Quantitative Strategies

Research Notes
Goldman
Sac hs
Trading and Hedging
Local Volatility
Iraj Kani
Emanuel Derman
Michael Kamal
August 1996
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Copyr ight 1996 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 afliat 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 afliat 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 dened 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 afliat 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 uct 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 ofce 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.
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
SUMMARY
This not e out lines a met hodology for hedging and t r ading
index volat ilit ies.
In t he bond wor ld, for war d r at es ar e t he ar bit r age-fr ee
int er est r at es at fut ur e t imes t hat can be locked in by
t r ading bonds t oday. Similar ly, in t he wor ld of index
opt ions, local volat ilit ies ar e t he ar bit r age-fr ee volat ilit ies
at fut ur e t imes and mar ket levels t hat can be locked in by
t r ading opt ions t oday. The dependence of local volat ilit y
on fut ur e t ime and index level is called t he local volat ilit y
sur face, and is t he analog of t he for war d yield cur ve. In
t his paper we show how t o hedge por t folios of index
opt ions against changes in implied volat ilit y by hedging
t hem against changes in fut ur e local volat ilit y. This is
analogous t o hedging bond por t folios against changes in
for war d r at es.
Eur odollar fut ur es on int er est r at es ar e t he best -suit ed
inst r ument for for war d-r at e hedging. Unfor t unat ely,
t her e ar e no liquid fut ur es on local index volat ilit y. So, we
will dene a volatility gadget, t he volat ilit y analog of a
Eur odollar fut ur es cont r act . A gadget is a small por t folio
of st andar d index opt ions t hat is sensit ive t o local index
volat ilit y only at a denit e fut ur e t ime and index level,
and, like a fut ur es cont r act , has an init ial pr ice of zer o.
We can cr eat e unique volat ilit y gadget s for each fut ur e
t ime and index level. By buying or selling suit able quan-
t it ies of gadget s, cor r esponding t o differ ent fut ur e t imes
and mar ket levels, we can hedge an index opt ion por t folio
against any changes in fut ur e local volat ilit y. This pr oce-
dur e is t heor et ically cost less. It can help r emove
unwant ed volat ilit y r isk, or help acquir e desir ed volat il-
it y exposur e, over any r ange of index levels and t imes
wher e we t hink fut ur e local volat ilit y changes ar e likely
t o occur.
________________________
Ir aj Kani (212) 902-3561
Emanuel Der man (212) 902-0129
Michael Kamal (212) 357-3722
Edit or ial: We ar e gr at eful t o Bar bar a Dunn for her car e-
ful r eview of t he manuscr ipt .
0
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Table of Contents
IMPLIED AND LOCAL VOLATILITIES...............................................................................1
NOTATION ......................................................................................................................2
THE ANALOGY BETWEEN LOCAL VOLATILITIES AND FORWARD RATES ..................... 2
INTRODUCING GADGETS: GADGETS FOR INTEREST RATES............................................4
HEDGING AGAINST FORWARD RATE CHANGES............................................................ 6
Hedging a Portfolio of Cashflows Against a Set of Forward Rates ..................8
Examples of Interest Rate Hedging Using Gadgets........................................ 10
HEDGING LOCAL VOLATILITIES: VOLATILITY GADGETS.............................................13
Relation to Forward Probability Distribution .................................................16
Constructing Finite Volatility Gadgets ...........................................................18
Using Volatility Gadgets to Hedge Against Local Volatility Changes .......... 19
AN EXPLICIT EXAMPLE............................................................................................... 22
An Example Using Finite Volatility Gadgets ..................................................27
CONCLUSIONS...............................................................................................................28
APPENDIX A: LOCAL VOLATILITY AND THE FORWARD EQUATION FOR STANDARD
OPTIONS ......................................................................................................................29
APPENDIX B: FORWARD PROBABILITY MEASURE ...................................................33
APPENDIX C: MATHEMATICS OF GADGETS ..............................................................35
1
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
If we t hink of t he implied volat ilit y of an index opt ion as t he mar ket s
est imat e of t he aver age fut ur e index volat ilit y dur ing t he life of t hat
opt ion, we can t hink of local volat ilit y as t he mar ket s est imat e of
index volat ilit y at a par t icular fut ur e t ime and mar ket level. The set
of implied volat ilit ies
K,T
for a r ange of st r ikes K and expir at ions T
const it ut es an implied volat ilit y sur face. We can ext r act fr om t his
sur face t he mar ket est imat e of t he local index volat ilit y
S,t
at a par -
t icular fut ur e t ime t and mar ket level S. The set
S,t
for a r ange of
index levels S and fut ur e t imes t const it ut es t he local volat ilit y sur -
face
1
.
Figur e 1 shows bot h t he implied and local volat ilit y sur faces for t he
S&P 500 index on May 17, 1996. The local volat ilit ies gener ally var y
mor e r apidly wit h mar ket level t han implied volat ilit ies var y wit h
st r ike. This behavior is obser ved whenever global quant it ies ar e
descr ibed in t er m of local ones; in t he int er est r at e wor ld t he for war d
r at e cur ve oft en displays mor e var iat ion t han t he cur ve of spot yields
which r epr esent t he aver age of for war d r at es.
We can ext r act local volat ilit ies fr om t he spect r um of available index
opt ions pr ices by means of implied models
2
. In t hese models, all
t r aded index opt ions pr ices const r ain a one-fact or equilibr ium pr o-
cess for t he fut ur e evolut ion of t he index pr ice so as t o be consist ent
wit h mar ket pr ices while disallowing any fut ur e ar bit r age oppor t uni-
1. See Der man, Kani and Zou [1995].
2. See for example, Der man and Kani [1994], Dupir e [1994] and Rubinst ein [1994].
IM PLIED AND LO CAL
VO LATILITIES
FIG URE 1. Volatility surfaces for S&P500 index options on May 17,
1996, (a) implied volatility surface, (b) local volatility surface.
2
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
t ies. In mat hemat ical t er ms, t he evolut ion over an innit esimal t ime
dt in an implied model is descr ibed by t he st ochast ic differ ent ial
equat ion:
( EQ 1)
wher e S = S (t) is t he index level at t ime t, is t he index's expect ed
r et ur n and dZ = dZ(t) is t he st andar d Wiener measur e. The inst ant a-
neous fut ur e volat ilit y (S,t) is assumed t o depend only on t he fut ur e
index level S and t ime t. This assumpt ion allows t he implied models
t o r emain pr efer ence-fr ee. The r equir ement t hat t he ar bit r age-fr ee
opt ions values fr om t his model mat ch mar ket pr ices complet ely xes
t he for m of t he local volat ilit y funct ion (S,t).
Implied models can be viewed as effective volat ilit y models. This is
because t he model aver ages out sour ces of var iat ion in volat ilit y
ot her t han index level and t ime. If t her e ar e ot her sour ces of var ia-
t ion, t he local volat ilit y in implied models is effect ively an aver age
over t hese var iat ions. Ther efor e, t he local volat ilit y funct ion (S,t) is
an expect at ion over all st ochast ic sour ces of uncer t aint y of inst ant a-
neous fut ur e volat ilit y at fut ur e st ock pr ice S and t ime t (see Appen-
dix A for mor e r igor ous denit ion), which can be comput ed fr om t he
spect r um of t r aded opt ion pr ices. Implied models ar e, in t his sense,
opt ions-wor ld analogs of int er est r at e models wit h static yield cur ves,
in which for war d r at es ar e assumed t o depend only on t he fut ur e
t ime, and can be dir ect ly implied fr om t he spect r um of t r aded bond
pr ices.
Much of t he past decades hist or y of yield cur ve modeling has been
concer ned wit h allowing for ar bit r age-fr ee st ochast ic var iat ions
about cur r ent for war d r at es. Similar ly, we can in pr inciple allow for
ar bit r age-fr ee st ochast ic var iat ion about cur r ent local volat ilit ies.
Thr oughout t his paper we will have t he need t o r efer t o secur it ies
and t o t heir values at a given t ime. In or der t o avoid confusion, we
adopt t he following convent ion. If a secur it y (or por t folio of secur it ies)
is denot ed by t he symbol P, we use t he symbol P(t) t o denot e it s value
as t ime t.
Much of t he r est of t his paper r elies on hacking a pat h t hr ough t he
volat ilit y for est t hat par allels t he r out e followed in t he int er est r at e
wor ld by user s of for war d r at es. The for war d r at e fr om one fut ur e
t ime t o anot her can be found fr om t he pr ices of bonds mat ur ing at
t hose t imes; similar ly, t he local volat ilit y at a fut ur e index level and
t ime is r elat ed t o opt ions expir ing in t hat neighbor hood.
dS
S
------ dt S t , ( )dZ + =
NOTATIO N
THE ANALO GY BETWEEN
LO CAL VO LATILITIES AND
FO RWARD RATES
3
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
Figur e 2 illust r at es t he similar it y bet ween int er est r at es and volat ili-
t ies. Figur e 2(a) illust r at es how t he innit esimal (cont inuously com-
pounded) for war d int er est r at e f
T
(t) bet ween t imes T-T and T can be
comput ed fr om t he pr ices B
T
(t) and B
TT
(t) of zer o-coupon bonds
mat ur ing at t imes T and TT r espect ively
( EQ 2)
or equivalent ly,
( EQ 3)
Figur e 2(b) shows a similar but mor e involved r elat ionship t he local
volat ilit y
K,T
in t he r egion near index level K and t ime T and opt ions
pr ices C
K,T
(t) for opt ions of st r ike K and expir at ion T. Assuming for
simplicit y t hat int er est r at es and dividend yields ar e zer o, t his r ela-
t ion is
( EQ 4)
f
T
B
T
B
T-T
C
K,T
C
K+K,T-T
C
K,T-T
C
K-K,T-T

K,T
FIG URE 2. Analogy between interest rate and volatility. (a) Forward
rates can be extracted from the current bond prices. (b) Local
volatilities can be extracted from the current option prices.
0 T - T T 0 T - T T
(a) (b)
B
T
t ( )
B
T T
t ( )
----------------------- f
T
T ( ) exp =
f
T
B
T T
t ( ) log B
T
t ( ) log
T
------------------------------------------------------------- =

K T ,
2
2
C
K T ,
t ( ) C
K T T ,
t ( )
T
-----------------------------------------------------
C
K K + T T ,
t ( ) 2C
K T T ,
t ( ) C
K K T T ,
t ( ) +
K
K
--------
. ,
| `
2
--------------------------------------------------------------------------------------------------------------------------
-------------------------------------------------------------------------------------------------------------------------- =
4
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
The numer at or in Equat ion 4 is r elat ed t o t he value of a posit ion in
an innit esimal calendar spr ead; t he denominat or is r elat ed t o t he
value of a posit ion in an innit esimal but t er y spr ead. Ther efor e,
local volat ilit y is r elat ed t o t he r at io of t he value of calendar t o but -
t er y spr eads
3
.
Appendix A shows t hat t he local var iance
2
K,T
is t he conditional
risk-neutral expectation of t he inst ant aneous fut ur e var iance of index
r et ur ns, given t hat t he index level at t he fut ur e t ime T is K. We can
also int er pr et t his measur e as a K-level, T-maturity forward-risk-
adjusted measure. This is analogous t o t he known r elat ionship
bet ween t he for war d and fut ur e spot r at es; f
T
is t he forward-risk-
adjusted expectation of t he inst ant aneous fut ur e spot r at e.
Many xed income invest or s choose t o analyze t he r isk of t heir por t -
folios in t er ms of sensit ivit y t o for war d r at es. Hedging t he por t folio
against changes in a par t icular for war d r at e r equir es t aking a posi-
t ion in a t r aded inst r ument whose pr esent value has an offset t ing
sensit ivit y t o t he same for war d r at e. It is oft en convenient for t he ini-
t ial hedge t o be cost less, as is t he case for fut ur es cont r act s. An inter-
est-rate gadget is a por t folio of bonds wit h zer o mar ket pr ice t hat has
sensit ivit y t o only one par t icular for war d r at e. You can t hink of it as
somet hing ver y much like a synt het ic fut ur es cont r act on for war d
r at es, const r uct ed fr om a por t folio of zer o-coupon bonds.
Figur e 3(a) displays t he const r uct ion of an innit esimal gadget
T
synt hesized fr om:
a long posit ion in B
T
, a zer o-coupon bond of mat ur it y T wit h value
B
T
(t) at t ime t, and
exp(-f
T
(0) T) unit s of a zer o-coupon bond B
T-T
of mat ur it y T T
of value B
T-T
(t),
wher e f
T
(0) is t he for war d r at e bet ween T-T and T at t he init ial t ime
t = 0. The value of t he innit esimal gadget
T
at t ime t is given by:
( EQ 5)
3. The cont inuous-t ime equat ion for local volat ilit y when r at es and yields ar e non-
zer o is given by:

2
K T ,
2
T
C
K T ,
r ( )K
K
C
K T ,
C
K T ,
+ +

' '

K
2
K
2
2

C
K T ,
=
INTRO DUC ING G ADG ETS:
G ADG ETS FO R INTEREST
RATES

T
t ( ) B
T
t ( ) f
T
0 ( )T [ ]B
T T
t ( ) exp =
5
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
Combining t his wit h Equat ion 2, we obt ain t he value of t he gadget in
t er ms of t he for war d r at e f
T
(t) at t ime t:
( EQ 6)
The init ial value (at t ime t = 0) of t his gadget is zer o, and it s value is
sensit ive only t o t he par t icular for war d r at e f
T
(t). As t ime elapses, it s
value r emains zer o as long as t he for war d r at e f
T
(t) does not change.
However, if f
T
(t) decr eases (incr eases), t he gadget value will cor r e-
spondingly incr ease (decr ease). In t his r espect , t he gadget s r esponse
t o changes in int er est r at es is similar t o t hat of a long posit ion in a
Eur odollar fut ur es cont r act . Aft er such a change, however, t he gadget
value becomes sensit ive t o r at es of all mat ur it ies less t han T.
Figur e 3(b) illust r at es t he const r uct ion of a nit e int er est -r at e gadget
( EQ 7)
consist ing of a long posit ion in a zer o coupon bond B
T1
and a shor t
posit ion in
1,2
zer o coupon bonds B
T2
. Since t he gadget is dened
t o have zer o init ial pr ice,
( EQ 8)
(a)
(b)
0 T - T T
0
T
2
T
1
B
T
B
T-T
B
T1
B
T2
f
T
f
T1,T2
FIG URE 3. Constructing gadgets from zero-coupon bonds. (a)
Innitesimal interest rate gadget
T
is sensitive only to the
innitesimal forward rate f
T
. (b)Finite gadget
T1,T2
is sensitive only
to the forward rate f
T1,T2
.
Innitesimal gadget
T
:
(1) long one zero-coupon bond of maturity T;
(ii) short exp(-f
T
T) zero coupon bonds of
maturity TT
Finite gadget
T1,T2
:
(1) long one zero-coupon bond of maturity T
1
;
(ii) short exp(-f
T1,T2
(T
2
-T
1
)) zero coupon
bonds of maturity T
2

T
t ( ) f
T
t ( )T [ ] exp f
T
0 ( )T [ ] exp { B
T T
t ( ) =

T1 T2 ,
B
T1

T1 T2 ,
B
T2
=

T1 T2 ,
B
T1
0 ( )
B
T2
0 ( )
----------------- f
T1 T2 ,
0 ( ) T
1
T
2
( ) [ ] exp = =
6
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman

T1,T2
is t he cur r ent forward discount factor
4
fr om t ime T
1
t o t ime T
2
:
it is t he discount ed value at t ime T
2
of one dollar paid at t ime T1,
using for war d r at es obt ained fr om t he cur r ent (t = 0) yield cur ve t o do
t he discount ing.
The pr ice of nit e gadget
T1,T2
at t ime t is given by
( EQ 9)
and is sensit ive only t o t he for war d r at e f
T1,T2
cor r esponding t o t he
int er val bet ween t imes T
1
and T
2
. Not e t hat t he -coefcient s t hat
dene t he gadget s compound, namely:
( EQ 10)
We can also view
T1,T2
as weight ed combinat ions of innit esimal
gadget s
t
, wit h weight s
T1,t
for all t imes t bet ween T
1
and T
2
.
Int er est r at e gadget s have well-dened sensit ivit ies t o t he changes in
t he r espect ive for war d r at es. For a small change f
T
(t ) , t he pr ice
change of t he innit esimal gadget is seen fr om Equat ion 5 t o be
( EQ 11)
Similar ly for a small change f
T1,T2
, t he pr ice change of a nit e gad-
get fr om Equat ion 7 is
( EQ 12)
Consider a por t folio consist ing of a single zer o-coupon bond B
T
t hat
mat ur es at t ime T. Inst ead of t hinking of t he gadget
T,T1
as t he por t -
folio , we can equivalent ly r eplicat e t he bond B
T
by means of t he bond B
T1
and a gadget :
( EQ 13)
4. The for war d discount fact or
T,
sat ises t he forward equation for
all and boundar y condit ion
T,T
= 1.
T,
can be viewed as t he propagator
(Gr eens funct ion) for t he backwar d diffusion in t ime effect ed by t he oper at or .
It also sat ises t he backwar d equat ion and hence can be viewed as
t he pr opagat or for t he diffusion for war d in t ime effect ed by t he oper at or .
T

f
T
+
. ,
| `

T ,
0 =
T
T

f
T
+
t

f
t

. ,
| `

T t ,
0 =
t

f
t

T1 T2 ,
t ( ) B
T1
t ( ) f
T1 T2 ,
0 ( ) T
1
T
2
( ) [ ]B
T2
t ( ) exp =
f
T1 T2 ,
t ( ) T
1
T
2
( ) [ ] f
T1 T2 ,
0 ( ) T
1
T
2
( ) [ ] exp exp { B
T2
t ( ) =

T1 T3 ,

T1 T2 ,

T2 T3 ,
=

T
t ( ) T f
T
f
T
t ( )T ( )B
T T
t ( ) exp =
T f
T
B
T
t ( ) =

T1 T2 ,
t ( ) T
1
T
2
( ) f
T1 T2 ,
B
T1
t ( ) =
HEDG ING AG AINST
FO RWARD RATE C HANG ES

T T1 ,
B
T

T T1 ,
B
T1
=
B
T

T T1 ,

T T1 ,
B
T1
+ =
7
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
A zer o coupon bond of mat ur it y T has exact ly t he same payoff as t he
gadget
T,T1
and a posit ion in
T,T1
zer o coupon bonds B
T1
of shor t er
mat ur it y T1.
We can now r eplace B
T1
by anot her gadget
T1,T2
and a posit ion in a
bond B
T2
of st ill shor t er mat ur it y T2, and so on, t o obt ain
( EQ 14)
wher e we have used Equat ion 10 t o compound t he
T,Tn
coefcient s.
The ver y last t er m B
t
is a zer o coupon bond wit h mat ur it y at t he
pr esent t ime t, and is t her efor e equal t o one dollar of cash.

T,T'
is t he cur r ent for war d discount fact or fr om t imes T t o T' . Equa-
t ion 14 shows t hat you can st at ically r eplicat e a zer o coupon bond by
holding an amount of cash equal t o it s pr esent value, and buying an
appr opr iat e por t folio of cost less int er est -r at e gadget s, each weight ed
by t he init ial for war d discount fact or s fr om bond mat ur it y t o gadget
mat ur it y. The por t folio of cost less gadget s allows you t o r einvest your
cash fr om each gadget expir at ion t o t he next at t he cur r ent for war d
r at e, wit h t he quant it y of gadget s available insur ing t he amount of
cash at each gadget expir at ion. In t his way you can lock in t he face
value of t he zer o coupon at nal mat ur it y
5
. Figur e 5 illust r at es t his
hedging scheme for a zer o-coupon bond using a simple diagr am.
The gadget s pr ovide t he r eplicat ion; conver sely, if you own t he zer o
coupon bond, you can hedge it against fut ur e moves in all for war d
r at es, once and for all, by t aking an offset t ing posit ion in t he set of
gadget s.
You can r eplicat e or hedge por t folios of fut ur e cashows by applying
t he pr ocedur e out lined above t o each of t hem, and aggr egat ing t he
posit ions in t he gadget s.
5. If you lack an int er mediat e gadget for some for war d per iod, you can r oll over your
cash at cur r ent for war d r at es only out t o t he st ar t of t hat gadget per iod. Dur ing
t he per iod spanned by t he missing gadget , you ar e unhedged, and t he cash may
gr ow at a r at e differ ent fr om t odays for war d r at e. Fr om t her e on, you have gad-
get s t o guar ant ee r olling over cash at t he for war d r at es again, but , since t he cash
gr ew at t he wr ong r at e for one per iod, t he face value t he gadget s hedge may not
mat ch t he cash you act ually have at t hat point .
B
T

T T1 ,

T T1 ,

T1 T2 ,

T T2 ,

T2 T3 ,

T T3 ,

T3 T4 ,
...
T t ,
B
t
+ + + + + =
8
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Consider anot her simple por t folio consist ing of a single cash ow C
T
at some xed t ime T in t he fut ur e. Figur e 6 shows how t o hedge t he
pr esent value of t his simple por t folio against a set of for war d r at es.
To say t hat we want our por t folio t o be hedged against a par t icular
set of for war d r at es, we mean t hat we want it s pr esent value t o
r emain unchanged if t hose, and only t hose, for war d r at es under go
some fut ur e change. We let V

denot e t he forward price at any fut ur e

at t ime < T is dened by V

=
T,
C
T
, i.e. t he discount ed value of t he
cashow t o t ime using t he pr evailing yield cur ve. Figur e 6(a) shows
t hat we can hedge our por t folio against t he for war d r at e f
T1,T2
by
t aking a shor t posit ion in V
T1
gadget s
T1,T2
. The quant it y V
T1
is
obser ved t o be independent of t he for war d r at e f
T1,T2
, so it will
r emain unchanged as long as f
T1,T2
is t he only for war d r at e along t he
cur ve t hat changes. Figur e 4(b) illust r at es hedging against t wo for -
war d r at es f
T1,T2
and f
T3,T4
, cor r esponding t o t wo differ ent t ime
int er vals along t he yield cur ve. Again, as long as all ot her for war d
r at es along t he yield cur ve r emain unchanged, we can hedge our
por t folio against bot h of t hese for war d r at es by t aking a shor t posi-
FIG URE 5. Converting a zero-coupon bond into a portfolio of cash and
gadgets. The vertical arrows represent the cashows of zero coupon
bonds. The lighter horizontal double arrows represent costless
interest-rate gadgets. The portfolios below are all equivalent in value
and sensitivity to forward rates.
zero coupon B
T
with
cashow at time T
zero coupon B
T1
with
cashow at T
1
and a gadget
zero coupon B
T2
with
cashow at T
2
and two gadgets
zero coupon B
T3
with
cashow at T
3
and three gadgets
cash and four gadgets
time
0 T T
1
T
2
T
3
He d g ing a Portfolio of
C a shows Ag a inst a Se t of
Forwa rd Ra te s
9
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
The same t echnique can be used for hedging against for war d r at es
cor r esponding t o any number of specied r egions along t he yield
cur ve.
We can ext end t his hedging scheme t o ar bit r ar y por t folios consist ing
of any number of cashows. Figur e 7 shows an example of a por t folio
wit h sever al cashows, some of which fall wit hin t he for war d r at e
r egions which we want t o hedge our por t folio against . To do t his we
will divide each r egion int o subr egions bet ween any t wo consecut ive
cashows, and t hen hedge our por t folio against t he for war d r at es
associat ed wit h t hese subr egions, by t aking a shor t posit ion in t heir
r espect ive int er est r at e gadget s. The cor r ect amount t o be shor t for
each gadget
TfTi
, cor r esponding t o t he int er val bet ween T
i
and T
f
, is
V
Tf
, t he for war d pr ice at t ime T
f
. Not e t hat t he for war d pr ice at any
t ime t wher e t her e is a cashow, must include t he value of t he cash-
ow at t hat point . In Figur e 7 we have shown t he t imes wher e t her e
is a cashow wit h an apost r ophe symbol.
0 T
2
T
1
T
f
T1,T2
c
T
0 T
2
T
1
T
f
T1,T2
c
T
f
T3,T4
V
T1
V
T1
V
T3
(a)
(b)
FIG URE 6. Hedging the present value of a single cash ow against the
changes in one or more forward rates along the yield curve. (a)
hedging against a single forward rate f
T1T2
, (b) hedging against two
forward rates f
T1,T2
and f
T3,T4
.
T
3
T
4
hedging against forward rate f
T1,T2
:
(1) long portfolio c
(ii) short V
T1
gadgets
T1,T2
hedging against forward rates f
T1,T2
and f
T3,T4
:
(1) long portfolio c
(ii) short V
T1
gadgets
T1,T2
(ii) short V
T3
gadgets
T3,T4
10
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Suppose we want t o hedge a por t folio of zer o coupon bonds against
changes in for war d r at es. Table 1 shows how a ve-year zer o coupon
bond wit h face value of $100 (t he t ar get ) can be per fect ly hedged
against changes in t he for war d r at e bet ween year 2 and year 3 by t ak-
ing a shor t posit ion in t he gadget
3,2
, as dened in Equat ion 7. The
example illust r at es t he changes in t he value of bot h t ar get zer o and
hedge for a t wo point change in t he for war d r at e.
Table 2 illust r at es how we can hedge t he same t ar get zer o wit h t wo
gadget s,
2,1
and
4,3
, against a change in t he for war d r at e bet ween
year 1 and year 2, and a change in t he for war d r at e bet ween year 3
and year 4. To be specic, we allow a simult aneous change of ve per -
cent age point s in t he for mer, and t hr ee per cent age point s in t he lat t er.
Finally, Table 3 cont ains a similar hedge for a t ar get por t folio of t wo
zer o coupon bonds, mat ur ing r espect ively in year 2 and year 5 wit h
face values of $100. The gadget s in t he hedge ar e used t o pr ot ect t he
value of t he por t folio against changes in t he same for war d r at es as in
Table 2.
FIG URE 7. Hedging the present value of an arbitrary portfolio of cash
ows against forward rates within several regions along the yield
curve. We must divide each region into smaller regions between
cashows and hedge against all of the corresponding forward rates.
0 T
2
T
1
T T
3
T
4
V
T1
V
T'1
V
T''1
V
T3
V
T'3
T'
1
T''
1
T'
3
f
T1T'1
f
T'1T''1 f
T''1T2
f
T3T'3
f
T'3T4
EXAM PLE O F INTEREST RATE
HEDG ING USING G ADG ETS
11
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
TABLE 1. Using gadgets to hedge a target ve-year zero coupon bond
against changes in the forward rate between two and three years.
Maturity
(years)
Zero
Yields
Initial
Forward
Rates
a
Zero
Prices
Target
Zero
Gadgets Gadgets
in
Hedge
Zeros
in
Hedge
Forward
Rates
Zero
Prices
Change in
Value of
Target Zeros
Change in
Value of
Gadgets Zeros
1 5.00% 5.00% $95.123
10
5.00% $95.123
2 5.25 5.50 90.032
21
0.7925 5.50 90.032
3 5.75 6.75 84.156
32
-0.8479 -0.8479 8.75 82.489 1.413
4 6.50 8.75 77.105
43
8.75 75.578
5 6.75 7.75 71.355 1
54
7.75 69.942 -1.413
total -1.413 1.413
a The forward rate corresponds to the one-year period ending at the corresponding maturity.
Final
TABLE 2. Using gadgets to hedge a target ve-year zero coupon bond
against changes in two forward rates. All rates are continuously
compounded and annual.
Maturity
(years)
Zero
Yields
Initial
Forward
Rates
a
Zero
Prices
Target
Zero
Gadgets Gadgets
in
Hedge
Zeros
in
Hedge
Forward
Rates
Zero
Prices
Change in
Value of
Target Zeros
Change in
Value of
Gadgets Zeros
1 5.00% 5.00% $95.123
10
0.7501 5.00% $95.123
2 5.25 5.50 90.032
21
-0.7925 -0.7925 8.50 87.372 2.109
3 5.75 6.75 84.156
32
0.8479 6.75 81.669 -2.109
4 6.50 8.75 77.105
43
-0.9254 -0.9254 13.75 71.177 5.486
5 6.75 7.75 71.355 1
54
7.75 65.869 -5.486
total -5.486 5.486
a The forward rate corresponds to the one-year period ending at the corresponding maturity.
Final
12
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
TABLE 3. Using gadgets to hedge a target portfolio of two and ve-year
zero coupon bonds against changes in two forward rates. All rates are
continuously compounded and annual.
Maturity
(years)
Zero
Yields
Initial
Forward
Rates
a
Zero
Prices
Target
Zero
Gadgets Gadgets
in
Hedge
Zeros
in
Hedge
Forward
Rates
Zero
Prices
Change in
Value of
Target Zeros
Change in
Value of
Gadgets Zeros
1 5.00% 5.00% $95.123
10
1.6966 5.00% $95.123
2 5.25 5.50 90.032 1
21
-1.7925 -1.7925 8.50 87.372 -2.660 4.768
3 5.75 6.75 84.156
32
0.8479 6.75 81.669 -2.109
4 6.50 8.75 77.105
43
-0.9254 -0.9254 13,75 71.177 5.486
5 6.75 7.75 71.355 1
54
7.75 65.869 -5.486
total -8.146 8.146
a The forward rate corresponds to the one-year period ending at the corresponding maturity.
Final
13
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
In t he same way as t he xed-income invest or s analyze t he int er est
r at e r isk of t heir por t folios using for war d r at es, index opt ions inves-
t or s should analyze t he volat ilit y r isk of t heir por t folios using local
volat ilit ies. We have seen t hat it is possible t o hedge xed-income
por t folios against local r at e changes along t he yield cur ve, by means
of int er est r at e gadget s. Similar ly, we can hedge index opt ion por t fo-
lios against local changes on t he volat ilit y sur face, using volatility
gadgets. A volat ilit y gadget is synt hesized fr om Eur opean st andar d
opt ions, just as an int er est r at e gadget is synt hesized fr om zer o-cou-
pon bonds. We can best illust r at e it s const r uct ion in a discr et e wor ld,
as shown in Figur e 8. This wor ld is descr ibed by an implied t r inomial
t r ee
6
wher e t he st ock pr ice at any t r ee node can move t o one of t hr ee
possible fut ur e values dur ing a t ime st ep. The locat ion of t he nodes in
t his kind of t r ee is gener ally at our disposal and can be chosen r at her
ar bit r ar ily. But , t hen t he t r ansit ion pr obabilit ies ar e complet ely con-
st r ained by t he r equir ement t hat all t he t r aded fut ur es (or for war ds)
and opt ions have pr ices at t he r oot of t he t r ee which mat ch t heir cur -
r ent mar ket pr ices. Figur e 8 shows a few nodes of t his t r ee at t imes
T-T and T. To keep our discussion gener al we leave t he locat ion of
t he nodes ar bit r ar y. The backward t r ansit ion pr obabilit ies p, q and 1-
p-q cor r espond, r espect ively, t o t he diffusion for war d in t ime fr om t he
6. See Der man, Kani and Chr iss [1996].
HEDG ING LO CAL VO LATILITIES:
VO LATILITY G ADG ETS
FIG URE 8. Synthesis of an innitesimal volatility gadget
K,T
using
standard options in a discrete world.
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . .
C
Ku,T-T
C
Kd,T-T
K, C
K, T
C
Km,T-T

d
p
q
1-p-q
K
u
K
d
K'
K''
, K
u
, K
m
,K
d
14
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
node K
m
at t ime T-T t o t hr ee nodes K
u
, K and K
d
at t ime T. These
pr obabilit ies gener ally var y fr om node t o node depending on t he
indexs local volat ilit y and gr owt h r at e t her e. They can be calculat ed
dir ect ly fr om t he known mar ket pr ices of st andar d Eur opean opt ions
on t he t r ee (see Foot not e 6). Our analysis begins at t ime t = 0, hence
in what follows t he pr obabilit y p r epr esent s p
(0)
et c.
An innit esimal volat ilit y gadget
K,T
in t his wor ld consist s of a long
posit ion in t he call opt ion C
K,T
, wit h st r ike K and expir ing at T, and a
shor t posit ion in
u
unit s of t he call opt ion ,
m
unit s of t he
call opt ion and
d
unit s of t he call opt ion , all expir -
ing at t he pr evious t ime per iod T-T. Let t ing C
K,T
(t,S ) denot e t he
pr ice at t ime t and index level S of a call opt ion wit h st r ike K and
expir at ion T, t he pr ice of t he innit esimal volat ilit y gadget
K,T
is
given by:

K,T
(t,S ) = C
K,T
(t,S ) -
u
C
Ku, T-T
(t,S ) ( EQ 15)
-
m
C
Km, T-T
(t,S )-
d
C
Kd, T-T
(t,S )
We choose t he weight s
u
,
m
and
d
of t he component opt ions so t hat
t he t ot al gadget value is zer o at ever y node at t ime T-T. This condi-
t ion is aut omat ically sat ised for t he node K
d
or any ot her node
st r ict ly below K
m
, like K'' in Figur e 8, because wit h t he index at
t hese nodes t he gadget s component opt ions all expir e out -of-t he-
money and t he t ar get opt ion has no chance of ever becoming in-t he-
money. However, t he t ar get opt ion and some or all of t he component
opt ions will have non-zer o values when t he index is at t he node K
m
or
at any node above it . Table 3 shows t hese values at nodes K
m
and K
u
,
and at an ar bit r ar y node K' above K
u
at t ime T-T.
TABLE 4. The values of the target option and the gadgets component
options at nodes K
m
, K
u
and K'.
Index Level
at Time
T - T
Tar get Opt ion
C
K,T
Gadget s Component Opt ions
C
Kd,T-T
C
Km,T-T
C
Ku,T-T
K
m
e
-r T
p (K
u
- K)
d
(K
m
- K
d
) 0 0
K
u
K
u
e
-T
- Ke
-r T

d
(K
u
- K
d
)
m
(K
u
- K
m
) 0
K' K'e
-T
- Ke
-r T

d
(K' - K
d
)
m
(K' - K
m
)
u
(K' - K
u
)
C
K
u
T T ,
C
K
m
T T ,
C
K
d
T T ,
15
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
The value of t he t ar get opt ion C
K,T
at t ime T-T when t he index is at
t he node K
m
is equal t o t he discount ed expect ed value of it s payoff
when it expir es, at t he next t ime per iod at t ime T. Since t he index has
pr obabilit y p of moving up fr om t he node K
m
t o t he node K
u
, wher e
t he t ar get opt ion expir es and pays t he amount K
u
- K, t he value of t he
t ar get opt ion at t he node K
m
is given by e
-rT
p(K
u
- K), as shown in
Table 4. In cont r ast , if at t ime T-T t he index ends up at t he node K
u
t hen t her e is no chance for t he t ar get opt ion t o expir e out of t he
money. At t his point t he value of t he t ar get opt ion will be t he same as
t he value of a for war d cont r act wit h deliver y pr ice K, and is equal t o
K
u
e
-T
- Ke
-rT
. Ther e is a similar sit uat ion when t he index is at t he
node K' wher e t he t ar get opt ion will be wor t h K'e
-T
- Ke
-rT
. All t hr ee
opt ions compr ising t he gadget expir e at t ime T-T. If at t his t ime t he
index is at any node st r ict ly below K
m
, all t hr ee opt ions will expir e
wor t hless. If t he index ends up at K
m
, only t he component opt ion
C
Kd,T-T
will be in t he money. Since t he weight of t his opt ion wit hin
t he gadget is
d
, t he t ot al value of t his component at t his node is

d
(K
m
- K
d
). Similar ly, if t he index ends up at K
u
, only t he t wo opt ions
C
Kd,T-T
and C
Km,T-T
will have non-zer o values, r espect ively equal t o

d
(K
u
- K
d
) and
m
(K
u
- K
m
). Last ly, wit h t he index at K' at t ime T-T,
all t hr ee component opt ions expir e in t he money wit h t heir values
shown in t he last r ow of Table 4.
Requir ing t hat t he gadget value must vanish at t he nodes K
m
, K
u
and K', we obt ain t hr ee equat ions const r aining t he weight s:

d
(K
m
- K
d
) = e
-rT
p (K
u
- K) ( EQ 16)

m
(K
u
- K
m
) +
d
(K
u
- K
d
) = K
u
e
-T
- K e
-rT

u
(K' - K
u
) +
m
(K' - K
m
) +
d
(K' - K
d
) = K' e
-T
- K e
-rT
The second and t hir d equat ions ar e obser ved t o be equivalent t o a
normalization condit ion

u
+
m
+
d
= e
-t
( EQ 17)
and a mean condit ion

u
K
u
+
m
K
m
+
d
K
d
= Ke
-rT
( EQ 18)
The r st equat ion, t he necessar y condit ion for vanishing of t he gad-
get pr ice when S = K
m
, can be used t o solve for t he unknown weight

d
in t er ms of t he backwar d diffusion pr obabilit y p:

d
= e
-rT
p (K
u
- K)/ (K
m
- K
d
) ( EQ 19)
16
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Using put s inst ead of calls we can get a similar expr ession for
u
:

u
= e
-rT
q (K - K
d
)/ (K
u
- K
m
) ( EQ 20)
Since, wit h t his choice of weight s, t he gadget value is zer o for all t he
nodes at t ime T-T, it will also be zer o for all of t he nodes at all ear -
lier t imes, including t he r st node which cor r esponds t o t he pr esent .
We can nor malize t he weight s by t he dividend fact or on t he r ight -
hand side of Equat ion 17, e.g
u
= e
t

u
et c. In t er ms of t he nor mal-
ized weight s, Equat ions 17 and 18 r ead:

u
+
m
+
d
= 1 ( EQ 21)

u
K
u
+
m
K
m
+
d
K
d
= Ke
-(r)T
( EQ 22)
Similar ly fr om Equat ions 19 and 20 we have:

d
= e
-(r)T
p (K
u
- K)/ (K
m
- K
d
) ,
u
= e
-(r)T
q (K - K
d
)/ (K
u
- K
m
) ( EQ 23)
Equat ion 21 has t he int er pr et at ion t hat s dene a pr obabilit y mea-
sur e, oft en called t he forward probability measure. This measur e
r elat es opt ions wit h differ ent st r ike and expir at ions. It r elat es longer
mat ur it y opt ions of a given st r ike t o opt ions of shor t er mat ur it y and
differ ent st r ikes. It can be ar gued t hat t his is t he pr obabilit y measur e
associat ed wit h a diffusion backwar d in t ime. Equat ion 22 shows t hat
t he backwar d diffusion has a dr ift r at e of t he same magnit ude, but
wit h opposit e sign, t o t hat of for war d diffusion. Finally, Equat ion 23
shows t hat t he for war d and backwar d diffusion pr obabilit ies for dif-
fusions t hr ough small t ime per iods ar e closely r elat ed. For example,
if t he node spacing is const ant t hr oughout t he (implied) t r inomial
t r ee and node pr ices ar e chosen t o gr ow along t he for war d cur ve, t he
t wo ar e obser ved fr om Equat ion 23 t o be ident ical. This is t r ue only
for innit esimal t ime per iods and does not gener ally hold for nit e
t ime per iods. Appendix B discusses t he r elat ionships bet ween for -
war d and backwar d measur es in mor e mat hemat ical det ail.
At t = 0 t he innit esimal volat ilit y gadget
K,T
has zer o pr ice, hence
fr om Equat ion 15 for all fut ur e levels S and t imes t ear lier t han T:
C
K,T
(t,S ) =
u
C
Ku, T-T
(t,S ) +
m
C
Km, T-T
(t,S ) +
d
C
Kd, T-T
(t,S ) ( EQ 24)
This r epr esent s a decomposition of an opt ion expir ing at T in t er ms of
opt ions expir ing at t he ear lier t ime T-T. Since t he coefcient s
u
,
m
Re la tion to Forwa rd Prob a b ility
Distrib ution

17
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
and
d
only explicit ly depend on t he local volat ilit y (and not on S or t),
t he same decomposit ion is also valid as long as local volat ilit y does
not change.
Convolut ion of backwar d diffusions for many small t ime st eps leads
t o backwar d diffusion for longer t ime per iods. This is illust r at ed in
Figur e 9 which shows t he r elat ionship bet ween an opt ion C
K,T
, of
st r ike K and expir at ion T, t o opt ions wit h var ious st r ikes K' expir ing
at an ear lier t ime T' . Let denot e t he weight of t he opt ion
C
K',T'
in t his decomposit ion of t he opt ion C
K,T
. In our implied t r ee
wor ld, t his weight does not depend on t he cur r ent t ime or t he cur r ent
index level, but it does depend on local volat ilit ies along var ious
pat hs which connect t he t wo point s (K,T) and (K',T'). J ust as befor e,
we can modify t he weight s by t he dividend fact or e
(T-T')
, i.e. by den-
ing . This modied weight can be
int er pr et ed as t he t r ansit ion pr obabilit y for backwar d diffusion fr om
t he level K at t ime T t o t he level K' at ear lier t ime T' .
A gener alizat ion of Equat ions 21-23 for nit e t ime int er vals can also
be given in t er ms of t he modied weight s as follows:
( EQ 25)
( EQ 26)
FIG URE 9. Convolution of backward diffusions through small time
steps leads to backward diffusion through longer time periods.
C
K,T
C
K1',T'
C
K2',T'
C
Kn',T'
.
.
.
.
.
.
.
.
.
. . . . . .
. . . . . .
. .. . . . .
. . . . . . .
K T K' T' , , , ( )
K T K' T' , , , ( ) e
T T' ( )
K T K' T' , , , ( ) =
K T K'
i
T' , , , ( )
i

1 =
K T K'
i
T' , , , ( )K'
i
i

Ke
r ( ) T T' ( )
=
18
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
Figur e 9 also illust r at es a nit e-t ime gener alizat ion of Equat ion 24.
At t = 0 t he call opt ion C
K,T
has t he same value, for all fut ur e levels
S and t imes t at (or befor e) ear lier t ime T' , as a por t folio consist ing of
opt ions C
K',T'
for all possible values of st r ike K', i.e.,
( EQ 27)
The same decomposit ion also holds t r ue as long as t he local volat ili-
t ies for all t he nodes shown in t his gur e do not change.
By combining sever al innit esimal volat ilit y gadget s we can for m
nit e volat ilit y gadget s of var ious shapes and sensit ivit ies t o differ -
ent r egions on t he local volat ilit y sur face. Figur e 10 illust r at es a few
examples of nit e volat ilit y gadget s const r uct ed in t his way.
Since all innit esimal gadget s ar e init ially cost less t hen ever y nit e
volat ilit y gadget will also be init ially cost less. A nit e gadget will
r emain cost less as long all local volat ilit ies in t he nodal r egion
dened by t hat gadget r emain unchanged. It s pr ice will change, how-
ever, as soon as any of t he local volat ilit ies in t his r egion changes.
K T K' T' , , , ( )
C
K T ,
t S , ( ) K T K'
i
T' , , , ( )C
K
i
' T' ,
t S , ( )
i

=
C onstruc ting Finite Vola tility
G a d g e ts
(a)
(b)
(c)
(d)
FIG URE 10. Combining innitesimal volatility gadgets to form various
nite volatility gadgets. Darker nodes represent long option positions
and lighter nodes represent the short option positions within the
gadget. Hollow nodes represent options for which the long and short
options precisely cancel, therefore, there is a zero net position for
these options in the gadget.
19
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
Figur e 11 depict s t he local volat ilit y r egion cor r esponding t o a nit e
volat ilit y gadget wit h an ar bit r ar y shape. The dar ker boundar y
point s r epr esent long opt ions and t he light er boundar y point s r epr e-
sent t he shor t opt ion posit ions. The pr ice of t he volat ilit y gadget is
only sensit ive t o t he var iat ions of local volat ilit ies wit hin t he dot t ed
nodal r egion, and is insensit ive t o changes in t he local volat ilit ies in
any ot her r egion of t he t r ee.
Suppose we want ed t o hedge a st andar d call opt ion C
K,T
, wit h st r ike
pr ice K and expir at ion T, against t he fut ur e changes of some local vol-
at ilit y (K*,T*), cor r esponding t o t he fut ur e level K* and t ime T*.
Figur e 12(a) shows how t o do t his wit hin t he cont ext of implied t r ino-
mial t r ees which we have been discussing. Analogous wit h t he int er -
est r at e case, we must shor t t he amount (K,T,K*,T*) of t he
innit esimal volat ilit y gadget
K*,T*
against t he long posit ion in
C
K,T
. This pr ocedur e will effect ively r emove t he sensit ivit y of t he
st andar d opt ion t o t he local volat ilit y (K*,T*). In addit ion, we do
t his at no cost since t he gadget
K*,T*
is init ially cost less. Figur e
12(b) shows t hat we can do t he same for any por t folio of st andar d
opt ions. The only differ ence in t his case is t hat we must shor t an
amount equal t o t he sum of t he weight s of t he innit esimal gadget s

i
(K
i
,T
i
,K*,T*) over all t he opt ions whose expir at ion T
i
is aft er T*.
FIG URE 11. Local volatility region corresponding to a nite volatility
gadget of an arbitrary shape.
Using Vola tility G a d g e ts to
He d g e Ag a inst Loc a l Vola tility
C ha ng e s
20
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
By past ing appr opr iat e number of innit esimal volat ilit y gadget s
t oget her we can cr eat e volat ilit y hedges against one or mor e nit e
r egions on t he volat ilit y sur face. Figur e 13 illust r at es sever al exam-
ples of t his const r uct ion. Figur e 13(a)-(b) show t his for a single st an-
dar d opt ion and Figur e 13(c) shows t his for ar bit r ar y por t folios of
st andar d opt ions. Finally, Figur e 13(d) shows t hat t he same can be
done when some of t he opt ions in t he por t folio fall wit hin t he local
volat ilit y r egions of int er est . This is analogous t o t he similar case in
int er est r at es wher e some of t he cashows fall wit hin t he for war d
r at e r egions of int er est , as was shown in Figur e 7.
C
K,T
(K,T,K*,T*)
0 T*-T* T* T

K*,T*
C
K,T

i
(K
i
,T
i
,K*,T*)

K*,T*
C
K1,T1
C
K2,T2
C
Ki ,Ti
(a)
(b)
FIG URE 12. Hedging portfolios of standard options against a local
volatility (K*,T*). (a) Hedging a single standard option C
K,T
. (b)
Hedging a portfolio of standard options with different strikes and
expirations.
21
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
FIG URE 13. Examples of hedging portfolios of standard options
against one or more regions on the local volatility surface.
C
K,T
(a)
C
K,T
(b)
C
K,T
C
K1,T1
C
K2,T2
C
Ki ,Ti
C
K4,T4
C
K3,T3
C
K,T
(d)
C
K1,T1
C
K2,T2
C
K4,T4
C
K3,T3
(c)
22
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
In t his sect ion we pr esent a simple example of local volat ilit y hedging
using a discr et e wor ld r epr esent ed by a one-year, four-per iod
(implied) t r inomial t r ee. The st at e space, r epr esent ing t he locat ion of
all t he nodes in t his t r ee, is shown in Figur e 14 below. We have
assumed t hat t he cur r ent index level is 100, t he dividend yield is 5%
per annum and t he annually compounded r iskless int er est r at e is
10% for all mat ur it ies. We have also assumed t hat implied volat ilit y
of an at -t he-money Eur opean call is 25%, for all expir at ions, and t hat
implied volat ilit y incr eases (decr eases) 0.5 per cent age point s wit h
ever y 10 point dr op (r ise) in t he st r ike pr ice. The st at e space of our
implied t r inomial is chosen, for simplicit y, t o coincide wit h nodes of a
one-year, four-per iod, 25% const ant volat ilit y CRR-t ype, t r inomial
t r ee. Figur e 15 shows backwar d t r ansit ion pr obabilit ies, Ar r ow-
Debr eu pr ices and local volat ilit ies at differ ent nodes of t his t r ee
7
.
7. This st at e space is const r uct ed by viewing t wo st eps of a CRR binomial t r ee, wit h
st ep size t/ 2, as one st ep of a t r inomial t r ee wit h st ep size t. Ther efor e, t he t hr ee
st at es S
u
, S
m
and S
d
ar e given by , and . See Der -
man, Kani and Chr iss [1996] for det ailed algor it hms used for comput ing t hese quan-
t it ies.
AN EXPLIC IT EXAM PLE
FIG URE 14. State space of a one-year, four-period implied trinomial
tree constructed using a constant volatility of 25%.
100.00
83.80
119.34 119.34 119.34 119.34
100.00 100.00 100.00 100.00
142.41 142.41 142.41
70.22 70.22 70.22
169.95
169.95
58.84 58.84
49.31
202.81
83.80 83.80 83.80
0 0.25 0.5 0.75 1
time
(years)
A
B
S
u
Se
2t
=
S
m
S = S
d
Se
2t
=
23
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
Using Equat ions 17-23, we can calculat e t he for war d t r ansit ion pr ob-
abilit ies for all t he nodes in t he t r ee. The r esult has been shown in
Figur e 16.
Suppose we want ed t o hedge a st andar d opt ion wit h st r ike K = 100
and expir ing in T = 1 year s against t he changes of t he local volat ilit y
at t he node A, cor r esponding t o level K* = 100 and t ime T* = 0.5
backward up-transition probability tree:
nodes show p
i
backward down-transition probability tree:
nodes show q
i
Arrow-Debreu price tree:
nodes show
i
local volatility tree:
nodes show (s
i
,t
n
)
0 0.25 0.5 0.75 1
time
0.259
0.241 0.236 0.233
0.259 0.255 0.255
0.214 0.209
0.392 0.358
0.160
0.425
0.287 0.296 0.304
1.000
0.176
0.253 0.257 0.239 0.220
0.488 0.362 0.294 0.249
0.060 0.094 0.112
0.068 0.071 0.083
0.012 0.028
0.026 0.029
0.011
0.002
0.194 0.207 0.235
0.241
0.220 0.213 0.209
0.241 0.236 0.236
0.188 0.181
0.400 0.359
0.123
0.438
0.274 0.285 0.294
0.249
0.240 0.236 0.235
0.249 0.247 0.247
0.224 0.221
0.313 0.298
0.188
0.327
0.264 0.269 0.272
FIG URE 15. Backward transition probability trees, Arrow-Debreu price
tree and local volatility tree for the example.
(years)
24
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
year s, in Figur e 14. To const r uct t he hedge we need t he weight s for
differ ent opt ions compr ising t he one-per iod gadget cor r esponding t o
t he local volat ilit y at t his node. The t r ees of weight s
u
,
m
and
d
ar e
shown in Figur e 17. The last gur e also shows t he t ot al weight s
(K,T,K',T').
.
We can use t his infor mat ion t o comput e t he composi-
t ion of t he gadget and t he number of gadget s r equir ed t o be hedged.
The r oot of t he gadget consist s of a long posit ion in (100,1,100,0.75)
= 0.498 call opt ions wit h st r ike 100 and mat ur ing in 9 mont hs. The
t hr ee leaves of t he gadget consist of shor t posit ions in, r espect ively,
(100,1,100,0.75)
u,100,0.75
= 0.096 calls wit h st r ike of 119.34,
(100,1,100,0.75)
m,100,0.75
= 0.248 calls wit h st r ike of 100, and
(100,1,100,0.75)
d,100,0.75
= 0.148 calls wit h st r ike 83.80, all expir -
FIG URE 16. Forward up- , middle- and down- transition probability
trees for the example.
forward up-transition probability tree:
nodes show
ui
forward down-transition probability tree:
nodes show
di
forward middle-transition probability tree:
0 0.25 0.5 0.75 1
time
0.000
0.000 0.000 0.176
0.000 0.200 0.195
0.000 0.000
1.000 0.331
0.000
1.000
0.236 0.244 1.000
1.00
0.434
0.000 0.715 0.545 0.552
1.000 0.494 0.504 0.504
0.000 0.747 0.604
0.000 0.669 0.280
0.000 0.811
0.000 0.637
0.000
0.000
0.414 0.756 0.000
0.000
1.000 0.285 0.278
0.000 0.306 0.301
1.000 0.253
0.000 0.000
1.000
0.000
0.350 0.000 0.000
(years)
0.000
0.150
0.174
0.195
0.227
0.298
0.363
1.000
0.000
0.189
0.246
0.275
0.301
0.339
0.423
0.000
0.000
1.000
nodes show
mi
25
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
FIG URE 17. Forward up- , middle- and down- weight trees for the
example.
up- option weight tree:
nodes show
ui
down- option weight tree:
nodes show
di
middle- option weight tree:
0 0.25 0.5 0.75 1
time
0.000
0.000 0.000 0.174
0.000 0.197 0.193
0.000 0.000
0.987 0.327
0.000
0.987
0.233 0.241 0.987
0.987
0.428
0.000 0.706 0.538 0.545
0.987 0.488 0.498 0.498
0.000 0.737 0.596
0.000 0.660 0.276
0.000
0.801
0.000 0.629
0.000
0.000
0.409 0.747 0.000
0.000
0.987 0.281 0.275
0.000 0.302 0.297
0.987 0.250
0.000 0.000
0.987
0.000
0.345 0.000 0.000
(years)
0.000
0.148
0.171
0.193
0.225
0.294
0.359
0.987
0.000
0.186
0.243
0.271
0.297
0.335
0.417
0.000
0.000
1.000
nodes show
mi
total gadget weight tree:
nodes show (,,
i
,
i
)
0.950
0.247 0.200 0.193
0.301 0.370 0.498
0.034 0.000
0.102 0.000
0.000
0.000
0.297 0.269 0.414
0.000
0.000
0.000
1.000
0.000
0.000
0.000
0.000
0.000
26
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
ing in 6 mont hs. Ther efor e, we can wr it e t he composit ion of t he vola-
t ilit y gadget as:

A
= 0.498*C
100,0.75
- 0.148*C
83.80,0.5
- 0.248C
100,0.5
- 0.096*C
119.34,0.5
( EQ 28)
Figur e 18 illust r at es t he composit ion of t he volat ilit y gadget and it s
per for mance against a 2% (inst ant aneous) change in t he local volat il-
FIG URE 18. Price trees for the volatility gadget
A
and the option C
K,T
before and after a 2% change in the local volatility
A.
gadget price tree:
before
gadget price tree:
after
option price tree:
0 0.25 0.5 0.75 1
time
0.000
0.000 0.000 9.622
0.000 0.000 0.000
0.000 21.104
0.000 0.000
34.806
0.000
0.000 0.000 0.000
11.146
0.000
23.275 21.744 20.171 19.336
9.447 7.410 4.809 0.000
43.460 42.952 42.412
0.000 0.000 0.000
70.139 69.649
0.000 0.000
0.000
102.811
0.000 1.391 2.744
0.136
0.081 0.000 9.622
0.183 0.376 0.000
0.000 21.104
0.000 0.000
34.806
0.000
0.000 0.000 0.112
(years)
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
before
option price tree:
after
11.282
23.355 21.744 20.171
9.630 7.786 4.809
43.460 42.952
0.000 0.000
70.139
0.000
0.000 1.391 2.856
69.649
42.412
19.336
0.000
0.000
0.000
0.000
0.000
102.911
gadget composition:
0.000
0.000 -0.096 0.000
0.000 -0.248 0.498
0.000 0.000
0.000 0.000
0.000
0.000
0.000 -0.148 0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
27
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
it y
A.
. It shows t hat t he change in t odays pr ice (i.e. at node (1,1)) of
t he volat ilit y gadget , when local volat ilit y
A
is changed by some
amount , pr ecisely offset s a similar change in t he opt ion pr ice. The
same holds at any node on t he t r ee befor e t ime T*.
To illust r at e t he use of nit e volat ilit y gadget s, let us t r y t o hedge t he
same opt ion against changes in local volat ilit y at bot h nodes A and B
of Figur e 14. In Figur e 19, we have shown t he composit ion and t he
An Exa mp le Using Finite
Vola tility G a d g e ts
FIG URE 19. Price trees for the volatility gadget
,
and the option
C
K,T
before and after a 3% change in the local volatility
A
and a 5%
change in the local volatility
B
.
gadget price tree:
before
gadget price tree:
after
option price tree:
0 0.25 0.5 0.75 1
time
0.000
0.000 0.000 9.622
0.000 0.000 0.000
0.000 25.554
0.000 0.000
44.567
0.000
0.000 0.000 0.000
11.146
0.000
23.275 21.744 20.171 19.336
9.447 7.410 4.809 0.000
43.460 42.952 42.412
0.000 0.000 0.000
70.139 69.649
0.000 0.000
0.000
102.811
0.000 1.391 2.744
0.315
0.348 0.430 9.622
0.385 0.566 0.000
0.000 25.554
0.000 0.000
44.567
0.000
0.000 0.000 0.168
(years)
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
before
option price tree:
after
11.462
23.622 22.174 20.171
9.832 7.976 4.809
43.460 42.952
0.000 0.000
70.139
0.000
0.000 1.391 2.912
69.649
42.412
19.336
0.000
0.000
0.000
0.000
0.000
102.911
gadget composition:
0.000
0.000 -0.200 0.193
0.000 -0.301 0.498
-0.034 0.000
0.000 0.000
0.000
0.000
0.000 -0.148 0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
28
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
per for mance of t he of t he nit e gadget in t his case. The gadget compo-
sit ion is given by

,
= 0.498*C
100,0.75
+ 0.193*C
119.34,0.75
- ( EQ 29)
0.148*C
83.80,0.5
- 0.301C
100,0.5
- 0.200*C
119.34,0.5
- 0.034*C
142.41,0.5
The gur e also shows t hat t his gadget per for ms well against a 3%
inst ant aneous change in t he local volat ilit y
A
and a simult aneous 5%
inst ant aneous change in t he local volat ilit y
B.
.
We can use t r aded inst r ument s t o hedge xed income por t folios against
t he fut ur e uncer t aint y in t he for war d r at es. We can synt hesize simple
bond por t folios, wit h zer o init ial pr ice, whose values ar e (init ially) sen-
sit ive only t o specic for war d r at es. We call t hese por t folios int er est
r at e gadget s. By t aking a posit ions in differ ent of int er est r at e gadget s,
cor r esponding t o differ ent fut ur e t imes, we can hedge our xed income
por t folio against t he fut ur e changes of t he for war d r at es in any r egion
along t he yield cur ve. Because gadget s have zer o mar ket pr ice, t his
pr ocedur e is t heor et ically cost less.
We can devise a similar met hod for hedging index opt ions por t folios
against fut ur e local volat ilit y changes. We can synt hesize zer o-cost vol-
at ilit y gadget s fr om t he st andar d index opt ions. By buying or selling
suit able amount s of volat ilit y gadget s, cor r esponding t o differ ent
fut ur e t imes and mar ket levels, we can hedge any por t folio of index
opt ions against t he local volat ilit y changes on any of r egions on t he vol-
at ilit y sur face. Again, t his pr ocedur e is t heor et ically cost less. It can be
used t o r emove an unwant ed volat ilit y r isk, or t o acquir e a desir ed vol-
at ilit y r isk over any r ange of fut ur e index pr ices and t ime.
C O NC LUSIO NS
29
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
This appendix pr ovides t he gener al denit ion for local volat ilit y. It
also der ives t he for war d equat ion for st andar d opt ions which allows
ext r act ion of t he local volat ilit y funct ion fr om t he st andar d opt ions
pr ices.
We assume
8
a r isk-neut r al index pr ice evolut ion gover ned by t he st o-
chast ic differ ent ial equat ion
dS
t
/ S
t
= r
t
dt +
t
dZ
t
( EQ 30)
wher e r
t
is t he r iskless r at e of r et ur n at t ime t, assumed t o be a det er -
minist ic funct ion of t ime, and
t
is t he inst ant aneous index volat ilit y
at t ime t, assumed t o follow some as yet unspecied st ochast ic pr o-
cess
9
. Z
t
is a st andar d Br ownian mot ion under t he r isk-neut r al mea-
sur e. Let E(.) = E
t
(.) denot e t he expect at ion, based on infor mat ion at
t ime t, wit h r espect t o t his measur e. This infor mat ion may include,
for example, t he index pr ice S
t
(or Z
t
) and t he values of n addit ional
independent st ochast ic fact or s W
i
t
, i = 1,...n, which gover n t he st o-
chast ic evolut ion of index volat ilit y
t
.
The payoff of t he st andar d Eur opean call opt ion, wit h st r ike K and
expir at ion T, is given by (S
T
- K)
+
. For mal applicat ion of It os lemma
wit h t his expr ession gives
d(S
T
- K)
+
= (S
T
- K) dS
T
+ 1/ 2
2
T
S
2
T
(S
T
- K) dT ( EQ 31)
wher e (.) is t he Heaviside funct ion and (.) is t he Dirac delta func-
t ion. Taking Expect at ions of bot h sides of t his r elat ion and using
Equat ion 30 leads t o
d E {(S
T
- K)
+
}= r
T
E {S
T
(S
T
- K) }dT + 1/ 2 E {
2
T
S
2
T
(S
T
- K) }dT ( EQ 32)
We can r ewr it e t he r st t er m in mor e familiar for m not ing t hat
E {S
T
(S
T
- K) } = E {(S
T
- K)
+
} + K E {(S
T
- K) } ( EQ 33)
The st andar d Eur opean call opt ion pr ice is given by t he r elat ion
10
C
K,T
= D
T
E{(S
T
- K)
+
} wher e D
T
denot es t he discount funct ion cor r e-
8. We will not pr esent any ar gument s for t he exist ence or uniqueness of t he r isk-
neut r al measur e her e and, inst ead, mer ely post ulat e it in or der t o pr esent t he expec-
t at ion denit ion of local volat ilit y. Equat ion 39 gives an alt er nat ive denit ion of local
volat ilit y which does not a priori r equir e t he exist ence of t his measur e.
9. Subject t o t he usual measur abilit y and int egr abilit y condit ions.
10. The dependence on t, S
t
and t he n st ochast ic fact or s W
i
t
(or possibly t heir past
values) at t ime t is implicit in t his and ot her expect at ions comput ed at t ime t.
APPENDIX A: Loc a l Vola tility
a nd the Forwa rd Eq ua tion for
Sta nd a rd O p tions
30
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
sponding t o mat ur it y T, i.e. . Differ ent iat ing once
wit h r espect t o K gives
dC
K,T
/ dK = D
T
E {(S
T
- K)} ( EQ 34)
Differ ent iat ing t wice wit h r espect t o K gives
d
2
C
K,T
/ dK
2
= D
T
E {(S
T
- K) } ( EQ 35)
while differ ent iat ing wit h r espect t o T leads t o
dC
K,T
/ dT = r
T
D
T
E {(S
T
- K)
+
} + D
T
dE{(S
T
- K)
+
}/ dT ( EQ 36)
Replacing t he last t er m fr om Equat ion 32 combined wit h Equat ions
33-36 we nd
dC
K,T
/ dT = r
T
K dC
K,T
/ dK + 1/ 2 K
2
E{
2
T
(S
T
- K)} ( EQ 37)
We dene local var iance
2
K,T
, cor r esponding t o level K and mat ur it y
T, as t he condit ional expect at ion of t he inst ant aneous var iance of
index r et ur n at t he fut ur e t ime T, cont ingent on index level S
T
being
equal t o K:

2
K,T
= E {
2
T
| S
T
= K } = E {
2
T
(S
T
- K) } / E {(S
T
- K) } ( EQ 38)
Local volatility
K,T
is t hen dened as t he squar e r oot of t he local
var iance,
K,T
= (
2
K,T
)
1/ 2
. Using Equat ion 35, we can r ewr it e Equa-
t ion 37 in t er ms of t he local volat ilit y funct ion:
dC
K,T
/ dT = r
T
K dC
K,T
/ dK + 1/ 2 K
2

2
K,T
d
2
C
K,T
/ dK
2
( EQ 39)
This is t he forward equation sat ised by t he st andar d Eur opean
opt ions. It is consist ent wit h Dupir es equat ion
11
when inst ant aneous
index volat ilit y is assumed t o be a funct ion of t he index level and
t ime, i.e. when
T
= (S
T
,T). In t his case

2
K,T
= E {
2
T
| S
T
= K } = E {
2
(S
T
,T) | S
T
= K }
2
(K, T) ( EQ 40)
Equat ion 39 can be used as an alt er nat ive denit ion of local volat il-
it y. This denit ion has t he added advant age t hat it does not r equir e
t he knowledge of a r isk-neut r al measur e and it is ent ir ely dened in
t er ms of t r aded opt ion pr ices. Viewed as a funct ion of fut ur e level K
and mat ur it y T,
K,T
denes t he local volat ilit y sur face. In gener al,
t her e is an implicit dependence of t his sur face on t ime t, index pr ice
S
t
, and var iables W
i
t
, i = 1, ..n, or possibly t heir past hist or ies. In t he
11. See, for inst ance, Dupir e [1994] or Der man and Kani [1994].
D
T
r u ( ) u d
t
T

[ ] exp =
31
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
specic case when
T
= (S
T
,T), t hough, t hese dependencies collec-
t ively disappear, and we ar e left wit h a static local volat ilit y sur face
whose shape r emains unchanged as t ime evolves. We can also t hink
of t hese as effective theories wher e expect at ions of fut ur e volat ilit y
have been t aken (at some point in t ime) and t he r esult ing local vola-
t ilit y sur face is assumed t o r emain xed for t he subsequent evolu-
t ion. In an effect ive t heor y, t he inst ant aneous index volat ilit y is t hen
effect ively only a funct ion of t he fut ur e index level and fut ur e t ime,
and no ot her sour ce of uncer t aint y.
In t he mor e gener al st ochast ic set t ing, we can descr ibe t he evolut ion
of
2
K,T
by t he st ochast ic differ ent ial equat ion
d
2
K,T
/
2
K,T
=
K,T
dt +
K,T
dZ
t
+
i
K,T
dW
t
i
( EQ 41)
The dr ift
K,T
and volat ilit y funct ions
K,T
and
i
K,T
ar e in gener al
funct ions of t ime t, index level S
t
and fact or values W
i
t
and t heir past
hist or ies. Ther e is also an implied summat ion over t he index i t his
equat ion. in As we can see fr om Equat ion 40, t he special case of a
t heor y wit h
T
= (S
T
,T) cor r esponds t o d
2
K,T
= 0, leading t o zer o
values for all t hese funct ions.
The expr ession in t he denominat or of Equat ion 38 descr ibes t he pr ob-
abilit y t hat t he index level at t ime T ar r ives at S
T
= K. Denot e t his
pr obabilit y
12
by P
K,T
P
K,T
= E {(S
T
- K) } ( EQ 42)
Now consider t he st ochast ic differ ent ial equat ion descr ibing t he evo-
lut ion of t his pr obabilit y
dP
K,T
/ P
K,T
=
K,T
dZ +
i
K,T
dW
i
( EQ 43)
The vanishing dr ift in t his equat ion r esult s fr om t he fact t hat P
K,T
is
a local mar t ingale. Because t he numer at or on t he r ight -hand-side of
Equat ion 38 is also a mar t ingale, a simple applicat ion of It os lemma
t o bot h sides of t hat equat ion under t he assumpt ion t hat t he Br own-
ian mot ions W
i
and Z ar e uncor r elat ed gives

K,T
+
K,T

K,T
+
i
K,T

i
K,T
= 0 ( EQ 44)
Using t his ident it y we can r ewr it e Equat ion 41 in anot her for m
12. Nor mally we would wr it e t his pr obabilit y as P(t,S
t
,K,T) wit h t he dependence on
W
i
t
(and possibly t he hist or y) implicit ly under st ood.
32
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
d
2
K,T
/
2
K,T
=
K,T
(dZ -
K,T
dt) +
i
K,T
(dW
i
-
i
K,T
dt) ( EQ 45)
We see t hat in t er ms of t he new measur es = dZ -
K,T
dt and =
dW
i
-
i
K,T
dt t he local var iance is a mar t ingale:
d
2
K,T
/
2
K,T
=
K,T
+
i
K,T
( EQ 46)
We call t his new measur e t he K-level, T-maturity forward risk-
adjusted measure in analogy wit h T-mat ur it y for war d r isk-neut r al
measur e in int er est r at es (see J amshidian [1993]).
Let t ing E
K,T
(.) denot e expect at ions wit h r espect t o t his measur e, we
can r ewr it e Equat ion 38 in a simpler for m:

2
K,T
= E
K,T
{
2
T
} ( EQ 47)
Ther efor e, in t he K-T for war d r isk-adjust ed measur e t he local var i-
ance
2
K,T
is t he expect at ion of fut ur e inst ant aneous var iance
2
T .
This is analogous t o a similar sit uat ion in int er est r at es wher e t he
for war d r at e f
T
is t he T-mat ur it y for war d r isk-adjust ed expect at ion of
t he fut ur e spot (shor t ) r at e at t ime T.
dZ

dW

i
dZ

dW

i
33
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
In t his appendix we dene t he concept of for war d pr obabilit y mea-
sur e when t he index pr ice evolut ion is gover ned by some gener al
(mult i-fact or ) st ochast ic volat ilit y pr ocess, as descr ibed in Equat ion
30. In Appendix A we have alr eady seen t he denit ion of backwar d
pr obabilit y measur e. Set t ing S = S
t
and using t he expanded not at ion
P(t,S,K,T) in place of P
K,T
and D
t,T
in place of D
T
, Equat ion 42 gives
P(t,S,K,T) = E
t,S
{(S
T
- K) } = ( EQ 48)
Ther e is an implicit dependence on st ochast ic fact or s W
i
and per haps
t heir hist or y at t ime t which we will collect ively denot e by
t
, t hus
. In an effect ive t heor y wher e
K,T
= (K, T)
t he local volat ilit y sur face is st at ic and r emains unchanged as t ime
elapses, t her efor e t her e is no dependence on
t
. In t his t heor y, Eur o-
pean opt ion pr ice C is r elat ed t o pr obabilit y P t hr ough t he r elat ion:
( EQ 49)
Also in an effect ive t heor y, P sat ises t he for war d Kolmogor ov equa-
t ions:
( EQ 50)
It also sat ises t he backwar d Kolmogor ov equat ion
( EQ 51)
and, for any such t hat , t he Chapman-Kolmogor ov r elat ion
( EQ 52)
Fixing
t
in a gener al t heor y has t he effect of r est r ict ing t he evolu-
t ion of index pr ice t o be based on volat ilit ies along t he par t icular sur -
face of local volat ilit ies cor r esponding t o
t
. This evolut ion
denes an effect ive t heor y for each w, in which index pr ice evolves
wit h an effect ive inst ant aneous volat ilit y funct ion
T
= (S
T
, T) =
and whose t r ansit ion pr obabilit y measur e is . Once we x

t
, and r est r ict t he evolut ion t o a par t icular local volat ilit y sur face,
APPENDIX B: Forwa rd Prob a b il-
ity M e a sure
1
D
t T ,
-----------
K
2
2

C
t S ,
K T ,
P t S K T , , , ( ) P

t
t S K T , , , ( ) =
C
t S ,
K T , D
t T ,
P t S t' S' , , , ( )C
t' S' ,
K T , S' d
0

=
S'
2
2

1
2
---
2
S' t' , ( )S'
2
P t S t' S' , , , ( )
. ,
| `
S'

r t' ( ) ( )S' P t S t' S' , , , ( ) ( )

P t S t' S' , , , ( ) =
1
2
---
2
S t , ( )S
2
S
2
2

P t S t' S' , , , ( ) r t ( ) ( )S
S

+ P t S t' S' , , , ( ) =
t

P t S t' S' , , , ( )
t' t t' T
P t S T K , , , ( ) P t S t' S' , , , ( )P t' S' T K , , , ( ) S' d
0

S
T
T ,
w
t

S
T
T ,
w
t
P
w
t
34
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
we ar e wit hin t he cont ext of effect ive t heor ies and all of Equat ions
49-52 apply.
In Appendix A, we showed t hat t he Eur opean opt ion pr ice sat ises
t he for war d equat ion, given by Equat ion 39. For a x t his can be
shown t o be equivalent t o exist ence of a for war d pr obabilit y densit y
funct ion (K,T,K',T') dened by t he r elat ion
( EQ 53)
and sat isfying t he for war d equat ion (for (K, T) =

K,T
):
( EQ 54)
as well as t he Chapman-Kolmogor ov equat ion for any such t hat
( EQ 55)
We can ar gue t hat denes t he t r ansit ion pr obabilit y densit y for t he
evolut ion backward in t ime along t he effect ive local volat ilit y sur face
dened by . we can also view (K,T,K',T') as t he propagator (or
gr eens funct ion) for t he diffusion backwar d in t ime associat ed wit h
t he differ ent ial oper at or . Fur t her mor e, it fol-
lows fr om Equat ion 53 t hat
( EQ 56)
C
t S ,
K T , e
T t ( )
K T K' T' , , , ( )C
t S ,
K' T' , K' d
0

=
1
2
---
2
K T , ( )
K
2
2

K T K' T' , , , ( ) r
T
( ) K
K

K T K' T' , , , ( )
T

K T K' T' , , , ( ) =
T

T T

T'
K T K' T' , , , ( ) K T K

, , , ( ) K

K' T' , , , ( ) K

d
0

=
T
1
2
---
2
K T ,
K
2
2

r
T
( )K
K

+
t S K T , , , ( ) e
T t ( )
S
2
2

C
t S ,
K T ,
=
35
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
In t his appendix we pr esent an alt er nat ive const r uct ion for gadget s
using t he diffusion equat ions sat ised by t r aded asset s. Fir st we con-
sider int er est r at e gadget s. The zer o-coupon bond pr ices sat isfy t he
for war d differ ent ial equat ion
( EQ 57)
Her e B
T
(t) is t he pr ice at t ime t of a T-mat ur it y zer o-coupon bond sat -
isfying t he t er minal condit ion B
T
(T) = 1. The explicit solut ion of
Equat ion 57 is t he familiar bond pr icing for mula ,
but we do not r equir e t his explicit for m for our ar gument s her e. Let

T,T'
(t) denot e t he gr eens funct ion associat ed wit h t he oper at or
expr ession . This means t hat
T,T
(t) = 1 for all T and ,
and t hat for all t imes
( EQ 58)
In t er ms of t he Gr eens funct ion t he solut ion of Equat ion 57 is given
by
( EQ 59)
Since B
t
(t) = 1, it follows t hat B
T
(t) =
T,t
(t). Equat ion 59 has an int er -
pr et at ion which is useful for const r uct ing int er est r at e gadget s. To
see t his const r uct a por t folio
T,T'
consist ing of a long one T-mat ur it y
bond, B
T
, and shor t
T,T'
( t = 0 ) of T' -mat ur it y bonds B
T'
( EQ 60)
We call t his por t folio t he int er est r at e gadget associat ed wit h t he
t ime int er val bet ween T' and T. Fr om Equat ion 59, t he gadget pr ice
at t ime t = 0 is zer o. It s pr ice will change, however, if (and only if) t he
for war d r at e f
T,T'
associat ed wit h int er val bet ween T' and T changes.
For T' = T - T wit h small T we obt ain innit esimal int er est r at e
gadget s
T
=
T,T-T
. We can const r uct nit e int er est r at e gadget s
fr om innit esimal ones. The nit e gadget
T,T'
, for inst ance, can be
mat hemat ically descr ibed as
( EQ 61)
The volat ilit y gadget s can be const r uct ed in a similar way. Tr aded
st andar d opt ion pr ices sat isfy t he for war d differ ent ial equat ion
APPENDIX C : M a the ma tic s of
G a d g e ts
T

f
T
t ( ) +
. ,
| `
B
T
t ( ) 0 =
B
T
t ( ) f
u
t ( ) u d
t
T

. ,
| `
exp =
T

f
T
t ( ) + t T
t T' T
T

f
T
t ( ) +
. ,
| `

T T' ,
t ( ) 0 =
B
T
t ( )
T T' ,
t ( )B
T'
t ( ) =

T T' ,
B
T

T T' ,
0 ( )B
T'
=

T T' ,

T u ,

u
u d
T'
T

=
36
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
( EQ 62)
wher e C
K,T
(t) is t he pr ice at t ime t of a K-st r ike, T-mat ur it y st andar d
Eur opean call opt ion wit h t he t er minal condit ion C
K,T
(T) = Max(S
T
-
K, 0). Cont r ar y t o t he int er est r at e case, an explicit solut ion of Equa-
t ion 62 is unavailable, but is not needed for our discussion her e. Let

K,T,K',T'
(t) denot e t he gr eens funct ion associat ed wit h t he oper at or
. Ther efor e
K,T,K',T
(t) = (K' - K) for all values
of K, K' and t imes , and fur t her mor e, for all
( EQ 63)
The solut ion of Equat ion 62 in t er ms of t his Gr eens funct ion can be
wr it t en as
( EQ 64)
Set t ing T' = t we see t hat . As
befor e, Equat ion 64 nds an int er pr et at ion in t er ms of volat ilit y gad-
get s. Consider a por t folio composed of a long posit ion in one K-st r ike
and T-mat ur it y (Eur opean) st andar d call opt ion and a shor t posit ion
in
K,T,K',T'
( t = 0 ) unit s of K'-st r ike and T'-mat ur it y st andar d call
opt ions for all values of K' and T' wit h , i.e
( EQ 65)
Set t ing T' = T - T we nd innit esimal volat ilit y gadget s
K,T
=

K,T,T-T
. The nit e gadget
K,T,T'
can be const r uct ed out of innit es-
imal gadget s, for mally as
( EQ 66)
In fact by combining innit esimal volat ilit y gadget s, we can con-
st r uct an innit e var iet y of nit e volat ilit y gadget s associat ed wit h
any given r egion of t he volat ilit y sur face. Let R be any such r egion
(not necessar ily connect ed) and dene a nit e volat ilit y gadget
R
associat ed wit h t hat r egion as
( EQ 67)
1
2
---
2
K T ,
t ( )K
2
K
2
2

r
T
( )K
K

. ,
| `
C
K T ,
t ( ) 0 =
1
2
---
2
K T ,
t ( )K
2
K
2
2

r ( )K
K

t T t T' T
1
2
---
2
K T ,
t ( )K
2
K
2
2

r ( )K
K

. ,
| `

K T K' T' , , ,
t ( ) 0 =
C
K T ,
t ( )
K T K' T' , , ,
t ( )C
K' T' ,
t ( ) K' d
0

=
C
K T ,
t ( )
K T K' T , , ,
t ( )Max S
t
K' 0 , ( )
0

=
t T' T

K T T' , ,
C
K T ,

K T K' T' , , ,
0 ( )C
K' T' ,
K' d
0

K T T' , ,

K T v u , , ,
0 ( )
v u ,
v d u d
0

T'
T

R

K T v u , , ,
0 ( )
v u ,
R

=
37
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
Viewing volat ilit y gadget s as a collect ion of st andar d opt ions, we can
ar gue t hat t he st andar d opt ions which compr ise
R
all lie on t he
boundar y of t he r egion R. A mat hemat ical way of seeing t his is by
not ing t hat for mally
K,T
= A
K,T
C
K,T
wher e A
K,T
is t he differ ent ial
oper at or in Equat ion 62. It can be shown t hat if A*
K,T
is t he dual of
t he oper at or A
K,T
t hen A*
K,T

K,T,K',T'
= 0 for all K' and . Our
asser t ion t hen follows fr om Equat ion 67 using int egr at ion by par t s.
T' T
38
QUANTITATIVE STRATEGIES RESEARCH NOTES
Sac hs
Goldman
REFERENCES
Der man, E. and I. Kani (1994). Riding on a Smile. RIS K 7 no 2, 32-
39.
Der man, E., I. Kani and N. Chr iss (1996). Implied Tr inomial Tr ees of
t he Volat ilit y Smile. Journal of Derivatives, Vol. 3, No. 4, 7-22.
Der man, E., I. Kani and J. Zou (1995). The Local Volat ilit y Sur face.
To appear in Financial Analyst Journal.
Dupir e, B. (1994). Pr icing wit h a Smile. RIS K 7 no 1, 18-20.
J amshidian, F. (1993). Opt ion and Fut ur es Valuat ion wit h Det er min-
ist ic Volat ilit ies, Mathematical Finance, Vol. 3, No. 2, 149-159.
Rubinst ein, M.E. (1994). Implied Binomial Tr ees, Journal of Finance,
69, 771-818.
39
QUANTITATIVE STRATEGIES RESEARCH NOTES
Goldman
Sac hs
SELECTED QUANTITATIVE STRATEGIES PUBLICATIONS
J une 1990 Understanding Guaranteed Exchange-Rate
Contracts In Foreign S tock Investments
Emanuel Der man, Piot r Kar asinski
and J effr ey S. Wecker
J anuar y 1992 Valuing and Hedging Outperformance Options
Emanuel Der man
Mar ch 1992 Pay-On-Exercise 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 anuar y 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
December 1995 The Local Volatility S urface
Emanuel Der man, Ir aj Kani and J oseph Z. Zou
Febr uar y 1996 Implied Trinomial Trees of the Volatility S mile
Emanuel Der man, Ir aj Kani and Neil Chr iss