Risk Neutral Probabilities and Option Prices

Philip Protter
School of Operations Research, Cornell
University
13
th
Annual CAP Workshop, Columbia
University
This talk is based on joint work with Jean
Jacod
1
I. Overview
`o Aritrac a--umjtion i- cquivacnt to cxi-tcncc ot a
li-k `cutra `ca-urc Q
lt Q i- uniquc thc markct i- complete
lt Q i- not uniquc, thc markct i- incomplete: thcrc
i- a choicc ot an innitc numcr ot -uch Q, and not a
contincnt caim- can c rcjicatcd
lcojc cicvc markct- arc incomjctc. it thcy thouht
thcy wcrc comjctc and modc- wcrc corrcct, u-in hi-tor-
ica voatiity woud c cquivacnt to u-in imjicd voati-
itic-
2
A -tandard jrocm how doc- onc choo-c onc -uch Q
Iour mcthod- jrojo-cd to datc. jo--iy morc
`inima martinac mca-urc- (Iomcr-Schwcitzcr)
`inima cntrojy mca-urc-
Lquiirium arumcnt-
Aritrariy choo-c onc and -tick with it, tor no ood
rca-on
\c jrojo-c an atcrnativc
3
The usual approach is as follows:
1 Lcin with thc ri-ky a--ct jricc jrocc-- X (X
t
)
t0
and a (jo--iy random) -avin- ratc.
2 Ly a chanc ot numcrairc arumcnt, a--umc intcrc-t
ratc- arc zcro.
3 Iind and choo-c a ri-k ncutra mca-urc Q.
! lt g(X
T
) i- a nancia (Lurojcan -tyc) dcrivativc at
timc T, thc jricc jrocc-- ot thc dcrivativc i- declared
to c E
Q
{g(X
t
)|F
t
} tor 0 t T
4
Our approach:
1 Lcin with thc ri-ky a--ct jricc jrocc-- and a (jo--iy
random) -avin- ratc.
2 `cxt a--umc thcrc arc markct ivcn jricc jrocc--c- tor
a arc numcr ot (Lurojcan -tyc) dcrivativc- ot thc
torm g(X
T
), whcrc T varic-.
3 Iind thc cocction Q ot ri-k ncutra mca-urc- that
makc oth thc jricc jrocc-- and a ot thc dcrivativc
jricc jrocc--c- oca martinac-.
! If there are enough derivative prices, the
cardinality of Q might be one;
` Alternatively the cardinality of Q might be
zero or (comjatiiity i--uc- arc -criou- hcrc)
5
Other Approaches; Some key ones follow:
H. Dengler and R. Jarrow(199) A -imjc ,umj
modc whcrc thc jricc jrocc-- i- jurc ,umj, with two
di-tinct ,umj -izc-. two ca ojtion- arc u-cd to com-
jctc thc markct
B. Dupire (199) A -imjc -tocha-tic voatiity modc,
whcrc thc -tock jricc varic- ut thc cxjiration timc T
i- xcd A lLL i- otaincd, and whcn (and it) -ovcd,
it ivc- a uniquc martinac mca-urc.
E. Derman and I. Kani (199) thc -tocha-tic
voatiity ca-c, whcrc thc -trikc jricc K varic- and T
i- xcd Smoothnc-- in K i- a--umcd, and y twicc dit-
tcrcntiatin otain a dcn-ity tor thc ca ojtion, cad-
in to a lLL, which whcn (and it) -ovcd cad- to a
martinac mca-urc choicc
M. Schweizer and J. Wissel (200o), to ajjcar
in Math Finance imjrovcd on jrcviou- idca- y u--
in dicrcnt maturitic-, crcatin a tcrm -tructurc ot
voatiitic-, within a Lrownian a-cd -tocha-tic voati-
ity tramcwork Thcy nd condition- tor cquivacnt
martinac mca-urc- to cxi-t, and a-o a condition tor
it to c uniquc
6
Thc -cmina jajcr- ot Eberlein, Jacod and Raible
in-jircd thc jajcr on which thi- tak i- a-cd
lndccd, thc tak i- ony tor thc continuou- ca-c, ut an
anaoou- (and morc intcrc-tin) thcory i- dcvcojcd in
our jajcr tor thc cncra ca-c (with ,umj-) a- wc
7
I. For simplicity of presentation, we consider
here the continuous paths case.
(, F, P, F), whcrc F (F
t
)
t0
li-ky a--ct jricc
X
t
X
0
+

t
0
a
s
ds +

iI

t
0

i
s
dW
i
s
,
with

t
0
(|a
s
| +

iI
|
i
s
|
2
)ds < a-, a t
\c want X
t
> 0 -o wc a--umc a and tactor a
t
X
t
a
t
and
i
t
X
t

i
t

`cxt wc add thc nancia dcrivativc- (ojtion-), away-

Lurojcan -tyc
8
Lct g (0, ) R
+
, c nonncativc and convcx
The price of the option g(X
T
) at time t with
expiration time T is P(T)
t
.
\c -ujjo-c thcrc arc ojtion- with dicrcnt cxjiration
timc-, with thc -amc g, and let T denote the set of
expiration times.
|0, T

| is your period of trading.

T i- a nitc -ct in jracticc and can c quitc -ma Lut
in thc -jirit ot L!` modc-, wc takc T to c an intcrva
in R
+
, or a countac dcn-c -u-ct ot an intcrva
9
We have two cases:
Full Models: T

and T (0, );
Partial Models: T

< and T |T
0
, ) with
T
0
> T

.

0 T

T
0

Lct P(T)
t
thc jricc ot an ojtion g(X
T
) at timc t
\c mu-t havc P(T)
T
g(X
T
)
\c a-o takc P(T)
t
g(X
T
) P(T)
T
tor t T
10
The Standard Approach:
Choo-c Q cquivacnt to P -uch that thc jricc jrocc--
X i- a Q martinac (or a Q oca martinac), with
E
Q
(g(X
T
)) < and
P(T)
t
E
Q
(g(X
T
)|F
t
) tor t T.
(P(T)
t
)
t0
i- thcn a Q martinac on |0, T|
Change from the standard approach:
`ow x t and con-idcr T P(T)
t
, on thc intcrva |t, )
Sincc X i- continuou-, and g i- convcx (and hcncc con-
tinuou-),
T g(X
T
) i- continuou-.
Sincc X i- a Q martinac and g i- convcx, T g(X
T
)
i- a continuou- Q -umartinac Lcncc
T P(T)
t
is non decreasing and
continuous for T t, Q a.s.
and hcncc a-o P a-
11
So it P(T)
t
arc ojtion jricc-, T P(T)
t
i- a- contin-
uou- and nondccrca-in
Assume T P(T)
t
is absolutely continuous on
(0, ). Thus:
P(T)
t
g(X
t
) +

T
t
f(t, s)ds
tor t T, whcrc f(t, s) F
t
, with f 0
\hat do thc jrocc--c- f(t, s) ook ikc
Ior t < u a--umc f can c cxjrc--cd with a dccomjo-i-
tion ot thc torm
f(t, u) f(0, u) +

t
0
(r, u)dr +

iI

t
0

i
(r, u)dW
i
r
.
12
Ior thc Black-Merton-Scholes model, ct
C(x, t) E{g(xe
U

2
t
2
)} whcrc U i- N(0, 1)
Onc ha- C

t
(x, 0)

2
2
x
2
g

(x) whcn g i- C
2
.
\hcn g i- convcx ut not C
2
, thcn
C(x, t) g(x) O(

t) a- t 0.
More generally, C

t
(x, 0) exists if g is C
2
at x, and
does not exist if g is not.
And it g(x) (x K)
+
, thcn

tC

t
(x, t)

2

2
1
{x=K}
x a- t 0.
Onc ha-
f(t, T) C

t
(X
t
, T t)
f(0, T) C

t
(X
0
, T)
(r, T) C

tt
(X
r
, T r) +
1
2
C

txx
(X
r
, T r)
2
X
2
r
(r, T) C

tx
(X
r
, T r)X
r
13
Denition 1. A full option model for (X, g) is a
family of processes P(T), T > 0, given by P(T)
t

g(X
t
) +

T
t
f(t, s)ds, where, for t < s,
f(t, s) f(0, s) +

t
0
(u, s)du +

t
0

i
(u, s)dW
u
,
and
1. f(0, s) 0, non-random, locally integrable in s
2. appropriate measurability of and
i
3. and
i
are such that the integrals make sense
4. f(t, s) 0 for all t < s
5.

T
t
f(t, s)ds < a.s. for all t T.
A-o, wc want to dcnc
(s)
t

t
0
(|(u, s)| +

iI
|
i
(u, s)|
2
)du.
`otc that (3) aovc i- cquivacnt to (s)
t
< a- tor
a t < s
14
(s)
t

t
0
(|(u, s)| +

iI
|
i
(u, s)|
2
)du.
A-o notc that t (s)
t
i- incrca-in, ut s (s)
t
i-
not incrca-in in cncra
Denition 2. A full option model for (X, g) is
1. regular if (s)
s
< a.s., for almost all s. (This
implies f(s, s) is well dened.)
2. fair if for all T,

T
t
(s)
t
ds < a.s.
3. strongly regular if for all T,

T
t
(s)
s
ds <
a.s.
Strony lcuar

Iair
lcuar
Relation to Black-Scholes
P(T) i- a tu ojtion modc, which i- always fair
lt g C
2
, it i- strongly regular
lt g C
2
, thcn it i- not even regular
15
Easier situation: Partial Models
Trading takes place up to time T

Expiration dates all have T T

0
, with T
0
> T

.
\c ony nccd to modc P(T) tor T T
0
, with
P(T)
t
P(T
0
)
t
+

T
T
0
f(t, s)ds.
So it wc know thc dynamic- ot P(T
0
), thc modc wi c
-jccicd y thc dynamic- ot t f(t, s) tor s > T
0

Denition 3. A (T

, T
0
) partial option model as-
sociated with (X, g) is a family of processes (P(T)
T T
0
), where f is as in our denition of full option
models, and for t T

:
P(T)
t
P(T
0
)
t
+

t
0

s
ds +

iI

t
0

i
s
dW
i
s
,
with the integrability condition:

0
(|
t
| +

iI
|
t
i
|
2
)dt < ,
and nally
t |0, T

| P(T
0
)
t
g(X
t
).
16
Thc model is fair it wc havc

T
0
(s)
T

ds < a- tor
a T > T
0

Thc notion- ot rcuar or -trony rcuar do not ajjy

II. Equivalent Local Martingale Measures
Thc Lcacn-Schachcrmaycr thcory -ay- that no ari-
trac i- cquivacnt to thc cxi-tcncc ot at ca-t onc mca-urc
Q, cquivacnt to P, makin thc jricc jrocc-- a oca mar-
tinac
\c divcrc a it Q is locally equivalent to P it Q
and P arc cquivacnt on cach F
t
, t <
Example: W and Z, whcrc Z
t
W
t
+ t on |0, )
M
loc
i- thc -ct ot a jroaiity mca-urc- Qocay cquiv-
acnt to P, undcr which X and P(T) tor a T T arc
Q local martingales
M
loc
(T

, T
0
) i- thc -ct ot a Qon (, F
T

) cquivacnt to P
on F
T

and undcr which X and P(T), tor a T T

, arc
Q local martingales on the time interval |0, T

|
17
\c can charactcrizc thc oca martinac mca-urc- -imjy,
a- toow-
Denition 4. Let (b
i
s
)
iI
be predictable processes with

iI

t
0
(b
i
s
)
2
ds < . Two families B and B

of such
processes are equivalent if
i I b
i
b
i
The set of equivalent classes of these families of pro-
cesses is .
Theorem 5. There is a one-to-one correspondence
between and the Q which are locally equivalent to
P.
Moreover if B (b
i
s
)
iI
is in then under Q, the pro-
cesses W
i
t
W
i
t

t
0
b
i
s
ds are independent Brownian
motions.

loc
i- thc -ct ot a B which corrc-jond to a jro-
aiity mca-urc in M
loc
Anaoou-y

loc
(T

, T
0
) i- thc
-ct ot a B

(T

) which corrc-jond to a jroaiity

mca-urc in M
loc
(T

, T
0
).
18
Theorem 6. For a strongly regular model and if g is
C
2
, the set
loc
is the set of all (b
i
) which satisfy
a
s
+

iI

i
s
b
i
s
0 (1)
f(s, s)
1
2
g

(X
s
)

iI
(
i
s
)
2
(2)
and for all T s:
(s, T) +

iI

i
(s, T)b
i
s
0. (3)
Theorem 7. For a (T

, T
0
) partial fair model, the set

loc
(T , T

) is the set of all (b

i
)

which satisfy (1)

above for s T

, (3) for s T

and T T
0
, and
nally, for s T

s
+

iI

i
s
b
i
s
0
19
III. Completeness for (T

, T
0
) partial models
Theorem 8. Let us be given a (T

, T
0
) partial fair
model such that the set M
loc
(T

, T
0
) . Then there
are two alternatives:
(a) Either for all s T

and the closed linear sub-

space of L(I) spanned by the vectors (
i
()
s
)
iI
,
i
()
s
)
iI
and (
i
(, s, T)
iI
for T T
0
, is equal to L(I) itself;
in this case M
loc
(T

, T
0
) is a singleton;
(b) Or, this property fails, and the set M
loc
(T

, T
0
) is
innite.
lt, tor cxamjc, thc coccicnt-
i
s
and
i
s
and
i
(s, T) arc
takcn continuou- in s and T, thcn thcrc i- a nicc vcr-ion
ot cvcrythin and condition (a) can be checked
Thc hard jart wi c chcckin that thc -ct M
loc
i- not
cmjty' \hcn a thc coccicnt- can c takcn to c con-
tinuou- in s and T, and it thcrc i- a countac dcn-c -u-
-ct D(s, ) ot (T
0
, ) -uch that thc vcctor-
s
(),
s
(),
and (, s, T) tor T D(, s) arc incary indcjcndcnt,
then M
loc
is not empty
20
IV. Full Regular Models
A tu modc dc-cric- what hajjcn- whcn timc incrca-c-
and rcachc- thc cxjiration timc- T T , and thu- some
compatibility between the stock price and the
option model is required, in addition to P(T)
t

g(X
t
) Thi- rc-trict- thc jo--ic ojtion modc- a--oci-
atcd with (X, g) it wc want thcm to c trcc ot aritrac
\c know trom thc Lack-`crton-Schoc- cxamjc that
thcrc arc -omc modc-, ut cxi-tcncc in cncra i- dci-
catc
lcca thc `cycr-Tanaka tormua
g(X
t
) g(X
0
) +

t
0

s
ds + M
t
+ L
t
,
whcrc M i- a oca martinac, i- ot cour-c intcrac,
and L i- an adajtcd, incrca-in, -inuar jrocc-- (c, a
oca timc) lt g i- C
2
thcn L 0, ut othcrwi-c L doc-
not vani-h
21
Theorem 9. a) If the process L does not vanish a.s.
for any full strongly regular option model associated
with (X, g), then the set M
loc
is empty;
b) If g is C
2
and there exists at least one locally equiv-
alent probability such that the price process X is a
local martingale, then there is a sequence of stop-
ping times (S
n
)
n1
a.s. and for each n a full
regular option model associated with (X
S
n
, g)
such that M
loc
is not empty;
c) If g is C
2
with at most linear growth and there exists
at least one locally equivalent probability such that X
is a martingale, then there exists a full regular op-
tion model associated with (X, g) such that M
loc
is not empty.
Theorem 10. Assume that g is of class C
2
. A lo-
cally equivalent probability Q cannot be in M
loc
for
two dierent full strongly regular models associated
with (X, g).
22
Caution: Thcorcm 10 doc- not mcan that thcrc i- a
-inc aritrac trcc -trony rcuar ojtion modc a--oci-
atcd with (X, g) Thcrc coud c -cvcra, cvcn innitcy
many, with cach onc corrc-jondin to a dicrcnt cquiva-
cnt oca martinac mca-urc
Example: Markov type stochastic volatility
models
Thc jricc jrocc-- i-
X
t
X
0
+

t
0
a(X
s
, Y
s
)ds +

t
0
(X
s
, Y
s
)dW
1
s
(!)
Y can c mutidimcn-iona
Y
t
Y
0
+

t
0
h(X
s
, Y
s
)ds +

iI

t
0
k
i
(X
s
, Y
s
)dW
i
s
+

t
0

R
H(X
s
, Y
s
, x)( )(ds, dx)
with thc coccicnt- nicc cnouh tor uniquc -tron -ou-
tion- ot oth cquation-, and a-o that Y i- -tron `arkov
23
In the classic paradigm, choo-c ocay cquivacnt Q
with X a oca martinac, and we are not concerned
with whether or not Y is a local martingale
under Q Thcn cvauatc thc jricc- P(T)
t
ot ojtion-
undcr Q These prices may be discontinuous
Amon thc many -uch Q, thcrc arc -omc that jrc-crvc
thc `arkov jrojcrty, namcy thc cxtrcma onc- in thc
convcx -ct ot a ocay cquivacnt jroaiity mca-urc-
Lct (L, D
L
) c thc innitc-ima cncrator ot (X, Y ) un-
dcr Q A--umc too that D
L
C
2

Thcn thc jricc P(T)

t
takc- thc torm (undcr Q)
P(T)
t
E{g(X
T
)|F
t
} Q
Tt
G(X
t
, Y
t
),
whcrc G(x, y) g(x). lt g C
2
, hcncc G a- wc, thcn
G D
L
, and thcrctorc
P(T)
t
g(X
t
) +

T
t
Q
st
LG(X
t
, Y
t
)ds.
So in thi- ca-c, f(t, s) Q
st
LG(X
t
, Y
t
)
(f(t, s) t s) i- a Q martinac which i- di-continu-
ou- a- -oon a- Y i-
24
\c concudc, attcr -omc work and u-in rc-ut- trom thc
di-continuou- ca-c, that wc havc a- many tu rcuar oj-
tion modc- a--ociatcd with (X, g) with no aritrac a-
thcrc arc ocay cquivacnt jroaiitic- tor which (X, Y )
i- a `arkov jrocc-- and X i- a oca martinac
\hcn g i- convcx ut not C
2
, thcn in cncra G D
L

Lut undcr -omc cijticity, Q

G i- in it tor a > 0 \c
can ct a tu ojtion modc, ut it wi not c rcuar
25