Gs Volatility Swaps

Goldman
Sachs
March 1999
Quantitative Strategies
Research Notes
More
Than You Ever Wanted To Know*
About
Volatility Swaps
Kresimir Demeterfi
Emanuel Derman
Michael Kamal
Joseph Zou
_____________
* But Less Than Can Be Said
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
Copyright 1999 Goldman, Sachs & Co. All rights reserved.

This material is for your private information, and we are not soliciting any action based upon it. This report is not to be construed as
an offer to sell or the solicitation of an offer to buy any security in any jurisdiction where such an offer or solicitation would be
illegal. Certain transactions, including those involving futures, options and high yield securities, give rise to substantial risk and are
not suitable for all investors. Opinions expressed are our present opinions only. The material is based upon information that we
consider reliable, but we do not represent that it is accurate or complete, and it should not be relied upon as such. We, our affiliates,
or persons involved in the preparation or issuance of this material, may from time to time have long or short positions and buy or
sell securities, futures or options identical with or related to those mentioned herein.
This material has been issued by Goldman, Sachs & Co. and/or one of its affiliates and has been approved by Goldman Sachs
International, regulated by The Securities and Futures Authority, in connection with its distribution in the United Kingdom and by
Goldman Sachs Canada in connection with its distribution in Canada. This material is distributed in Hong Kong by Goldman Sachs
(Asia) L.L.C., and in Japan by Goldman Sachs (Japan) Ltd. This material is not for distribution to private customers, as defined by
the rules of The Securities and Futures Authority in the United Kingdom, and any investments including any convertible bonds or
derivatives mentioned in this material will not be made available by us to any such private customer. Neither Goldman, Sachs & Co.
nor its representative in Seoul, Korea is licensed to engage in securities business in the Republic of Korea. Goldman Sachs
International or its affiliates may have acted upon or used this research prior to or immediately following its publication. Foreign
currency denominated securities are subject to fluctuations in exchange rates that could have an adverse effect on the value or price
of or income derived from the investment. Further information on any of the securities mentioned in this material may be obtained
upon request and for this purpose persons in Italy should contact Goldman Sachs S.I.M. S.p.A. in Milan, or at its London branch
office at 133 Fleet Street, and persons in Hong Kong should contact Goldman Sachs Asia L.L.C. at 3 Garden Road. Unless governing
law permits otherwise, you must contact a Goldman Sachs entity in your home jurisdiction if you want to use our services in
effecting a transaction in the securities mentioned in this material.
Note: Options are not suitable for all investors. Please ensure that you have read and understood the current options
disclosure document before entering into any options transactions.
-2
Goldman
Sachs
SUMMARY
Volatility swaps are forward contracts on future realized
stock volatility. Variance swaps are similar contracts on variance, the square of future volatility. Both of these instruments
provide an easy way for investors to gain exposure to the
future level of volatility.
Unlike a stock option, whose volatility exposure is contaminated by its stock-price dependence, these swaps provide pure
exposure to volatility alone. You can use these instruments to
speculate on future volatility levels, to trade the spread
between realized and implied volatility, or to hedge the volatility exposure of other positions or businesses.
In this report we explain the properties and the theory of both
variance and volatility swaps, first from an intuitive point of
view and then more rigorously. The theory of variance swaps
is more straightforward. We show how a variance swap can be
theoretically replicated by a hedged portfolio of standard
options with suitably chosen strikes, as long as stock prices
evolve without jumps. The fair value of the variance swap is
the cost of the replicating portfolio. We derive analytic formulas for theoretical fair value in the presence of realistic volatility skews. These formulas can be used to estimate swap
values quickly as the skew changes.
We then examine the modifications to these theoretical
results when reality intrudes, for example when some necessary strikes are unavailable, or when stock prices undergo
jumps. Finally, we briefly return to volatility swaps, and show
that they can be replicated by dynamically trading the more
straightforward variance swap. As a result, the value of the
volatility swap depends on the volatility of volatility itself.
_________________
Kresimir Demeterfi
Emanuel Derman
Michael Kamal
Joseph Zou
(212) 357-4611
(212) 902-0129
(212) 357-3722
(212) 902-9794
_________________
Acknowledgments: We thank Emmanuel Boussard, Llewellyn Connolly, Rustom Khandalavala, Cyrus Pirasteh, David
Rogers, Emmanuel Roman, Peter Selman, Richard Sussman,
Nicholas Warren and several of our clients for many discussions and insightful questions about volatility swaps.
_________________
Editorial: Barbara Dunn
-1
Goldman
Sachs

Table of Contents
INTRODUCTION ......................................................................................... 1
Volatility Swaps ........................................................................... 1
Who Can Use Volatility Swaps? ................................................. 2
Variance Swaps ............................................................................ 3
Outline .......................................................................................... 4
I. REPLICATING VARIANCE SWAPS: FIRST STEPS ...................................... 6
The Intuitive Approach ............................................................... 6
Trading Realized Volatility with a Log Contract ..................... 11
The Vega, Gamma and Theta of a Log Contract ...................... 11
Imperfect Hedges ...................................................................... 13
The Limitations of the Intuitive Approach .............................. 13
II. REPLICATING VARIANCE SWAPS: GENERAL RESULTS ....................... 15
Valuing and Pricing the Variance Swap.................................. 17
III. AN EXAMPLE OF A VARIANCE SWAP ................................................ 20
IV. EFFECTS OF THE VOLATILITY SKEW ................................................ 23
Skew Linear in Strike ............................................................... 23
Skew Linear in Delta ................................................................ 25
V. PRACTICAL PROBLEMS WITH REPLICATION ....................................... 27
Imperfect Replication Due to Limited Strike Range ............... 27
The Effect of Jumps on a Perfectly Replicated Log Contract .. 29
The Effect of Jumps When Replicating With a
Finite Strike Range.............................................................. 32
VI. FROM VARIANCE TO VOLATILITY CONTRACTS ................................. 33
Dynamic Replication of a Volatility Swap ............................... 34
CONCLUSIONS AND FUTURE INNOVATIONS ............................................ 36
APPENDIX A: REPLICATING LOGARITHMIC PAYOFFS .............................. 37
APPENDIX B: SKEW LINEAR IN STRIKE .................................................. 40
APPENDIX C: SKEW LINEAR IN DELTA ................................................... 44
APPENDIX D: STATIC AND DYNAMIC REPLICATION OF A
VOLATILITY SWAP .................................................. 48
REFERENCES .......................................................................................... 50
Goldman
Sachs
INTRODUCTION
A stocks volatility is the simplest measure of its riskiness or uncertainty. Formally, the volatility R is the annualized standard deviation
of the stocks returns during the period of interest, where the subscript
R denotes the observed or realized volatility. This note is concerned
with volatility swaps and other instruments suitable for trading volatility1.
Why trade volatility? Just as stock investors think they know something about the direction of the stock market, or bond investors think
they can foresee the probable direction of interest rates, so you may
think you have insight into the level of future volatility. If you think
current volatility is low, for the right price you might want to take a
position that profits if volatility increases.
Investors who want to obtain pure exposure to the direction of a stock
price can buy or sell short the stock. What do you do if you simply want
exposure to a stocks volatility?
Stock options are impure: they provide exposure to both the direction
of the stock price and its volatility. If you hedge the options according
to Black-Scholes prescription, you can remove the exposure to the stock
price. But delta-hedging is at best inaccurate because the real world
violates many of the Black-Scholes assumptions: volatility cannot be
accurately estimated, stocks cannot be traded continuously, transactions costs cannot be ignored, markets sometimes move discontinuously and liquidity is often a problem. Nevertheless, imperfect as they
are, until recently options were the only volatility vehicle available.
Volatility Swaps
The easy way to trade volatility is to use volatility swaps, sometimes

called realized volatility forward contracts, because they provide pure
exposure to volatility (and only to volatility).
A stock volatility swap is a forward contract on annualized volatility.
Its payoff at expiration is equal to
( R K vol ) N
(EQ 1)
where R is the realized stock volatility (quoted in annual terms) over

the life of the contract, Kvol is the annualized volatility delivery price,
and N is the notional amount of the swap in dollars per annualized volatility point. The holder of a volatility swap at expiration receives N
dollars for every point by which the stocks realized volatility R has
1. For a discussion of volatility as an asset class, see Derman, Kamal,

Kani, McClure, Pirasteh, and Zou (1996).
Goldman
Sachs
exceeded the volatility delivery price Kvol. He or she is swapping a fixed

volatility Kvol for the actual (floating) future volatility R.
The delivery price Kvol is typically quoted as a volatility, for example
30%. The notional amount is typically quoted in dollars per volatility
point, for example, N = $250,000/(volatility point). As with all forward
contracts or swaps, the fair value of volatility at any time is the delivery price that makes the swap currently have zero value.
The procedure for calculating the realized volatility should be clearly
specified with respect to the following aspects:
The source and observation frequency of stock or index prices for
example, using daily closing prices of the S&P 500 index;
The annualization factor in moving from daily or hourly observations to annualized volatilities for example, using 260 business
days per year as a multiplicative factor in computing annualized
variances from daily returns; and
Whether the standard deviation of returns is calculated by subtracting the sample mean from each return, or by assuming a zero mean.
The zero mean method is theoretically preferable, because it corresponds most closely to the contract that can be replicated by options
portfolios. For frequently observed prices, the difference is usually
negligible.
Who Can Use Volatility
Swaps?
Volatility has several characteristics that make trading attractive. It is

likely to grow when uncertainty and risk increase. As with interest
rates, volatilities appear to revert to the mean; high volatilities will
eventually decrease, low ones will likely rise. Finally, volatility is often
negatively correlated with stock or index level, and tends to stay high
after large downward moves in the market. Given these tendencies,
several uses for volatility swaps follow.
Directional Trading of Volatility Levels. Clients who want to speculate on the future levels of stock or index volatility can go long or
short realized volatility with a swap. This provides a much more direct
method than trading and hedging options. For example, if you foresee a
rapid decline in political and financial turmoil after a forthcoming election, a short position in volatility might be appropriate.
Trading the Spread between Realized and Implied Volatility
Levels. As we will show later, the fair delivery price Kvol of a volatility
swap is a value close to the level of current implied volatilities for
options with the same expiration as the swap. Therefore, by unwinding
Goldman
Sachs
the swap before expiration, you can trade the spread between realized
and implied volatility.
Hedging Implicit Volatility Exposure. There
nesses that are implicitly short volatility:
are
several
busi-
Risk arbitrageurs or hedge funds often take positions which assume

that the spread between stocks of companies planning mergers will
narrow. If overall market volatility increases, the merger may
become less likely and the spread may widen.
Investors following active benchmarking strategies may require
more frequent rebalancing and greater transactions expenses during volatile periods.
Portfolio managers who are judged against a benchmark have tracking error that may increase in periods of higher volatility.
Equity funds are probably short volatility because of the negative
correlation between index level and volatility. As global equity correlations have increased, diversification across countries has become a
less effective portfolio hedge. Since volatility is one of the few
parameters that tends to increase during global equity declines, a
long volatility hedge may be appropriate, especially for financial
businesses.
Variance Swaps
Although options market participants talk of volatility, it is variance,

or volatility squared, that has more fundamental theoretical significance. This is so because the correct way to value a swap is to value the
portfolio that replicates it, and the swap that can be replicated most
reliably (by portfolios of options of varying strikes, as we show later) is
a variance swap.
A variance swap is a forward contract on annualized variance, the
square of the realized volatility. Its payoff at expiration is equal to
2
( R K var ) N
(EQ 2)
where R is the realized stock variance (quoted in annual terms) over

the life of the contract, Kvar is the delivery price for variance, and N is
the notional amount of the swap in dollars per annualized volatility
point squared. The holder of a variance swap at expiration receives N
2
dollars for every point by which the stocks realized variance R has
exceeded the variance delivery price Kvar.
Goldman
Sachs
Though theoretically simpler, variance swaps are less commonly

traded, and so their quoting conventions vary. The delivery price Kvar
can be quoted as a volatility squared, for example (30%)2. Similarly, for
example, the notional amount can be expressed as $100,000/(one volatility point)2. The fair value of variance is the delivery price that makes
the swap have zero value.
Outline
Most of this note will focus on the theory and properties of variance
swaps, which provide similar volatility exposure to straight volatility
swaps. Because of its fundamental role, variance can serve as the basic
building block for constructing other volatility-dependent instruments.
At the end, we will return to a discussion of the additional risks
involved in replicating and valuing volatility swaps.
Section I presents an intuitive, Black-Scholes-based account of the fundamental strategy by which a variance swap can be replicated and valued. First, we show that the hedging of a (slightly) exotic stock option,
the log contract, provides a payoff equal to the variance of the stocks
returns under a fairly wide set of circumstances. Then, we explain how
this exotic option itself can be replicated by a portfolio of standard
stock options with a range of strikes, so that their market prices determine the cost of the variance swap. We also provide insight into the
swap by showing, from a variety of viewpoints, how the apparently
complex hedged log contract produces an instrument with the simple
constant exposure to the realized variance of a variance swap.
Section II derives the same results much more rigorously and generally, without depending on the full validity of the Black-Scholes model.
Though more difficult, this presentation is capable of much greater
generalization.
In Section III, we provide a detailed numerical example of the valuation of a variance swap. Some practical issues concerning the choice of
strikes are also discussed.
The fair value of the variance swap is determined by the cost of the
replicating portfolio of options. This cost, especially for index options, is
significantly affected by the volatility smile or skew. Therefore, we
devote Section IV to the effects of the skew. In particular, for a skew
linear in strike or linear in delta, we derive theoretical formulas that
allow us to simply determine the approximate effect of the skew on the
fair value of index variance swaps, without detailed numerical computation. The formulas and the intuition they provide are beneficial in
rapidly estimating the effect of changes in the skew on swap values.
Goldman
Sachs
The fair value of a variance swap is based on (1) the ability to replicate
a log contract by means of a portfolio of options with a (continuous)
range of strikes, and (2) on classical options valuation theory, which
assumes continuous stock price evolution. In practice, not all strikes
are available, and stock prices can jump. Section V discusses the effects
of these real limitations on pricing.
Finally, Section VI explains the risks involved in replicating a volatility
contract. Since variance can be replicated relatively simply, it is useful
to regard volatility as the square root of variance. From this point of
view, volatility is itself a square-root derivative contract on variance.
Thus, a volatility swap can be dynamically hedged by trading the
underlying variance swap, and its value depends on the volatility of the
underlying variance that is, on the volatility of volatility.
Four appendices cover some more advanced mathematics. In Appendix
A, we derive the details of the replication of a log contract by means of
a continuum of option strikes. It also shows how the replication can be
approximated in practice when only a discrete set of strikes are available. In Appendix B, we derive the approximate formulas for the value
of an index variance contract in the presence of a volatility skew that
varies linearly with strike. In Appendix C, we derive the analogous formulas for a skew varying linearly with the delta exposure of the
options. Appendix D provides additional insight into the static and
dynamic hedging of a volatility swap using the variance as an underlyer.
Goldman
Sachs
REPLICATING
VARIANCE SWAPS:
FIRST STEPS
In this section, we explain the replicating strategy that captures realized variance. The cost of implementing that strategy is the fair value
of future realized variance.
The Intuitive
Approach: Creating a
Portfolio of Options
Whose Variance Sensitivity is Independent
of Stock Price
We approach variance replication by building on the readers assumed

familiarity with the standard Black-Scholes model. In the next section,
we shall provide a more general proof that you can replicate variance,
even when some of the Black-Scholes assumptions fail, as long as the
stock price evolves continuously that is, without jumps.
We ease the development of intuition by assuming here that the riskless interest rate is zero. Suppose at time t you own a standard call
option of strike K and expiration T, whose value is given by the BlackScholes formula C BS ( S, K , ) , where S is the current stock price,
is the time to expiration (T t), is the return volatility of the stock,
2
is the stocks variance, and v = is the total variance of the

stock to expiration. (We have written the option value as a function of
in order to make clear that all its dependence on both volatility
and time to expiration is expressed in the combined variable .)
We will call the exposure of an option to a stocks variance V ; it measures the change in value of the position resulting from a change in
variance2. Figure 1a shows a graph of how V varies with stock price S,
for each of three different options with strikes 80, 100 and 120. For
each strike, the variance exposure V is largest when the option is at
the money, and falls off rapidly as the stock price moves in or out of the
money. V is closely related to the time sensitivity or time decay of the
option, because, in the Black-Scholes formula with zero interest rate,
2
options values depend on the total variance .

If you want a long position in future realized variance, a single option
is an imperfect vehicle: as soon as the stock price moves, your sensitivity to further changes in variance is altered. What you want is a portfo-
2.
Here, we define the sensitivity V =
C BS
S exp ( d 1 2 )
= ----------- -------------------------------- , where
2
2
log ( S K ) + ( ) 2
d 1 = ------------------------------------------------------ . We will sometimes refer to V as variance

vega. Note that d1 depends only on the two combinations S/K and
. V decreases extremely rapidly as S leaves the vicinity of the
strike K.
Goldman
Sachs
lio whose sensitivity to realized variance is independent of the stock

price S.
To obtain a portfolio that responds to volatility or variance independent of moves in the stock price, you need to combine options of many
strikes. What combination of strikes will give you such undiluted variance exposure?
Figure 1b shows the variance exposure for the portfolio consisting of all
three option strikes in Figure 1a. The dotted line represents the sum of
equally weighted strikes; the solid line represents the sum with
weights in inverse proportion to the square of their strike. Figures 1c, e
and g show the individual sensitivities to variance of increasing numbers of options, each panel having the options more closely spaced. Figures 1d, f and h show the sensitivity for the equally-weighted and
strike-weighted portfolios. Clearly, the portfolio with weights inversely
2
proportional to K produces a V that is virtually independent of stock

price S, as long as S lies inside the range of available strikes and far
from the edge of the range, and provided the strikes are distributed
evenly and closely.
Appendix A provides a mathematical derivation of the requirement
2
that options be weighted inversely proportional to K in order to

achieve constant V. You can also understand this intuitively. As the
stock price moves up to higher values, each additional option of higher
strike in the portfolio will provide an additional contribution to V proportional to that strike. This follows from the formula in footnote 2,
and you can observe it in the increasing height of the V -peaks in Figure 1a. An option with higher strike will therefore produce a V contribution that increases with S. In addition, the contributions of all
options overlap at any definite S. Therefore, to offset this accumulation
of S-dependence, one needs diminishing amounts of higher-strike
2
options, with weights inversely proportional to K .

If you own a portfolio of options of all strikes, weighted in inverse proportion to the square of the strike level, you will obtain an exposure to
variance that is independent of stock price, just what is needed to trade
variance. What does this portfolio of options look like, and how does
trading it capture variance?
Consider the portfolio ( S, ) of options of all strikes K and a single
2
expiration , weighted inversely proportional to K . Because out-of-
Goldman
Sachs
FIGURE 1. The variance exposure, V, of portfolios of call options of different

strikes as a function of stock price S. Each figure on the left shows the
individual V i contributions for each option of strike Ki. The corresponding
figure on the right shows the sum of the contributions, weighted two
different ways; the dotted line corresponds to an equally-weighted sum of
options; the solid line corresponds to weights inversely proportional to K2,
and becomes totally independent of stock price S inside the strike range
strikes: 80,100,120
equally
weighted
(a)
(b)
20
60
100
140
180
weighted
inversely
proportional
to square
of strike
20
60
100
140
180
20
60
100
140
180
20
60
100
140
180
20
60
100
140
180
strikes 60 to 140
spaced 20 apart
(c)
(d)
20
60
100
140
180
strikes 60 to 140
spaced 10 apart
(e)
(f)
20
60
100
140
180
strikes 20 to 180
spaced 1 apart
(g)
(h)
20
60
100
140
180
Goldman
Sachs
the-money options are generally more liquid, we employ put options

P ( S, K , ) for strikes K varying continuously from zero up to some
reference price S*, and call options C ( S, K , ) for strikes varying
continuously from S* to infinity3. You can think of S* as the approximate at-the-money forward stock level that marks the boundary
between liquid puts and liquid calls.
At expiration, when t = T, one can show that the sum of all the payoff
values of the options in the portfolio is simply
ST S*
S T
( S T , 0 ) = -------------------- log -------
S*
S*
(EQ 3)
where log( ) denotes the natural logarithm function, and ST is the terminal stock price.
Similarly, at time t you can sum all the Black-Scholes options values to
show that the total portfolio value is
2
S S*
S

( S, ) = ---------------- log ------ + -------- S *
S*
2
(EQ 4)
where S is the stock price at time t. Note how little the value of the
portfolio before expiration differs from its value at expiration at the
same stock price. The only difference is the additional value due to half
2
the total variance .

Clearly, the variance exposure of is
V = --2
(EQ 5)
To obtain an initial exposure of $1 per volatility point squared, you

need to hold (2/T) units of the portfolio . From now on, we will use to
refer to the value of this new portfolio, namely
3. Formally, the expression for the portfolio is given by

( S, ) =
1
1
-------C
( S, K , ) + -------2 P ( S, K , )
2
K
K
K > S*
K < S*
Goldman
Sachs
2
2 S S*
S
( S, ) = ---- ---------------- log ------ + ----
T
T S*
S*
(EQ 6)
The first term in the payoff in Equation 6, (S S*)/S*, describes 1/S*

forward contracts on the stock with delivery price S*. It is not really an
option; its value represents a long position in the stock (value S) and a
short position in a bond (value S*), which can be statically replicated,
once and for all, without any knowledge of the stocks volatility. The
second term, log ( S S * ) , describes a short position in a log contract4
with reference value S*, a so-called exotic option whose payoff is proportional to the log of the stock at expiration, and whose correct hedging depends on the volatility of the stock. All of the volatility sensitivity
of the weighted portfolio of options we have created is contained in the
log contract.
Figure 2 graphically illustrates the equivalence between (1) the
summed, weighted payoffs of puts and calls and (2) a long position in a
forward contract and a short position in a log contract.
An example that illustrates the equivalence at expiration
between (1) (ST,0) the weighted sum of puts and calls, with weight
inversely proportional to the square of the strike, and (2) the payoff of a
long position in a forward contract and a short position in a log contract.
(a) Individual contributions to the payoff from put options with all integer
strikes from 20 to 99, and call options with all integer strikes from 100 to 180.
(b) The payoff of 1/100 of a long position in a forward contract with
delivery price 100 and one short position in a log contract with reference
value 100.
FIGURE 2.
ST
ST S*
-------------------- log ------S*
S*
( ST, 0 )
20
60
100
(a)
140
180
20
60
100
140
180
(b)
4. The log contract was first discussed in Neuberger (1994). See also Neuberger (1996).
10
Goldman
Sachs
Trading Realized
Volatility with a Log
Contract
For now, assume that we are in a Black-Scholes world where the

implied volatility I is the estimate of future realized volatility. If you
take a position in the portfolio , the fair value you should pay at time
t = 0 when the stock price is S0 is
S 0
2
2 S0 S*
0 = ---- ------------------- log ------- + I
T
S*
S *
At expiration, if the realized volatility turns out to have been R , the
initial fair value of the position captured by delta-hedging would have
been
S 0
2
2 S0 S*
0 = ---- ------------------- log ------- + R
T
S*
S *
The net payoff on the position, hedged to expiration, will be
2
payoff = ( R I )
(EQ 7)
Looking back at Equation 2, you will see that by rehedging the position
in log contracts, you have, in effect, been the owner of a position in a
2
variance swap with fair strike Kvar = I and face value $1. You will
have profited (or lost) if realized volatility has exceeded (or been
exceeded by) implied volatility.
The Vega, Gamma
and Theta of a Log
Contract
In Equation 6 we showed that, in a Black-Scholes world with zero

interest rates and zero dividend yield, the portfolio of options whose
variance vega is independent of the stock price S can be written
(T t) 2
2 S S*
S
( S, , t, T ) = ---- ---------------- log ------ + -----------------
S *
T
T S*
The ( S S * ) term represents a long position in the stock and a short
position in a bond, both of which can be statically hedged with no
dependence on volatility. In contrast, the log( ) term needs continual
dynamic rehedging. Therefore, let us concentrate on the log contract
term alone, whose value at time t for a logarithmic payoff at time T is
(T t) 2
2
S
L ( S, , t, T ) = ---- log ------ + -----------------
S *
T
T
(EQ 8)
11
Goldman
Sachs
The sensitivities of the value of this portfolio are precisely appropriate

for trading variance, as we now show.
The variance vega of the portfolio in Equation 8 is
Tt
V = ------------
T
(EQ 9)
The exposure to variance is equal to 1 at t = 0, and decreases linearly to

zero as the contract approaches expiration.
The time decay of the log contract, the rate at which its value changes
if the stock price remains unchanged as time passes, is
1 2
= ----
T
(EQ 10)
The contract loses time value at a constant rate proportional to its variance, so that at expiration, all the initial variance has been lost.
The log contracts exposure to stock price is
21
= ---- ---TS
shares of stock. That is, since each share of stock is worth S, you need a
constant long position in $(2/T) worth of stock to be hedged at any time.
The gamma of the portfolio, the rate at which the exposure changes as
the stock price moves, is
2 1
= ---- ------T S2
(EQ 11)
Gamma is a measure of the risk of hedging an option. The log contracts gamma, being the sum of the gammas of a portfolio of puts and
calls, is a smoother function of S than the sharply peaked gamma of a
single option.
Equations 10 and 11 can be combined to show that
2
1
+ --- S 2 = 0
2
(EQ 12)
Equation 12 is the essence of the Black-Scholes options pricing theory.

It states that the disadvantage of negative theta (the decrease in value
with time to expiration) is offset by the benefit of positive gamma (the
curvature of the payoff).
12
Goldman
Sachs
Imperfect Hedges
It takes an infinite number of strikes, appropriately weighted, to replicate a variance swap. In practice, this isnt possible, even when the
stock and options market satisfy all the Black-Scholes assumptions:
there are only a finite number of options available at any maturity. Figure 1 illustrates that a finite number of strikes fails to produce a uniform V as the stock price moves outside the range of the available
strikes. As long as the stock price remains within the strike range,
trading the imperfectly replicated log contract will allow variance to
accrue at the correct rate. Whenever the stock price moves outside, the
reduced vega of the imperfectly replicated log contract will make it less
responsive than a true variance swap.
Figure 3 shows how the variance vega of a three-month variance swap
is affected by imperfect replication. Figure 3a shows the ideal variance
vega that results from a portfolio of puts and calls of all strikes from
zero to infinity, weighted in inverse proportion to the strike squared.
Here the variance vega is independent of stock price level, and
decreases linearly with time to expiration, as expected for a swap
2
whose value is proportional to the remaining variance at any time.

Figure 3b shows strikes from $75 to $125, uniformly spaced $1 apart.
Here, deviation from constant variance vega develops at the tail of the
strike range, and the deviation is greater at earlier times. Finally, Figure 3c shows the vega for strikes from $20 to $200, spaced $10 apart.
Now, although the range of strikes is greater, the coarser spacing
causes the vega surface to develop corrugations between strike values
that grow more pronounced closer to expiration.
The Limitations of the
Intuitive Approach
A variance swap has a payoff proportional to realized variance. In this

section, assuming the Black-Scholes world for stock and options markets, we have shown that the dynamic, continuous hedging of a log contract produces a payoff whose value is proportional to future realized
variance. We have also shown that you can use a portfolio of appropriately weighted puts and calls to approximate a log contract.
The somewhat intuitive derivations in this section have assumed that
interest rates and dividend yields are zero, but it is not hard to generalize them. We have also assumed that all the Black-Scholes assumptions hold. In practice, in the presence of an implied volatility skew, it
is difficult to extend these argument clearly. Therefore, we move on to a
more general and rigorous derivation of the value of variance swaps
based on replication. Many of the results will be similar, but the conditions under which they hold, and the correct answers when they do not
hold, will be more easily understandable.
13
Goldman
Sachs
FIGURE 3. The variance vega, V , of a portfolio of puts and calls, weighted
(a)
inversely proportional to the square of the strike level, and chosen to

replicate a three-month variance swap. (a) An infinite number of strikes.
(b) Strikes from $75 to $125, uniformly spaced $1 apart. (c) Strikes from $20
to $200, uniformly spaced $10 apart.
140
120
100
0.2
80
0.1
60
(b)
140
120
100
0.2
80
0.1
60
(c)
140
120
100
0.2
14
80
0.1
60
Goldman
Sachs
REPLICATING
VARIANCE SWAPS:
GENERAL RESULTS
In the previous section, we explained how to replicate a variance swap

by means of a portfolio of options whose payoffs approximate a log contract. Although our explanation depended on the validity of the BlackScholes model, the result that the dynamic hedging of a log contract
captures realized volatility holds true more generally. The only
assumption we will make about the future underlyer evolution is that
it is diffusive, or continuous this means that no jumps are allowed.
(In a later section, we will consider the effects of discontinuous stock
price movements on the success of the replication.) Therefore, we
assume that the stock price evolution is given by
dS t
---------- = ( t, ... )dt + ( t, ... )dZ t
St
(EQ 13)
Here, we assume that the drift and the continuously-sampled volatility are arbitrary functions of time and other parameters. These
assumptions include, but are not restricted to, implied tree models in
which the volatility is a function ( S, t ) of stock price and time only.
For simplicity of presentation, we assume the stock pays no dividends;
allowing for dividends does not significantly alter the derivation.
The theoretical definition of realized variance for a given price history
is the continuous integral
1 T
V = ---- 2 ( t, ) dt
T 0
(EQ 14)
This is a good approximation to the variance of daily returns used in

the contract terms of most variance swaps.
Conceptually, valuing a variance forward contract or swap is no different than valuing any other derivative security. The value of a forward contract F on future realized variance with strike K is the
expected present value of the future payoff in the risk-neutral world:
F = E [ e rT ( V K ) ]
(EQ 15)
Here r is the risk-free discount rate corresponding to the expiration

date T, and E[ ] denotes the expectation.
The fair delivery value of future realized variance is the strike K var
for which the contract has zero present value:
K var = E [ V ]
(EQ 16)
15
Goldman
Sachs
If the future volatility in Equation 13 is specified, then one approach

for calculating the fair price of variance is to directly calculate the riskneutral expectation
T
1
K var = ---- E
2 ( t, ) dt
T
0
(EQ 17)
No one knows with certainty the value of future volatility. In implied

tree models5, the so-called local volatility ( S, t ) consistent with all
current options prices is extracted from the market prices of traded
stock options. You can then use simulation to calculate the fair variance Kvar as the average of the experienced variance along each simulated path consistent with the risk-neutral stock price evolution given
of Equation 13, where the drift is set equal to the riskless rate.
The above approach is good for valuing the contract, but it does not
provide insight into how to replicate it. The essence of the replication
strategy is to devise a position that, over the next instant of time, generates a payoff proportional to the incremental variance of the stock
during that time6.
By applying Itos lemma to log S t , we find
1
d ( log S t ) = --- 2 dt + dZ t
(EQ 18)
Subtracting Equation 18 from Equation 13, we obtain

dS t
1 2
---------- d ( log S t ) = --- dt
2
St
(EQ 19)
in which all dependence on the drift has cancelled. Summing Equation 19 over all times from 0 to T, we obtain the continuously-sampled
variance
5. See, for example, Derman and Kani (1994), Dupire (1994) and Derman,
Kani and Zou (1996).
6. This approach was first outlined in Derman, Kamal, Kani, and Zou
(1996). For an alternative discussion, see Carr and Madan (1998).
16
Goldman
Sachs
1 2
V ---- dt
T
(EQ 20)
T dS t
ST
2
= ------------ log ------T 0 St
S0
This mathematical identity dictates the replication strategy for variance. The first term in the brackets can be thought of as the net outcome of continuous rebalancing a stock position so that it is always
instantaneously long 1/S t shares of stock worth $1. The second term
represents a static short position in a contract which, at expiration,
pays the logarithm of the total return. Following this continuous rebalancing strategy captures the realized variance of the stock from inception to expiration at time T. Note that no expectations or averages have
been taken Equation 20 guarantees that variance can be captured no
matter which path the stock price takes, as long as it moves continuously.
Valuing and Pricing
the Variance Swap
Equation 20 provides another method for calculating the fair variance.

Instead of averaging over future variances, as in Equation 17, one can
take the expected risk-neutral value of the right-hand side of Equation
20 to obtain the cost of replication directly, so that
ST
T dS t
2
K var = ---- E
--------- log ------T
S0
0 St
(EQ 21)
The expected value of the first term in Equation 21 accounts for the
cost of rebalancing. In a risk-neutral world with a constant risk-free
rate r, the underlyer evolves according to:
dS t
--------- = rdt + ( t, )dZ
St
(EQ 22)
so that the risk-neutral price of the rebalancing component of the hedging strategy is given by
E
T dS t
0 -------St
= rT
(EQ 23)
This equation represents the fact that a shares position, continuously

rebalanced to be worth $1, has a forward price that grows at the riskless rate.
17
Goldman
Sachs
As there are no actively traded log contracts for the second term in
Equation 21, one must duplicate the log payoff, at all stock price levels
at expiration, by decomposing its shape into linear and curved components, and then duplicating each of these separately. The linear component can be duplicated with a forward contract on the stock with
delivery time T; the remaining curved component, representing the
quadratic and higher order contributions, can be duplicated using standard options with all possible strike levels and the same expiration
time T.
For practical reasons we want to duplicate the log payoff with liquid
options that is, with a combination of out-of-the-money calls for high
stock values and out-of-the-money puts for low stock values. We introduce a new arbitrary parameter S* to define the boundary between
calls and puts. The log payoff can then be rewritten as
ST
ST
S*
log ------- = log ------- + log ------S0
S*
S0
(EQ 24)
The second term log ( S * S 0 ) is constant, independent of the final stock

price ST, so only the first term has to be replicated.
The following mathematical identity, which holds for all future values
of ST, suggests the decomposition of the log-payoff:
ST S*
ST
log ------- = -------------------S*
S*
S*
1
-------- Max ( K S T , 0 ) d K
K2
- Max ( S T K , 0 ) d K
S ------K2
(forward contract)
(put options)
(EQ 25)
(call options)
Equation 25 represents the decomposition of a log payoff into a portfolio consisting of:
a short position in ( 1 S * ) forward contracts struck at S*;
a long position in ( 1 K 2 ) put options struck at K, for all strikes
from 0 to S*; and
a similar long position in ( 1 K 2 ) call options struck at K, for all
strikes from S* to .
All contracts expire at time T. Figure 4 shows this decomposition schematically.
18
Goldman
Sachs
The fair value of future variance can be related to the initial fair value
of each term on the right hand side of Equation 21. By using the identities in Equations 23 and 25, we obtain
S*
S 0 rT
2
K var = ---- rT ------- e 1 log ------T
S0
S*
+ e rT
+ e rT
S*
1
-------- P ( K ) d K
K2
(EQ 26)
- C ( K ) dK
S ------
2
K
*
where P(K) and (C(K)), respectively, denote the current fair value of a
put and call option of strike K. If you use the market prices of these
options, you obtain an estimate of the current market price of future
variance.
This approach to the fair value of future variance is the most rigorous
from a theoretical point of view, and makes less assumptions than our
intuitive treatment in the section on page 6. Equation 26 makes precise the intuitive notion that implied volatilities can be regarded as the
markets expectation of future realized volatilities. It provides a direct
connection between the market cost of options and the strategy for capturing future realized volatility, even when there is an implied volatility
skew and the simple Black-Scholes formula is invalid.
FIGURE 4. Replication of the log payoff. (a) The payoff of a short position in a
log contract at expiration. (b) Dashed line: the linear payoff at expiration of
a forward contract with delivery price S*; Solid line: the curved payoff of
calls struck above S* and puts struck below S* . Each option is weighted by
the inverse square of its strike. The sum of the payoffs for the dashed and
solid lines provide the same payoff as the log contract.
portfolio of options
ST
log ------S*
ST S*
-------------------S*
S*
S*
(a)
(b)
19
Goldman
Sachs
AN EXAMPLE OF A
VARIANCE SWAP
We now present a detailed practical example. Suppose you want to

price a swap on the realized variance of the daily returns of some hypothetical equity index. The fair delivery variance is determined by the
cost of the replicating strategy discussed in the previous section. If you
could buy options of all strikes between zero and infinity, the fair variance would be given by Equation 26 with some choice of S*, say S* = S0.
In practice, however, only a small set of discrete option strikes is available, and using Equation 26 with only a few strikes leads to appreciable errors. Here we suggest a better approximation.
We start with the definition of fair variance given by Equation 21,
which can be written as
ST
S* ST S*
T dS t S T S *
2
K var ---- E
--------- -------------------- log ------- + -------------------- log ------T
S*
S*
S*
S0
0 St
Taking expectations, this becomes

S*
S 0 rT
2
K var = ---- rT ------- e 1 log ------- + e rT CP
T
S0
S*
(EQ 27)
where CP is the present value of the portfolio of options with payoff

at expiration given by
S T
2 ST S*
f ( S T ) = ---- -------------------- log -------
T S*
S*
(EQ 28)
Suppose that you can trade call options with strikes Kic such that
K 0 = S * < K 1c < K 2c < K 3c < ... and put options with strikes Kip such
that K 0 = S * > K 1 p > K 2 p > K 3 p > ...
In Appendix A we derive the formula that determines how many
options of each strike you need in order to approximate the payoff f(ST)
by piece-wise linear options payoffs. The procedure in Appendix A
guarantees that these payoffs will always exceed or match the value of
the log contract, but never be worth less. Once these weights are calculated, CP is obtained from
CP =
w ( K ip )P ( S, K ip ) + w ( K ic )C ( S, K ic )
i
(EQ 29)
We now illustrate this procedure with a concrete numerical example.
20
TABLE 1. The portfolio of European-style put and call options used for
PUTS
calculating the cost of capturing realized variance in the presence of the

implied volatility skew with a discrete set of options strikes.
CALLS
Goldman
Sachs
Weight
Value
per
Option
Strike
Volatility
Contribution
50
30
163.04
0.000002
0.0004
55
29
134.63
0.00003
0.0035
60
28
113.05
0.0002
0.0241
65
27
96.27
0.0013
0.1289
70
26
82.98
0.0067
0.5560
75
25
72.26
0.0276
1.9939
80
24
63.49
0.0958
6.0829
85
23
56.23
0.2854
16.0459
90
22
50.15
0.7384
37.0260
95
21
45.00
1.6747
75.3616
100
20
20.98
3.3537
70.3615
100
20
19.63
4.5790
89.8691
105
19
36.83
2.2581
83.1580
110
18
33.55
0.8874
29.7752
115
17
30.69
0.2578
7.9130
120
16
28.19
0.0501
1.4119
125
15
25.98
0.0057
0.1476
130
14
24.02
0.0003
0.0075
135
13
22.27
0.000006
0.0001
TOTAL COST
419.8671
Assume that the index level S0 is 100, the continuously compounded

annual riskless interest rate r is 5%, the dividend yield is zero, and the
maturity of the variance swap is three months (T = 0.25. Suppose that
21
Goldman
Sachs
you can buy options with strikes in the range from 50 to 150, uniformly
spaced 5 points apart. We assume that at-the-money implied volatility
is 20%, with a skew such that the implied volatility increases by 1 volatility point for every 5 point decrease in the strike level. In Table 1 we
provide the list of strikes and their corresponding implied volatilities.
We then show the weights, the value of each individual option and the
contribution of each strike level to the total cost of the portfolio. At the
bottom of the table we show the total cost of the options portfolio,
CP = 419.8671 . It is clear from Table 1 that most of the cost comes
from options with strikes near the spot value. Although the number of
options which are far out of the money is large, their value is small and
contributes little to the total cost.
The cost of capturing variance is now simply calculated using Equation
2
27 with the result K var = ( 20.467 ) . This is not strictly the fair variance; because the procedure of approximating the log contract in
Appendix A always over-estimates the value of the log contract, this
value is higher than the true theoretical value for the fair variance
obtained by approximating the log contract with a continuum of
strikes. In Figure 5 we illustrate the cost of variance as function of the
spacing between strikes, for two cases, with and without a volatility
skew. You can see that as the spacing between strikes approaches zero,
the cost of capturing variance approaches the theoretically fair variance.
Convergence of Kvar, the cost of capturing variance with a
discrete set of strikes, towards the fair value of variance as a function of
K, the spacing between strikes. The line with square symbols shows the
convergence for no skew, with all implied volatilities at the same value of
20%. The theoretical fair variance for K = 0 is then (20)2 = 400. The line
with diamond symbols shows similar convergence to a higher fair
variance of about 402, the extra contribution coming from the effect of
the skew.
FIGURE 5.
420
K var
415
410
405
400
0
3
DK
22
Goldman
Sachs
EFFECTS OF THE
VOLATILITY SKEW
The general strategy discussed in the previous section can be used to

determine the fair variance and the hedging portfolio from the set of
available options and their implied volatilities. Here we discuss the
effects of a volatility skew on the fair variance. We assume that there is
no term structure and consider two different skew parameterizations,
both of which resemble typical index skews. The first is a skew that
varies linearly with the strike of the option, the second a skew that
varies linearly with the Black-Scholes delta. In both cases we will compare the numerically correct value of fair variance, computed from
Equation 26, with an approximate analytic formula that we derive.
This formula provides a good rule of thumb for a quick estimate of the
impact of the volatility skew on the fair variance.
Skew Linear in Strike
We first consider a skew for which the implied volatility varies linearly
with strike, so that
K SF
( K ) = 0 b -----------------S
(EQ 30)
Here 0 is the implied volatility of an option struck at the forward. The

steepness of the skew is determined by the slope b, with a positive
value indicating a higher volatility for strikes below the forward. Note
that this parametrization cannot hold for all strikes, because, for a
large enough value of K, the implied volatility would become negative7.
A value of b = 0.2 means that the implied volatility corresponding to a
strike 10% below the forward, for example, is 2 volatility points higher
than 0. In Appendix B we derive the following approximate formula
for the fair variance of the contract with time to expiration T:
2
K var 0 ( 1 + 3Tb 2 + .... )
(EQ 31)
The skew increases the value of the fair variance above the at-themoney-forward level of volatility, and the size of the increase is proportional to time to maturity and the square of the skew slope. (Note that
b in Equation 30 has the same dimension as volatility, so that b2T is a
dimensionless parameter, and therefore a natural candidate for the
order of magnitude of the percentage correction to Kvar. Note also that
there is no term b T in Equation 31. This approximation works best
for short maturities and skews that are not too steep.
7. Note that for large values of K, where this parameterization is invalid,

the options prices in Equation 26 are negligible and therefore do not
affect the value of the fair variance.
23
Goldman
Sachs
In Table 2 we compare the exact results for fair variance, computed

numerically, with the approximate values given by the analytic formula in Equation 31.
TABLE 2. Comparison of the exact fair variance, computed numerically, with
the approximate analytic formula of Equation 31. We assume 0 = 30%, S =

100, the continuously compounded annual discount rate r = 5%, zero
dividend yield, and use strikes evenly spaced one point apart from K = 10
to K = 200 to replicate the log payoff.
T = 3 months
Skew Slope
b
Exact
Value
T = 1 year
Analytic
Approximation
Exact
Value
Analytic
Approximation
0.0
( 30.01 )
( 30.00 )
( 29.97 )
( 30.00 )
0.1
( 30.01 )
( 30.11 )
( 30.05 )
( 30.45 )
0.2
( 30.22 )
( 30.44 )
( 30.82 )
( 31.75 )
0.3
( 30.65 )
( 30.99 )
( 32.33 )
( 33.81 )
Figure 6 contains a graph of these results. We see excellent agreement

in the case of the three-month variance swap, and reasonable agreement for one year.
1150
1150
1100
1100
1050
1050
Kvar
Kvar
FIGURE 6. Comparison of the exact value of fair variance, Kvar, with the
approximate value from the formula of Equation 31, as a function of the
skew slope b. The thin line with squares shows the exact values obtained
by replicating the log-payoff. The thick line depicts the approximate
value given by Equation 31. (a) three-month variance swap. (b) one-year
variance swap.
1000
950
900
24
1000
950
0.1
0.2
0.3
900
0.1
0.2
(a)
(b)
0.3
Next we consider a skew that varies linearly with the Black-Scholes

delta of the option, so that:
1
( p ) = 0 + b p + ---
(EQ 32)
Here p is the Black-Scholes exposure of a put option, given by

p = N ( d 1 ) , where d1 is defined in Footnote 2, 0 is the implied
volatility of a 50-delta put option and b is the slope of the skew that
is, the change in the skew per unit delta. This parameter b is not the
same as the b in the previous section. In particular, there is an implicit
dependence on the time to expiration in the formula of Equation 32,
because of the p term, which was absent from Equation 30. Since p is
bounded, the implied volatility is always positive provided b < 20.
This restriction is irrelevant, since Equation 32 leads to arbitrage violation before b reaches this limit. In practice, this parameterization
leads to more realistic skews than those produced by the linear-strike
formula.
Appendix C presents a detailed derivation of the following approximate
formula for the fair variance of the contract with time to expiration T:
2
2
1
1b
K var 0 1 + ------- b T + ------ ------ + ....
12 2
(EQ 33)
Here, in contrast to the skew linear in strike, the first-order correction

is of magnitude b T , because a variation linear in delta about the atthe-money-forward strike is not equivalent to a variation linear in
strike.
FIGURE 7. (a) A volatility skew that varies linearly in delta. (b) The
corresponding skew plotted as a function of strike. We have assumed that
the stock price S is 100, the continuously compounded annual discount
rate r is 5%, the term to maturity is three months, and the skew slope is 0.2.
40
40
35
35
30
30
Skew Linear in Delta
Goldman
Sachs
25
25
20
-1
-0.75
-0.5
(a)
-0.25
20
60
80
100
120
140
(b)
25
Goldman
Sachs
First, we convert the skew by delta in Equation 32 into a skew by

strike, as displayed in Figure 7. Again, we compare the exact results
computed according to Appendix A with the approximate values given
by Equation 33. In Table 3 we compare the results for fair variance,
computed numerically, with the approximate values given by the analytic formula in Equation 33. The analytic formula works very well for
the three-month variance swap, and truly impressively for the oneyear swap, as displayed in Figure 8.
TABLE 3. Comparison of the fair variance, computed numerically, with the
approximate analytic formula of Equation 33. We assume 0 = 30%, S = 100,

the continuously compounded annual discount rate r = 5%, zero dividend
yield, and use strikes evenly spaced one point apart from K = 10 to K = 200
to replicate the log payoff.
T = 3 months
Skew Slope
b
Exact
Value
T = 1 year
Analytic
Approximation
Exact
Value
Analytic
Approximation
0.0
( 30.01 )
( 30.00 )
( 29.97 )
( 30.00 )
0.1
( 30.61 )
( 30.62 )
( 31.06 )
( 31.03 )
0.2
( 31.49 )
( 31.60 )
( 32.42 )
( 32.40 )
0.3
( 32.64 )
( 32.93 )
( 34.06 )
( 34.06 )
FIGURE 8. Comparison of the exact value of fair variance, Kvar, with the
1150
1150
1100
1100
Kvar
Kvar
approximate value from the formula of Equation 33, as a function of the

skew slope b. The thin line with squares shows the exact values obtained
by replicating the log-payoff. The thick line depicts the approximate
value given by Equation 33. (a) Three-month variance swap. (b) Oneyear variance swap.
1050
1000
1000
950
950
900
0.1
0.2
b
(a)
26
1050
0.3
900
0.1
0.2
b
(b)
0.3
Goldman
Sachs
PRACTICAL PROBLEMS
WITH REPLICATION
We have shown in Equation 20 that a variance swap is theoretically

equivalent to a dynamically adjusted, constant-dollar exposure to the
stock, together with a static long position in a portfolio of options and a
forward that together replicate the payoff of a log contract. This portfolio strategy captures variance exactly, provided the portfolio of options
contains all strikes between zero and infinity in the appropriate weight
to match the log payoff, and provided the stock price evolves continuously.
Two obvious things can go wrong. First, you may be able to trade only a
limited range of options strikes, insufficient to accurately replicate the
log payoff. Second, the stock price may jump. Both of these effects
cause the strategy to capture a quantity that is not the true realized
variance. We will focus on the effects of these two limitations below,
though other practical issues, like liquidity, may also corrupt the ideal
strategy.
Imperfect Replication
Due to Limited Strike
Range
Variance replication requires a log contract. Since log contracts are not
traded in practice, we replicate the payoff with traded standard options
in a limited strike range. Because these strikes fail to duplicate the log
contract exactly, they will capture less than the true realized variance.
Therefore, they have lower value than that of a true log contract, and
so produce an inaccurate, lower estimate of the fair variance.
In Table 4 below we show how the estimated value of fair variance is
affected by the range of strikes that make up the replicating portfolio.
The fair variances are estimated from (1) a replicating portfolio with a
narrow range of strikes, ranging from 75% to 125% of the initial spot
level, and (2) a portfolio with a wide range of strikes, from 50% to 200%
of the initial spot level. In both cases the strikes are uniformly spaced,
one point apart. (The fair variance is calculated according to Equation
26, except that the integrals are replaced by sums over the available
option strikes whose weights are chosen according to the procedure of
Appendix A). We assume here that implied volatility is 25% per year
for all strikes, with no volatility skew, so that all options are valued at
the same implied volatility. We also assume a continuously compounded annual interest rate of 5%.
For both expirations, the wide strike range accurately approximates
the actual square of the implied volatility. However, the narrow strike
range underestimates the fair variance, more dramatically so for
longer expirations.
27
Goldman
Sachs
TABLE 4. The effect of strike range on estimated fair variance.
Expiration
Wide strike range

(50% - 200%)
Narrow strike range

(75% - 125%)
Three-month
( 25.0 ) 2
( 24.9 ) 2
One-year
( 25.0 ) 2
( 23.0 ) 2
In the section entitled Replicating Variance Swaps: First Steps on page 6, we have
already discussed one approach to understanding why the narrow
strike range fails to capture variance. As shown in Figure 3, the vega
and gamma of a limited strike range both fall to zero when the index
moves outside the strike range, and the strategy then fails to accrue
realized variance as the stock price moves. Consequently, the estimated variance is lower than the true fair value for both expirations
above, and the reduction in value is greater for the one-year case. Over
a longer time period it is more likely that the stock price will evolve
outside the strike range.
In essence, capturing variance requires owning the full log contract,
whose duplication demands an infinite range of strikes. If you own a
limited number of strikes, still appropriately weighted, you pay less
than the full value, and, when the stock price evolves into regions
where the curvature of the portfolio is insufficiently large, you capture
less than the full realized variance, even if no jumps occur and the
stock always moves continuously. In order to keep capturing variance,
you need to maintain the curvature of the log contract at the current
stock price, whatever value it takes.
A simpler way of understanding why a narrow strike range leads to a
lower fair variance is to compare the payoff of the narrow-strike replicating portfolio at expiration to the terminal payoff that the portfolio is
attempting to replicate, that is, the nonlinear part of the log payoff:
ST
ST S0
-------------------- log ------S0
S0
(EQ 34)
Figure 9 displays the mismatch between the two payoffs. The narrowstrike option portfolio matches the curved part of the log payoff well at
stock price levels between the range of strikes, that is, from 75 to 125.
Beyond this range, the option portfolio payoff remains linear, always
growing less rapidly than the nonlinear part of the log contract. The
lack of curvature (or gamma, or vega) in the options portfolio outside
28
Goldman
Sachs
FIGURE 9. Comparison of the terminal payoff of the narrow-strike replicating
Terminal payoff
portfolio (dashed line) and the nonlinear part of the log-payoff (solid line).
50
100
ST
150
200
the narrow strike range is responsible for the inability to capture variance.
The Effect of Jumps
on a Perfectly Replicated Log Contract
When the stock price jumps, the log contract may no longer capture
realized volatility, for two reasons. First, if the log contract has been
approximately replicated by only a finite range of strikes, a large jump
may take the stock price into a region in which variance does not
accrue at the right rate. Second, even with perfect replication, a discontinuous stock-price jump causes the variance-capture strategy of Equation 20 to capture an amount not equal to the true realized variance. In
reality, both these effects contribute to the replication error. In this section, we focus only on the second effect and examine the effects of
jumps assuming that the log-payoff can be replicated perfectly with
options.
For the sake of discussion, from now on we will assume that we are
short the variance swap, which we will hedge by following a discrete
version of the variance-capture strategy
2
V = ---T
S i
ST
- log ------ -----------Si 1

S0
(EQ 35)
i=1
where S i = S i S i 1 is the change in stock price between successive

observations. Rather than continuously rebalance as the stock price
moves, we instead adjust the exposure to (2/T) dollars worth of stock
only when a new stock price is recorded for updating the realized variance.
Because of the additive properties of the logarithm function, the terminal log payoff is equivalent to a daily accumulation of log payoffs:
29
Goldman
Sachs
2
V = ---T
i=1
S i
Si
------------- log ------------Si 1
Si 1
(EQ 36)
Suppose that all but one of the daily price changes are well-behaved
that is, all changes are diffusive, except for a single jump event. We
characterize the jump by the parameter J , the percentage jump downwards, from S S ( 1 J ) ; a jump downwards of 10% corresponds to J
= 0.1. A jump up corresponds to a value J < 0.
The contribution of this one jump to the variance is easy to isolate,
because variance is additive; the total (un-annualized) realized variance for a zero-mean contract is the sum
1
V = ---T
2
2
S i
S i
1
1 S 2
------------------------=
+ ---- --------
S
T
T S jump
S i 1
no jumps i 1
(EQ 37)
The contribution of the jump to the realized total variance is given by:
1 S 2
J2
---- --------
= ------T S jump
T
(EQ 38)
On the other hand, the impact of the jump on the quantity captured by
our variance replication strategy in Equation 36 is
Si
2 S i
---- ------------- log -------------
T Si 1
S i 1
jump
2
= ---- [ J log ( 1 J ) ]
T
(EQ 39)
In the limit that the jump size J is small enough to be regarded as part
of a continuous stock evolution process, the right hand side of Equation
39 does reduce to the contribution of this (now small) move to the true
realized variance. It is only because J is not small that the variance
capture strategy is inaccurate. Therefore, the replication error, or the
P&L (profit/loss) due to the jump for a short position in a variance
swap hedged by a long position in a variance-capture strategy is
2
J2
P&L due to jump = ---- [ J log ( 1 J ) ] ------T
T
(EQ 40)
To understand this result better, it is helpful to expand the log function

as a series in J:
J2 J3
log ( 1 J ) = J + ------- + ------- +
2
3
30
(EQ 41)
The leading contribution to the replication error is then

2J3
P&L due to jump = --- ------- +
3T
(EQ 42)
The quadratic contribution of the jump is the same for the variance
swap as it is for the variance-capture strategy, and has no impact on
the hedging mismatch. The leading correction is cubic in the jump size
J and has a different sign for upwards or downwards jumps. A large
move downwards (J > 0) leads to a profit for the (short variance swap)(long variance-capture strategy), while a large move upwards (J < 0)
leads to a loss. Furthermore, a large move one day, followed by a large
move in the opposite direction the next day would tend to offset each
other. Figure 10 shows the impact of the jump on the strategy for a
range of jump values. Note that the simple cubic approximation of
Equation 42 correctly predicts the sign of the P&L for all values of the
jump size.
FIGURE 10. he impact of a single jump on the profit or loss of a short position
in a variance swap and a long position in the variance replication strategy,

as given by Equation 40 as a function of (downward) jump size for T=1 year.
0.006
Impact of jump
Goldman
Sachs
0.004
0.002
0
-0.002
-0.004
-20
-10
10
20
Jump size (downward)
There is an analogy between the cancellation of the quadratic jump

term in variance replication and the linear jump term in options replication. When you are long an option you are long linear, quadratic and
higher-order dependence on the stock price. If you are also short the
options delta-hedge, then the linear dependence of the net position
cancels, leaving only the quadratic and higher-order dependencies.
Because the leading-order term is quadratic, large moves in either
direction benefit the position; this is precisely why hedged long options
positions capture variance. In contrast, in the case of variance replication considered here, the variance replication strategy is long quadratic, cubic and higher-order terms in the stock price, while the
position in the variance swap is short only the quadratic dependence.
Now the quadratic term in the net position cancels, leaving only cubic
and higher-order dependencies on the jump size. Since the leading
31
Goldman
Sachs
term is cubic, the direction of the jump determines whether there is a

net profit or loss.
Table 5 displays the profit or loss due to jumps of varying sizes for
three-month and one-year variance swaps.
TABLE 5. The profit/loss due to a single jump for a short variance swap with a
notional value of $1 per squared variance point, that is hedged with the
variance replication strategy of Equation 36 for T=1 year.
Jump size and

direction
One-year
J = 15% (down)
101.5
25.4
J = 10% (down)
28.8
7.2
J = 5% (down)
3.5
0.9
3.2
0.8
J = 10% (up)
24.8
6.2
J = 20% (up)
80.9
20.2
J = 5% (up)
The Effect of Jumps

When Replicating With
a Finite Strike Range
Three-month
In practice, both the effects of jumps and the risks of log replication
with only a limited strike range cause the strategy to capture a quantity different from the true realized variance of the stock price. The
combined effect of both these risks is harder to characterize because
they interact with one another in a complicated manner.
Consider again a short position in a variance contract that is being
hedged by the variance-capture strategy. Suppose that a downward
jump occurs, large enough to move the stock price outside the range of
option strikes. If the log-payoff were replicated perfectly, the constantdollar exposure would cancel the linear part of the stock price change,
and lead to a convexity gain. Although the log-payoff is not being replicated perfectly, there is still a convexity gain from the jump, but it is
smaller in size. However, after this jump, with the stock price now outside the strike range, the vega and gamma of the replicating portfolio
are now too low to accrue sufficient variance, even if no further jumps
occur. In this scenario, the gain from the jump has to be balanced
against the subsequent failure of the hedge to capture the smooth variance. The net results will depend on the details of the scenario.
In contrast, a large move upwards will be doubly damaging: there will
be convexity loss due to the jump and the hedge will not capture variance if the jump takes the index outside the strike range.
32
Goldman
Sachs
FROM VARIANCE TO
VOLATILITY
CONTRACTS
For most of this note we have focused on valuing and replicating variance swaps. But most market participants prefer to quote levels of volatility rather than variance, and so we now consider volatility swaps.
There is no simple replication strategy for synthesizing a volatility
swap; it is variance that emerges naturally from hedged options trading. The replication strategy for the variance swap makes no assumptions about the level of future volatility, other than assuming that the
stock price evolves continuously (without jumps). Changes in volatility
have no effect on the strategy, which still captures the total variance
over the life of the log contract. In contrast, as we will show, the replication strategy for a volatility swap is fundamentally different; it is
affected by changes in volatility and its value depends on the volatility
of future realized volatility. In essence, from a contingent claims or
derivatives point of view, variance is the primary underlyer and all
other volatility payoffs, such as volatility swaps, are best regarded as
derivative securities on the variance as underlyer. From this perspective, volatility itself is a nonlinear function (the square root) of variance and is therefore more difficult, both theoretically and practically,
to value and hedge.
To illustrate the issues involved, lets consider a naive strategy:
approximate a volatility swap by statically holding a suitably chosen
variance contract. In order to approximate a volatility swap struck at
K vol , which has payoff R K vol , we can use the approximation
2
2
1
R K vol ---------------- ( R K vol )
2K vol
(EQ 43)
This means that 1 ( 2K vol ) variance contracts with strike K vol can
approximate a volatility swap with a notional $1/(vol point), for realized volatilities near K vol . With this choice, the variance and volatility
payoffs agree in value and volatility sensitivity (the first derivative
with respect to R ) when R = K vol . Naively, this would also imply
that the fair price of future volatility (the strike for which the volatility
swap has zero value) is simply the square root of fair variance K var :
K vol =
K var
(naive estimate)
(EQ 44)
In Figure 11 we compare the two sides of Equation 43 for Kvol = 30%

for different values of the realized volatility. We see that the actual volatility swap and the approximating variance swap differ appreciably
33
Goldman
Sachs
FIGURE 11. Payoff of a volatility swap (straight line) and variance swap
(curved line) as a function of realized volatility, for K vol = 30% .
10
payoff
5
0
20
30
40
-5
-10
-15
only when the future realized volatility moves away from K vol ; you
cannot fit a line everywhere with a parabola.
The naive estimate of Equations 43 and 44 is not quite correct. With
this choice, the variance swap payoff is always greater than the volatility swap payoff. The mismatch between the variance and volatility
swap payoffs in Equation 43, is the
2
1
convexity bias = ---------------- ( R K vol )
2K vol
This square is always positive, so that with this choice of the fair delivery price for volatility, the variance swap always outperforms the volatility swap. To avoid this arbitrage, we should correct our naive
estimate to make the fair strike for the volatility contract lower than
the square root of the fair strike for a variance contract, so that
K vol < K var . In this way, the straight line in Figure 11 will shift to
the left and will not always lie below the parabola.
In order to estimate the size of the convexity bias, and therefore the
fair strike for the volatility swap, it is necessary to make an assumption about both the level and volatility of future realized volatility. In
Appendix D we estimate the expected hedging mismatch and static
hedging parameters under the assumption that future realized volatility is normally distributed.
Dynamic Replication
of a Volatility Swap
34
In principle, some of the risks inherent in the static approximation of a

volatility swap by a variance swap could be reduced by dynamically
trading new variance contracts throughout the life of the volatility
swap. This dynamic replication of a volatility swap by means of vari-
Goldman
Sachs
ance swaps would (in principle) produce the payoff of a volatility swap
independent of the moves in future volatility. This is closely analogous
to replicating a curved stock option payoff by means of delta-hedging
using the linear underlying stock price. In practice, of course, there is
no market in variance swaps liquid enough to provide a usable underlyer.
In the same way that the appropriate option hedge ratio depends on
the assumed future volatility of the stock, the dynamic replication of a
volatility swap requires a model for the volatility of volatility. Taking
the analogy further, one could imagine that the strategy would call for
holding at every instant a variance-delta equivalent of variance contracts to hedge a volatility derivative.
The practical implementation of these ideas requires an arbitrage-free
model for the stochastic evolution of the volatility surface. Due to the
complexity of the mathematics involved, it is only very recently that
such models have been developed [see for example Derman and Kani
(1998) and Ledoit (1998)]. When there is a liquid market in variance
swaps, these models may be useful in hedging volatility swaps and
other variance derivatives.
35
Goldman
Sachs
CONCLUSIONS AND
FUTURE INNOVATIONS
We have tried to present a comprehensive and didactic account of both

the principles and methods used to value and hedge variance swaps.
We have explained both the intuitive and the rigorous approach to replication. In markets with a volatility skew (the real world for most
swaps of interest), the intuitive approach loses its footing. Here, using
the rigorous approach, one can still value variance swaps by replication. Remarkably, we have succeeded in deriving analytic approximations that work well for the swap value under commonly used skew
parameterizations. These formulas enable traders to update price
quotes quickly as the market skew changes.
There are at least two areas where further development is welcome.
First, our ability to effectively price and hedge volatility swaps is still
limited. To fully implement a replication strategy for volatility swaps,
we need a consistent stochastic volatility model for options. Much work
remains to be done in this area.
Second, some market participants prefer to enter a capped variance
swap or volatility swap that limits the possible loss on the position. The
capped variance swap has embedded in it an option on realized variance. The development of a truly liquid market in volatility swaps, forwards or futures would lead to the possibility of trading and hedging
volatility options. Once again, this requires a consistent model for stochastic volatility.
36
Goldman
Sachs
APPENDIX A:
REPLICATING
LOGARITHMIC
PAYOFFS
In this Appendix we derive several results concerning the replication of

a logarithmic payoff with portfolios of standard options.
Constant Vega
Requires Options
Weighted Inversely
Proportional to the
Square of the Strike
Consider a portfolio of standard options
(S) =
( K )O ( S, K , v ) dK
(A 1)
where O(S,K,v) represents a standard Black-Scholes option of strike K

2
and total variance v = when the stock price is S.

Vega, the sensitivity to the total variance of an individual option O in
this portfolio, is given by
VO =
K
(O) = Sf -----, v
v
S
where
2
1 exp ( d 1 2 )
f ( S, K , v ) = ----------- -------------------------------2 v
2
and
ln ( S K ) + v 2
d 1 = ----------------------------------------- .
v
The variance sensitivity of the whole portfolio is therefore
K
V ( S ) = ( K )Sf -----, v d K
S
(A 2)
The sensitivity of vega to S is

V
S
S[ S
( xS ) ] f ( x, v ) d x
= S [ 2 ( xS ) + xS' ( xS ) ] f ( x, v ) d x
0
37
Goldman
Sachs
where, in the second line of the above equation, we changed the integration variable to x = K S .
We want vega to be independent of S, that is
V
S
= 0 , which implies
that
2 + K
= 0
K
The solution to this equation is

const
= -------------2
K
Log Payoff Replication
with a Discrete Set of
Options
(A 3)
It was shown in the main text that the realized variance is related to
trading a log contract. Since there is no log-contract traded, we want to
represent it in terms of standard options. It is useful to subtract the
linear part (corresponding to the forward contract) and look at the
function
ST
2 ST S*
f ( S T ) = ---- -------------------- log ------T
S*
S*
(A 4)
where S* is some reference price. In practice, only a discrete set of

option strikes is available for replicating f ( S T ) , and we need to determine the number of options for each strike. Assume that you can trade
call options with strikes
K 0 = S * < K 1c < K 2c < K 3c < ...
and put options with strikes
K 0 = S * > K 1 p > K 2 p > K 3 p > ...
We can approximate f ( S T ) with a piece-wise linear function as in Figure 11. The first segment to the right of S* is equivalent to the payoff of
a call option with strike K0. The number of options is determined by
the slope of this segment:
f ( K 1c ) f ( K 0 )
w c ( K 0 ) = ----------------------------------------K 1c K 0
38
(A 5)
Goldman
Sachs
FIGURE 12. Log-payoff and options portfolio at maturity.
K2 p
K 1p
K0
K 1c
K2 c
Similarly, the second segment looks like a combination of calls with

strikes K 0 and K 1c . Given that we already hold w c ( K 0 ) options with
strike K 0 we need to hold w c ( K 1 ) calls with strike K 1 where
f ( K 2c ) f ( K 1c )
w c ( K 1 ) = ------------------------------------------- w c ( K 0 )
K 2c K 1c
(A 6)
Continuing in this way we can build the entire payoff curve one step at
the time. In general, the number of call options of strike K n, c is given
by
f ( K n + 1 , c ) f ( K n, c )
w c ( K n, c ) = --------------------------------------------------------
K n + 1, c K n, c
n1
w c ( K i, c )
(A 7)
i=0
The other side of the curve can be built using put options:
f ( K n + 1 , p ) f ( K n, p )
w p ( K n, p ) = -----------------------------------------------------------
K n, p K n + 1, p
n1
w c ( K i, p )
(A 8)
i=0
39
Goldman
Sachs
APPENDIX B:
SKEW LINEAR IN STRIKE
Here, we derive a formula which gives the approximate value of the

variance swap when the skew is linear in strike. We parameterize the
implied volatility by
K SF
( K ) = 0 b -----------------S
(B 1)
rT
where S F = S 0 e
is the forward value corresponding to the current
spot, 0 is at-the-money forward implied volatility and b is the slope of
the skew.
We start with the general expression for the fair variance discussed in
the main text:
S*
S 0 rT
2
K var = ---- rT ------- e 1 log ------- +
T
S0
S*
rT S *
1
-------- P ( K , ( b ) ) d K + e
K2
(B 2)
rT
- C ( K , ( b ) ) dK
S ------
K2
1
We now expand option prices as a power series in b around a flat

implied volatility ( b = 0 ),
2
1 2C
C
C ( K , ( b ) ) = C ( K , 0 ) + b
+ --- b
+ ...
b b = 0 2 b2
b=0
2
P ( K , ( b ) ) = P ( K , 0 ) + b
(B 3)
1 2P
P
+ --- b
+ ...
b b = 0 2 b2
b=0
Using this expansion we can formally write an expansion of fair variance in powers of b as follows:
1 C
2
2 rT S * 1 P
--------------dK +
dK +
K var = 0 + b ---- e
T
0 K 2b
2
b
S* K
b=0
b=0
2
2
1 C
1 2 2 rT S * 1 P
--- b ---- e
--------------dK +
d K + ...
0 K2 2
2
2
2 T
S* K b
b b = 0
b=0
40
(B 4)
Goldman
Sachs
Here 0 is the fair variance in the flat world where volatility is constant and is given by Equation B2 with ( b ) replaced by 0 .
The derivatives which enter Equation B4 are given by
2
P
P
=
,
b b = 0 b b = 0
0
C
C
=
,
b b = 0 b b = 0
0
b b = 0

0
2
b b = 0
C
2
b=0
b=0
The derivatives with respect to volatility are easily calculated using

the Black-Scholes formula
S T d1 2
P
C
=
= ------------e

2
0
0
2
P
2
S T d 1 d 1 2
= ------------d 1
e
2 0
2
(B 5)
SF
1 2
log ------- + --- 0 T
K 2
d 1 = ------------------------------------------0 T
where, for the model we are considering here
K
= ------- 1
SF
b b = 0
(B 6)
The fact that call and put options have the same vega in the BlackScholes framework makes it possible to combine the integrals in Equation B4 into one integral from 0 to :
2
d1 2
2
2 rT S T 1 K
dK
K var = 0 b ---- e ------------ -------- ------- 1 e
T
2 0 K 2 S
F
2 d 1 d 1 2
1 2 2 rT S T 1 K
------------ ---------- b ---- e

1 d1
e
d K + ...
2 0 K 2 ------2 T
0
SF
(B 7)
41
Goldman
Sachs
To evaluate these integrals, one can, for example, change the integraSF 1
2
tion variable to z = log ------- + --- v 0 v 0 d 1 , where v 0 = 0 T , and
K 2
then write Equation B7 as
2
2
dz
1 e v 0 z v 0 2 e z 2 ---------- +
K var = 0 b 2 0
2
v0 z + v0 2
2 v0 z v0 2
2
z 2 dz
-----------
b
e
+e
2 ( z v 0 z )e
The term linear in b vanishes and the term quadratic in b has coeffi2
cient 3 0 T , so that
2
K var = 0 ( 1 + 3Tb 2 + .... )
(B 8)
We now present an alternative derivation of this result. We start with

the fundamental definition of the fair delivery variance as the expected
value of future realized variance, i.e.
1 T 2
K var = E ---- ( S, t ) dt
T 0
(B 9)
This can be evaluated approximately as follows. First, we use the relation between implied and local volatility:
+ 2rK
---- + 2
T
K
T
2 ( S, t ) = -----------------------------------------------------------------------------------------------------------------------2
2
2
2 1 1
d 1 T + --- -------------- + d 1
K
K
K T
K
K 2
(B 10)
K = S
T = t
S
Denote x = ------- 1 . Equation B10 can be written as
SF
0 bx 2br ( 1 + x )t
2 ( S, t ) = ---------------------------------------------------------------------------------------------------------------------------------2
2
2
1
1
( 1 + x ) t b td 1 + ------------------- ------------------------ bd 1
0 bx ( 1 + x ) t
42
(B 11)
Goldman
Sachs
where
log ( 1 + x ) 1
d 1 = ------------------------------ + --- ( 0 bx ) t
( 0 bx ) t 2
(B 12)
We expand 2 ( S, t ) in powers of x and calculate the expected value in

a lognormal world with volatility 0 using
E[ x] = 0
2
0 t
2
3 0 t
E[ x ] = e
E[ x ] = e
1
2
3e
0 t
+2
...
Expected values of higher powers of x are easily calculated using
2
2
( n n ) 0 t 2
S n
E -------
= e
S F
After averaging over the stock price distribution, we average over time
and, finally, expand the result in powers of the skew slope b . Tedious
calculation leads to the relation
2
K var = 0 ( 1 + 3Tb 2 + .... )

It is reassuring that these two very different methods lead to the same
approximation formula.
43
Goldman
Sachs
APPENDIX C:
SKEW LINEAR IN DELTA
Here we consider the case where implied volatility varies linearly with
delta. Such a skew can be parameterized in terms of p , the delta of a
European-style put, as
1
( p ) = 0 + b p + ---
(C 1)
where 0 is the implied volatility of options with p = 1 2 (the 50delta volatility). (We could also parametrize the skew in terms of the
1
call delta as ( c ) = 0 + b c --- .)
2
To derive the formula for the fair variance we follow the same procedure as in Appendix B, starting with Equation B2. One important difference is that now implied volatility is nonlinear in b (since p
depends implicitly on b ) so that second derivatives have an additional
term:
2
P
2
b b = 0
b b = 0

0
2
C
2
P
+
b2
b=0
0
b=0
2
+
b=0
(C 2)
C
b2
0
b=0
Other derivatives we need are easily calculated:
1
= p + --b b = 0
2
2
b b = 0
1 p
= 2 p + ---
2 0
where
p = N ( d1 )
and
44
(C 3)
Goldman
Sachs
SF
log ------K 1
d 1 = ---------------- + --- 0 T
0 T 2
Combining these relations, the fair variance can be written as
K var =
2 rT S T 1
1 d1 2
b ---- e ------------ -------- p + --- e
dK
T
2 K 2
2
2
0
2
1 2 2 rT S T 1
1 2 d 1 d 1 2
--- b ---- e
------------ -------- p + --- d 1
e
dK
2 K 2
2 T
0
2
0
(C 4)
2
1
1 p d 1 2
2 -------- p + ---
e
dK
2 0
K 2
Again, integrals can be evaluated by changing the integration variable

SF 1
2
to z = log ------- + --- v 0 v 0 d 1 , where v 0 = 0 T , so that
K 2
K var =
2
0
z 2
v0 z v0 2
dz
1
----------- +
b 2 0
N ( z ) --- e
e
2
2
2
v 0 z v 0 2 z 2 dz
1 2 2
----------N ( z ) --- ( z v 0 z )e
e
2
2
v 0 z v 0 2 z 2 dz
1
N ( z ) --- ( z v 0 )e
e -----2
2
All these integrals can be evaluated exactly. Since we are eventually

2
interested in expanding the result in powers of v 0 = 0 T , one can first

expand e
v0 z v0 2
in powers of z and integrate term by term. It is also

1
useful to note that N ( z ) --- is antisymmetric in z to simplify calcula2
tions. In addition the following results are useful:
45
Goldman
Sachs
2n az 2
( 2 )
1
b
1 dz
------- arctan ------N ( bz ) --- ----------- = -------------2 a n a
2 2
a
2n + 1 az 2
( 2 ) b
1
1 dz
-----------------------N ( bz ) --- ----------- = -----------------n
2 2
2 2 a a a + b 2
(C 5)
n
2
n
( 2 )
1
a + 2b
1 2 dz
------------------------------------- ---
------------=
N ( bz ) --arctan
2 a n a
2
4
2
a
2n az 2
2n + 1 az 2
n
n
b
1
( 2 ) b
1 2 dz
------------------------ arctan --------------------
N ( bz ) --- ----------- = -------------------32
n
2
2
2
( 2 )
a a a + b2
a+b
After evaluating all integrals we find the final answer to be

2
2 T
1 2
K var = 0 + b 0 ---- + ------b + ....
12
Two- Slope Model
(C 6)
Our calculations can easily be generalized to the model where the slope
of the skew is different for put and call options, i.e.
1
p ( p ) = 0 + b p p + ---
2
1
c ( c ) = 0 + b c c ---
for
1
--- p 0
2
for
1
0 c --2
(C 7)
We now briefly sketch the derivation emphasizing only the differences

with the above detailed calculations. We start with the same fundamental expression:
S*
S 0 rT
2
K var = ---- rT ------- e 1 log ------- +
T
S0
S*
46
rT S *
1
-------- P ( K , p ( b p ) ) d K + e
K2
(C 8)
rT
- C ( K , c ( bc ) ) dK
S ------
2
K
*
Goldman
Sachs
We use different implied volatility parameterizations for put and call

options, as given by Equation C7. Note that we should choose S* so
that
2
S* = S F e
0 T 2
This ensures that we use the put (call) parameterization in Equation

C7 for strikes below (above) S*. We expand put option prices in powers
of b p and call option prices in powers of b c . Evaluating all integrals as
above we find
2
K var =
2
0
b p + bc
1 b p + bc
1
T
+ --- 0 ( b p b c ) + 0 ------------------- 0 ---- + ------ ------------------- + ....
12 2
4
2
(C 9)
Obviously, for b p = b c this reduces to the result for single slope given
in Equation C6. Note that by changing the sign of b c we turn the
implied skew into a smile.
47
Goldman
Sachs
APPENDIX D:
STATIC AND DYNAMIC
REPLICATION OF A
VOLATILITY SWAP
We have argued that volatility swaps are fundamentally different from

variance swaps and that, unlike the variance swap, there is no simple
replicating strategy to synthetically create a volatility swap.
In the section From Variance to Volatility Contracts on page 33, we showed that
attempting to create a volatility swap from a variance swap by means
of a buy-and-hold strategy invariably leads to misreplication, since
this amounts to trying to fit a linear payoff (the volatility payoff) with a
quadratic payoff (the variance swap).
Given a view on both the direction and volatility of future volatility, we
will show that it is possible to pick the strike and notional size of a
variance contract to match the payoff of a volatility contract, on average, as closely as possible. The extent of the replication mismatch will
depend on how close the realized volatility is to its expected value.
2 ), while the target
The hedging instrument is the realized variance ( T
of the replication is the realized volatility ( T ). We want to approximate the volatility as a function of the variance by writing
2
T a T + b
(D 1)
and choose a and b to minimize the expected squared deviation of the

two sides of Equation D1:
2 b )2 ]
min E [ ( T a T
(D 2)
Differentiation leads to the following equations for the coefficients a

and b :
2
E [ T ] = aE [ T ] + b
(D 3)
3 ] = aE [ 4 ] + bE [ 2 ]
E [ T
T
T
The distribution of future volatility could be assumed to be normal,

with mean and standard deviation :
T N ( , )
(D 4)
This model only makes sense if the probability of negative volatilities

is negligible. This strategy will replicate only on average; the expected
squared replication error is given by:
48
Goldman
Sachs
2 ) )2 ]
min E [ ( T a T b ) ] = Var ( T ) [ 1 ( corr ( T , T
(D 5)
For realized volatilities distributed normally as in Equation D4, the

hedging coefficients are
1
a = ------------------- 2
2 + ------
b = ---------------- 2
2 + ------2
(D 6)
and the expected squared replication error is:

2
2
2
min E [ ( T a T b ) ] = -------------------2 2
1 + --------- 2
(D 7)
49
Goldman
Sachs
REFERENCES
Carr, P. and D. Madan (1998). Towards a Theory of Volatility Trading,

in Volatility: New Estimation Techniques for Pricing Derivatives,
edited by R. Jarrow, 417-427.
Derman, E. and I. Kani (1994). Riding on a Smile, RISK 7, No. 2, 32-39.
Derman, E., I. Kani and J. Zou (1996). The Local Volatility Surface,
Financial Analyst Journal, July/August, 25-36.
Derman, E., M. Kamal, I. Kani, and J. Zou (1996). Valuing Contracts
with Payoffs Based on Realized Volatility, Global Derivatives (July)
Quarterly Review, Equity Derivatives Research, Goldman, Sachs & Co.
Derman, E., M. Kamal, I. Kani, J. McClure, C. Pirasteh, and J. Zou
(1996). Investing in Volatility, Quantitative Strategies Research Notes,
October, Goldman, Sachs & Co.
Derman, E. and I. Kani (1998). Stochastic Implied Trees: Arbitrage
Pricing with Stochastic Term and Strike Structure of Volatility, International Journal of Theoretical and Applied Finance, Vol. 1, No. 1, 61110.
Dupire, B. (1994). Pricing with a Smile, RISK 7, No. 1, 18-20.
Ledoit, O. and P. Santa-Clara (1998). Relative Price of Options with
Stochastic Volatility. Working Paper, The Anderson Graduate Schools
of Management, University of California, Los Angeles.
Neuberger, A. (1994). The Log Contract: A new instrument to hedge
volatility, Journal of Portfolio Management, Winter, 74-80.
Neuberger, A. (1996). The Log Contract and Other Power Contracts, in
The Handbook of Exotic Options, edited by I. Nelken, 200-212.
50
Goldman
Sachs
SELECTED QUANTITATIVE STRATEGIES PUBLICATIONS

June 1990
Understanding Guaranteed Exchange-Rate

Contracts In Foreign Stock Investments
Emanuel Derman, Piotr Karasinski and Jeffrey Wecker
Jan. 1992
Valuing and Hedging Outperformance Options

Emanuel Derman
Mar. 1992
Pay-On-Exercise Options
Emanuel Derman and Iraj Kani
June 1993
The Ins and Outs of Barrier Options

Jan. 1994
The Volatility Smile and Its Implied Tree

May 1994
Static Options Replication

Emanuel Derman, Deniz Ergener and Iraj Kani
May 1995
Enhanced Numerical Methods for Options

with Barriers
Emanuel Derman, Iraj Kani, Deniz Ergener
and Indrajit Bardhan
Dec. 1995
The Local Volatility Surface: Unlocking the

Information in Index Option Prices
Emanuel Derman, Iraj Kani and Joseph Z. Zou
Feb. 1996
Implied Trinomial Trees of the Volatility Smile

Emanuel Derman, Iraj Kani and Neil Chriss
Apr. 1996
Model Risk
Emanuel Derman,
Aug. 1996
Trading and Hedging Local Volatility

Iraj Kani, Emanuel Derman and Michael Kamal
Oct. 1996
Investing in Volatility
Emanuel Derman, Michael Kamal, Iraj Kani,
John McClure, Cyrus Pirasteh and Joseph Zou
51
Goldman
Sachs
52
Apr. 1997
Is the Volatility Skew Fair?

Emanuel Derman, Michael Kamal, Iraj Kani
and Joseph Zou
Apr. 1997
Stochastic Implied Trees: Arbitrage Pricing with

Stochastic Term and Strike Structure of Volatility
Sept. 1997
The Patterns of Change in Implied Index Volatilities

Michael Kamal and Emanuel Derman
Nov. 1997
Predicting the Response of Implied Volatility to Large

Index Moves: An October 1997 S&P Case Study
Emanuel Derman and Joe Zou
Sept. 1998
How to Value and Hedge Options on Foreign Indexes

Kresimir Demeterfi
Jan. 1999
Regimes of Volatility: Some Observations on the

Variation of S&P 500 Implied Volatilities
Emanuel Derman

Gs Volatility Swaps

Uploaded by

Copyright:

Available Formats

Gs Volatility Swaps

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Gs Volatility Swaps

Uploaded by

Copyright:

Available Formats

What is the purpose of volatility swaps and variance swaps?

What is the purpose of volatility swaps and variance swaps?

What are some of the publications by the authors on quantitative strategies mentioned in the notes?

What are some of the publications by the authors on quantitative strategies mentioned in the notes?

Goldman

QUANTITATIVE STRATEGIES RESEARCH NOTES

Copyright 1999 Goldman, Sachs & Co. All rights reserved.

QUANTITATIVE STRATEGIES RESEARCH NOTES

QUANTITATIVE STRATEGIES RESEARCH NOTES

QUANTITATIVE STRATEGIES RESEARCH NOTES

The easy way to trade volatility is to use volatility swaps, sometimes

where R is the realized stock volatility (quoted in annual terms) over

1. For a discussion of volatility as an asset class, see Derman, Kamal,

QUANTITATIVE STRATEGIES RESEARCH NOTES

exceeded the volatility delivery price Kvol. He or she is swapping a fixed

Volatility has several characteristics that make trading attractive. It is

QUANTITATIVE STRATEGIES RESEARCH NOTES

Risk arbitrageurs or hedge funds often take positions which assume

Although options market participants talk of volatility, it is variance,

where R is the realized stock variance (quoted in annual terms) over

QUANTITATIVE STRATEGIES RESEARCH NOTES

Though theoretically simpler, variance swaps are less commonly

QUANTITATIVE STRATEGIES RESEARCH NOTES

QUANTITATIVE STRATEGIES RESEARCH NOTES

We approach variance replication by building on the readers assumed

is the stocks variance, and v = is the total variance of the

options values depend on the total variance .

Here, we define the sensitivity V =

QUANTITATIVE STRATEGIES RESEARCH NOTES

lio whose sensitivity to realized variance is independent of the stock

proportional to K produces a V that is virtually independent of stock

that options be weighted inversely proportional to K in order to

options, with weights inversely proportional to K .

expiration , weighted inversely proportional to K . Because out-of-

QUANTITATIVE STRATEGIES RESEARCH NOTES

FIGURE 1. The variance exposure, V, of portfolios of call options of different

QUANTITATIVE STRATEGIES RESEARCH NOTES

the-money options are generally more liquid, we employ put options

the total variance .

To obtain an initial exposure of $1 per volatility point squared, you

3. Formally, the expression for the portfolio is given by

QUANTITATIVE STRATEGIES RESEARCH NOTES

The first term in the payoff in Equation 6, (S S*)/S*, describes 1/S*

QUANTITATIVE STRATEGIES RESEARCH NOTES

For now, assume that we are in a Black-Scholes world where the

In Equation 6 we showed that, in a Black-Scholes world with zero

QUANTITATIVE STRATEGIES RESEARCH NOTES

The sensitivities of the value of this portfolio are precisely appropriate

The exposure to variance is equal to 1 at t = 0, and decreases linearly to

Equation 12 is the essence of the Black-Scholes options pricing theory.

QUANTITATIVE STRATEGIES RESEARCH NOTES

whose value is proportional to the remaining variance at any time.

A variance swap has a payoff proportional to realized variance. In this

QUANTITATIVE STRATEGIES RESEARCH NOTES

FIGURE 3. The variance vega, V , of a portfolio of puts and calls, weighted

inversely proportional to the square of the strike level, and chosen to

QUANTITATIVE STRATEGIES RESEARCH NOTES

In the previous section, we explained how to replicate a variance swap

This is a good approximation to the variance of daily returns used in

The first term in the payoff in Equation 6, (S S)/S, describes 1/S*