Professional Documents
Culture Documents
Page 1 of 38
Lecture 4:
More on The Smile: Arbitrage Bounds,
Valuation Problems, Models
Summary of Lecture 3:
Hedging at different volatilities
Hedging at an arbitrary constant volatility, assuming (unrealistically) that all
volatilities are known:
T
1 r ( t t0 )
2
2
2
PV ( P&L ) = V h V i + --- e
h S ( r h )dt
2
t0
If you hedge at realized, your total P&L is deterministic but has uncontrollable
random changes along the way to expiration.
If you hedge at implied, the P&L predictable at each instant, but path-dependent and thus the total P&L is unknown.
Hedging discretely
If you hedge discretely at the wrong volatility, hedging more often doesnt
decrease your error because of the random nature of the P&L from terms linear
in the stock price movements.
If you hedge at the right volatility, with h = r = i , then the hedging error
goes to zero like n
0.5
2/14/10
C
------
n
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 2 of 38
C
at time t to the
S
C
S
0.5
) , converges to zero.
2/14/10
0.5
0.5
) , diverges.
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 3 of 38
This Lecture:
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 4 of 38
V
V
1 2 2 V 2
dt + dS + --- S
Z dt dS S N
=
2
t
S
2
S
2
V
V
1 2 2 V 2
Z dt S N
=
dt + ( Sdt + SZ dt ) + --- S
2
S
t
2
S
2
1 2 2 V 2
V
V
V
= SZ dt + --- S
Z + S + dt S N
2
S
t
2
S
where we have set the dividend yield D and the riskless rate r to zero, N is the
number of shares traded to rehedge the initially riskless portfolio at the next
interval, and the modulus sign reflects the fact that transactions costs are paid
for both buying and selling shares.
Now we hedge the initial portfolio by choosing as usual =
V ( S, t ) . After
S
V ( S + S, t + t ) V ( S, t )
S
S
2
V
S
V
S
SZ t
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 5 of 38
E [ N ] V SE Z
2
S
2
--- V S t
2
S
t =
with an average transactions cost obtained by multiplying the above by the cost
S per share, to yield the cost
2
2 V
2
--- 2 S t
S
The expected value of the change in the P&L is therefore given by
2
1 2 2 V 2 V
2 V
2
dE [ P&L ] = E --- S
Z
+
------S
dt
2
2
t
t
2
S
S
2
1 2 2 V 2 V
2 V
2
--- S
Z
+
------S
dt
2
t
t S 2
2
S
This isnt riskless, but rather stochastic. We are going to assume, as does Wilmott, that even though the portfolio isnt riskless, the holder of this not-quitehedged portfolio would expect to earn the riskless rate. In that case, since the
value of the hedged portfolio is V S
V
, the expected value of the portfolio a
S
V
time dt later should be r V S dt .
S
Inserting this expression into the LHS of the equation above leads to the equation
2
V 1 2 2 V 2
2
V
2
+ --- S
Z -------- V S + rS rV = 0
2
t 2
t 2
S
S
S
Eq.4.1
.
2
S
Because of the nonlinearity, the sum of two solutions to the equation is not necessarily a solution too; you cannot assume that the transactions costs for a portfolio of options is the sum of the transactions costs for hedging each option in
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 6 of 38
V
S
V 1 2 2 V 2
V
Z + rS rV = 0
+ --- S
2
t 2
S
S
Eq.4.2
where
2
2
2
= 2 -------t
When you sell the option you must ask for money because hedging it is going
to cost you.
The effective volatility is
2
-------t
Eq.4.3
For very small t this expression diverges and the approximation becomes
invalid.
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 7 of 38
Pre-crash
20
V olatility
18
16
14
0.95
0.975
Post-crash
20
1.025
Strike/Index
1.05
18
16
14
0.95
0.975
1.025
1.05
Strike/Index
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 8 of 38
S&P
September 27,
1995.
Strike
Oct. 1 2007
Jan. 24 2008
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 9 of 38
short-term implieds
move more
than long-term
negative correlation
during crisis
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 10 of 38
20
T
A
M
28
26
FO
R
25
30
24
22
20
18
30
14
85
90
95
100
105
110
115
120
80
85
90
SixMonth Options
22
20
16
26
24
22
20
90
95
100
105
110
115
TH
85
14
80
120
28
O
R
EP
22
20
18
85
90
95
100
95
100
105
110
115
120
115
120
30
30
24
90
ThreeYear Options
TwoYear Options
26
85
28
26
24
22
20
18
105
110
115
120
80
85
90
95
100
105
110
FourYear Options
FiveYear Options
32
LE
30
26
IL
IS
120
18
16
14
115
28
IS
18
24
22
20
18
80
120
TI
24
28
115
26
30
110
28
80
105
LE
30
30
80
100
OneYear Options
32
16
IT
95
IN
80
16
15
The slope of
implied volatility
against strike as a
percentage of spot
is negative, even
for long maturities, though not as
steep as for short
maturities.
35
TO
out-of-the-money
puts have higher
implied BlackScholes volatilities than out-ofthe-money calls.
(Why?)
ThreeMonth Options
34
32
OneMonth Options
40
28
26
24
22
20
18
85
90
95
100
105
110
115
120
80
85
90
95
100
105
110
Lines represent the sample averages of the implied volatility quotes plotted against the xed moneyness levels dened as strike prices as percentages of the spot level. Different panels are for options at different maturities. Data are daily from May 31, 1995, to May 31, 2005, spanning
2,520 business days for each series. The 12 lines in each panel represent the 12 equity indexes listed in Exhibit 1.
2/14/10
Lecture4.2010.fm
Moneyness, d
NKY
Moneyness, d
FTS
Moneyness, d
AEX
L
A
TO
25
26
27
28
29
30
31
32
33
34
35
25
26
27
28
29
30
31
32
33
34
14
16
18
20
22
24
Moneyness, d
E
C
U
D
OMX
Moneyness, d
HSI
Moneyness, d
PR
E
R
2
short maturity
long maturity
ALO
18
20
22
24
26
28
30
IS
TH
20
3
22
24
26
28
30
32
20
22
24
26
28
30
32
34
Moneyness, d
SMI
Moneyness, d
LE
C
I
T
R
IBE
Moneyness, d
CAC
16
18
20
22
24
26
28
30
32
22
24
26
28
30
32
IN
20
3
25
30
35
Moneyness, d
SPX
Moneyness, d
MIB
Moneyness, d
Y
N
G
E
L
T
A
R
O
F
DAX
Lines denote the sample averages of the implied volatility quotes, plotted against a standard measure of moneyness d = ln(K/S)/( ) where
K, S, and denote the strike price, the spot index level, and the time to maturity in years, respectively. The term represents a mean volatility
level for each equity index, proxied by the sample average of the implied volatility quotes underlying each equity index. For each equity index,
we plot the implied volatility smirks at the 8 different maturities in the same panel. The maturities for each line are 1 month, 3 months, 6
months, 1 year, 2 years, 3 years, 4 years, and 5 years. The length of the line shrinks with increasing maturity, with the longest line representing
the shortest maturity (1 month). The 12 panels correspond to the 12 equity indexes.
22
23
24
25
26
27
28
29
30
31
32
16
18
20
22
24
26
28
30
32
24
26
28
30
32
34
36
38
40
2/14/10
Average Implied Volatility, %
EXHIBIT 3
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 11 of 38
Strike
FIGURE 4.2. Implied Volatility as a Function of log -------------- ( )
Spot
related to d1
When plotted against the number of standard deviations between the log of the
strike and the log of the spot price for a lognormal process, the slope of the
skew actually increases with expiration. Whatever is happening to cause this
doesnt fade away with future time.
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 12 of 38
Sample Average
Sample Average
0.28
0.26
0.24
0.22
term structure of
implied volatility is
roughly flat
0.2
0.18
0.16
0.5
1.5
2.5
3.5
4.5
Maturity in Years
c0
0.16
volatility of volatility
decreases with expiration,
suggesting mean reversion
or stationarity for the
instantaneous volatility
evolution
Standard Deviation
0.14
0.12
0.1
0.08
0.06
0.04
0.5
1.5
2.5
3.5
4.5
Maturity in Years
c
0
daily autocorrelation of
implied volatility is large,
and larger for longer
maturities
[excitement or depression
tends to continue]
0.998
0.996
Autocorrelation
0.994
0.992
0.99
0.988
0.986
0.984
0.98
0.978
1
1.5
2.5
3.5
4.5
Maturity in Years
0.5
EP
0.982
The cross-correlation between volatility level and slope of the skew is large.
2/14/10
Lecture4.2010.fm
Page 13 of 38
R
EP
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
EXHIBIT 5
TO
0.4
IL
LE
G
A
L
0.1
0.2
0
0.2
Corr( c , c )
0 1
Corr(c ,c )
0.3
0.2
0.4
0.5
IS
0.6
0.7
0.4
IT
0.6
0.8
0.8
0.9
0.5
1.5
2.5
Maturity in Years
3.5
4.5
0.5
1.5
2.5
3.5
4.5
Maturity in Years
Lines denote the cross-correlation estimates between the volatility level proxy (c0) and the volatility smirk slope proxy (c1). The left panel measures the correlation based on daily estimates, the right panel measures the correlation based on daily changes of the estimates.
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 14 of 38
2/14/10
Volatilities are steepest for small expirations as a function of strike, shallower for longer expirations.
Low strike volatilities are usually higher than high-strike volatilities, but
high strike volatilities can also increase.
Shocks across the surface are highly correlated. There are a small number
of principal components or driving factors. Well study these effects more
closely later in the course.
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 15 of 38
one year:
slope ~ 5 volatility pt. per 25% change in strike
Indexes generally have a negative skew. The slope here for a one-year option is
0.05
of order 5 volatility points per 250 S&P points, or about ---------- = 0.0002 . Note
250
that the slope for a 3-month option is about twice as much, which roughly con( ln K S )
firms the idea that the smile depends on --------------------- , because a four-fold
( )
decrease in time to expiration then implied a doubling of the slope of the smile.
The magnitude of the slope of the one-month option volatility is about 23 volatility points per 250 S&P points, or about 0.001.
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 16 of 38
USD/EUR
JPY/USD
weak
USD
strong Euro
strong
USD
weak peso
9.85
123.67
The smiles are more symmetric for equally powerful currencies, less so for
unequal ones. Equally powerful currencies are likely to move up or down.
There are investors for whom a move down in the dollar is painful, but there
are investors for whom a move down in the yen, i.e. up in the dollar, is equally
painful. Hence, there is a motive for symmetry. FX smiles tend to be more
symmetric and resemble a real smile.
Equity index smiles tend to be skewed to the downside. The big painful move
for an index is a downward move, and needs the most protection. Upward
moves hurt almost no-one. An option on index vs. cash is very different and
much more asymmetric than an option on JPY vs. USD.
Single-stock smiles tend to be more symmetric than index smiles. Single stock
prices can move dramatically up or down. Indexes like the S&P when they
move dramatically, move down.
Interest-rate or swaption volatility, which we will not consider much in this
course, tend to be more skewed and less symmetric, with higher implied volatilities corresponding to lower interest rate strikes. This can be partially understood by the tendency of interest rates to move normally rather than
lognormally as rates get low.
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 17 of 38
1200
1150
1100
1050
1000
950
900
850
800
750
700
650
S&P
45
40
35
30
at-the-money volatility
25
20
ATM
11-02-98
10-01-98
09-01-98
08-03-98
07-01-98
06-01-98
05-01-98
04-01-98
03-02-98
02-02-98
01-02-98
12-01-97
11-03-97
10-01-97
09-01-97
15
INDEX
Here volatility goes up as the index goes down, and vice versa, but the volatility plotted is the at-the-money volatility ( S, t, S, T ) which is the implied
volatility of a different option each day, because as the index level S changes
the atm strike level changes. ATM volatility is therefore not the volatility of a
particular option you own.
If the indexs negative smile doesnt move as time passes and the index level
changes, then at-the-money volatility will go up when the index goes down
simply because the atm strike moves down with index level, and lower strikes
have higher implied volatilities. Thus, some of the apparent correlation in the
figure above would occur even if ( S, t, K, T ) didnt change with S at all.
How much of the correlation is true co-movement and not incidental?
A NOTE ABOUT FIGURES OF SPEECH: People in the market often talk
about how volatility changed. One must be very careful in speaking about
volatility because there are so many different kinds of volatility. There is realized volatility , at-the-money volatility, and implied volatility for a definite
strike, = ( S, t ;K, T ) which can vary with S,t and K,T. When you talk
about the change in , what are you keeping fixed?
For example, at-the-money volatility is atm = ( S, t ;S, T ) which constrains
strike to equal spot. When you talk about how this moves, its a very different
quantity from volatility of an option with a fixed strike. Its a little like the difference between talking about the yield of the 2016 bond and the yield of the
ten-year constant maturity bond over time. Those are different things: one ages
and the other doesnt.
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 18 of 38
The index skews variation with time and with market level
1400
1200
1000
6
800
4
25D-50D
SPX
600
2
400
0
200
05/04/99
03/16/99
01/26/99
12/08/98
10/20/98
09/01/98
07/14/98
05/26/98
04/07/98
02/17/98
12/30/97
11/11/97
09/23/97
08/05/97
06/17/97
04/29/97
03/11/97
01/21/97
12/03/96
10/15/96
08/27/96
07/09/96
05/21/96
04/02/96
02/13/96
12/26/95
11/07/95
09/19/95
08/01/95
06/13/95
04/25/95
03/07/95
01/17/95
11/29/94
10/11/94
08/23/94
0
07/05/94
05/17/94
-2
Date
Page 1
More recently, notice how skew varies with atm implied volatility.
S&P atm implied vol and risk reversal
between
-25 delta put and
25 delta call
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 19 of 38
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 20 of 38
Since depends on both strike and expiration, you can compare the
implied volatilities of differing expirations and strikes as a function of single variable.
Using the wrong quoting convention can distort the simplicity of the underlying dynamics. Perhaps the Black-Scholes model uses the wrong dynamics for
stocks and therefore the smile looks peculiar in that quoting convention. Thats
the underlying hope behind advanced models of the smile.
An example is the case of a stock price that undergoes arithmetic (rather than
geometric) Brownian motion. A constant arithmetic volatility then corresponds
to a variable lognormal volatility that varies inversely with the level of the
stock price. Plotting lognormal volatility against stock price would obscure the
simplicity of the underlying evolution.
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
4.6
Page 21 of 38
d ln ( S ) = ------ dt + dZ
2
2
where dZ = dt .
Then
2
S
ln -----t = ------ t + t
2
S0
Eq.4.4
K
= P ------ t + t > ln ----
2
S0
2
ln K S 0 ------ t
2
= P > ----------------------------------------------- t
= P [ Z > d2 ] = P [ Z < d2 ]
= N ( d2 )
For small t this is approximately equal to N ( d 1 ) or the delta of a call,
which is therefore approximately the risk-neutral probability of the option finishing in the money at expiration.
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 22 of 38
d 1, 2
S
ln --K
= ---------- ---------- and = T t
2
d1
y2
y2
exp ----- dy + exp ----- dy
2
2
0
d1
1
--- + ---------2
2
So, at the money,
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 23 of 38
1
1
--- + ---------- ---------- 0.5 + ( 0.4 ) ( 0.5 )
2
2 2
As an example, for a typical annual volatility of 0.2 (20%) and an expiration of
one year, we have 0.5 + 0.04 = 0.54 . (Check for yourself on a BlackScholes calculator that this is approximately the correct delta for an at-themoney option at this volatility.)
Now suppose we move slightly out of the money, so that K = S + S where
S
S
S is small. Then ln --------------- = ln ( 1 + S S ) ------ and so
S + S
S
( S ) S
( S ) S
d 1 ------------------ + ---------- = ------------------ + ---------2
2
where ( S ) S is the fractional move in the strike away from the money, and
is the square root of the variance over the life of the option.
Then for a slightly out-of-the-money option, a fraction ( S ) S away from the
at-the-money level,
d1
1
1
1 ( S ) S
--- + ---------- --- + ---------- ---------- ------------------
2
2 2
2 2
Lets look at a real example. Suppose ( S ) S = 0.01, a 1% move away from
the at-the-money. And also assume T = 1 year and = 0.2.
( 0.4 ) ( 0.01 )
Then 0.54 ---------------------------- = 0.54 0.02 = 0.52
0.2
Thus, decreases by two basis points for every 1% that the strike moves out of
the money for a one-year 20%-volatility call option.
The difference between a 50-delta and a 25-delta option therefore corresponds
to about a 12% or 13% move in the strike price.
The move S to higher index levels necessary to decrease the delta of an initially at-the-money call by 0.29 from 0.54 to 0.25 is approximately given by
1 ( S ) S
-------- ------------------ 0.29or ( S ) S = 0.29 2 0.29 2.5 0.2 0.15
2
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 24 of 38
Thus the strike of the 25-delta call is about 115. Actually its about 117 if you
use the exact Black-Scholes formula to compute deltas.
2
and a one-basis point change in corresponds to a change in ( S ) S of about
0.025 .
The key variable here is the percent move in stock price divided by the square
root of the annual variance. For a greater volatility or time to expiration, the
terminal distribution of the stock is broader, and you need a bigger move in the
strike to get to the same .
Example for a 1-month call with zero interest rates, 20% volatility
Delta
Delta decreases by 0.069 as
strike increases by 1%
Strike
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 25 of 38
r ( T t )
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 26 of 38
2. For the same expiration, options prices satisfy two constraints on their
derivatives:
2
C
C
> 0,
< 0 and
2
K
K
Proof: Look at payoff of a call spread and a butterfly.
payoff
call spread
C(K) - C(K+dK) > 0
butterfly
C(K-dK) - 2 C(K) + C(K+dK) > 0
C(K) - C(K+dK)
K+dK
K-dK
K+dK
The values of these portfolios are always positive, since they have positive
payoff, and are respectively proportional to the first and second derivatives
of the call price w/r/t strike in the limit as the dK 0 . Therefore these
derivatives must be positive, for actual call prices, independent of a model.
There are similar constraints on European put prices:
2
P
P
>0
> 0 and
2
K
K
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 27 of 38
P
C
> 0 put limits on the slope of the smile.
< 0 and
K
K
These constraint are true in the Black-Scholes formula with strike-independent
volatility. Now suppose you parameterize actual market call and put prices in
terms of the Black-Scholes formula, and so allow the implied volatility to vary
with strike (and expiration). Then, if volatility were to increase (decrease) with
strike level in the Black-Scholes formula, a too rapid increase (decrease) in
volatility could offset the natural decrease (increase) with strike for a call (put)
and so cause the call (put) price actually increase (decrease) with strike level.
These limits on call- and put-price slopes sets respective limits on the positive
and negative slope of the skew, as illustrated schematically below for call and
put prices at some fixed index level S.
The constraints on
implied
volatility
implied
volatility at
index level S
strike
Eq.4.5
C BS
r
e N ( d2 )
N ( d2 )
K
------------ = -----------------------------------=
------------------------- r
K
K N' ( d 2 )
C BS
Ke
N' ( d 2 )
Assume that volatility is small and strike price is at-the-money forward so that
1
S F = K . Then d 2 0 , N ( d 2 ) 0.5 and N' ( d 2 ) ---------- , so that
2
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 28 of 38
1
--- ----------K
2K
Eq.4.6
( 1.25 )
--------------K K
Eq.4.7
or
We can also examine the limits for asymptotically short and long expirations.
From Equation 4.5 we see that
N ( d2 )
-------------------------- K K N' ( d )
2
1 2
O(
)
K
as 0 .
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 29 of 38
1 2
).
1 1
1
----------- --------------- O ------ ------ O ---
K K N' ( d 2 )
as the time to expiration gets large. (To prove the line above we have made use
of the asymptotic relation ( N ( d 2 ) ) N' ( d 2 ) O (
0.5
) as .)
Thus, the slope of the smile can decrease with time to expiration no more
1
slowly than O ( ) .
Reference: Arbitrage Bounds on the Implied Volatility Strike and Term Structures of European-Style Options. Hardy M. Hodges, Journal of Derivatives,
Summer 1996, pp. 23-35.
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 30 of 38
The Black-Scholes model assumes constant volatility and cannot account for
the pattern of options prices at a given time. It attributes a different underlying
stock implied volatility to each option with a different strike, but that implied
volatility is not really the volatility of the option; it is the volatility of the stock,
and in the model, a stock must have one lognormal volatility of returns which
cannot know about the options strike. There cannot be many GBM volatilities
for the same stock.
Therefore, Black-Scholes is often simply being used as a quoting mechanism,
rather than a valuation mechanism, similar to the way in which yield to maturity is used in quoting rather than calculating bond or mortgage prices.
What problems does this cause?
Eq.4.8
2
We can plausibly estimate
order
2/14/10
, so that
K
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 31 of 38
0.02
---------- 0.0002
S K 100
where, as we saw earlier, the right hand side of the equation above is the typical order of the magnitude for the S&P 500 skew.
Then, the mismatch between the true and the Black-Scholes owing to the
skew, using Equation 4.8, is approximately
C BS
C
=
400 0.0002 = 0.08
S
S
For a not uncommon daily index move of 1%, or about 10 S&P 500 index
points, we can compare the mismatch in the P&L from using the wrong delta to
the incremental P&L expected in the Black-Scholes model with perfect hedging.
The mismatch in the P&L from being long or short 0.08 of a share when the
index moves 10 points is about 0.8 index points.
The incremental P&L from the curvature in a position in a perfectly hedged atthe-money one-year option in a Black-Scholes world with a not atypical volatility of 0.2 when the index moves 10 points is of order
2
1 S
1
S
-------- -------------- -------- --------- ( 50 ) = 0.25 points
2
2
200
S T
The mismatch in can cause a large distortion in the incremental P&L from
hedging at each step.
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 32 of 38
payoff
V(S)
(1/dK) call spreads
K
K + dK
One can approximately replicate V with a call spread, a long position in a call
with strike K and expiration T and a short position in a call with strike K + dK
and the same expiration. In the limit as dK 0 the call spreads payoff converges to that of the exotic option. (In practice, this is often how exotic traders
hedge themselves, choosing a small value of dK .)
The value of the call spread at stock price S and time to time t is
C BS ( S, K + dK, t, T, ( K ) ) + C BS ( K, S, t, T, ( K ) )
d
------------------------------------------------------------------------------------------------------------------------------ C BS ( S, K, t, T, ( K ) )
dK
dK
where C BS ( S, K, t, T, ( K ) ) is the market price of the call quoted in BlackScholes terms, with ( K ) incorporating the dependence of implied volatility
on K. The total derivative with respect to K includes the change of all variables
with K, including that of the implied volatility.
We can estimate the current value V ( S, K, t, T, ( K ) ) if we know how call
prices vary with strike K:
V ( S, K, t, T ) =
C BS C BS
d
C BS ( S, K, t, T, ( K ) ) =
K
K
dK
C BS
1
1
= N ( d 2 ) = N --- = --- ---------- --- = 0.46
2
K
2
2 2
C BS S T
---------- 400
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 33 of 38
The non-zero slope of the skew adds about 16% to the value of the option. This
is a significant difference.
Why does the skew add to the value of the derivative V? Because its a call
spread, and a negative skew means there is less volatility at higher strikes, and
therefore less risk-neutral probability of the stock price moving upwards;
therefore the second, higher leg of the call spread is worth less than when there
is no skew.
How can we fix it or extend Black-Scholes to match the skew and allow us
to calculate all these quantities correctly? What changes can we make? Or,
how, as we did in the above example, can we tread carefully and so avoid our
lack of knowledge about the right model and still get reasonable estimates of
value? Those are the questions we will tackle later in the course.
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 34 of 38
2/14/10
Expectation of changes in volatility as the market approaches certain significant levels can give rise to skew structure. For example, investors perception of support or resistance levels in currencies and interest rates
suggests that realized volatility will decrease as those levels are
approached.
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 35 of 38
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 36 of 38
Eq.4.9
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
S = AB
Page 37 of 38
assets - liabilities
dA
------- = dZ
A
dS
dA
AdZ
(S + B)
------ = ------- = -------------- = -----------------dZ
S
S
S
S
S = ( 1 + B S )
So, stock volatility increases as stock price decreases.
Constant Elasticity of Variance (CEV) models, developed by Cox and Ross
soon after the Black-Scholes model appeared, are the earliest local volatility
models:
dS = ( S, t )dt + S dZ
Here = 1 corresponds to the usual lognormal case, and = 0 to normal
evolution. You need to be negative and quite large in magnitude to account
for the magnitude of the negatively sloped equity index skew.
The conceptual difference between these parametric models and local volatility
models is that the CEV and leverage models are parametric models that have
only a few parameters and cannot fit an arbitrary smile, whereas local volatility
models are non-parametric, so that ( S, t ) must be numerically calibrated to
the observed smile at time t.
V =
E [ dWdZ ] = dt
If you are allowed to replicate options through dynamic trading only in the
stock and the bond markets, and volatility itself is stochastic, then options markets are not complete and perfect replication of the options payoff isnt possible. The principle of no riskless arbitrage will not lead to a unique price and
you need to know the market price of risk or a utility function for risk and
reward to price options; thats not a preference-free method, and less reliable
2/14/10
Lecture4.2010.fm
E4718 Spring 2010: Derman: Lecture 4:More on The Smile: Arbitrage Bounds, Valuation Problems, Models
Page 38 of 38
than either static or dynamic replication, and wed like to avoid it here. (Of
course, just because we prefer to avoid it from a theoretical point of view
doesnt mean that the market itself doesnt operate that way.)
If you can trade options, and if you know (or rather assume that you know) the
stochastic process for option prices or volatility as well as the stochastic process stock prices, then you can hedge one options exposure to volatility with
another option, you can derive an arbitrage-free formula for options values.
But assuming that you know all this is, as Rebonato says, a tall order.
An additional objection to stochastic volatility models is that they assume that
the correlation is constant. In fact, correlations are stochastic too, with perhaps a greater variance than volatility. Choosing constant may be too
extreme an assumption.
But, you must start somewhere if you want to get anywhere. So, model we
must. Later in the course well study stochastic volatility models.
2/14/10
Lecture4.2010.fm