of strike.+ . there is an implicit dependence on the time to expiration in the formula of Equation 32. (a) A volatility skew that varies linearly in delta. this parameterization leads to more realistic skews than those produced by the linearstrike formula. Σ 0 is the implied volatility of a “50delta” put option and b is the slope of the skew – that is. This parameter b is not the same as the b in the previous section.5 0. the continuously compounded annual discount rate r is 5%. where d1 is deﬁned in Footnote 2..25 0 Σ 30 25 20 60 80 100 120 140 ∆p K (a) (b) 25 .. since Equation 32 leads to arbitrage violation before b reaches this limit.75 0. in contrast to the skew linear in strike. which was absent from Equation 30. In practice. We have assumed that the stock price S is 100. Since ∆p is bounded. 40 35 40 35 Σ 30 25 20 1 0. Appendix C presents a detailed derivation of the following approximate formula for the fair variance of the contract with time to expiration T: 2 2 1 1b . and the skew slope is 0. the change in the skew per unit delta. This restriction is irrelevant. In particular. because a variation linear in delta about the atthemoneyforward strike is not equivalent to a variation linear in strike... the ﬁrstorder correction is of magnitude b T . the term to maturity is three months. the implied volatility is always positive provided b < 2Σ0.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES Skew Linear in Delta Next we consider a skew that varies linearly with the BlackScholes delta of the option. given by ∆ p = – N ( – d 1 ) .K var ≈ Σ 0 1 + . because of the ∆p term.b T + . 2 (EQ 32) Here ∆p is the BlackScholes exposure of a put option. 12 Σ 2 π 0 (EQ 33) Here. FIGURE 7.
and use strikes evenly spaced one point apart from K = 10 to K = 200 to replicate the log payoff.60 ) ( 32. the continuously compounded annual discount rate r = 5%.49 ) ( 32.2 0. as displayed in Figure 8. Comparison of the fair variance. Kvar.3 1050 1000 950 900 0. 1150 1100 1150 1100 Kvar Kvar 1050 1000 950 900 0. computed numerically. S = 100. with the approximate analytic formula of Equation 33.93 ) 2 2 2 2 ( 29. In Table 3 we compare the results for fair variance.3 ( 30. with the approximate values given by the analytic formula in Equation 33. with the approximate value from the formula of Equation 33. Again. (a) Threemonth variance swap.00 ) ( 31.40 ) ( 34.62 ) ( 31. (b) Oneyear variance swap.01 ) ( 30.06 ) 2 2 2 2 FIGURE 8.97 ) ( 31.3 b (a) b (b) 26 . computed numerically.2 0.2 0. we compare the exact results computed according to Appendix A with the approximate values given by Equation 33. The thin line with squares shows the exact values obtained by replicating the logpayoff. We assume Σ0 = 30%.0 0. T = 3 months Skew Slope b Exact Value Analytic Approximation 2 2 2 2 T = 1 year Exact Value Analytic Approximation 2 2 2 2 0.42 ) ( 34.1 0. we convert the skew by delta in Equation 32 into a skew by strike. zero dividend yield.06 ) ( 32.06 ) ( 30. as a function of the skew slope b. as displayed in Figure 7. The analytic formula works very well for the threemonth variance swap.1 0.61 ) ( 31. TABLE 3. and truly impressively for the oneyear swap.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES First. The thick line depicts the approximate value given by Equation 33.03 ) ( 32.64 ) ( 30. Comparison of the exact value of fair variance.00 ) ( 30.1 0.
so that all options are valued at the same implied volatility. Both of these effects cause the strategy to capture a quantity that is not the true realized variance. (The fair variance is calculated according to Equation 26. First. provided the portfolio of options contains all strikes between zero and inﬁnity in the appropriate weight to match the log payoff. may also corrupt the ideal strategy. 27 . and (2) a portfolio with a wide range of strikes. lower estimate of the fair variance. This portfolio strategy captures variance exactly. In both cases the strikes are uniformly spaced. Second. the stock price may jump. The fair variances are estimated from (1) a replicating portfolio with a narrow range of strikes. the narrow strike range underestimates the fair variance. they will capture less than the true realized variance. We also assume a continuously compounded annual interest rate of 5%. ranging from 75% to 125% of the initial spot level. together with a static long position in a portfolio of options and a forward that together replicate the payoff of a log contract. constantdollar exposure to the stock. Two obvious things can go wrong. more dramatically so for longer expirations. insufﬁcient to accurately replicate the log payoff. In Table 4 below we show how the estimated value of fair variance is affected by the range of strikes that make up the replicating portfolio. though other practical issues. and so produce an inaccurate. they have lower value than that of a true log contract. Since log contracts are not traded in practice. one point apart. the wide strike range accurately approximates the actual square of the implied volatility. However. except that the integrals are replaced by sums over the available option strikes whose weights are chosen according to the procedure of Appendix A). from 50% to 200% of the initial spot level.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES PRACTICAL PROBLEMS WITH REPLICATION We have shown in Equation 20 that a variance swap is theoretically equivalent to a dynamically adjusted. we replicate the payoff with traded standard options in a limited strike range. you may be able to trade only a limited range of options strikes. Therefore. Imperfect Replication Due to Limited Strike Range Variance replication requires a log contract. Because these strikes fail to duplicate the log contract exactly. with no volatility skew. like liquidity. For both expirations. We will focus on the effects of these two limitations below. We assume here that implied volatility is 25% per year for all strikes. and provided the stock price evolves continuously.
the nonlinear part of the log payoff: ST ST – S0 . we have already discussed one approach to understanding why the narrow strike range fails to capture variance. the vega and gamma of a limited strike range both fall to zero when the index moves outside the strike range. or vega) in the options portfolio outside 28 . whose duplication demands an inﬁnite range of strikes. If you own a limited number of strikes. you pay less than the full value. The effect of strike range on estimated fair variance.– log S0 S0 (EQ 34) Figure 9 displays the mismatch between the two payoffs.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES TABLE 4. always growing less rapidly than the nonlinear part of the log contract.9 ) 2 ( 23. As shown in Figure 3. even if no jumps occur and the stock always moves continuously. the estimated variance is lower than the true fair value for both expirations above.125%) Threemonth Oneyear ( 25. and the strategy then fails to accrue realized variance as the stock price moves. that is. from 75 to 125. that is. and. In essence. Beyond this range.200%) Narrow strike range (75% . The narrowstrike option portfolio matches the curved part of the log payoff well at stock price levels between the range of strikes.0 ) 2 ( 25. when the stock price evolves into regions where the curvature of the portfolio is insufﬁciently large. A simpler way of understanding why a narrow strike range leads to a lower fair variance is to compare the payoff of the narrowstrike replicating portfolio at expiration to the terminal payoff that the portfolio is attempting to replicate. Consequently. the option portfolio payoff remains linear. Expiration Wide strike range (50% . In order to keep capturing variance. whatever value it takes. you need to maintain the curvature of the log contract at the current stock price.0 ) 2 In the section entitled Replicating Variance Swaps: First Steps on page 6. still appropriately weighted. Over a longer time period it is more likely that the stock price will evolve outside the strike range. you capture less than the full realized variance. The lack of curvature (or gamma. and the reduction in value is greater for the oneyear case. capturing variance requires owning the full log contract.0 ) 2 ( 24.
both these effects contribute to the replication error. The Effect of Jumps on a Perfectly Replicated Log Contract When the stock price jumps. Because of the additive properties of the logarithm function. the log contract may no longer capture realized volatility. from now on we will assume that we are short the variance swap. In this section.– log Si – 1 S0 N ∆S i ST (EQ 35) where ∆S i = S i – S i – 1 is the change in stock price between successive observations. a large jump may take the stock price into a region in which variance does not accrue at the right rate. Comparison of the terminal payoff of the narrowstrike replicating portfolio (dashed line) and the nonlinear part of the logpayoff (solid line). First. for two reasons. if the log contract has been approximately replicated by only a ﬁnite range of strikes. Second. we instead adjust the exposure to (2/T) dollars worth of stock only when a new stock price is recorded for updating the realized variance. even with perfect replication. the terminal log payoff is equivalent to a daily accumulation of log payoffs: 29 . Rather than continuously rebalance as the stock price moves. we focus only on the second effect and examine the effects of jumps assuming that the logpayoff can be replicated perfectly with options. Terminal payoff 50 100 ST 150 200 the narrow strike range is responsible for the inability to capture variance. a discontinuous stockprice jump causes the variancecapture strategy of Equation 20 to capture an amount not equal to the true realized variance.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES FIGURE 9. which we will hedge by following a discrete version of the variancecapture strategy 2 V = T i=1 ∑ . In reality. For the sake of discussion.
– log .[ – J – log ( 1 – J ) ] T (EQ 39) jump In the limit that the jump size J is small enough to be regarded as part of a continuous stock evolution process. + . A jump up corresponds to a value J < 0.+ … 2 3 (EQ 41) 30 . the total (unannualized) realized variance for a zeromean contract is the sum 1 V = T ∑ 2 2 ∆S i ∆S i 1 1 ∆S 2 .  T T S jump S i – 1 S i – 1 no jumps ∑ (EQ 37) The contribution of the jump to the realized total variance is given by: 1 ∆S 2 J2 . a jump downwards of 10% corresponds to J = 0.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES 2 V = T ∑ i=1 N ∆S i Si .1. the percentage jump downwards. because variance is additive. all changes are diffusive.+ . the replication error. or the P&L (proﬁt/loss) due to the jump for a short position in a variance swap hedged by a long position in a variancecapture strategy is 2 J2 P&L due to jump = . from S → S ( 1 – J ) . T Si – 1 S i – 1 2 = . The contribution of this one jump to the variance is easy to isolate.  = T S jump T (EQ 38) On the other hand.– log Si – 1 Si – 1 (EQ 36) Suppose that all but one of the daily price changes are wellbehaved – that is. = . It is only because J is not small that the variance capture strategy is inaccurate.[ – J – log ( 1 – J ) ] – T T (EQ 40) To understand this result better. . Therefore. except for a single jump event. We characterize the jump by the parameter J . the right hand side of Equation 39 does reduce to the contribution of this (now small) move to the true realized variance. the impact of the jump on the quantity captured by our variance replication strategy in Equation 36 is Si 2 ∆S i . it is helpful to expand the log function as a series in J: J2 J3 – log ( 1 – J ) = J + .
Furthermore. If you are also short the option’s deltahedge.. leaving only cubic and higherorder dependencies on the jump size. as given by Equation 40 as a function of (downward) jump size for T=1 year.004 0. cubic and higherorder terms in the stock price. a large move one day.P&L due to jump = . and has no impact on the hedging mismatch. The leading correction is cubic in the jump size J and has a different sign for upwards or downwards jumps. the variance replication strategy is long quadratic. this is precisely why hedged long options positions capture variance. large moves in either direction beneﬁt the position. while a large move upwards (J < 0) leads to a loss. Note that the simple cubic approximation of Equation 42 correctly predicts the sign of the P&L for all values of the jump size. then the linear dependence of the net position cancels.004 20 10 0 10 20 Jump size (downward) There is an analogy between the cancellation of the quadratic jump term in variance replication and the linear jump term in options replication. FIGURE 10.+ … 3T (EQ 42) The quadratic contribution of the jump is the same for the variance swap as it is for the variancecapture strategy. in the case of variance replication considered here. A large move downwards (J > 0) leads to a proﬁt for the (short variance swap)(long variancecapture strategy). 0.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES The leading contribution to the replication error is then 2J3 . Figure 10 shows the impact of the jump on the strategy for a range of jump values. In contrast. Since the leading 31 . When you are long an option you are long linear. leaving only the quadratic and higherorder dependencies.002 0 0.002 0. Because the leadingorder term is quadratic. followed by a large move in the opposite direction the next day would tend to offset each other. he impact of a single jump on the proﬁt or loss of a short position in a variance swap and a long position in the variance replication strategy.006 Impact of jump 0. while the position in the variance swap is short only the quadratic dependence. quadratic and higherorder dependence on the stock price. Now the quadratic term in the net position cancels.
8 3.5 −3. Suppose that a downward jump occurs. large enough to move the stock price outside the range of option strikes. but it is smaller in size. even if no further jumps occur. with the stock price now outside the strike range. that is hedged with the variance replication strategy of Equation 36 for T=1 year. The net results will depend on the details of the scenario. both the effects of jumps and the risks of log replication with only a limited strike range cause the strategy to capture a quantity different from the true realized variance of the stock price.2 −24.5 28.4 7. However. a large move upwards will be doubly damaging: there will be convexity loss due to the jump and the hedge will not capture variance if the jump takes the index outside the strike range. the gain from the jump has to be balanced against the subsequent failure of the hedge to capture the smooth variance. the constantdollar exposure would cancel the linear part of the stock price change. The combined effect of both these risks is harder to characterize because they interact with one another in a complicated manner.8 −6.9 −0. the vega and gamma of the replicating portfolio are now too low to accrue sufﬁcient variance. and lead to a convexity gain.8 −80. Table 5 displays the proﬁt or loss due to jumps of varying sizes for threemonth and oneyear variance swaps.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES term is cubic. after this jump. TABLE 5.2 0. the direction of the jump determines whether there is a net proﬁt or loss. Although the logpayoff is not being replicated perfectly. Jump size and direction Threemonth Oneyear J = 15% (down) J = 10% (down) J = 5% (down) J = −5% (up) J = −10% (up) J = −20% (up) The Effect of Jumps When Replicating With a Finite Strike Range 101. If the logpayoff were replicated perfectly. In contrast. there is still a convexity gain from the jump.2 In practice.9 25. Consider again a short position in a variance contract that is being hedged by the variancecapture strategy.2 −20. 32 . The proﬁt/loss due to a single jump for a short variance swap with a notional value of $1 per squared variance point. In this scenario.
to value and hedge. the replication strategy for a volatility swap is fundamentally different. which still captures the total variance over the life of the log contract.( σ R – K vol ) 2K vol 2 (EQ 43) This means that 1 ⁄ ( 2K vol ) variance contracts with strike K vol can approximate a volatility swap with a notional $1/(vol point). for realized volatilities near K vol . which has payoff σ R – K vol . In order to approximate a volatility swap struck at K vol . it is variance that emerges naturally from hedged options trading. both theoretically and practically. as we will show. other than assuming that the stock price evolves continuously (without jumps). from a contingent claims or derivatives point of view. the variance and volatility payoffs agree in value and volatility sensitivity (the ﬁrst derivative with respect to σ R ) when σ R = K vol . In essence. and so we now consider volatility swaps. let’s consider a naive strategy: approximate a volatility swap by statically holding a suitably chosen variance contract. The replication strategy for the variance swap makes no assumptions about the level of future volatility. In contrast. variance is the primary underlyer and all other volatility payoffs. such as volatility swaps. it is affected by changes in volatility and its value depends on the volatility of future realized volatility. We see that the actual volatility swap and the approximating variance swap differ appreciably 33 . are best regarded as derivative securities on the variance as underlyer. From this perspective. But most market participants prefer to quote levels of volatility rather than variance. Changes in volatility have no effect on the strategy. To illustrate the issues involved. There is no simple replication strategy for synthesizing a volatility swap. With this choice. Naively. this would also imply that the fair price of future volatility (the strike for which the volatility swap has zero value) is simply the square root of fair variance K var : K vol = K var (naive estimate) (EQ 44) In Figure 11 we compare the two sides of Equation 43 for Kvol = 30% for different values of the realized volatility.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES FROM VARIANCE TO VOLATILITY CONTRACTS For most of this note we have focused on valuing and replicating variance swaps. we can use the approximation 2 2 1 σ R – K vol ≈ . volatility itself is a nonlinear function (the square root) of variance and is therefore more difﬁcult.
Payoff of a volatility swap (straight line) and variance swap (curved line) as a function of realized volatility. The naive estimate of Equations 43 and 44 is not quite correct. you cannot ﬁt a line everywhere with a parabola. the variance swap always outperforms the volatility swap. the variance swap payoff is always greater than the volatility swap payoff. Dynamic Replication of a Volatility Swap In principle. for K vol = 30% . In Appendix D we estimate the expected hedging mismatch and static hedging parameters under the assumption that future realized volatility is normally distributed. the straight line in Figure 11 will shift to the left and will not always lie below the parabola. it is necessary to make an assumption about both the level and volatility of future realized volatility. In this way. The mismatch between the variance and volatility swap payoffs in Equation 43. so that with this choice of the fair delivery price for volatility. In order to estimate the size of the convexity bias. some of the risks inherent in the static approximation of a volatility swap by a variance swap could be reduced by dynamically trading new variance contracts throughout the life of the volatility swap.( σ R – K vol ) 2K vol This square is always positive. To avoid this arbitrage. With this choice. This dynamic replication of a volatility swap by means of vari 34 . 10 5 payoff 0 5 20 30 40 10 15 σR only when the future realized volatility moves away from K vol . is the 2 1 convexity bias = . and therefore the fair strike for the volatility swap. we should correct our naive estimate to make the fair strike for the volatility contract lower than the square root of the fair strike for a variance contract.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES FIGURE 11. so that K vol < K var .
there is no market in variance swaps liquid enough to provide a usable underlyer. The practical implementation of these ideas requires an arbitragefree model for the stochastic evolution of the volatility surface. of course. When there is a liquid market in variance swaps. it is only very recently that such models have been developed [see for example Derman and Kani (1998) and Ledoit (1998)]. In practice. This is closely analogous to replicating a curved stock option payoff by means of deltahedging using the linear underlying stock price. 35 . one could imagine that the strategy would call for holding at every instant a “variancedelta” equivalent of variance contracts to hedge a volatility derivative. Taking the analogy further. In the same way that the appropriate option hedge ratio depends on the assumed future volatility of the stock.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES ance swaps would (in principle) produce the payoff of a volatility swap independent of the moves in future volatility. these models may be useful in hedging volatility swaps and other variance derivatives. the dynamic replication of a volatility swap requires a model for the volatility of volatility. Due to the complexity of the mathematics involved.
we have succeeded in deriving analytic approximations that work well for the swap value under commonly used skew parameterizations. Second. this requires a consistent model for stochastic volatility.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES CONCLUSIONS AND FUTURE INNOVATIONS We have tried to present a comprehensive and didactic account of both the principles and methods used to value and hedge variance swaps. Remarkably. Here. In markets with a volatility skew (the real world for most swaps of interest). The capped variance swap has embedded in it an option on realized variance. using the rigorous approach. To fully implement a replication strategy for volatility swaps. we need a consistent stochastic volatility model for options. First. Once again. our ability to effectively price and hedge volatility swaps is still limited. There are at least two areas where further development is welcome. one can still value variance swaps by replication. 36 . some market participants prefer to enter a capped variance swap or volatility swap that limits the possible loss on the position. These formulas enable traders to update price quotes quickly as the market skew changes. We have explained both the intuitive and the rigorous approach to replication. forwards or futures would lead to the possibility of trading and hedging volatility options. Much work remains to be done in this area. the intuitive approach loses its footing. The development of a truly liquid market in volatility swaps.
the sensitivity to the total variance of an individual option O in this portfolio. is given by VO = τ where 1 exp ( – d 1 ⁄ 2 ) f ( S.v) represents a standard BlackScholes option of strike K and total variance v = σ τ when the stock price is S. K . v S ∂v ∫ (A 2) 0 The sensitivity of vega to S is ∂V Π ∂S ∞ = τ ∫ ∂ S[ S 0 ∞ ∂ 2 ρ ( xS ) ] f ( x. v ) = . v ) d x 0 ∫ 37 .. v The variance sensitivity of the whole portfolio is therefore K V Π ( S ) = τ ρ ( K )Sf .K..Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES APPENDIX A: REPLICATING LOGARITHMIC PAYOFFS Constant Vega Requires Options Weighted Inversely Proportional to the Square of the Strike In this Appendix we derive several results concerning the replication of a logarithmic payoff with portfolios of standard options.2 v 2π and ln ( S ⁄ K ) + v ⁄ 2 d 1 = . v ) dK 0 (A 1) where O(S. Consider a portfolio of standard options ∞ Π(S) = ∫ ρ ( K )O ( S.. v ) d x = τ S [ 2ρ ( xS ) + xSρ' ( xS ) ] f ( x. v d K S ∞ 2 2 ∂ K (O) = τSf . Vega. K .
– log T S* S* (A 4) where S* is some reference price.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES where.. that is that 2ρ + K The solution to this equation is const ρ = 2 K Log Payoff Replication with a Discrete Set of Options (A 3) ∂V Π ∂S = 0 . which implies ∂ρ = 0 ∂K It was shown in the main text that the realized variance is related to trading a log contract. The number of options is determined by the slope of this segment: f ( K 1c ) – f ( K 0 ) w c ( K 0 ) = K 1c – K 0 (A 5) 38 . We want vega to be independent of S. only a discrete set of option strikes is available for replicating f ( S T ) .. and we need to determine the number of options for each strike.. and put options with strikes K 0 = S * > K 1 p > K 2 p > K 3 p > . Since there is no logcontract traded. In practice. We can approximate f ( S T ) with a piecewise linear function as in Figure 11. we changed the integration variable to x = K ⁄ S .. The ﬁrst segment to the right of S* is equivalent to the payoff of a call option with strike K0. in the second line of the above equation. Assume that you can trade call options with strikes K 0 = S * < K 1c < K 2c < K 3c < . we want to represent it in terms of standard options. It is useful to subtract the linear part (corresponding to the forward contract) and look at the function ST 2 ST – S* f ( S T ) = ..
– K n + 1.– w c ( K 0 ) K 2c – K 1c (A 6) Continuing in this way we can build the entire payoff curve one step at the time. c is given by f ( K n + 1 . c – K n. c ) = . p ) (A 8) 39 . Logpayoff and options portfolio at maturity. c ) w c ( K n. c n–1 i=0 ∑ w c ( K i. c ) – f ( K n. Given that we already hold w c ( K 0 ) options with strike K 0 we need to hold w c ( K 1 ) calls with strike K 1 where f ( K 2c ) – f ( K 1c ) w c ( K 1 ) = . the number of call options of strike K n. p ) w p ( K n. K2 p K 1p K0 K 1c K2 c Similarly. p – K n + 1. p ) – f ( K n. p n–1 i=0 ∑ w c ( K i. p ) = . the second segment looks like a combination of calls with strikes K 0 and K 1c . In general.– K n.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES FIGURE 12. c ) (A 7) The other side of the curve can be built using put options: f ( K n + 1 .
Σ0 ) + b 1 2∂P ∂P + .b . Σ ( b ) ) dK K2 1 * We now expand option prices as a power series in b around a ﬂat implied volatility ( b = 0 ).+ T S0 S* e rT S * (B 2) ∫0 1 . Σ 0 is atthemoney forward implied volatility and b is the slope of the skew.. Σ0 ) + b + . 0 K2 2 2 2 2 T S* K ∂ b ∂b b = 0 b=0 (B 4) ∫ ∫ 40 .. Σ ( b ) ) = C ( K . We start with the general expression for the fair variance discussed in the main text: S* S 0 rT 2 K var = .P ( K .e – 1 – log .b + . Σ ( b ) ) d K + e K2 rT ∞ ∫S .e dK + d K + .Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES APPENDIX B: SKEW LINEAR IN STRIKE Here... we derive a formula which gives the approximate value of the variance swap when the skew is linear in strike.e T 0 K 2∂b 2∂b S* K b=0 b=0 ∫ ∫ 2 2 ∞ 1 ∂C 1 2 2 rT S * 1 ∂ P . rT – .C ( K . 1 2∂C ∂C C ( K ... Σ ( b ) ) = P ( K . ∂ b b = 0 2 ∂ b2 b=0 2 2 (B 3) Using this expansion we can formally write an expansion of fair variance in powers of b as follows: ∞ 1 ∂C 2 2 rT S * 1 ∂P dK + dK + K var = Σ 0 + b .b + . We parameterize the implied volatility by K – SF Σ ( K ) = Σ 0 – b S F rT (B 1) where S F = S 0 e is the forward value corresponding to the current spot. ∂ b b = 0 2 ∂ b2 b=0 P ( K .
Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES Here Σ 0 is the fair variance in the “ﬂat world” where volatility is constant and is given by Equation B2 with Σ ( b ) replaced by Σ 0 . 1 ∂ Σ0 2π 0 K 2 2 T SF 2 (B 7) ∫ 41 .b ... . ∂b b = 0 ∂Σ Σ ∂b b = 0 0 ∂ P ∂b b = 0 ∂ C ∂b b = 0 2 2 2 2 2 = ∂ P ∂Σ Σ 0 ∂ C ∂Σ Σ 0 2 2 2 2 ∂Σ ∂ b ∂Σ ∂ b 2 b=0 2 b=0 = The derivatives with respect to volatility are easily calculated using the BlackScholes formula S T –d1 ⁄ 2 ∂P ∂C = = e ∂Σ Σ ∂Σ Σ 2π 0 0 2 ∂ P ∂Σ Σ 0 2 2 = ∂ C ∂Σ Σ 0 2 2 S T ∂d 1 – d 1 ⁄ 2 = – d 1 e 2π ∂ Σ 0 2 (B 5) SF 1 2 log  + .– 1 S ∂b b = 0 F (B 6) The fact that call and put options have the same vega in the BlackScholes framework makes it possible to combine the integrals in Equation B4 into one integral from 0 to ∞ : 2 –d1 ⁄ 2 2 2 rT S T ∞ 1 K dK – K var = Σ 0 – b . ∂b b = 0 ∂Σ Σ ∂b b = 0 0 ∂C ∂Σ ∂C = .e –1 e d K + ...Σ 0 T K 2 d 1 = Σ0 T where. for the model we are considering here K ∂Σ = – .– 1 e T 2π 0 K 2 S F ∫ ∂d 1 – d 1 ⁄ 2 1 2 2 rT S T ∞ 1 K 2d .e . The derivatives which enter Equation B4 are given by ∂P ∂Σ ∂P = .
. ) 2 (B 8) 2 We now present an alternative derivation of this result. 1 T 2 K var = E . Equation B10 can be written as SF Σ 0 – bx – 2br ( 1 + x )t σ 2 ( S.. t ) dt T 0 ∫ (B 9) This can be evaluated approximately as follows.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES To evaluate these integrals. and K 2 then write Equation B7 as 2 ∞ 2 dz 1 – e v 0 z – v 0 ⁄ 2 e – z ⁄ 2 . +  K var = Σ 0 – b 2Σ 0 –∞ 2π ∫ 2 – v0 z + v0 ⁄ 2 2 ∞ v0 z – v0 ⁄ 2 2 – z ⁄ 2 dz .. t ) = 2 2 2 1 1 ( 1 + x ) t – b td 1 + .σ ( S.+ d 1 K ∂ K Σ K T ∂ K ∂ K 2 (B 10) K = S T = t S Denote x = . for example. .+ . we use the relation between implied and local volatility: ∂Σ ∂Σ Σ + 2rK . one can. b e +e – 2 ( z – v 0 z )e 2π –∞ ∫ The term linear in b vanishes and the term quadratic in b has coefﬁcient 3Σ 0 T .+ 2 ∂T ∂K T σ 2 ( S.v 0 ⁄ v 0 ≡ d 1 .– 1 . change the integraSF 1 2 tion variable to z = log . t ) = 2 2∂ Σ ∂Σ 2 ∂Σ 2 1 1 – d 1 T + .– bd 1 Σ 0 – bx ( 1 + x ) t (B 11) 42 . where v 0 = Σ 0 T . First.. so that K var = Σ 0 ( 1 + 3Tb 2 + . i. We start with the fundamental deﬁnition of the fair delivery variance as the expected value of future realized variance.e.
( Σ 0 – bx ) t 2 (B 12) We expand σ 2 ( S. Expected values of higher powers of x are easily calculated using 2 2 ( n – n )Σ 0 t ⁄ 2 S n E  = e S F 2 3 Σ0 t 2 –1 – 3e Σ0 t 2 2 3Σ 0 t +2 After averaging over the stock price distribution.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES where – log ( 1 + x ) 1 d 1 = . we average over time and... t ) in powers of x and calculate the expected value in a lognormal world with volatility Σ 0 using E[ x] = 0 E[ x ] = e E[ x ] = e .( Σ 0 – bx ) t . ﬁnally. expand the result in powers of the skew slope b .+ . ) It is reassuring that these two very different methods lead to the same approximation formula. 2 43 ... Tedious calculation leads to the relation K var = Σ 0 ( 1 + 3Tb 2 + ..
as 1 Σ ( ∆ p ) = Σ 0 + b ∆ p + .) 2 To derive the formula for the fair variance we follow the same procedure as in Appendix B. One important difference is that now implied volatility is nonlinear in b (since ∆ p depends implicitly on b ) so that second derivatives have an additional term: ∂ P ∂b b = 0 ∂ C ∂b b = 0 2 2 2 2 = ∂ P ∂Σ Σ 0 ∂ C ∂Σ Σ 0 2 2 2 2 ∂Σ ∂ b ∂Σ ∂ b 2 ∂P ∂ Σ + ∂ Σ Σ ∂ b2 b=0 0 b=0 + b=0 2 2 = ∂C ∂ Σ ∂ Σ Σ ∂ b2 0 b=0 2 (C 2) Other derivatives we need are easily calculated: ∂Σ 1 = ∆ p + ∂b b = 0 2 ∂ Σ ∂b b = 0 where ∆ p = – N ( –d1 ) and 2 2 1 ∂∆ p = 2 ∆ p + . 2 ∂ Σ 0 (C 3) 44 .Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES APPENDIX C: SKEW LINEAR IN DELTA Here we consider the case where implied volatility varies linearly with delta. Such a skew can be parameterized in terms of ∆ p . 2 (C 1) where Σ 0 is the implied volatility of options with ∆ p = – 1 ⁄ 2 (the “50delta volatility”). the delta of a Europeanstyle put. starting with Equation B2. . (We could also parametrize the skew in terms of the 1 call delta as Σ ( ∆ c ) = Σ 0 + b ∆ c – .
e .Σ 0 T Σ0 T 2 Combining these relations..+ . integrals can be evaluated by changing the integration variable SF 1 2 to z = log .∆ p + . ∆ p + . one can ﬁrst expand e in powers of z and integrate term by term.+ .b . d 1 e dK  2π K 2 2 T ∂ Σ0 2 0 ∫ (C 4) 2 1 1 ∂∆ p – d 1 ⁄ 2 – 2 . ∆ p + .e e 2 2π –∞ K var = 2 Σ0 ∫ b 2 ∞ –∞ ∞ ∫ 2 v 0 z – v 0 ⁄ 2 – z ⁄ 2 dz 1 2 2 N ( z ) – . e dK 2 ∂ Σ 0 K 2 0 ∞ ∫ Again. + – b 2Σ 0 N ( z ) – .( z – v 0 z )e e 2 2π –2 v 0 z – v 0 ⁄ 2 – z 2 dz 1 N ( z ) – . where v 0 = Σ 0 T .Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES SF log K 1 d 1 = . In addition the following results are useful: v0 z – v0 ⁄ 2 2 45 .. e dK – T 2π K 2 2 ∞ ∫ 2 0 ∞ 2 1 2 2 rT S T 1 1 2 ∂d 1 – d 1 ⁄ 2 . Since we are eventually interested in expanding the result in powers of v 0 = Σ 0 T .( z – v 0 )e e 2π 2 –∞ ∫ All these integrals can be evaluated exactly.is antisymmetric in z to simplify calcula2 tions. It is also 1 useful to note that N ( z ) – .v 0 ⁄ v 0 ≡ d 1 . the fair variance can be written as 2 Σ0 K var = 2 rT S T 1 1 –d1 ⁄ 2 – b .e . so that K 2 2 ∞ –z ⁄ 2 v0 z – v0 ⁄ 2 dz 1 .
rT – ..+ . i.+ . We start with the same fundamental expression: S* S 0 rT 2 K var = ...= . . Σ p ( b p ) ) d K + e K2 rT ∞ ∫S .C ( K .– .= n 2 2π 2 2π ∂ a a a + b 2 (C 5) n n ∫z 0 ∞ 2n – az ⁄ 2 2 e n 2 n ( –2 ) ∂ 1 a + 2b 1 2 dz π . 2 1 – .P ( K . N ( bz ) – .= 3⁄2 n 2 2 2π ( 2π ) ∂ a a a + b2 a+b After evaluating all integrals we ﬁnd the ﬁnal answer to be 2 2 T 1 2 b K var = Σ 0 + bΣ 0 .= 2π ∂ a n a 2 2π a n n ∫ 0 ∞ z 2n + 1 – az ⁄ 2 2 e ( –2 ) b ∂ 1 1 dz N ( bz ) – .. arctan .e. π 12 (C 6) Two.arctan N ( bz ) – .+ T S0 S* e rT S * (C 8) ∫0 1 ..e – 1 – log ..≤ ∆ p ≤ 0 2 1 0 ≤ ∆ c ≤ 2 for for (C 7) We now brieﬂy sketch the derivation emphasizing only the differences with the above detailed calculations. 1 Σ p ( ∆ p ) = Σ 0 + b p ∆ p + . 2 1 Σ c ( ∆ c ) = Σ 0 + b c ∆ c – .Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES ∞ ∫z 0 ∞ 2n – az ⁄ 2 2 e ( –2 ) ∂ 1 b 1 dz .N ( bz ) – 2π ∂ a n a 2 4 2π a ∫ 0 z 2n + 1 – az ⁄ 2 2 e n n b 1 ( –2 ) b ∂ 1 2 dz . arctan . Σc ( bc ) ) dK K2 1 * 46 .Slope Model Our calculations can easily be generalized to the model where the slope of the skew is different for put and call options.
+ . Note that by changing the sign of b c we turn the implied skew into a smile. We expand put option prices in powers of b p and call option prices in powers of b c . Σ 0 . Evaluating all integrals as above we ﬁnd 2 Σ0 K var = b p + bc 1 b p + bc 1 T + .Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES We use different implied volatility parameterizations for put and call options.Σ 0 ( b p – b c ) + Σ 0 ... Note that we should choose S* so that S* = S F e –Σ0 T ⁄ 2 2 This ensures that we use the put (call) parameterization in Equation C7 for strikes below (above) S*.. + . for b p = b c this reduces to the result for single slope given in Equation C6.. 12 2 4 2 π 2 2 (C 9) Obviously. 47 . as given by Equation C7.
while the target of the replication is the realized volatility ( Σ T ). 2 The hedging instrument is the realized variance ( Σ T ).Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES APPENDIX D: STATIC AND DYNAMIC REPLICATION OF A VOLATILITY SWAP We have argued that volatility swaps are fundamentally different from variance swaps and that. we showed that attempting to create a volatility swap from a variance swap by means of a “buyandhold” strategy invariably leads to misreplication. since this amounts to trying to ﬁt a linear payoff (the volatility payoff) with a quadratic payoff (the variance swap). In the section From Variance to Volatility Contracts on page 33. on average. with mean Σ and standard deviation σ Σ : Σ T ∼ N ( Σ. unlike the variance swap. We want to approximate the volatility as a function of the variance by writing Σ T ≈ aΣ T + b 2 (D 1) and choose a and b to minimize the expected squared deviation of the two sides of Equation D1: 2 min E [ ( Σ T – aΣ T – b ) 2 ] (D 2) Differentiation leads to the following equations for the coefﬁcients a and b : E [ Σ T ] = aE [ Σ T ] + b 3 4 2 E [ Σ T ] = aE [ Σ T ] + bE [ Σ T ] 2 (D 3) The distribution of future volatility could be assumed to be normal. there is no simple replicating strategy to synthetically create a volatility swap. as closely as possible. we will show that it is possible to pick the strike and notional size of a variance contract to match the payoff of a volatility contract. Given a view on both the direction and volatility of future volatility. the expected squared replication error is given by: 48 . The extent of the replication mismatch will depend on how close the realized volatility is to its expected value. This strategy will replicate only on average. σ Σ ) (D 4) This model only makes sense if the probability of negative volatilities is negligible.
Σ T ) ) 2 ] 2 2 (D 5) For realized volatilities distributed normally as in Equation D4. the hedging coefﬁcients are 1 a = 2 σΣ 2Σ + Σ Σ b = 2 σΣ 2 + Σ2 and the expected squared replication error is: 2 σΣ 2 2 min E [ ( Σ T – aΣ T – b ) ] = 2Σ 2 1 + 2 σΣ (D 6) (D 7) 49 .Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES 2 min E [ ( Σ T – aΣ T – b ) ] = Var ( Σ T ) [ 1 – ( corr ( Σ T .
50 . I. E. and J. M. 1820. (1996). Goldman. Nelken.. Winter. RISK 7. M. 1. Kamal. edited by R. 7480. SantaClara (1998). Journal of Portfolio Management. P. The Log Contract: A new instrument to hedge volatility. Global Derivatives (July) Quarterly Review. Derman. Kamal. (1994). Kani and J. Jarrow. O. Derman. The Local Volatility Surface. and J.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES REFERENCES Carr. Kani (1998). Zou (1996). Sachs & Co. University of California. Investing in Volatility. E. No. Vol. July/August. A. Relative Price of Options with Stochastic Volatility. Working Paper. A. Pricing with a Smile.. Dupire. Pirasteh. Madan (1998). and I. Los Angeles. edited by I. The Anderson Graduate Schools of Management. Derman. Kani. J. 1. Ledoit. Derman. and D. Equity Derivatives Research. No. in The Handbook of Exotic Options. E. 2. October. Goldman. Valuing Contracts with Payoffs Based on Realized Volatility. Neuberger.. E. 2536. Derman. C. No. I. 1. Stochastic Implied Trees: Arbitrage Pricing with Stochastic Term and Strike Structure of Volatility. B. 417427. I. Zou (1996). Riding on a Smile. Towards a Theory of Volatility Trading. in Volatility: New Estimation Techniques for Pricing Derivatives. International Journal of Theoretical and Applied Finance. Kani. and I. Kani (1994). McClure. E. Neuberger. The Log Contract and Other Power Contracts. (1994). and P. 3239. 200212. RISK 7. Financial Analyst Journal. Quantitative Strategies Research Notes. Zou (1996). Sachs & Co. 61110.
1996 Oct. Trading and Hedging Local Volatility Iraj Kani. Michael Kamal. Iraj Kani. Cyrus Pirasteh and Joseph Zou Jan. 1996 Apr.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES SELECTED QUANTITATIVE STRATEGIES PUBLICATIONS June 1990 Understanding Guaranteed ExchangeRate Contracts In Foreign Stock Investments Emanuel Derman. John McClure. Iraj Kani and Joseph Z. Piotr Karasinski and Jeffrey Wecker Valuing and Hedging Outperformance Options Emanuel Derman PayOnExercise Options Emanuel Derman and Iraj Kani The Ins and Outs of Barrier Options Emanuel Derman and Iraj Kani The Volatility Smile and Its Implied Tree Emanuel Derman and Iraj Kani Static Options Replication Emanuel Derman. 1992 Mar. 1996 51 . 1992 June 1993 Jan. 1996 Aug. 1995 Feb. Deniz Ergener and Iraj Kani Enhanced Numerical Methods for Options with Barriers Emanuel Derman. Iraj Kani and Neil Chriss Model Risk Emanuel Derman. Zou Implied Trinomial Trees of the Volatility Smile Emanuel Derman. Emanuel Derman and Michael Kamal Investing in Volatility Emanuel Derman. 1994 May 1994 May 1995 Dec. Deniz Ergener and Indrajit Bardhan The Local Volatility Surface: Unlocking the Information in Index Option Prices Emanuel Derman. Iraj Kani.
1999 52 . Michael Kamal.Goldman Sachs QUANTITATIVE STRATEGIES RESEARCH NOTES Apr. 1997 Sept. Iraj Kani and Joseph Zou Stochastic Implied Trees: Arbitrage Pricing with Stochastic Term and Strike Structure of Volatility Emanuel Derman and Iraj Kani The Patterns of Change in Implied Index Volatilities Michael Kamal and Emanuel Derman Predicting the Response of Implied Volatility to Large Index Moves: An October 1997 S&P Case Study Emanuel Derman and Joe Zou How to Value and Hedge Options on Foreign Indexes Kresimir Demeterﬁ Regimes of Volatility: Some Observations on the Variation of S&P 500 Implied Volatilities Emanuel Derman Apr. 1998 Jan. 1997 Sept. 1997 Nov. 1997 Is the Volatility Skew Fair? Emanuel Derman.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.