This action might not be possible to undo. Are you sure you want to continue?
64 Wilmott magazine
Riaz Ahmad
Course Director for CQF, 7city, London
Paul Wilmott
Wilmott Associates, London
Which Free Lunch Would
You Like Today, Sir?: Delta
Hedging, Volatility Arbitrage
and Optimal Portfolios
1 Introduction
There are many thousands of papers on forecasting volatility using a host
of increasingly sophisticated, even NobelPrizewinning, statistical tech
niques. A possible goal of these is, presumably, to help one exploit mis
pricings in derivatives, and so profit from volatility arbitrage. There is a
similar order of magnitude of papers on various, and again, increasingly
complicated, volatility models calibrated to vanilla option prices and used
to price fancy exotic contracts. The former set of papers usually stop short
of the application of the volatility forecasts to options and how to actually
make that volatility arbitrage profit via delta hedging. And the second set
blind us with science without ever checking the accuracy of the volatility
models against data for the underlying. (A marvelous exception to this is
the paper by Schoutens, Simons & Tistaert (2004) which rather exposes the
Emperor for his lack of attire.) In this paper we are going to do none of this
clever stuff. No, we are going right back to basics.
In this paper we address the obvious question of how to make money
from volatility arbitrage. We are going to keep the model and analysis
very simple, hardly straying from the Black–Scholes world at all. We are
going to analyze different deltahedging strategies in a world of constant
volatility. Or more accurately, three constant volatilities: Implied, actual
and hedging.
Much of what we will examine is the profit to be made hedging
options that are mispriced by the market. This is the subject of how to
Abstract: In this paper we examine the statistical properties of the profit to be made from hedging vanilla options that are mispriced by the market and/or hedged using a
delta based on different volatilities. We derive formulas for the expected profit and the variance of profit for single options and for portfolios of options on the same underly
ing. We suggest several ways to choose optimal portfolios.
Wilmott magazine 65
delta hedge when your estimate of future actual volatility differs from
that of the market as measured by the implied volatility (Natenberg,
1994). Since there are two volatilities in this problem, implied and actual,
we have to study the effects of using each of these in the classical delta
formula (Black & Scholes, 1973). But why stop there? Why not use a
volatility between implied and actual, or higher than both or lower? We
will look at the profit or loss to be made hedging vanilla options and
portfolios of options with different ‘hedging volatilities.’
We will see how you can hedge using a delta based on actual volatility
or on implied volatility, or on something different. Whichever hedging
volatility you use you will get different risk/return profiles. Part of what
follows repeats the excellent work of Carr (2005) and Henrard (2003). Carr
derived the expression for profit from hedging using different volatili
ties. Henrard independently derived these results and also performed
simulations to examine the statistical properties of the possibly path
dependent profit. He also made important observations on portfolios of
options and on the role of the asset’s growth rate in determining the
profit. Our paper extends their analyses in several directions.
Other relevant literature in this area includes the paper by Carr &
Verma (2005) which expands on the problem of hedging using the
implied volatility but with implied volatility varying stochastically.
Dupire (2005) discusses the advantages of hedging using volatility based
on the realized quadratic variation in the stock price. Related ideas
applied to the hedging of barrier options and Asians can be found in
Forde (2005).
In Section 2 we set up the problem by explaining the role that volatil
ity plays in hedging. In Section 3 we look at the marktomarket profit
and the final profit when hedging using actual volatility. In Section 4 we
then examine the marktomarket and total profit made when hedging
using implied volatility. This profit is path dependent. Sections 3 and 4
repeat the analyses of Carr (2005) and Henrard (2003). Because the final
profit depends on the path taken by the asset when we hedge with
implied volatility we look at simple statistical properties of this profit. In
Section 4.1 we derive a closedform formula for the expected total profit
and in Section 4.2 we find a closedform formula for the variance of this
profit.
In Section 5 we look at hedging with volatilities other than just
implied or actual, examining the expected profit, the standard deviation
of profit as well as minimum and maximum profits. In Section 6 we look
at the advantages and disadvantages of hedging using different volatili
ties. For the remainder of the paper we focus on the case of hedging
using implied volatility, which is the more common market practice.
Portfolios of options are considered in Section 7, and again we find
closedform formulas for the expectation and variance of profit. To find
the full probability distribution of total profit we could perform simula
tions (Henrard, 2003) or solve a threedimensional differential equation.
We outline the latter approach in Section 8. This is to be preferred gener
ally since it will be faster than simulations, therefore making portfolio
optimizations more practical. In Section 9 we outline a portfolio selection
method based on exponential utility. In Section 10 we briefly mention
portfolios of options on many underlyings, and draw conclusions and
comment on further work in Section 11. Technical details are contained in
an appendix.
Some of the work in this paper has been used successfully by a volatil
ity arbitrage hedge fund.
Before we start, we present a simple plot of the distributions of
implied and realized volatilities, Figure 1. This is a plot of the distribu
tions of the logarithms of the VIX and of the rolling 30day realized SPX
volatility using data from 1990 to mid 2005. The VIX is an implied volatil
ity measure based on the SPX index and so you would expect it and the
realized SPX volatility to bear close resemblance. However, as can be seen
in the figure, the implied volatility VIX seems to be higher than the real
ized volatility. Both of these volatilities are approximately lognormally
distributed (since their logarithms appear to be Gaussian), especially the
realized volatility. The VIX distribution is somewhat truncated on the
left. The mean of the realized volatility, about 15%, is significantly lower
than the mean of the VIX, about 20%, but its standard deviation is
greater.
2 Implied versus Actual, Delta Hedging
but Using which Volatility?
Actual volatility is the amount of ‘noise’ in the stock price, it is the coef
ficient of the Wiener process in the stock returns model, it is the amount
^
TECHNICAL ARTICLE 3
Figure 1: Distributions of the logarithms of the VIX and the rolling realized
SPX volatility, and the normal distributions for comparison.
66 Wilmott magazine
of randomness that ‘actually’ transpires. Implied volatility is how the
market is pricing the option currently. Since the market does not have
perfect knowledge about the future these two numbers can and will be
different.
Imagine that we have a forecast for volatility over the remaining life
of an option, this volatility is forecast to be constant, and further assume
that our forecast turns out to be correct.
We shall buy an underpriced option and delta hedge to expiry. But
which delta do you choose? Delta based on actual or implied volatility?
Scenario: Implied volatility for an option is 20%, but we believe that
actual volatility is 30%. Question: How can we make money if our forecast
is correct? Answer: Buy the option and delta hedge. But which delta do
we use? We know that
= N(d
1
)
where
N(x) =
1
√
2π
x
−∞
e
−
s
2
2
ds
and
d
1
=
ln(S/E) +
r +
1
2
σ
2
(T − t)
σ
√
T − t
.
We can all agree on S, E, T − t and r (almost), but not on σ. So should we
use σ = 0.2 or 0.3?
In what follows we use σ to denote actual volatility and ˜ σ to represent
implied volatility, both assumed constant.
3 Case 1: Hedge with Actual Volatility, σ
By hedging with actual volatility we are replicating a short position in a
correctly priced option. The payoffs for our long option and our short repli
cated option will exactly cancel. The profit we make will be exactly the
difference in the Black–Scholes prices of an option with 30% volatility
and one with 20% volatility. (Assuming that the Black–Scholes assump
tions hold.) If V(S, t; σ) is the Black–Scholes formula then the guaranteed
profit is
V(S, t; σ) − V(S, t; ˜ σ).
But how is this guaranteed profit realized? Let us do the analysis on a
marktomarket basis.
In the following, superscript ‘a’ means actual and ‘i’ denotes implied,
these can be applied to deltas and option values. For example,
a
is the
delta using the actual volatility in the formula. V
i
is the theoretical
option value using the implied volatility in the formula. Note also that V,
, and are all simple, known, Black–Scholes formulas.
The model is the classical
dS = µS dt + σS dX.
Now, set up a portfolio by buying the option for V
i
and hedge with
a
of the stock. The values of each of the components of our portfolio are
shown in the following table, Table 1.
Leave this hedged portfolio ‘overnight,’ and come back to it the next
‘day.’ The new values are shown in Table 2. (We have included a continu
ous dividend yield in this.)
Therefore we have made, mark to market,
dV
i
−
a
dS − r(V
i
−
a
S) dt −
a
DS dt.
Because the option would be correctly valued at V
a
we have
dV
a
−
a
dS − r(V
a
−
a
S) dt −
a
DS dt = 0.
So we can write the marktomarket profit over one time step as
dV
i
− dV
a
+ r(V
a
−
a
S) dt − r(V
i
−
a
S) dt
= dV
i
− dV
a
− r(V
i
− V
a
) dt = e
rt
d
e
−rt
(V
i
− V
a
)
.
That is the profit from time t to t + dt. The present value of this profit at
time t
0
is
e
−r(t−t0 )
e
rt
d
e
−rt
(V
i
− V
a
)
= e
rt0
d
e
−rt
(V
i
− V
a
)
.
So the total profit from t
0
to expiration is
e
rt0
T
t0
d
e
−rt
(V
i
− V
a
)
= V
a
− V
i
.
This confirms what we said earlier about the guaranteed total profit by
expiration.
Component Value
Option V
i
Stock −
a
S
Cash −V
i
+
a
S
TABLE 1: PORTFOLIO
COMPOSITION AND
VALUES, TODAY.
Component Value
Option V
i
+ dV
i
Stock −
a
S −
a
dS
Cash (−V
i
+
a
S)(1 + r dt) −
a
DS dt
TABLE 2: PORTFOLIO COMPOSITION
AND VALUES, TOMORROW.
^
Wilmott magazine 67
We can also write that one time step marktomarket profit (using
Itô’s lemma) as
i
dt +
i
dS +
1
2
σ
2
S
2
i
dt −
a
dS − r(V
i
−
a
S) dt −
a
DS dt
=
i
dt + µS(
i
−
a
) dt +
1
2
σ
2
S
2
i
dt − r(V
i
− V
a
) dt
+ (
i
−
a
)σS dX −
a
DS dt
= (
i
−
a
)σS dX + (µ + D)S(
i
−
a
) dt +
1
2
σ
2
− ˜ σ
2
S
2
i
dt
(using Black–Scholes with σ = ˜ σ)
=
1
2
(σ
2
− ˜ σ
2
)S
2
i
dt + (
i
−
a
)((µ − r + D)S dt + σS dX).
The conclusion is that the final profit is guaranteed (the difference
between the theoretical option values with the two volatilities) but how
that is achieved is random, because of the dX term in the above. On a
marktomarket basis you could lose before you gain. Moreover, the mark
tomarket profit depends on the real drift of the stock, µ. This is illustrat
ed in Figure 2. The figure shows several simulations of the same delta
hedged position. Note that the final P&L is not exactly the same in each
case because of the effect of hedging discretely, we hedged ‘only’ 1000
times for each realization. The option is a oneyear European call, with a
strike of 100, at the money initially, actual volatility is 30%, implied is
20%, the growth rate is 10% and interest rate 5%.
When S changes, so will V. But these changes do not cancel each other
out. This leaves us with a dX in our marktomarket P&L and from a risk
management point of view this is not ideal.
There is a simple analogy for this behavior. It is similar to owning a
bond. For a bond there is a guaranteed outcome, but we may lose on a
marktomarket basis in the meantime.
4 Case 2: Hedge with Implied Volatility, σ
∼
Compare and contrast now with the case of hedging using a delta based
on implied volatility. By hedging with implied volatility we are balancing
the random fluctuations in the marktomarket option value with the
fluctuations in the stock price. The evolution of the portfolio value is
then ‘deterministic’ as we shall see.
Buy the option today, hedge using the implied delta, and put any cash
in the bank earning r. The marktomarket profit from today to tomorrow is
dV
i
−
i
dS − r(V
i
−
i
S) dt −
i
DS dt
=
i
dt +
1
2
σ
2
S
2
i
dt − r(V
i
−
i
S) dt −
i
DS dt
=
1
2
σ
2
− ˜ σ
2
S
2
i
dt.
(1)
Observe how the profit is deterministic, there are no dX terms. From a
risk management perspective this is much better behaved. There is
another advantage of hedging using implied volatility, we do not even
need to know what actual volatility is. To make a profit all we need to
know is that actual is always going to be greater than implied (if we are
buying) or always less (if we are selling). This takes some of the pressure
off forecasting volatility accurately in the first place.
Integrate the present value of all of these profits over the life of the
option to get a total profit of
1
2
σ
2
− ˜ σ
2
T
t0
e
−r(t−t0 )
S
2
i
dt.
This is always positive, but highly path dependent. Being path
dependent it will depend on the drift µ. If we start off at the money
and the drift is very large (positive or negative) we will find our
selves quickly moving into territory where gamma and hence
expression (1) is small, so that there will be not much profit to be
made. The best that could happen would be for the stock to end up
close to the strike at expiration, this would maximize the total prof
it. This path dependency is shown in Figure 3. The figure shows sev
eral realizations of the same deltahedged position. Note that the
lines are not perfectly smooth, again because of the effect of hedg
ing discretely. The option and parameters are the same as in the
previous example.
The simple analogy is now just putting money in the bank. The
P&L is always increasing in value but the end result is random.
Carr (2005) and Henrard (2003) show the more general result
that if you hedge using a delta based on a volatility σ
h
then the PV
of the total profit is given by
V(S, t; σ
h
) − V(S, t; ˜ σ) +
1
2
σ
2
− σ
2
h
T
t0
e
−r(t−t0 )
S
2
h
dt, (2)
TECHNICAL ARTICLE 3
Figure 2: P&L for a deltahedged option on a marktomarket basis, hedged using
actual volatility.
68 Wilmott magazine
where the superscript on the gamma means that it uses the Black–
Scholes formula with a volatility of σ
h
.
From this equation it is easy to see that the final profit is bounded by
V(S, t; σ
h
) − V(S, t; ˜ σ)
and
V(S, t; σ
h
) − V(S, t; ˜ σ) +
E
σ
2
− σ
2
h
e
−r(T−t0 )
√
T − t
0
σ
h
√
2π
.
The righthand side of the latter expression above comes from maximiz
ing S
2
h
. This maximum occurs along the path ln(S/E) + (r − D − σ
2
h
/2)
(T − t) = 0, that is
S = E exp
−
r − D − σ
2
h
/2
(T − t)
.
4.1 The expected profit after hedging using implied
volatility
When you hedge using delta based on implied volatility the profit each
‘day’ is deterministic but the present value of total profit by expiration is
path dependent, and given by
1
2
σ
2
− ˜ σ
2
T
t0
e
−r(s−t0 )
S
2
i
ds.
Introduce
I =
1
2
σ
2
− ˜ σ
2
t
t0
e
−r(s−t0 )
S
2
i
ds.
Since therefore
dI =
1
2
σ
2
− ˜ σ
2
e
−r(t−t0 )
S
2
i
dt
we can write down the following partial differential equation for
the real expected value, P(S, I, t), of I
∂P
∂t
+
1
2
σ
2
S
2
∂
2
P
∂S
2
+ µS
∂P
∂S
+
1
2
σ
2
− ˜ σ
2
e
−r(t−t0 )
S
2
i
∂P
∂I
= 0,
with
P(S, I, T) = I.
Look for a solution of this equation of the form
P(S, I, t) = I + F(S, t)
so that
∂F
∂t
+
1
2
σ
2
S
2
∂
2
F
∂S
2
+ µS
∂F
∂S
+
1
2
σ
2
− ˜ σ
2
e
−r(t−t0 )
S
2
i
= 0.
The source term can be simplified to
E
σ
2
− ˜ σ
2
e
−r(T−t0 )
e
−d
2
2
/2
2 ˜ σ
√
2π(T − t)
,
where
d
2
=
ln(S/E) + (r − D −
1
2
˜ σ
2
)(T − t)
σ
√
T − t
.
Change variables to
x = ln(S/E) +
2
σ
2
µ −
1
2
σ
2
τ and τ =
σ
2
2
(T − t),
where E is the strike and T is expiration, and write
F(S, t) = w(x, τ ).
The resulting partial differential equation is a then simpler.
Result 1: After some manipulations we end up with the expected profit
initially (t = t
0
, S = S
0
, I = 0) being the single integral
Figure 3: P&L for a deltahedged option on a marktomarket basis, hedged using
implied volatility.
^
Wilmott magazine 69
F(S
0
, t
0
) =
Ee
−r(T−t0 )
(σ
2
− ˜ σ
2
)
2
√
2π
T
t0
1
σ
2
(s − t
0
) + ˜ σ
2
(T − s)
exp
−
ln(S
0
/E) +
µ −
1
2
σ
2
(s − t
0
) +
r − D −
1
2
˜ σ
2
(T − s)
2
2(σ
2
(s − t
0
) + ˜ σ
2
(T − s))
ds.
Derivation: See Appendix.
Results are shown in the following figures.
In Figure 4 is shown the expected profit versus the growth rate µ.
Parameters are S = 100, σ = 0.3, r = 0.05, D = 0, E = 110, T = 1, ˜ σ = 0.2.
Observe that the expected profit has a maximum. This will be at the
growth rate that ensures, roughly speaking, that the stock ends up close
to at the money at expiration, where gamma is largest. In the figure is
also shown the profit to be made when hedging with actual volatility.
For most realistic parameter regimes the maximum expected profit
hedging with implied is similar to the guaranteed profit hedging with
actual.
In Figure 5 is shown expected profit versus E and µ. You can see how
the higher the growth rate the larger the strike price at the maximum.
The contour map is shown in Figure 6.
The effect of skew is shown in Figure 7. Here we have used a linear
negative skew, from 22.5% at a strike of 75, falling to 17.5% at the 125
strike. The atthemoney implied volatility is 20% which in this case is the
actual volatility. This picture changes when you divide the expected prof
it by the price of the option (puts for lower strikes, call for higher), see
Figure 8. There is no maximum, profitability increases with distance
away from the money. Of course, this does not take into account the risk,
the standard deviation associated with such trades.
TECHNICAL ARTICLE 3
Figure 4: Expected profit, hedging using implied volatility, versus growth rate µ;
S = 100, σ = 0.4, r = 0.05, D = 0, E = 110, T = 1, ~σ = 0.2. The dashed line is the prof
it to be made when hedging with actual volatility.
Figure 5: Expected profit, hedging using implied volatility,
versus growth rate µ and strike E; S = 100, σ = 0.4, r = 0.05,
D = 0, T = 1, ~σ = 0.2.
Figure 6: Contour map of expected profit, hedging using implied
volatility, versus growth rate µ and strike E; S = 100, σ = 0.4, r = 0.05,
D = 0, T = 1, ~σ = 0.2.
70 Wilmott magazine
4.2 The variance of profit after hedging using implied
volatility
Once we have calculated the expected profit from hedging using implied
volatility we can calculate the variance in the final profit. Using the above
notation, the variance will be the expected value of I
2
less the square of
the average of I. So we will need to calculate v(S, I, t) where
∂v
∂t
+
1
2
σ
2
S
2
∂
2
v
∂S
2
+ µS
∂v
∂S
+
1
2
σ
2
− ˜ σ
2
e
−r(t−t0 )
S
2
i
∂v
∂I
= 0,
with
v(S, I, T) = I
2
.
The details of finding this function v are rather messy, but a solution can
be found of the form
v(S, I, t) = I
2
+ 2I H(S, t) + G(S, t).
Result 2: The initial variance is G(S
0
, t
0
) − F(S
0
, t
0
)
2
, where
G(S
0
, t
0
) =
E
2
(σ
2
− ˜ σ
2
)
2
e
−2r(T−t0 )
4πσ ˜ σ
T
t0
T
s
e
p(u,s;S0 ,t0 )
√
s − t
0
√
T − s
σ
2
(u − s) + ˜ σ
2
(T − u)
1
σ
2
(s − t
0
)
+
1
˜ σ
2
(T − s)
+
1
σ
2
(u − s) + ˜ σ
2
(T − u)
du ds
(3)
where
p(u, s; S
0
, t
0
) = −
1
2
(x + α(T − s))
2
˜ σ
2
(T − s)
−
1
2
(x + α(T − u))
2
σ
2
(u − s) + ˜ σ
2
(T − u)
+
1
2
x + α(T − s)
˜ σ
2
(T − s)
+
x + α(T − u)
σ
2
(u − s) + ˜ σ
2
(T − u)
2
1
σ
2
(s − t
0
)
+
1
˜ σ
2
(T − s)
+
1
σ
2
(u − s) + ˜ σ
2
(T − u)
and
x = ln(S
0
/E) +
µ −
1
2
σ
2
(T − t
0
), and α = µ −
1
2
σ
2
− r + D +
1
2
˜ σ
2
.
Derivation: See Appendix.
In Figure 9 is shown the standard deviation of profit versus growth
rate, S = 100, σ = 0.4, r = 0.05, D = 0, E = 110, T = 1, ˜ σ = 0.2. Figure 10
shows the standard deviation of profit versus strike, S = 100, σ = 0.4,
r = 0.05, D = 0, µ = 0.1, T = 1, ˜ σ = 0.2.
Note that in these plots the expectations and standard deviations
have not been scaled with the cost of the options.
In Figure 11 are shown expected profit divided by cost versus standard
deviation divided by cost, as both strike and expiration vary. In these
S = 100, σ = 0.4, r = 0.05, D = 0, µ = 0.1, ˜ σ = 0.2. To some extent,
although we emphasize only some, these diagrams can be interpreted in a
classical meanvariance manner. The main criticism is, of course, that we
are not working with normal distributions, and, furthermore, there is no
downside, no possibility of any losses.
Figure 12 completes the earlier picture for the skew, since it now con
tains the standard deviation.
Figure 7: Effect of skew, expected profit, hedging using implied volatility,
versus strike E; S = 100, µ = 0, σ = 0.2, r = 0.05, D = 0, T = 1.
Figure 8: Effect of skew, ratio of expected profit to price, hedging using
implied volatility, versus strike E; S = 100, µ = 0, σ = 0.2, r = 0.05, D = 0, T = 1.
^
Wilmott magazine 71
TECHNICAL ARTICLE 3
Figure 9: Standard deviation of profit, hedging using implied volatility,
versus growth rate µ; S = 100, σ = 0.4, r = 0.05, D = 0, E = 110, T = 1, ~σ =
0.2. The expected profit is also shown.
Figure 10: Standard deviation of profit, hedging using implied volatility, ver
sus strike E; S = 100, σ = 0.4, r = 0.05, D = 0, µ = 0, T = 1, ~σ = 0.2. The expected
profit is also shown.
Figure 11: Scaled expected profit versus scaled standard deviation; S = 100,
σ = 0.4, r = 0.05, D = 0, µ = 0.1, ~σ = 0.2. Four different expirations, varying strike.
Figure 12: Effect of skew, ratio of expected profit to price, and ratio of
standard deviation to price, versus strike E; S = 100, µ = 0, σ = 0.2, r = 0.05,
D = 0, T = 1.
72 Wilmott magazine
5 Hedging with Different Volatilities
We will briefly examine hedging using volatilities other than actual or
implied, using the general expression for profit given by (2).
The expressions for the expected profit and standard deviations now
must allow for the V(S, t; σ
h
) − V(S, t; ˜ σ), since the integral of gamma
term can be treated as before if one replaces ˜ σ with σ
h
in this term.
Results are presented in the next sections.
5.1 Actual volatility = Implied volatility
For the first example let’s look at hedging a long position in a correctly
priced option, so that σ = ˜ σ. We will hedge using different volatilities,
σ
h
. Results are shown in Figure 13. The figure shows the expected profit
and standard deviation of profit when hedging with various volatilities.
The chart also shows minimum and maximum profit. Parameters are
E = 100, S = 100, µ = 0, σ = 0.2, r = 0.1, D = 0, T = 1, and ˜ σ = 0.2.
With these parameters the expected profit is small as a fraction of the
market price of the option ($13.3) regardless of the hedging volatility. The
standard deviation of profit is zero when the option is hedged at the
actual volatility. The upside, the maximum profit is much greater than
the downside. Crucially all of the curves have zero value at the
actual/implied volatility.
5.2 Actual volatility > Implied volatility
In Figure 14 is shown the expected profit and standard deviation of profit
when hedging with various volatilities when actual volatility is greater
than implied. The chart again also shows minimum and maximum profit.
Parameters are E = 100, S = 100, µ = 0, σ = 0.4, r = 0.1, D = 0, T = 1,
and ˜ σ = 0.2. Note that it is possible to lose money if you hedge at below
implied, but hedging with a higher volatility you will not be able to lose
until hedging with a volatility of approximately 75%. The expected profit
is again insensitive to hedging volatility.
5.3 Actual volatility < Implied volatility
In Figure 15 is shown properties of the profit when hedging with various
volatilities when actual volatility is less than implied. We are now selling
the option and delta hedging it. Parameters are E = 100, S = 100, µ = 0,
σ = 0.4, r = 0.1, D = 0, T = 1, and ˜ σ = 0.2. Now it is possible to lose
money if you hedge at above implied, but hedging with a lower volatility
you will not be able to lose until hedging with a volatility of approxi
mately 10%. The expected profit is again insensitive to hedging volatility.
The downside is now more dramatic than the upside.
6 Pros and Cons of Hedging with each
Volatility
Given that we seem to have a choice in how to delta hedge it is instructive
to summarize the advantages and disadvantages of the possibilities.
6.1 Hedging with actual volatility
Pros: The main advantage of hedging with actual volatility is that you
know exactly what profit you will get at expiration. So in a classical
risk/reward sense this seems to be the best choice, given that the expected
Figure 13: Expected profit, standard deviation of profit, minimum and
maximum, hedging with various volatilities. E = 100, S = 100, µ = 0, σ = 0.2,
r = 0.1, D = 0, T = 1, ~σ = 0.2.
Figure 14: Expected profit, standard deviation of profit, minimum and
maximum, hedging with various volatilities. E = 100, S = 100, µ = 0, σ = 0.4,
r = 0.1, D = 0, T = 1, ~σ = 0.2.
^
Wilmott magazine 73
profit can often be insensitive to which volatility you choose to hedge
with whereas the standard deviation is always going to be positive away
from hedging with actual volatility.
Cons: The P&L fluctuations during the life of the option can be daunt
ing, and so less appealing from a ‘local’ as opposed to ‘global’ risk man
agement perspective. Also, you are unlikely to be totally confident in
your volatility forecast, the number you are putting into your delta for
mula. However, you can interpret the previous two figures in terms of
what happens if you intend to hedge with actual but don’t quite get it
right. You can see from those that you do have quite a lot of leeway before
you risk losing money.
6.2 Hedging with implied volatility
Pros: There are three main advantages to hedging with implied volatility.
The first is that there are no local fluctuations in P&L, you are continual
ly making a profit. The second advantage is that you only need to be on
the right side of the trade to profit. Buy when actual is going to be high
er than implied and sell if lower. Finally, the number that goes into the
delta is implied volatility, and therefore easy to observe.
Cons: You don’t know how much money you will make, only that it is
positive.
6.3 Hedging with another volatility
You can obviously balance the pros and cons of hedging with actual and
implied by hedging with another volatility altogether. See Dupire (2005)
for work in this area.
In practice which volatility one uses is often determined by whether
one is constrained to mark to market or mark to model. If one is able to
mark to model then one is not necessarily concerned with the daytoday
fluctuations in the marktomarket profit and loss and so it is natural to
hedge using actual volatility. This is usually not far from optimal in the
sense of possible expected total profit, and it has no standard deviation of
final profit. However, it is common to have to report profit and loss based
on market values. This constraint may be imposed by a risk management
department, by prime brokers, or by investors who may monitor the
marktomarket profit on a regular basis. In this case it is more usual to
hedge based on implied volatility to avoid the daily fluctuations in the
profit and loss.
We can begin to quantify the ‘local’ risk, the daily fluctuations in
P&L, by looking at the random component in a portfolio hedged using a
volatility of σ
h
. The standard deviation of this risk is
σS 
i
−
h

√
dt. (4)
Note that this expression depends on all three volatilities.
Figure 16 shows the two deltas (for a call option), one using implied
volatility and the other the hedging volatility, six months before expira
tion. If the stock is far in or out of the money the two deltas are similar
and so the local risk is small. The local risk is also small where the two
deltas cross over. This ‘sweet spot’ is at
ln(S/E) + (r − D + ˜ σ
2
/2)(T − t)
˜ σ
√
T − t
=
ln(S/E) + (r − D + σ
h
2
/2)(T − t)
σ
h
√
T − t
,
TECHNICAL ARTICLE 3
Figure 15: Expected profit, standard deviation of profit, minimum and
maximum, hedging with various volatilities. E = 100, S = 100, µ = 0, σ = 0.2,
r = 0.1, D = 0, T = 1, ~σ = 0.4.
Figure 16: Deltas based on implied volatility and hedging volatility.
E = 100, S = 100, r = 0.1, D = 0, T = 0.5, ~σ = 0.2, σ
h
= 0.3.
74 Wilmott magazine
that is,
S = E exp
−
T − t
˜ σ − σ
h
˜ σ(r − D + σ
h
2
/2) − σ
h
(r − D + ˜ σ
2
/2)
.
Figure 17 shows a threedimensional plot of expression (4), without
the
√
dt factor, as a function of stock price and time. Figure 18 is a con
tour map of the same. Parameters are E = 100, S = 100, σ = 0.4, r = 0.1,
D = 0, T = 1, ˜ σ = 0.2, σ
h
= 0.3.
In the spirit of the earlier analyses and formulas we would ideally like
to be able to quantify various statistical properties of the local markto
market fluctuations. This will be the subject of future work.
For the remainder of this paper we will only consider the case of
hedging using a delta based on implied volatility, although the ideas can
be easily extended to the more general case.
7 Portfolios when Hedging with Implied
Volatility
A natural extension to the above analysis is to look at portfolios of
options, options with different strikes and expirations. Since only an
option’s gamma matters when we are hedging using implied volatility,
calls and puts are effectively the same since they have the same gamma.
The profit from a portfolio is now
1
2
¸
k
q
k
σ
2
− ˜ σ
2
k
Tk
t0
e
−r(s−t0 )
S
2
i
k
ds,
where k is the index for an option, and q
k
is the quantity of that option.
Introduce
I =
1
2
¸
k
q
k
σ
2
− ˜ σ
2
k
t
t0
e
−r(s−t0 )
S
2
i
k
ds, (5)
as a new state variable, and the analysis can proceed as before. Note that
since there may be more than one expiration date since we have several
different options, it must be understood in Equation (5) that
i
k
is zero for
times beyond the expiration of the option.
The governing differential operator for expectation, variance, etc. is
then
∂
∂t
+
1
2
σ
2
S
2
∂
2
∂S
2
+ µS
∂
∂S
+
1
2
¸
k
σ
2
− ˜ σ
2
k
e
−r(t−t0 )
S
2
i
k
∂
∂I
= 0,
with final condition representing expectation, variance, etc.
7.1 Expectation
Result 3: The solution for the present value of the expected profit
(t = t
0
, S = S
0
, I = 0) is simply the sum of individual profits for each
option,
F(S
0
, t
0
) =
¸
k
q
k
E
k
e
−r(Tk −t0 )
(σ
2
− ˜ σ
2
k
)
2
√
2π
Tk
t0
1
σ
2
(s − t
0
) + ˜ σ
2
k
(T
k
− s)
exp
−
ln(S
0
/E
k
) +
µ −
1
2
σ
2
(s − t
0
) +
r − D −
1
2
˜ σ
2
k
(T
k
− s)
2
2(σ
2
(s − t
0
) + ˜ σ
2
k
(T
k
− s))
ds.
Derivation: See Appendix.
Figure 17: Local risk as a function of stock price and
time to expiration. E = 100, S = 100, σ = 0.4, r = 0.1, D = 0,
T = 1, ~σ = 0.2, σ
h
= 0.3.
Figure 18: Contour map of local risk as a function of stock price and time to
expiration. E = 100, S = 100, σ = 0.4, r = 0.1, D = 0, T = 1, ~σ = 0.2, σ
h
= 0.3.
^
Wilmott magazine 75
7.2 Variance
Result 4: The variance is more complicated, obviously, because of the cor
relation between all of the options in the portfolio. Nevertheless, we can
find an expression for the initial variance as G(S
0
, t
0
) − F(S
0
, t
0
)
2
where
G(S
0
, t
0
) =
¸
j
¸
k
q
j
q
k
G
jk
(S
0
, t
0
)
where
G
jk
(S
0
, t
0
) =
E
j
E
k
(σ
2
− ˜ σ
2
j
)(σ
2
− ˜ σ
2
k
)e
−r(Tj −t0 )−r(Tk −t0 )
4πσ ˜ σ
k
min(Tj ,Tk )
t0
Tj
s
e
p(u,s;S0 ,t0 )
√
s − t
0
√
T
k
− s
σ
2
(u − s) + ˜ σ
2
j
(T
j
− u)
1
σ
2
(s − t
0
)
+
1
˜ σ
2
k
(T
k
− s)
+
1
σ
2
(u − s) + ˜ σ
2
j
(T
j
− u)
du ds (6)
where
p(u, s; S
0
, t
0
) = −
1
2
(ln(S
0
/E
k
) + ¯ µ(s − t
0
) + ¯ r
k
(T
k
− s))
2
˜ σ
2
k
(T
k
− s)
−
1
2
(ln(S
0
/E
j
) + ¯ µ(u − t
0
) + ¯ r
j
(T
j
− u))
2
σ
2
(u − s) + ˜ σ
2
j
(T
j
− u)
+
1
2
ln(S
0
/E
k
) + ¯ µ(s − t
0
) + ¯ r
k
(T
k
− s)
˜ σ
2
k
(T
k
− s)
+
ln(S
0
/E
j
) + ¯ µ(u − t
0
) + ¯ r
j
(T
j
− u)
σ
2
(u − s) + ˜ σ
2
j
(T
j
− u)
2
1
σ
2
(s − t
0
)
+
1
˜ σ
2
k
(T
k
− s)
+
1
σ
2
(u − s) + ˜ σ
2
j
(T
j
− u)
and
¯ µ = µ −
1
2
σ
2
, ¯ r
j
= r − D −
1
2
˜ σ
2
j
and ¯ r
k
= r − D −
1
2
˜ σ
2
k
.
Derivation: See Appendix.
7.3 Portfolio optimization possibilities
There is clearly plenty of scope for using the above formulas in portfolio
optimization problems. Here we give one example.
The stock is currently at 100. The growth rate is zero, actual volatility
is 20%, zero dividend yield and the interest rate is 5%. Table 3 shows the
available options, and associated parameters. Observe the negative skew.
The outofthemoney puts are overvalued and the outofthemoney calls
are undervalued. (The ‘Profit Total Expected’ row assumes that we buy a
single one of that option.)
Using the above formulas we can find the portfolio that maximizes or
minimizes target quantities (expected profit, standard deviation, ratio of
profit to standard deviation). Let us consider the simple case of maximiz
ing the expected return, while constraining the standard deviation to be
one. This is a very natural strategy when trying to make a profit from
volatility arbitrage while meeting constraints imposed by regulators, bro
kers, investors etc. The result is given in Table 4.
The payoff function (with its initial delta hedge) is shown in Figure 19.
This optimization has effectively found an ideal risk reversal trade. This
portfolio would cost −$0.46 to set up, i.e. it would bring in premium. The
expected profit is $6.83.
TECHNICAL ARTICLE 3
A B C D E
Type Put Put Call Call Call
Strike 80 90 100 110 120
Expiration 1 1 1 1 1
Volatility, Implied 0.250 0.225 0.200 0.175 0.150
Option Price, Market 1.511 3.012 10.451 5.054 1.660
Option Value, Theory 0.687 2.310 10.451 6.040 3.247
Profit Total Expected −0.933 −0.752 0.000 0.936 1.410
TABLE 3: AVAILABLE OPTIONS.
A B C D E
Type Put Put Call Call Call
Strike 80 90 100 110 120
Quantity −2.10 −2.25 0 1.46 1.28
TABLE 4: AN OPTIMAL PORTFOLIO.
Figure 19: Payoff with initial delta hedge for optimal portfolio; S = 100,
µ = 0, σ = 0.2, r = 0.05, D = 0, T = 1. See text for additional parameters
and information.
Payoff function (with initial delta hedge)
−30
−25
−20
−15
−10
−5
0
5
10
15
60 70 80 90 100 110 120 130 140
Asset
76 Wilmott magazine
Because the state variable representing the profit, I, is not normally
distributed a portfolio analysis based on mean and variance is open to
criticism. So now we shall look at other ways of choosing or valuing a
portfolio.
8 Other Optimization Strategies
Rather than choose an option or a portfolio based on mean and variance
it might be preferable to examine the probability density function for I.
The main reason for this is the observation that I is not normally distrib
uted. Mathematically the problem for the cumulative distribution func
tion for the final profit I
can be written as C(S
0
, 0, t
0
; I
) where C(S, I, t; I
)
is the solution of
∂C
∂t
+
1
2
σ
2
S
2
∂
2
C
∂S
2
+ µS
∂C
∂S
+
1
2
¸
k
σ
2
− ˜ σ
2
k
e
−r(t−t0 )
S
2
i
k
∂C
∂I
= 0,
subject to the final condition
C(S, I, T
max
; I
) = H(I
− I),
where T
max
is the expiration of the longest maturity option and H(·) is the
Heaviside function. The same equation, with suitable final conditions,
can be used to choose or optimize a portfolio of options based on criteria
such as the following.
• Maximize probability of making a profit greater than a specified
amount or, equivalently, minimize the probability of making less
than a specified amount
• Maximize profit at a certain probability threshold, such as 95% (a
ValueatRisk type of optimization, albeit one with no possibility of a
loss)
Constraints would typically need to be imposed on these optimization
problems, such as having a set budget and/or a limit on number of posi
tions that can be held.
In the spirit of maximizing expected growth rate (Kelly Criterion) we
could also examine solutions of the above threedimensional partial differ
ential equation having final condition being a logarithmic function of I.
9 Exponential Utility Approach
Rather than relying on means and variances, which could be criticized
because we are not working with a Gaussian distribution for I, or solving
a differential equation in three dimensions, which may be slow, there is
another possibility, and one that has neither of these disadvantages. This
is to work within a utility theory framework, in particular using con
stant absolute risk aversion with utility function
−
1
η
e
−ηI
.
The parameter η is then a person’s absolute risk aversion.
The governing equation for the expected utility, U, is then
∂U
∂t
+
1
2
σ
2
S
2
∂
2
U
∂S
2
+ µS
∂U
∂S
+
1
2
¸
k
σ
2
− ˜ σ
2
k
e
−r(t−t0 )
S
2
i
k
∂U
∂I
= 0,
with final condition
U(S, I, T
max
) = −
1
η
e
−ηI
.
where T
max
is the expiration of the longest maturity option.
We can look for a solution of the form
U(S, I, t) = −
1
η
e
−ηI
Q (S, t),
so that
∂Q
∂t
+
1
2
σ
2
S
2
∂
2
Q
∂S
2
+ µS
∂Q
∂S
−
ηQ
2
¸
k
σ
2
− ˜ σ
2
k
e
−r(t−t0 )
S
2
i
k
= 0,
with final condition
Q (S, T
max
) = 1.
Being only a twodimensional equation this will be very quick to solve
numerically. One can then pose and solve various optimal portfolio prob
lems. We shall not pursue this in this paper.
10 Many Underlyings and Portfolios
of Options
Although methodologies based on formulas and/or partial differential
equations are the most efficient when the portfolio only has a small
number of underlyings, one must use simulation techniques when there
are many underlyings in the portfolio.
Because we know the instantaneous profit at each time, for a given
stock price, via simple option pricing formulas for each option’s gamma,
we can very easily simulate the P&L for an arbitrary portfolio. All that is
required is a simulation of the real paths for each of the underlyings.
10.1 Dynamics linked via drift rates
Although option prices are independent of real drift rates, only depend
ing on riskneutral rates, i.e. the riskfree interest adjusted for dividends,
the profit from a hedged mispriced option is not. As we have seen above
the profit depends crucially on the growth rate because of the path
dependence. As already mentioned, ideally we would like to see the stock
following a path along which gamma is large since this gives us the high
est profit. When we have many underlyings we really ought to model the
drift rates of all the stocks as accurately as possible, something that is not
usually considered when simply valuing options in complete markets. In
the above example we assumed constant growth rates for each stock, but
^
Wilmott magazine 77
in reality there may be more interesting, interacting dynamics at work.
Traditionally, in complete markets the sole interaction between stocks
that need concern us is via correlations in the dX
i
terms, that is, a rela
tionship at the infinitesimal timescale. In reality, and in the context of
the present work, there will also be longer timescale relationships oper
ating, that is, a coupling in the growth rates. This can be represented by
dS
i
= µ
i
(S
1
, . . . , S
n
) dt + σ
i
S
i
dX
i
.
11 Conclusions and Further Work
This paper has expanded on the work of Carr and Henrard in terms of
final formulas for the statistical properties of the profit to be made hedg
ing mispriced options. We have also indicated how more sophisticated
portfolio construction techniques can be applied to this problem rela
tively straightforwardly. We have concentrated on the case of hedging
using deltas based on implied volatilities because this is the most com
mon in practice, giving marktomarket profit the smoothest behavior.
Appendix: Derivation of Results
Preliminary results
In the following derivations we often require the following simple
results.
First,
∞
−∞
e
−ax
2
dx =
π
a
. (7)
Second, the solution of
∂w
∂τ
=
∂
2
w
∂x
2
+ f (x, τ )
that is initially zero and is zero at plus and minus infinity is
1
2
√
π
∞
−∞
τ
0
f (x
, τ
)
√
τ − τ
e
−(x−x
)
2
/4(τ −τ
)
dτ
dx
. (8)
Finally, the transformations
x = ln(S/E) +
2
σ
2
µ −
1
2
σ
2
τ and τ =
σ
2
2
(T − t),
turn the operator
∂
∂t
+
1
2
σ
2
∂
2
∂S
2
+ µS
∂
∂S
into
1
2
σ
2
−
∂
∂τ
+
∂
2
∂x
2
. (9)
Result 1: Expectation, single option
The equation to be solved for F(S, t) is
∂F
∂t
+
1
2
σ
2
S
2
∂
2
F
∂S
2
+ µS
∂F
∂S
+
1
2
σ
2
− ˜ σ
2
e
−r(t−t0 )
S
2
i
= 0,
with zero final and boundary conditions. Using the above changes of
variables this becomes F(S, t) = w(x, τ ) where
∂w
∂τ
=
∂
2
w
∂x
2
+
E
σ
2
− ˜ σ
2
e
−r(T−t0 )
e
−d
2
2
/2
σ ˜ σ
√
πτ
where
d
2
=
σ
˜ σ
x −
2
σ
2
µ −
1
2
σ
2
τ +
2
σ
2
r − D −
1
2
˜ σ
2
τ
√
2τ
.
The solution of this problem is, using (8),
1
2π
E
σ
2
− ˜ σ
2
e
−r(T−t0 )
σ ˜ σ
∞
−∞
τ
0
1
√
τ
1
√
τ − τ
exp
−
(x − x
)
2
4(τ − τ
)
−
σ
2
4˜ σ
2
τ
x
−
2
σ
2
µ −
1
2
σ
2
τ
+
2
σ
2
r − D −
1
2
˜ σ
2
τ
2
dτ
dx
.
If we write the argument of the exponential function as
−a(x
+ b)
2
+ c
we have the solution
1
2π
E
σ
2
− ˜ σ
2
e
−r(T−t0 )
σ ˜ σ
τ
0
1
√
τ
1
√
τ − τ
∞
−∞
exp
−a(x
+ b)
2
+ c
dx
dτ
=
1
2
√
π
E
σ
2
− ˜ σ
2
e
−r(T−t0 )
σ ˜ σ
τ
0
1
√
τ
1
√
τ − τ
1
√
a
exp (c) dτ
.
It is easy to show that
a =
1
4(τ − τ
)
+
σ
2
4 ˜ σ
2
τ
and
c = −
σ
2
4˜ σ
2
τ
(τ − τ
)
x −
2τ
σ
2
µ −
1
2
σ
2
− r + D +
1
2
˜ σ
2
2
1
τ − τ
+
σ
2
˜ σ
2
τ
.
TECHNICAL ARTICLE 3
78 Wilmott magazine
With
s − t =
2
σ
2
τ
we have
c = −
ln(S/E) +
µ −
1
2
σ
2
(s − t) +
r − D −
1
2
˜ σ
2
(T − s)
2
2(σ
2
(s − t) + ˜ σ
2
(T − s))
.
From this follows Result 1, that the expected profit initially (t = t
0
,
S = S
0
, I = 0) is
Ee
−r(T−t0 )
(σ
2
− ˜ σ
2
)
2
√
2π
T
t0
1
σ
2
(s − t
0
) + ˜ σ
2
(T − s)
exp
−
ln(S
0
/E) +
µ −
1
2
σ
2
(s − t
0
) +
r − D −
1
2
˜ σ
2
(T − s)
2
2(σ
2
(s − t
0
) + ˜ σ
2
(T − s))
ds.
Result 2: Variance, single option
The problem for the expectation of the square of the profit is
∂v
∂t
+
1
2
σ
2
S
2
∂
2
v
∂S
2
+ µS
∂v
∂S
+
1
2
σ
2
− ˜ σ
2
e
−r(t−t0 )
S
2
i
∂v
∂I
= 0, (10)
with
v(S, I, T) = I
2
.
A solution can be found of the form
v(S, I, t) = I
2
+ 2I H(S, t) + G(S, t).
Substituting this into Equation (10) leads to the following equations
for H and G (both to have zero final and boundary conditions):
∂H
∂t
+
1
2
σ
2
S
2
∂
2
H
∂S
2
+ µS
∂H
∂S
+
1
2
σ
2
− ˜ σ
2
e
−r(t−t0 )
S
2
i
= 0;
∂G
∂t
+
1
2
σ
2
S
2
∂
2
G
∂S
2
+ µS
∂G
∂S
+
σ
2
− ˜ σ
2
e
−r(t−t0 )
S
2
i
H = 0.
Comparing the equations for H and the earlier F we can see that
H = F =
Ee
−r(T−t0 )
(σ
2
− ˜ σ
2
)
2
√
2π
T
t
1
σ
2
(s − t) + ˜ σ
2
(T − s)
exp
−
ln(S/E) +
µ −
1
2
σ
2
(s − t) +
r − D −
1
2
˜ σ
2
(T − s)
2
2(σ
2
(s − t) + ˜ σ
2
(T − s))
ds.
Notice in this that the expression is a present value at time t = t
0
, hence
the e
−r(T−t0 )
term at the front. The rest of the terms in this must be kept as
the running variables S and t.
Returning to variables x and τ , the governing equation for G(S, t) =
w(x, τ ) is
∂w
∂τ
=
∂
2
w
∂x
2
+
2
σ
2
Eσ
σ
2
− ˜ σ
2
e
−r(T−t0 )
e
−d
2
2
/2
4˜ σ
√
πτ
E
σ
2
− ˜ σ
2
e
−r(T−t0 )
2σ ˜ σ
√
π
τ
0
1
√
τ
1
√
τ − τ
1
√
a
exp (c) dτ
(11)
where
d
2
=
σ
˜ σ
x −
2
σ
2
µ −
1
2
σ
2
+ r − D −
1
2
˜ σ
2
τ
√
2τ
,
and a and c are as above.
The solution is therefore
1
2
√
π
2
σ
2
Eσ
σ
2
− ˜ σ
2
e
−r(T−t0 )
4˜ σ
√
π
E
σ
2
− ˜ σ
2
e
−r(T−t0 )
2σ ˜ σ
√
π
∞
−∞
τ
0
f (x
, τ
)e
−d
2
2
/2
√
τ − τ
e
−(x−x
)
2
/4(τ −τ
)
dτ
dx
.
where now
f (x
, τ
) =
1
√
τ
τ
0
1
√
τ
1
√
τ
− τ
1
√
a
exp (c) dτ
and in a and c all τ s become τ
s and all τ
s become τ
s, and in d
2
all τ s
become τ
s and all xs become x
s.
The coefficient in front of the integral signs simplifies to
1
8π
3/2
E
2
σ
2
− ˜ σ
2
2
e
−2r(T−t0 )
σ
2
˜ σ
2
.
The integral term is of the form
∞
−∞
τ
0
τ
0
· · · dτ
dτ
dx
,
with the integrand being the product of an algebraic term
1
√
τ
√
τ
√
τ − τ
√
τ
− τ
√
a
and an exponential term
exp
−
1
2
d
2
2
−
(x − x
)
2
4(τ − τ
)
+ c
.
This exponent is, in full,
−
1
4τ
σ
2
˜ σ
2
x
−
2
σ
2
µ −
1
2
σ
2
τ
+
2
σ
2
r − D −
1
2
˜ σ
2
τ
2
−
(x − x
)
2
4(τ − τ
)
−
σ
2
4˜ σ
2
τ
(τ
− τ
)
x
−
2τ
σ
2
µ −
1
2
σ
2
− r + D +
1
2
˜ σ
2
2
1
τ
− τ
+
σ
2
˜ σ
2
τ
.
This can be written in the form
−d(x
+ f )
2
+ g,
where
d =
1
4
σ
2
˜ σ
2
1
τ
+
1
4
1
τ − τ
+
1
4
σ
2
σ
2
(τ
− τ
) + ˜ σ
2
τ
and
g = −
σ
2
4˜ σ
2
τ
x −
2ατ
σ
2
2
−
σ
2
4(σ
2
(τ
− τ
) + ˜ σ
2
τ
)
x −
2ατ
σ
2
2
+
1
4
σ
2
˜ σ
2
τ
x −
2ατ
σ
2
+
σ
2
(σ
2
(τ
− τ
) + ˜ σ
2
τ
)
x −
2ατ
σ
2
2
σ
2
˜ σ
2
1
τ
+
1
τ − τ
+
σ
2
σ
2
(τ
− τ
) + ˜ σ
2
τ
,
where
α = µ −
1
2
σ
2
− r + D +
1
2
˜ σ
2
.
Using Equation (7) we end up with
1
4π
3/2
E
2
σ
2
− ˜ σ
2
2
e
−2r(T−t0 )
σ
2
˜ σ
2
τ
0
τ
0
1
√
τ
√
τ
√
τ − τ
√
τ
− τ
√
a
π
d
exp(g)dτ
dτ
.
Changing variables to
τ =
σ
2
2
(T − t), τ
=
σ
2
2
(T − s), and τ
=
σ
2
2
(T − u),
and evaluating at S = S
0
, t = t
0
, gives the required Result 2.
Result 3: Expectation, portfolio of options
This expression follows from the additivity of expectations.
Wilmott magazine 79
TECHNICAL ARTICLE 3
W
Result 4: Variance, portfolio of options
The manipulations and calculations required for the analysis of the port
folio variance are similar to that for a single contract. There is again a
solution of the form
v(S, I, t) = I
2
+ 2I H(S, t) + G(S, t).
The main differences are that we have to carry around two implied
volatilities, ˜ σ
j
and ˜ σ
k
and two expirations, T
j
and T
k
. We will find that the
solution for the variance is the sum of terms satisfying diffusion equa
tions with source terms like in Equation (11). The subscript ‘k’ is then
associated with the gamma term, and so appears outside the integral in
the equivalent of (11), and the subscript ‘j’ is associated with the integral
and so appears in the integrand.
There is one additional subtlety in the derivations and that concerns
the expirations. We must consider the general case T
j
= T
k
. The integra
tions in (6) must only be taken over the intervals up until the options
have expired. The easiest way to apply this is to use the convention that
the gammas are zero after expiration. For this reason the s integral is
over t
0
to min(T
j
, T
k
).
■ Black, F & Scholes, M 1973 The pricing of options and corporate liabilities. Journal of
Political Economy 81 637–59.
■ Carr, P 2005 FAQs in option pricing theory. J. Derivatives.
■ Carr, P & Verma, A 2005 Potential Problems in Market Models of Implied Volatility.
Working Paper.
■ Dupire, B 2005 Exploring Volatility Derivatives: New Advances in Modelling. Presented
at Global Derivatives 2005 Paris.
■ Forde, M 2005 Static hedging of barrier options under local volatility with r 6 = q , and
modelindependent hedging of Asian options. PhD thesis Bristol University.
■ Henrard, M 2003 Parameter risk in the Black and Scholes model. Risk and Insurance
0310002, Economics Working Paper Archive at WUSTL.
■ Natenberg, S 1994 Option Volatility and Pricing. McGraw–Hill.
■ Schoutens, W, Simons, E & Tistaert, J 2004 A perfect calibration! Now what? Wilmott
magazine March 2004 66–78.
ACKNOWLEDGMENTS
We would like to thank Hyungsok Ahn and Ed Thorp for their input on the practical
application of our results and on portfolio optimization and Peter Carr for his encyclopedic
knowledge of the literature.
REFERENCES
However. In Section 6 we look at the advantages and disadvantages of hedging using different volatilities. Our paper extends their analyses in several directions. Sections 3 and 4 repeat the analyses of Carr (2005) and Henrard (2003).’ We will see how you can hedge using a delta based on actual volatility or on implied volatility. and again we find closedform formulas for the expectation and variance of profit. but its standard deviation is greater. In Section 2 we set up the problem by explaining the role that volatility plays in hedging. 2 Implied versus Actual.2 we find a closedform formula for the variance of this profit. Before we start. In Section 5 we look at hedging with volatilities other than just implied or actual. the implied volatility VIX seems to be higher than the realized volatility. In Section 9 we outline a portfolio selection Figure 1: Distributions of the logarithms of the VIX and the rolling realized SPX volatility. He also made important observations on portfolios of options and on the role of the asset’s growth rate in determining the profit. We outline the latter approach in Section 8. or higher than both or lower? We will look at the profit or loss to be made hedging vanilla options and portfolios of options with different ‘hedging volatilities. Carr derived the expression for profit from hedging using different volatilities. This is to be preferred generally since it will be faster than simulations. In Section 10 we briefly mention portfolios of options on many underlyings. 1973). we have to study the effects of using each of these in the classical delta formula (Black & Scholes. and draw conclusions and comment on further work in Section 11. Figure 1.TECHNICAL ARTICLE 3 delta hedge when your estimate of future actual volatility differs from that of the market as measured by the implied volatility (Natenberg. 2003) or solve a threedimensional differential equation. it is the amount Wilmott magazine 65 ^ . In Section 4. or on something different. method based on exponential utility. about 20%. Dupire (2005) discusses the advantages of hedging using volatility based on the realized quadratic variation in the stock price. Because the final profit depends on the path taken by the asset when we hedge with implied volatility we look at simple statistical properties of this profit. about 15%. 1994). Delta Hedging but Using which Volatility? Actual volatility is the amount of ‘noise’ in the stock price. is significantly lower than the mean of the VIX. In Section 3 we look at the marktomarket profit and the final profit when hedging using actual volatility. But why stop there? Why not use a volatility between implied and actual.1 we derive a closedform formula for the expected total profit and in Section 4. Part of what follows repeats the excellent work of Carr (2005) and Henrard (2003). The VIX is an implied volatility measure based on the SPX index and so you would expect it and the realized SPX volatility to bear close resemblance. Other relevant literature in this area includes the paper by Carr & Verma (2005) which expands on the problem of hedging using the implied volatility but with implied volatility varying stochastically. the standard deviation of profit as well as minimum and maximum profits. Since there are two volatilities in this problem. Henrard independently derived these results and also performed simulations to examine the statistical properties of the possibly pathdependent profit. To find the full probability distribution of total profit we could perform simulations (Henrard. as can be seen in the figure. Related ideas applied to the hedging of barrier options and Asians can be found in Forde (2005). This is a plot of the distributions of the logarithms of the VIX and of the rolling 30day realized SPX volatility using data from 1990 to mid 2005. especially the realized volatility. Technical details are contained in an appendix. examining the expected profit. we present a simple plot of the distributions of implied and realized volatilities. Portfolios of options are considered in Section 7. therefore making portfolio optimizations more practical. it is the coefficient of the Wiener process in the stock returns model. Whichever hedging volatility you use you will get different risk/return profiles. For the remainder of the paper we focus on the case of hedging using implied volatility. and the normal distributions for comparison. implied and actual. This profit is path dependent. Some of the work in this paper has been used successfully by a volatility arbitrage hedge fund. The VIX distribution is somewhat truncated on the left. In Section 4 we then examine the marktomarket and total profit made when hedging using implied volatility. which is the more common market practice. The mean of the realized volatility. Both of these volatilities are approximately lognormally distributed (since their logarithms appear to be Gaussian).
But how is this guaranteed profit realized? Let us do the analysis on a marktomarket basis. Question: How can we make money if our forecast is correct? Answer: Buy the option and delta hedge. Component Option Stock Cash Value V i + dV i − a S− (−V i + a a where 1 N(x) = √ 2π x −∞ e− 2 ds s2 and ln(S/E) + r + 1 σ 2 (T − t) d1 = . t. σ By hedging with actual volatility we are replicating a short position in a correctly priced option. So should we use σ = 0.’ The new values are shown in Table 2. 3 Case 1: Hedge with Actual Volatility. and are all simple. T − t and r (almost). The payoffs for our long option and our short replicated option will exactly cancel. σ ) − V(S. TOMORROW. This confirms what we said earlier about the guaranteed total profit by expiration. dV i − a dS − r(V i − a S) dt − a DS dt. So the total profit from t0 to expiration is T ert 0 t0 d e−rt (V i − V a ) = V a − V i . Since the market does not have perfect knowledge about the future these two numbers can and will be different. both assumed constant. Imagine that we have a forecast for volatility over the remaining life of an option. For example. But which delta do we use? We know that = N(d1 ) TABLE 1: PORTFOLIO COMPOSITION AND VALUES.’ and come back to it the next ‘day.) If V(S. superscript ‘a’ means actual and ‘i’ denotes implied. That is the profit from time t to t + dt . The profit we make will be exactly the difference in the Black–Scholes prices of an option with 30% volatility and one with 20% volatility.) Therefore we have made. σ ) is the Black–Scholes formula then the guaranteed profit is V(S. Note also that V. a is the delta using the actual volatility in the formula. set up a portfolio by buying the option for V i and hedge with a of the stock. E. TODAY. Table 1.2 or 0.3? ˜ In what follows we use σ to denote actual volatility and σ to represent implied volatility. V i is the theoretical option value using the implied volatility in the formula. mark to market. Component Option Stock Cash Value Vi − aS −V i + a S TABLE 2: PORTFOLIO COMPOSITION AND VALUES. √ 2 σ T−t dS S)(1 + r dt) − a DS dt We can all agree on S. The model is the classical dS = µS dt + σ S dX. But which delta do you choose? Delta based on actual or implied volatility? Scenario: Implied volatility for an option is 20%. The present value of this profit at time t0 is e−r(t−t0 ) ert d e−rt (V i − V a ) = ert 0 d e−rt (V i − V a ) . and further assume that our forecast turns out to be correct. (Assuming that the Black–Scholes assumptions hold. this volatility is forecast to be constant. t. known. but we believe that actual volatility is 30%. t. ˜ Now. (We have included a continuous dividend yield in this. Wilmott magazine 66 . Because the option would be correctly valued at V a we have dV a − a dS − r(V a − a S) dt − a DS dt = 0. So we can write the marktomarket profit over one time step as dV i − dV a + r(V a − i a a i S) dt − r(V i − a rt a S) dt −rt = dV − dV − r(V − V ) dt = e d e (V i − V a ) . Black–Scholes formulas.of randomness that ‘actually’ transpires. but not on σ . σ ). We shall buy an underpriced option and delta hedge to expiry. Leave this hedged portfolio ‘overnight. Implied volatility is how the market is pricing the option currently. The values of each of the components of our portfolio are shown in the following table. these can be applied to deltas and option values. . In the following.
Integrate the present value of all of these profits over the life of the option to get a total profit of 1 2 σ − σ2 ˜ 2 T t0 e−r(t−t0 ) S2 i dt. The option is a oneyear European call. This leaves us with a dX in our marktomarket P&L and from a risk management point of view this is not ideal. but highly path dependent. implied is 20%. σ ) + ˜ 1 2 σ − σh2 2 T t0 ^ e−r(t−t0 ) S2 h dt. with a strike of 100. To make a profit all we need to know is that actual is always going to be greater than implied (if we are buying) or always less (if we are selling). On a marktomarket basis you could lose before you gain. this would maximize the total profit. and put any cash in the bank earning r. because of the dX term in the above. Carr (2005) and Henrard (2003) show the more general result that if you hedge using a delta based on a volatility σh then the PV of the total profit is given by Figure 2: P&L for a deltahedged option on a marktomarket basis. Buy the option today. Note that the lines are not perfectly smooth. If we start off at the money and the drift is very large (positive or negative) we will find ourselves quickly moving into territory where gamma and hence expression (1) is small. The option and parameters are the same as in the previous example. hedge using the implied delta. The P&L is always increasing in value but the end result is random. Wilmott magazine V(S. so that there will be not much profit to be made. but we may lose on a marktomarket basis in the meantime. The figure shows several realizations of the same deltahedged position. By hedging with implied volatility we are balancing the random fluctuations in the marktomarket option value with the fluctuations in the stock price. Moreover. we hedged ‘only’ 1000 times for each realization. µ. hedged using actual volatility. at the money initially. t. again because of the effect of hedging discretely. There is another advantage of hedging using implied volatility. For a bond there is a guaranteed outcome. σ i )σ S dX + (µ + D)S( − ) dt + 1 2 ˜ σ 2 − σ 2 S2 dt ˜ (using Black–Scholes with σ = σ ) = 1 (σ 2 − σ 2 )S2 ˜ 2 i dt + ( i − a )((µ − r + D)S dt + σ S dX). we do not even need to know what actual volatility is. But these changes do not cancel each other out. This is always positive. DS dt dt + µS( − − a a − 1 2 2 σ S 2 dt − r(V − V ) dt a )σ S dX − DS dt i ∼ 4 Case 2: Hedge with Implied Volatility. The marktomarket profit from today to tomorrow is dV i − = = 1 2 i i dS − r(V i − dt + 1 σ 2 S2 2 i i 2 2 i S) dt − i DS dt i dt − r(V i − dt. S) dt − i DS dt (1) ˜ σ −σ S 2 Observe how the profit is deterministic. (2) 67 . Note that the final P&L is not exactly the same in each case because of the effect of hedging discretely. This takes some of the pressure off forecasting volatility accurately in the first place. actual volatility is 30%. Being path dependent it will depend on the drift µ. Compare and contrast now with the case of hedging using a delta based on implied volatility. σh ) − V(S. the marktomarket profit depends on the real drift of the stock.TECHNICAL ARTICLE 3 We can also write that one time step marktomarket profit (using Itô’s lemma) as i dt + = +( =( i i i i dS + 1 σ 2 S2 2 i i a dt − ) dt + a a dS − r(V i − i a S) dt − i a a There is a simple analogy for this behavior. The best that could happen would be for the stock to end up close to the strike at expiration. so will V . When S changes. This is illustrated in Figure 2. t. there are no dX terms. The evolution of the portfolio value is then ‘deterministic’ as we shall see. The simple analogy is now just putting money in the bank. It is similar to owning a bond. From a risk management perspective this is much better behaved. the growth rate is 10% and interest rate 5%. This path dependency is shown in Figure 3. The conclusion is that the final profit is guaranteed (the difference between the theoretical option values with the two volatilities) but how that is achieved is random. The figure shows several simulations of the same deltahedged position.
I = 0) being the single integral Wilmott magazine 68 . From this equation it is easy to see that the final profit is bounded by V(S. so that 1 1 2 ∂F ∂F ∂2F + σ 2 S2 2 + µS + σ − σ 2 e−r(t−t0 ) S2 ˜ ∂t 2 ∂S ∂S 2 i = 0. t.1 The expected profit after hedging using implied volatility When you hedge using delta based on implied volatility the profit each ‘day’ is deterministic but the present value of total profit by expiration is path dependent. and given by 1 2 σ − σ2 ˜ 2 T t0 where E is the strike and T is expiration. that is S = E exp − r − D − σh2 /2 (T − t) . t) Figure 3: P&L for a deltahedged option on a marktomarket basis. I. √ σ T−t The righthand side of the latter expression above comes from maximizing S2 h . 2 2 4. I. Since therefore dI = 1 2 σ 2 − σ 2 e−r(t−t0 ) S2 ˜ i dt we can write down the following partial differential equation for the real expected value. where the superscript on the gamma means that it uses the Black– Scholes formula with a volatility of σh . t. Look for a solution of this equation of the form P(S. σ ) ˜ The source term can be simplified to E σ 2 − σ 2 e−r(T−t0 ) e−d2 /2 ˜ . σh ) − V(S. t. √ σh 2π where d2 = ˜ ln(S/E) + (r − D − 1 σ 2 )(T − t) 2 . The resulting partial differential equation is a then simpler.Introduce I= 1 2 σ − σ2 ˜ 2 t t0 e−r(s−t0 ) S2 i ds. t. t) = I + F(S. T) = I. ∂I with P(S. σ ) + ˜ . This maximum occurs along the path ln(S/E) + (r − D − σh2 /2) (T − t) = 0 . of I 1 1 2 ∂P ∂P ∂2P + σ 2 S2 2 + µS + σ − σ 2 e−r(t−t0 ) S2 ˜ ∂t 2 ∂S ∂S 2 i ∂P = 0. I. S = S0 . and write F(S. hedged using implied volatility. τ ). P(S. √ 2σ 2π(T − t) ˜ 2 and √ E σ 2 − σh2 e−r(T−t0 ) T − t0 V(S. e−r(s−t0 ) S2 i ds. Result 1: After some manipulations we end up with the expected profit initially (t = t0 . σh ) − V(S. t) = w(x. Change variables to x = ln(S/E) + 2 σ2 µ− 1 2 σ2 σ τ and τ = (T − t). t).
where gamma is largest. Figure 4: Expected profit. hedging using implied volatility. Results are shown in the following figures. roughly speaking.2 .05. r = 0.3 . You can see how the higher the growth rate the larger the strike price at the maximum. ~ = 0. D = 0. E = 110 . This will be at the growth rate that ensures. The contour map is shown in Figure 6. versus growth rate µ. r = 0. The atthemoney implied volatility is 20% which in this case is the actual volatility. σ = 0. Wilmott magazine Figure 6: Contour map of expected profit.05 . S = 100. strike.TECHNICAL ARTICLE 3 F(S0 . from 22. For most realistic parameter regimes the maximum expected profit hedging with implied is similar to the guaranteed profit hedging with actual. this does not take into account the risk. In Figure 5 is shown expected profit versus E and µ. profitability increases with distance away from the money. r = 0. This picture changes when you divide the expected profit by the price of the option (puts for lower strikes. In the figure is also shown the profit to be made when hedging with actual volatility.4.4. r = 0.05. call for higher). hedging using implied volatility. see Figure 8. ~ = 0. the standard deviation associated with such trades. falling to 17. σ S = 100. versus growth rate µ and strike E. Here we have used a linear negative skew. σ = 0.2. The effect of skew is shown in Figure 7.5% at the 125 Figure 5: Expected profit. 69 ^ .5% at a strike of 75. Derivation: See Appendix. σ D = 0. that the stock ends up close to at the money at expiration. T = 1. hedging using implied volatility. σ = 0. t0 ) = Ee−r(T−t0 ) (σ 2 − σ 2 ) ˜ √ 2 2π T t0 1 σ 2 (s − t0 ) + σ 2 (T − s) ˜ 2 exp − ln(S0 /E) + µ − 1 σ 2 (s − t0 ) + r − D − 1 σ 2 (T − s) ˜ 2 2 2(σ 2 (s − t0 ) + σ 2 (T − s)) ˜ ds. S = 100. ~ = 0. ˜ Parameters are S = 100 .2. Observe that the expected profit has a maximum. T = 1. versus growth rate µ and strike E.4.2. σ D = 0. The dashed line is the profit to be made when hedging with actual volatility. There is no maximum. In Figure 4 is shown the expected profit versus the growth rate µ. T = 1. σ = 0. σ = 0. Of course. T = 1.05. E = 110. D = 0.
So we will need to calculate v(S. ˜ 2 2 2 Derivation: See Appendix. r = 0. σ = 0. as both strike and expiration vary.2. σ = 0.S0 . Figure 10 shows the standard deviation of profit versus strike.2.1 . these diagrams can be interpreted in a classical meanvariance manner. r = 0. s. D = 0. where G(S0 . S0 . since it now contains the standard deviation. Result 2: The initial variance is G(S0 . S = 100 . t0 ) = E2 (σ 2 − σ 2 )2 e−2r(T−t0 ) ˜ 4π σ σ ˜ √ √ T t0 s T ep(u. 4. S = 100. hedging using implied volatility. t0 )2 . I. Figure 12 completes the earlier picture for the skew. Using the above 70 . of course.notation. expected profit. In Figure 11 are shown expected profit divided by cost versus standard deviation divided by cost. E = 110 . To some extent. I. hedging using implied volatility. S = 100. µ = 0. µ = 0.4 . σ = 0.t0 ) Figure 7: Effect of skew.05 . versus strike E.05 . T = 1. µ = 0. D = 0.2 The variance of profit after hedging using implied volatility Once we have calculated the expected profit from hedging using implied volatility we can calculate the variance in the final profit. ˜ although we emphasize only some. σ = 0.s. D = 0 .4 . ∂I with v(S.2. versus strike E. σ = 0. and α = µ − 1 σ 2 − r + D + 1 σ 2 . t) + G(S. t).05. T = 1. µ = 0. there is no downside. r = 0. furthermore. r = 0.2 .1. ratio of expected profit to price. but a solution can be found of the form v(S.05 .05. D = 0. no possibility of any losses. σ = 0.2. σ = 0. r = 0. D = 0. In these S = 100 . ˜ Note that in these plots the expectations and standard deviations have not been scaled with the cost of the options. T) = I2 . In Figure 9 is shown the standard deviation of profit versus growth ˜ rate. I. The main criticism is. the variance will be the expected value of I2 less the square of the average of I. The details of finding this function v are rather messy. t0 ) = − 1 1 (x + α(T − s))2 (x + α(T − u))2 − 2 (T − s) 2 (u − s) + σ 2 (T − u) 2 σ ˜ 2σ ˜ 2 x + α(T − u) x + α(T − s) + 2 2 (T − s) 2 (T − u) 1 σ ˜ σ (u − s) + σ ˜ + 1 1 1 2 + 2 + 2 σ 2 (s − t0 ) σ (T − s) ˜ σ (u − s) + σ 2 (T − u) ˜ and x = ln(S0 /E) + µ − 1 σ 2 (T − t0 ). and. S = 100 . T = 1. that we are not working with normal distributions. t) where 1 1 2 ∂v ∂v ∂2v + σ 2 S2 2 + µS + σ − σ 2 e−r(t−t0 ) S2 ˜ ∂t 2 ∂S ∂S 2 i ∂v = 0. T = 1. t) = I2 + 2I H(S.4. where s − t0 T − s σ 2 (u − s) + σ 2 (T − u) ˜ 1 1 1 + 2 + 2 σ 2 (s − t0 ) σ (T − s) ˜ σ (u − s) + σ 2 (T − u) ˜ du ds (3) p(u. t0 ) − F(S0 . σ = 0. Wilmott magazine Figure 8: Effect of skew.
T = 1. Figure 12: Effect of skew. S = 100. S = 100. Four different expirations. Wilmott magazine 71 ^ . r = 0. ~ = 0. r = 0. The expected profit is also shown. σ = 0.2. S = 100.2. hedging using implied volatility. σ = 0.05. D = 0. ~ = 0. versus strike E. ~ = 0.4. and ratio of standard deviation to price.4. r = 0. E = 110. T = 1. hedging using implied volatility. r = 0. σ versus growth rate µ.2.05. D = 0.1.05. σ Figure 10: Standard deviation of profit. varying strike. S = 100. D = 0.4. µ = 0. µ = 0. σ = 0. The expected profit is also shown.2. verσ sus strike E. T = 1. Figure 11: Scaled expected profit versus scaled standard deviation.05.TECHNICAL ARTICLE 3 Figure 9: Standard deviation of profit. σ = 0. ratio of expected profit to price. µ = 0. D = 0.
D = 0 . The standard deviation of profit is zero when the option is hedged at the actual volatility.1. σ = 0.2. σ = 0. hedging with various volatilities.2.1.1 . 5. Crucially all of the curves have zero value at the actual/implied volatility. σ = 0. ~ = 0. The downside is now more dramatic than the upside.4 . t.2.1 Actual volatility = Implied volatility For the first example let’s look at hedging a long position in a correctly ˜ priced option. standard deviation of profit. T = 1 . σ = 0. so that σ = σ . µ = 0 . We will hedge using different volatilities.2 Actual volatility > Implied volatility In Figure 14 is shown the expected profit and standard deviation of profit when hedging with various volatilities when actual volatility is greater than implied. D = 0.3 Actual volatility < Implied volatility In Figure 15 is shown properties of the profit when hedging with various volatilities when actual volatility is less than implied. σ = 0. σ ) . the maximum profit is much greater than the downside. given that the expected Wilmott magazine 72 . We are now selling the option and delta hedging it. and σ = 0. µ = 0. and σ = 0. E = 100.2. Parameters are E = 100 . t.1 Hedging with actual volatility Figure 13: Expected profit.2 .4.4 .1. D = 0 . σ r = 0. σ h . Results are shown in Figure 13. S = 100 . The chart again also shows minimum and maximum profit. S = 100. hedging with various volatilities. The chart also shows minimum and maximum profit. Pros: The main advantage of hedging with actual volatility is that you know exactly what profit you will get at expiration.1 . Parameters are E = 100 . µ = 0. since the integral of gamma ˜ term can be treated as before if one replaces σ with σh in this term. minimum and maximum. 5. µ = 0. The figure shows the expected profit and standard deviation of profit when hedging with various volatilities.2. T = 1 . 6. So in a classical risk/reward sense this seems to be the best choice. r = 0. Parameters are E = 100 . 5. S = 100 . The upside. Figure 14: Expected profit. r = 0. T = 1. 6 Pros and Cons of Hedging with each Volatility Given that we seem to have a choice in how to delta hedge it is instructive to summarize the advantages and disadvantages of the possibilities.5 Hedging with Different Volatilities We will briefly examine hedging using volatilities other than actual or implied. T = 1. Note that it is possible to lose money if you hedge at below implied. ˜ and σ = 0. T = 1. σh ) − V(S. standard deviation of profit. ~ = 0. D = 0. Now it is possible to lose ˜ money if you hedge at above implied.2. S = 100. The expressions for the expected profit and standard deviations now ˜ must allow for the V(S. µ = 0. but hedging with a higher volatility you will not be able to lose until hedging with a volatility of approximately 75%. σ r = 0.3) regardless of the hedging volatility. S = 100 . Results are presented in the next sections. r = 0. using the general expression for profit given by (2). minimum and maximum. The expected profit is again insensitive to hedging volatility. ˜ With these parameters the expected profit is small as a fraction of the market price of the option ($13. D = 0. but hedging with a lower volatility you will not be able to lose until hedging with a volatility of approximately 10%. The expected profit is again insensitive to hedging volatility. E = 100.
2. profit can often be insensitive to which volatility you choose to hedge with whereas the standard deviation is always going to be positive away from hedging with actual volatility. and it has no standard deviation of final profit. standard deviation of profit. you are continually making a profit. D = 0. If the stock is far in or out of the money the two deltas are similar and so the local risk is small. √ √ σ T−t ˜ σh T − t 2 Figure 15: Expected profit. the daily fluctuations in P&L. This is usually not far from optimal in the sense of possible expected total profit. S = 100. In this case it is more usual to hedge based on implied volatility to avoid the daily fluctuations in the profit and loss. Cons: You don’t know how much money you will make. and therefore easy to observe. by looking at the random component in a portfolio hedged using a volatility of σ h . one using implied volatility and the other the hedging volatility. minimum and maximum. only that it is positive. The second advantage is that you only need to be on the right side of the trade to profit. However. D = 0. six months before expiration. or by investors who may monitor the marktomarket profit on a regular basis. If one is able to mark to model then one is not necessarily concerned with the daytoday fluctuations in the marktomarket profit and loss and so it is natural to hedge using actual volatility. ~ = 0. The local risk is also small where the two deltas cross over. σ = 0. σ E = 100. We can begin to quantify the ‘local’ risk. T = 0. However. 6. σ h = 0. r = 0. Cons: The P&L fluctuations during the life of the option can be daunting. (4) Note that this expression depends on all three volatilities. µ = 0. The first is that there are no local fluctuations in P&L. Buy when actual is going to be higher than implied and sell if lower.4. 73 ^ . This ‘sweet spot’ is at ln(S/E) + (r − D + σ h /2)(T − t) ln(S/E) + (r − D + σ 2 /2)(T − t) ˜ = . by prime brokers. Figure 16 shows the two deltas (for a call option).2. T = 1. you can interpret the previous two figures in terms of what happens if you intend to hedge with actual but don’t quite get it right. ~ = 0. the number that goes into the delta is implied volatility. S = 100.TECHNICAL ARTICLE 3 In practice which volatility one uses is often determined by whether one is constrained to mark to market or mark to model. See Dupire (2005) for work in this area. The standard deviation of this risk is √ σ S  i − h  dt. you are unlikely to be totally confident in your volatility forecast. You can see from those that you do have quite a lot of leeway before you risk losing money. Wilmott magazine Figure 16: Deltas based on implied volatility and hedging volatility.1.1.2 Hedging with implied volatility Pros: There are three main advantages to hedging with implied volatility. σ r = 0. E = 100. hedging with various volatilities. Finally. 6. and so less appealing from a ‘local’ as opposed to ‘global’ risk management perspective. it is common to have to report profit and loss based on market values.5. Also. This constraint may be imposed by a risk management department. the number you are putting into your delta formula.3 Hedging with another volatility You can obviously balance the pros and cons of hedging with actual and implied by hedging with another volatility altogether.3.
E = 100. t0 ) = k qk Ek e−r(Tk −t0 ) (σ 2 − σk2 ) ˜ √ 2 2π Tk t0 1 σ 2 (s − t0 ) + σk2 (Tk − s) ˜ 2 exp − ˜ ln(S0 /Ek ) + µ − 1 σ 2 (s − t0 ) + r − D − 1 σk2 (Tk − s) 2 2 2(σ 2 (s − t0 ) + σk2 (Tk − s)) ˜ ds. ˜ In the spirit of the earlier analyses and formulas we would ideally like to be able to quantify various statistical properties of the local marktomarket fluctuations. ~ = 0. as a function of stock price and time. although the ideas can be easily extended to the more general case. 7. Wilmott magazine 74 .that is. Figure 18 is a contour map of the same. D = 0. I = 0 ) is simply the sum of individual profits for each option. Since only an option’s gamma matters when we are hedging using implied volatility. σ h = 0. D = 0. E = 100. σ = 0. D = 0. (5) Figure 17: Local risk as a function of stock price and time to expiration. as a new state variable. ∂I with final condition representing expectation.2. S = E exp − T−t 2 σ (r − D + σ h /2) − σ h (r − D + σ 2 /2) ˜ ˜ σ − σh ˜ . and the analysis can proceed as before. S = 100. options with different strikes and expirations. σ T = 1.3.2 . r = 0. T = 1. variance. Introduce I= 1 2 t qk σ 2 − σk2 ˜ k t0 e−r(s−t0 ) S2 i k ds. Note that since there may be more than one expiration date since we have several i different options.2. σ = 0. T = 1. The profit from a portfolio is now 1 2 qk σ 2 − σk2 ˜ k t0 Tk e−r(s−t0 ) S2 i k ds. S = 100 . 7 Portfolios when Hedging with Implied Volatility A natural extension to the above analysis is to look at portfolios of options.4. it must be understood in Equation (5) that k is zero for times beyond the expiration of the option. ~ = 0. Parameters are E = 100 . etc.1. F(S0 . σ = 0. where k is the index for an option. r = 0. Figure 18: Contour map of local risk as a function of stock price and time to σ expiration. variance.3. Figure 17 shows a threedimensional plot of expression (4). This will be the subject of future work.3. σ h = 0. Derivation: See Appendix. S = S0 . σ h = 0. etc. σ = 0. S = 100. and qk is the quantity of that option.1.1. without √ the dt factor. is then 1 1 ∂ ∂ ∂2 + σ 2 S2 2 + µS + ∂t 2 ∂S ∂S 2 σ 2 − σk2 e−r(t−t0 ) S2 ˜ k i k ∂ = 0.4. The governing differential operator for expectation. For the remainder of this paper we will only consider the case of hedging using a delta based on implied volatility.4. calls and puts are effectively the same since they have the same gamma.1 Expectation Result 3: The solution for the present value of the expected profit (t = t0 . r = 0.
See text for additional parameters and information. volatility arbitrage while meeting constraints imposed by regulators. The payoff function (with its initial delta hedge) is shown in Figure 19. A B Put 90 −2.200 10.t0 ) t0 min (Tj .s.310 −0. (The ‘Profit Total Expected’ row assumes that we buy a single one of that option. r = 0. Let us consider the simple case of maximizing the expected return. ratio of profit to standard deviation).247 1. t0 )2 where G(S0 . The growth rate is zero.936 E Call 120 1 0. S0 .10 Payoff function (with initial delta hedge) ¯ ¯ ln(S0 /Ek ) + µ(s − t0 ) + rk (Tk − s) σk2 (Tk − s) ˜ ¯ ¯ ln(S0 /Ej ) + µ(u − t0 ) + rj (Tj − u) 2 + 2 (u − s) + σ 2 (T − u) σ ˜j j 1 + 1 1 1 2 + 2 + 2 σ 2 (s − t0 ) σk (Tk − s) ˜ σ (u − s) + σj2 (Tj − u) ˜ Asset 90 100 110 120 130 140 and µ=µ− ¯ 1 2 σ . and associated parameters. we can find an expression for the initial variance as G(S0 .2 Variance Result 4: The variance is more complicated. Implied Option Price. The expected profit is $6.012 2.) Using the above formulas we can find the portfolio that maximizes or minimizes target quantities (expected profit.410 where Gjk (S0 .2. This optimization has effectively found an ideal risk reversal trade. The result is given in Table 4. σ = 0. Market Option Value. it would bring in premium. Table 3 shows the available options. Nevertheless. t0 ) = q j qk Gjk (S0 . Observe the negative skew.175 5.e.05.28 where p(u. t0 ) − F(S0 . T = 1. while constraining the standard deviation to be one.225 3.46 to set up. The stock is currently at 100. ˜ ˜ 2 2 −15 −20 −25 −30 Derivation: See Appendix. 7. standard deviation.000 D Call 110 1 0. The outofthemoney puts are overvalued and the outofthemoney calls are undervalued.T k ) s Tj √ s − t0 Tk − s σ 2 (u − s) + σj2 (Tj − u) ˜ du ds (6) 1 1 1 + 2 + 2 σ 2 (s − t0 ) σk (Tk − s) ˜ σ (u − s) + σj2 (Tj − u) ˜ TABLE 4: AN OPTIMAL PORTFOLIO. t0 ) j k TABLE 3: AVAILABLE OPTIONS.TECHNICAL ARTICLE 3 7. zero dividend yield and the interest rate is 5%.83. 2 ¯ ¯ rj = r − D − 1 σj2 and rk = r − D − 1 σk2 . D = 0. µ = 0. t0 ) = − − ¯ 1 (ln(S0 /Ek ) + µ(s − t0 ) + rk (Tk − s))2 ¯ 2 2 σk (Tk − s) ˜ ¯ ¯ 1 (ln(S0 /Ej ) + µ(u − t0 ) + rj (Tj − u))2 2 (u − s) + σ 2 (T − u) 2 σ ˜j j 15 10 5 0 60 −5 −10 70 80 Type Strike Quantity Put 80 −2. This portfolio would cost −$0.250 1. because of the correlation between all of the options in the portfolio.451 10.752 C Call 100 1 0. brokers.040 0.054 6.451 0. obviously. actual volatility is 20%.S 0 .150 1. A Type Strike Expiration Volatility. 75 ^ . s.46 E Call 120 1.687 −0. t0 ) = Ej Ek (σ 2 − σj2 )(σ 2 − σk2 )e−r(Tj −t0 )−r(Tk −t0 ) ˜ ˜ 4π σ σk ˜ √ e p(u.511 0.660 3. Theory Profit Total Expected Put 80 1 0.3 Portfolio optimization possibilities There is clearly plenty of scope for using the above formulas in portfolio optimization problems. i.933 B Put 90 1 0.25 C Call 100 0 D Call 110 1. Here we give one example. investors etc. This is a very natural strategy when trying to make a profit from Wilmott magazine Figure 19: Payoff with initial delta hedge for optimal portfolio. S = 100.
such as 95% (a ValueatRisk type of optimization. ∂I so that 1 ηQ ∂Q ∂Q ∂2Q + σ 2 S2 2 + µS − ∂t 2 ∂S ∂S 2 σ 2 − σk2 e−r(t−t0 ) S2 ˜ k i k subject to the final condition C(S. η 10. I. Mathematically the problem for the cumulative distribution function for the final profit I can be written as C(S0 . • Maximize probability of making a profit greater than a specified amount or. where Tmax is the expiration of the longest maturity option and H(·) is the Heaviside function. Tmax . ∂I 8 Other Optimization Strategies Rather than choose an option or a portfolio based on mean and variance it might be preferable to examine the probability density function for I. the profit from a hedged mispriced option is not. can be used to choose or optimize a portfolio of options based on criteria such as the following. and one that has neither of these disadvantages. with final condition Q (S. the riskfree interest adjusted for dividends. We can look for a solution of the form 1 U(S. η where Tmax is the expiration of the longest maturity option. or solving a differential equation in three dimensions. which could be criticized because we are not working with a Gaussian distribution for I. with suitable final conditions. I. Constraints would typically need to be imposed on these optimization problems. via simple option pricing formulas for each option’s gamma. in particular using constant absolute risk aversion with utility function − 1 −ηI e . Tmax ) = 1. I ) = H(I − I). 9 Exponential Utility Approach Rather than relying on means and variances. 76 . So now we shall look at other ways of choosing or valuing a portfolio.e. In the above example we assumed constant growth rates for each stock. Because we know the instantaneous profit at each time. As we have seen above the profit depends crucially on the growth rate because of the path dependence. U . As already mentioned. η ∂C = 0. such as having a set budget and/or a limit on number of positions that can be held. Tmax ) = − 1 −ηI e . This is to work within a utility theory framework.Because the state variable representing the profit. i. there is another possibility. something that is not usually considered when simply valuing options in complete markets. I. we can very easily simulate the P&L for an arbitrary portfolio. The main reason for this is the observation that I is not normally distributed. 0. The same equation. The governing equation for the expected utility. All that is required is a simulation of the real paths for each of the underlyings. but Wilmott magazine The parameter η is then a person’s absolute risk aversion. We shall not pursue this in this paper. one must use simulation techniques when there are many underlyings in the portfolio. only depending on riskneutral rates. In the spirit of maximizing expected growth rate (Kelly Criterion) we could also examine solutions of the above threedimensional partial differential equation having final condition being a logarithmic function of I. for a given stock price. albeit one with no possibility of a loss) Being only a twodimensional equation this will be very quick to solve numerically. t0 . One can then pose and solve various optimal portfolio problems. is not normally distributed a portfolio analysis based on mean and variance is open to criticism.1 Dynamics linked via drift rates Although option prices are independent of real drift rates. When we have many underlyings we really ought to model the drift rates of all the stocks as accurately as possible. ideally we would like to see the stock following a path along which gamma is large since this gives us the highest profit. which may be slow. = 0. I. I. 10 Many Underlyings and Portfolios of Options Although methodologies based on formulas and/or partial differential equations are the most efficient when the portfolio only has a small number of underlyings. equivalently. is then 1 1 ∂U ∂U ∂2U + σ 2 S2 2 + µS + ∂t 2 ∂S ∂S 2 ˜ σ 2 − σk2 e−r(t−t0 ) S2 k i k ∂U = 0. minimize the probability of making less than a specified amount • Maximize profit at a certain probability threshold. I ) where C(S. t. t). t) = − e−ηI Q (S. I ) is the solution of 1 1 ∂C ∂C ∂2C + σ 2 S2 2 + µS + ∂t 2 ∂S ∂S 2 σ 2 − σk2 e−r(t−t0 ) S2 ˜ k i k with final condition U(S.
. with zero final and boundary conditions. interacting dynamics at work. This can be represented by dSi = µi (S1 . Appendix: Derivation of Results Preliminary results In the following derivations we often require the following simple results. that is. 2 2 ˜ 1 E σ 2 − σ 2 e−r(T−t0 ) = √ σσ ˜ 2 π 1 1 1 √ √ √ exp (c) dτ . a relationship at the infinitesimal timescale. ˜ 1 E σ 2 − σ 2 e−r(T−t0 ) 2π σσ ˜ exp − + 2 σ2 ∞ −∞ 0 τ 1 1 √ √ τ −τ τ 2 σ2 µ− 1 2 σ τ 2 σ2 (x − x )2 − 4(τ − τ ) 4σ 2 τ ˜ r−D− 1 2 σ τ ˜ 2 2 x − e −∞ −ax 2 dx = π . and in the context of the present work. single option The equation to be solved for F(S. . We have concentrated on the case of hedging using deltas based on implied volatilities because this is the most common in practice. in complete markets the sole interaction between stocks that need concern us is via correlations in the dXi terms. 77 ^ . the transformations 2 x = ln(S/E) + 2 σ 1 σ2 µ − σ 2 τ and τ = (T − t). a τ −τ τ It is easy to show that a= σ2 1 + 4(τ − τ ) 4σ 2 τ ˜ turn the operator 1 ∂ ∂ ∂2 + σ 2 2 + µS ∂t 2 ∂S ∂S and σ 4σ 2 τ (τ − τ ) ˜ 2 into ∂ ∂2 1 2 σ − + 2 2 ∂τ ∂x Wilmott magazine . a coupling in the growth rates. Traditionally. τ ) where E σ 2 − σ 2 e−r(T−t0 ) e−d2 /2 ˜ ∂2w ∂w = + √ 2 ∂τ ∂x σσ πτ ˜ 2 11 Conclusions and Further Work This paper has expanded on the work of Carr and Henrard in terms of final formulas for the statistical properties of the profit to be made hedging mispriced options. Second. τ ) ∂τ ∂x2 If we write the argument of the exponential function as −a(x + b)2 + c we have the solution that is initially zero and is zero at plus and minus infinity is 1 √ ∞ −∞ 0 τ 2 π f (x .TECHNICAL ARTICLE 3 in reality there may be more interesting. τ ) −(x−x )2 /4(τ −τ ) e dτ dx . giving marktomarket profit the smoothest behavior. . there will also be longer timescale relationships operating. . a (7) dτ dx . t) is 1 1 2 ∂F ∂F ∂2F + σ 2 S2 2 + µS + σ − σ 2 e−r(t−t0 ) S2 ˜ ∂t 2 ∂S ∂S 2 i = 0. Using the above changes of variables this becomes F(S. ∞ The solution of this problem is. In reality. that is. x− c=− 2τ σ2 µ− (9) 1 2 1 σ − r + D + σ2 ˜ 2 2 σ2 1 + 2 τ −τ σ τ ˜ 2 . First. Result 1: Expectation. t) = w(x. where σ d2 = σ ˜ x− 2 σ2 µ− 1 2 2 σ τ+ 2 2 σ √ 2τ r−D− 1 2 σ τ ˜ 2 . using (8). Sn ) dt + σi Si dXi . √ τ −τ (8) ˜ 1 E σ 2 − σ 2 e−r(T−t0 ) 2π σσ ˜ τ 0 1 1 √ √ τ −τ τ τ 0 ∞ −∞ exp −a(x + b)2 + c dx dτ Finally. the solution of ∂2w ∂w = + f (x. We have also indicated how more sophisticated portfolio construction techniques can be applied to this problem relatively straightforwardly.
that the expected profit initially (t = t0 . t) = I2 + 2I H(S. τ )e−d2 /2 −(x−x )2 /4(τ −τ ) e dτ dx . single option The problem for the expectation of the square of the profit is 1 1 2 ∂v ∂v ∂2v + σ 2 S2 2 + µS + σ − σ 2 e−r(t−t0 ) S2 ˜ ∂t 2 ∂S ∂S 2 i ∂v = 0. 1 σ 2 (s − t) + σ 2 (T − s) ˜ 2 with the integrand being the product of an algebraic term 1 √ √ √ √ √ τ τ τ −τ τ −τ a exp − ˜ ln(S/E) + µ − 1 σ 2 (s − t) + r − D − 1 σ 2 (T − s) 2 2 2(σ 2 (s − t) + σ 2 (T − s)) ˜ ds. ∂I (10) ∞ −∞ 0 τ f (x . 1 1 1 √ √ √ exp (c) dτ a τ −τ τ Substituting this into Equation (10) leads to the following equations for H and G (both to have zero final and boundary conditions): 1 1 2 ∂H ∂H ∂2H + σ 2 S2 2 + µS + σ − σ 2 e−r(t−t0 ) S2 ˜ ∂t 2 ∂S ∂S 2 1 ∂G ∂G ∂2G + σ 2 S2 2 + µS + σ 2 − σ 2 e−r(t−t0 ) S2 ˜ ∂t 2 ∂S ∂S i i and in a and c all τ s become τ s and all τ s become τ s. τ ) is ˜ ˜ ∂2w 2 Eσ σ 2 − σ 2 e−r(T−t0 ) e−d2 /2 E σ 2 − σ 2 e−r(T−t0 ) ∂w = + 2 √ √ ∂τ ∂x2 σ 4σ π τ ˜ 2σ σ π ˜ 2 we have c=− ˜ ln(S/E) + µ − 1 σ 2 (s − t) + r − D − 1 σ 2 (T − s) 2 2 2(σ 2 (s − t) + σ 2 (T − s)) ˜ 2 (11) .With s−t= 2 τ σ2 Returning to variables x and τ . √ τ −τ 2 with v(S. I. the governing equation for G(S. √ 2τ 1 σ 2 (s − t0 ) + σ 2 (T − s) ˜ 2 ln(S0 /E) + µ − 1 σ 2 (s − t0 ) + r − D − 1 σ 2 (T − s) ˜ 2 2 2(σ 2 (s − t0 ) + σ 2 (T − s)) ˜ ds. The rest of the terms in this must be kept as the running variables S and t. where now 1 f (x . τ ) = √ τ τ 0 A solution can be found of the form v(S. t). and an exponential term Notice in this that the expression is a present value at time t = t0 . S = S0 . I = 0) is Ee−r(T−t0 ) (σ 2 − σ 2 ) ˜ √ 2 2π exp − T t0 where σ d2 = σ ˜ x− 2 σ2 µ− 1 2 1 σ + r − D − σ2 τ ˜ 2 2 . The integral term is of the form ∞ −∞ 0 τ 0 τ Comparing the equations for H and the earlier F we can see that H=F= Ee−r(T−t0 ) (σ 2 − σ 2 ) ˜ √ 2 2π T t · · · dτ dτ dx . The solution is therefore ˜ ˜ 2 Eσ σ 2 − σ 2 e−r(T−t0 ) E σ 2 − σ 2 e−r(T−t0 ) √ √ 2 2 πσ 4σ π ˜ 2σ σ π ˜ 1 √ Result 2: Variance. The coefficient in front of the integral signs simplifies to ˜ 1 E2 σ 2 − σ 2 e−2r(T−t0 ) . 1 (x − x )2 +c . H = 0. and in d2 all τ s become τ s and all xs become x s. exp − d2 − 2 2 4(τ − τ ) Wilmott magazine 78 . 3/2 8π σ 2σ 2 ˜ 2 = 0. 0 τ 1 1 1 √ √ √ exp (c) dτ a τ −τ τ From this follows Result 1. and a and c are as above. t) = w(x. hence the e−r(T−t0 ) term at the front. t) + G(S. T) = I2 . I.
2 and τ = σ2 (T − u). M 2005 Static hedging of barrier options under local volatility with r 6 = q . There is again a solution of the form v(S. where α = µ − 1 σ 2 − r + D + 1 σ 2. 2 + 1 4 x− 2ατ 2ατ σ2 x− + 2 2 σ (σ (τ − τ ) + σ 2 τ ) ˜ σ2 2 2 1 σ σ 1 + + 2 σ2 τ ˜ τ −τ σ (τ − τ ) + σ 2 τ ˜ . ■ Natenberg. There is one additional subtlety in the derivations and that concerns the expirations. J. 2 τ = σ2 (T − s). σ − 4σ 2 τ (τ − τ ) ˜ 2 x − 1 2 1 σ − r + D + σ2 ˜ 2 2 σ2 1 + 2 τ −τ σ τ ˜ . σj and σk and two expirations. ■ Schoutens. We must consider the general case Tj = Tk . S 1994 Option Volatility and Pricing. ■ Henrard. Risk and Insurance 0310002. Economics Working Paper Archive at WUSTL. P & Verma. 2 and evaluating at S = S0 . ■ Carr. The subscript ‘k’ is then associated with the gamma term. t) = I2 + 2I H(S. PhD thesis Bristol University. ■ Carr. ■ Forde. Simons. Tj and Tk . REFERENCES ■ Black. Result 3: Expectation. t). t = t0 . ■ Dupire. and the subscript ‘j’ is associated with the integral and so appears in the integrand. For this reason the s integral is over t0 to min(Tj . portfolio of options − 2 (x − x )2 4(τ − τ ) The manipulations and calculations required for the analysis of the portfolio variance are similar to that for a single contract. The easiest way to apply this is to use the convention that the gammas are zero after expiration. M 1973 The pricing of options and corporate liabilities. The integrations in (6) must only be taken over the intervals up until the options have expired. in full. We will find that the solution for the variance is the sum of terms satisfying diffusion equations with source terms like in Equation (11). and so appears outside the integral in the equivalent of (11). Derivatives. where d= 1 σ2 1 1 1 1 σ2 + + 4 σ2 τ ˜ 4τ −τ 4 σ 2 (τ − τ ) + σ 2 τ ˜ and g=− σ2 4σ 2 τ ˜ σ2 σ 2τ ˜ x− 2ατ σ2 2 − σ2 4(σ 2 (τ − τ ) + σ 2 τ ) ˜ x− 2ατ σ2 2 The main differences are that we have to carry around two implied ˜ ˜ volatilities. gives the required Result 2. W Wilmott magazine 79 . ACKNOWLEDGMENTS We would like to thank Hyungsok Ahn and Ed Thorp for their input on the practical application of our results and on portfolio optimization and Peter Carr for his encyclopedic knowledge of the literature.TECHNICAL ARTICLE 3 This exponent is. This can be written in the form −d(x + f )2 + g. A 2005 Potential Problems in Market Models of Implied Volatility. I. M 2003 Parameter risk in the Black and Scholes model. − 1 σ2 4τ σ 2 ˜ x − 2 σ2 µ− 1 2 2 σ τ + 2 2 σ 2τ σ2 µ− r−D− 1 2 σ τ ˜ 2 2 Result 4: Variance. J 2004 A perfect calibration! Now what? Wilmott magazine March 2004 66–78. portfolio of options This expression follows from the additivity of expectations. Presented at Global Derivatives 2005 Paris. Journal of Political Economy 81 637–59. P 2005 FAQs in option pricing theory. Working Paper. t) + G(S. and modelindependent hedging of Asian options. F & Scholes. d Changing variables to τ = σ2 (T − t). W. McGraw–Hill. B 2005 Exploring Volatility Derivatives: New Advances in Modelling. Tk ). ˜ 2 2 Using Equation (7) we end up with ˜ 1 E2 σ 2 − σ 2 e−2r(T−t0 ) 4π 3/2 σ 2σ 2 ˜ τ τ 1 √ √ √ √ √ τ τ τ −τ τ −τ a 0 0 2 π exp(g)dτ dτ . E & Tistaert.