Professional Documents
Culture Documents
May 2007
Abstract
We review here the theoretical approaches as well as the practical use
of volatility derivatives. A clear focus will be made on variance swaps,
for historical and logical - the most liquid of any volatility instruments -
reasons. Instead of going into all the details of the calculations, we will
try to point out the useful papers on different subtopics.
Contents
1 Variance and volatility swaps 1
1.1 Replication via Vanilla options . . . . . . . . . . . . . . . . . . . 2
1.2 Comments on the replication result . . . . . . . . . . . . . . . . . 3
3 Volatility derivatives 9
3.1 Options on realised variance . . . . . . . . . . . . . . . . . . . . . 9
3.2 VIX and options on the VIX . . . . . . . . . . . . . . . . . . . . 9
3.3 Forward-Started options and Cliquets . . . . . . . . . . . . . . . 10
3.4 Advanced Volatility Derivatives . . . . . . . . . . . . . . . . . . . 11
∗ a.jacquier@ems.bbk.ac.uk.
1
1 Variance and volatility swaps
Variance swaps are products with the following payoff at maturity :
1 T 2
Π=N σ dt − K
T 0 t
Integrating twice by parts and using the Put-Call parity relationship (we refer
to formula (11.1) for details), we obtain :
F ∞
E [g (ST ) /Ft ] = g (F ) + P (K) g (K) dK + C (K) g (K) dK
0 F
we eventually get
∞
T S∗
1 2 S0 rT S∗ dK dK
E σt2 dt = rT − e − 1 − log +e rT
P (K) + C (K)
T 0 T S∗ S0 0 K2 S∗ K
2
3
– Gatheral approach
Interestingly, Gatheral (see in the book, page 139) points out two
things: first, that the fair value of the variance swap can indeed be
expressed as a weighted average of implied variance ; then he presents
a different results for the impact of the skew : according to him, the
slope of the skew has no impact, whereas the curvature of the smile
does affect (in the same direction) the price of the variance swap. He
indeed gets the following formula :
1 T 2
E σ dt = dzN (z) σBS
2
(z)
T 0 t R
4
This means that a big move upwards (J > 0) leads to a loss for the variance
swap seller, and vice versa. For more details on this, see [14] (Section 5).
• Convexity adjustment for volatility swaps
Volatility√ swaps look like variance swap. The payoff of a volatility swap
isN
√
vol ( V − Kvol ). At first, we can use the following approximation :
E x ≈ E(x). Therefore, we can replicate a volatility swap (Kvol ,Nvol )
using a variance swap (Kvar ,Nvar ), where Nvol is noted in units of currency
per volatility point, and Nvar in units of currency per variance point. We
must take care that : 2
Kvar = Kvol
Nvol
Nvar = 2Kvol
But, in fact, there is a convexity bias between the volatility swap and the
variance swap due to the non linearity of the square root function.
As explained in Brockhaus and Long [1], we can have a much better ap-
proximation of a volatility swap using the results found
√ with a variance
swap. Given the square-root function F : x → x, the second order
Taylor expansion around x0 gives us
1
F (x) ≈ F (x0 ) + F (x0 )(x − x0 ) + F (x0 )(x − x0 )2
2
√ x − x0 1 (x − x0 )2
F (x) ≈ x0 + √ −
2 x0 8 x30
x + x0 (x − x0 )2
F (x) ≈ √ −
2 x0 8 x30
Applying this formula with x = v and x0 = E[v] and taking expectations,
we have
√ V[v]
E v ≈ E[v] −
8 E3 [v]
This formula gives us the approximate value of the convexity bias, which
is thus equal to √V[v]3 .
8 E [v]
The convexity bias needs either a specified model for the dynamics of
the volatility or an assumption on the level of the forward volatility and
variance. In terms of replication, there is no static hedging. But, as in the
tradional Black-Scholes framework, we can produce a dynamic hedging
for these volatility swaps, using variance swaps all along the life time of
the swaps so that it becomes instantaneously independent on the moves
5
in the volatility. Mathematically, we want to find (α, β) minimizing the
function Φ defined as :
2
Φ (α, β) = E σt − ασt2 − β
This two-equalities result hence needs the forward levels of volatility and
variance. This explains why variance swaps are nowadays much more
liquid than volatility swaps. Furthermore, despite the fact that we can’t
calculate a mark-to-market price for volatility swap without a specified
stochastic volatility model, Morokoff, Akesson and Zhou established in
[18] the existence of lower and upper bounds for such products, using
arbitrage arguments. Carr and Lee [5] also found an interesting result
: the volatility swap admits a robust supereplication using a variance
swap and has a lower bound equal to the at-the-money implied volatility
(ATMIV). They also provide an approximation using this ATMIV and the
total variance which is
a2 +σ̄2
4+ 02 0T
E0 σ̄T ≈ a0
4 + a20 T
where a0 represents the at-the-money implied volatility at time 0.
In a similar way to variance swaps, the hedge for volatility swaps isn’t
static but has to be dynamically rebalanced using the log-contracts. How-
ever, the large number of options needed to synthetically create this hedge
makes the transaction costs too much.
For practical as well as technical details on the convexity adjustment, see
[21].
• P&L and ”Cash” Gamma
Let us consider an option Vt (for example a Vanilla Call option or a Vari-
ance Swap). Using Black-Scholes method for constructing a delta-hedged
portfolio, one can easily show that the P&L of such a portfolio on the
period [t, t + dt] reads
2
1 2 dSt
P &L[t,t+dt] = ΓSt − σt dt
2
2 St
6
Proof : We briefly sketch the proof
We have
1
P &L[t,t+dt] = Delta P&L+Gamma P&L+Theta P&L = ∆ (∆S)+ Γ (∆S)2 +Θdt
2
But Θ ≈ − 12 ΓS 2 σimp
2 2
, where σimp stands for the implied volatility. Hence,
when delta-hedging the portfolio,
2
1 2
1 ∆S
P &L[t,t+dt] = Γ (∆S) − σimp 2
S 2 dt = S 2 Γ − σimp
2
dt
2 2 S
The first-term in the bracket corresponds to the realised variance, and the
second term to the implied variance. For more insights on this, see [4]
(part 8)
Now, the Gamma of the Variance Swap reads (see [17] for the greeks of a
Variance Swap)
2
ΓVarSwap = 2
St T
Plugging this into the P&L formula, we find
2
1 ∆S
P &L[t,t+dt] = − σimp dt
2
T S
This means that the P&L of a Variance Swap over a small period of time
does not depend on the level of the stock price. This is why it is called a
constant ”Cash” Gamma. In fact, the P&L, as we can see it, only depends
on the difference between implied and realised variance.
• A finite-difference approach for thepricing of Variance Swaps has also been
developped, see [16].
7
With this definition, Gamma Swaps will be replicated exactly like Variance
1
Swaps, but with a weighting scheme equal to K instead of K12 . Practically
speaking, Gamma Swaps have a constant ”Share” Gamma , and, as they are
weighted by the level of the stock, they do take into account the possibility of
jumps.
We refer to [3] for details on this product. Its replication is very similar to that
of the Gamma Swap.
From a pricing point of view, the replication scheme is almost the same than the
one used for pure Variance Swaps : the weights are identical, but the integrals
are truncated at the levels specified by the barriers(down or up).
General references
8
• A.Sepp, Variance swaps under no conditions, (Risk, March 2007)
• P.Carr, K.Lewis, Corridor variance swaps, [7]
3 Volatility derivatives
3.1 Options on realised variance
The natural extension of Variance Swaps is an option on the realised variance,
basically a call option, with payoff
1 T 2
σ dt − K
T 0 t
+
9
Carr & Lee [5] found a replication strategy for such an option under zero cor-
relation between the spot price and the volatility. In particular, they were able
to replicate any payoff of the form of the exponential of the quadratic variation.
See also [6] for very recent results on the replication of volatility derivatives,
using variance and volatility swaps as hedging instruments.
For pricing methods of options on realised variance under Levy processes, we
refer the reader to [9].
Where
σ = V100
IX
F : forward index level derived from index option prices
Ki : Strike of the ith option
Q (Ki ) : Midpoint of the bid-ask spread for an option with strike Ki
Now, let us consider a stock price with spot volatility σt . Then one can derive
the following formula :
t+τ
Q 1
V IXt = Et
2 2
σs ds
τ t
10
More advanced results on lower and upper bounds for VIX Futures have been
obtained in [8] and [11]. In particular, the bounds obtained by Carr and Wu in
[8] can be expressed in terms of a portfolio of European options (for the upper
bound) and of a Forward-tart ATM call option (for the lower bound), which are
directly observable.
References
[1] O. Brockhaus, D. Long. Volatility swaps made simple, Risk Magazine, Jan-
uary, 2000.
[2] H. Buehler. Stochastic volatility models and products, Risk Training Course,
Hong Kong, 2004
[3] H. Buehler. Volatility Markets, Consistent modeling, hedging and practical
implementation, PhD Thesis, 2006
[4] P. Carr. FAQ’s in option pricing theory, Courant Institute, NYU, 2002
[5] P. Carr, R. Lee. Robust replication of volatility derivatives, Courant Insti-
tute, NYU, Stanford University, 2003
[6] P. Carr, R. Lee. Realized Volatility and Variance: Options via Swaps, Risk,
2007
[7] P. Carr, R. Lee. Hedging variance options on continuous semimartingales,
Courant Institute, NYU, Stanford University, 2006
[7] P. Carr, K. Lewis. Corridor variance swaps, Risk, February 20004, 2006
[8] P. Carr, L. Wu. A tale of two indices, Journal of Derivatives, Spring 2006.
[9] P. Carr, H. Geman, D. Madan, M. Yor. Pricing options on realised variance,
Finance and Stochastics, 2005, issue 4
[10] N. Chriss, W. Morokoff. Market Risk for Volatility and Variance Swaps,
Risk, July 1999
11
[11] B. Dupire. Model free results on volatility derivatives, Bloomberg, 2006
[12] Chicago Board Options Exchange. VIX-CBOE Volatility Index, 2003,
http://www.cboe.com/micro/vix/vixwhite.pdf
[13] P. Friz, J. Gatheral. Valuation of volatility derivatives as an inverse prob-
lem, Quantitative Finance, 2005
[14] K. Demeterfi, E. Derman, M. Kamal, J. Zou. More than you ever wanted
to know about variance swaps, Goldman Sachs, Quantitative Strategies Re-
search Notes, 1999.
[15] Y. Lin. Pricing VIX Futures on Affine Stochastic Volatility Models with Si-
multaneous Sate-Dependent Jumps both in the S&P 00 Price and Variance
Processes, Working paper, 2006.
[16] T. Little, V. Pant. A finite difference method for variance swap, Working
paper, 2001.
[17] D.E. Kuenzi. Variance swaps and non constant Vega, Risk, October 2005.
[18] W. Morokoff, F. Akesson, Y. Zhou. Risk management of volatility and vari-
ance swaps, Risk Quantitative Modeling Notes, Goldman Sachs, 1999.
[19] A. Neuberger. The log contract, Journal of Portfolio Management, 20(2):74-
80, 1994.
[20] W. Schoutens. Moment swaps, Working paper, 2005.
[21] C. Youssfi. Convexity adjustment for volatility swaps, Merril Lynch presen-
tation, 2005.
[22] Y. Zhu, J.E. Zhang. Variance term structure and VIX Futures pricing,
International Journal of Theoretical and Applied Finance, Vol. 10, No. 1
(2007) 111-127.
12