This action might not be possible to undo. Are you sure you want to continue?

**Peter Carr Dilip Madan
**

Morgan Stanley College of Business and Management

1585 Broadway, 6th Floor University of Maryland

New York, NY 10036 College Park, MD 20742

(212) 761-7340 (301) 405-2127

carrp@ms.com dbm@mbs.umd.edu

Current Version: January 30, 2002

We thank the participants of presentations at Boston University, the NYU Courant Institute, M.I.T., Mor-

gan Stanley, and the Risk 1997 Congress. We would also like to thank Marco Avellaneda, Joseph Cherian,

Stephen Chung, Emanuel Derman, Raphael Douady, Bruno Dupire, Ognian Enchev, Chris Fernandes,

Marvin Friedman Iraj Kani, Keith Lewis, Harry Mendell, Lisa Polsky, John Ryan, Murad Taqqu, Alan

White, and especially Robert Jarrow for useful discussions. They are not responsible for any errors.

I Introduction

Much research has been directed towards forecasting the volatility

1

of various macroeconomic variables

such as stock indices, interest rates and exchange rates. However, comparatively little research has been

directed towards the optimal way to invest given a view on volatility. This absence is probably due to the

belief that volatility is diﬃcult to trade. For this reason, a small literature has emerged which advocates

the development of volatility indices and the listing of ﬁnancial products whose payoﬀ is tied to these

indices. For example, Gastineau[16] and Galai[15] propose the development of option indices similar in

concept to stock indices. Brenner and Galai[4] propose the development of realized volatility indices and the

development of futures and options contracts on these indices. Similarly, Fleming, Ostdiek and Whaley[14]

describe the construction of an implied volatility index (the VIX), while Whaley[27] proposes derivative

contracts written on this index. Brenner and Galai[5, 6] develop a valuation model for options on volatility

using a binomial process, while Grunbichler and Longstaﬀ[18] instead assume a mean reverting process in

continuous time.

In response to this hue and cry, some volatility contracts have been listed. For example, the OMLX,

which is the London based subsidiary of the Swedish exchange OM, has launched volatility futures at the

beginning of 1997. At this writing, the Deutsche Terminborse (DTB) recently launched its own futures

based on its already established implied volatility index. Thus far, the volume in these contracts has been

disappointing.

One possible explanation for this outcome is that volatility can already be traded by combining static

positions in options on price with dynamic trading in the underlying. Neuberger[24] showed that by

delta-hedging a contract paying the log of the price, the hedging error accumulates to the diﬀerence

between the realized variance and the ﬁxed variance used in the delta-hedge. The contract paying the

log of the price can be created with a static position in options as shown in Breeden and Litzenberger[3].

Independently of Neuberger, Dupire[11] showed that a calendar spread of two such log contracts pays

1

In this article, the term “volatility” refers to either the variance or the standard deviation of the return on an investment.

1

the variance between the 2 maturities, and developed the notion of forward variance. Following Heath,

Jarrow and Morton[19](HJM), Dupire modelled the evolution of the term structure of this forward variance,

thereby developing the ﬁrst stochastic volatility model in which the market price of volatility risk does not

require speciﬁcation, even though volatility is imperfectly correlated with the price of the underlying.

The primary purpose of this article is to review three methods which have emerged for trading realized

volatility. The ﬁrst method reviewed involves taking static positions in options. The classic example is

that of a long position in a straddle, since the value usually

2

increases with a rise in volatility. The second

method reviewed involves delta-hedging an option position. If the investor is successful in hedging away

the price risk, then a prime determinant of the proﬁt or loss from this strategy is the diﬀerence between the

realized volatility and the anticipated volatility used in pricing and hedging the option. The ﬁnal method

reviewed for trading realized volatility involves buying or selling an over-the-counter contract whose payoﬀ

is an explicit function of volatility. The simplest example of such a volatility contract is a vol swap. This

contract pays the buyer the diﬀerence betweeen the realized volatility

3

and the ﬁxed swap rate determined

at the outset of the contract

4

.

A secondary purpose of this article is to uncover the link between volatility contracts and some recent

path-breaking work by Dupire[12] and by Derman, Kani, and Kamal[10](henceforth DKK). By restricting

the set of times and price levels for which returns are used in the volatility calculation, one can synthesize

a contract which pays oﬀ the “local volatility”, i.e. the volatility which will be experienced should the

underlying be at a speciﬁed price level at a speciﬁed future date. These authors develop the notion of

forward local volatility, which is the ﬁxed rate the buyer of the local vol swap pays at maturity in the

event the speciﬁed price level is reached. Given a complete term and strike structure of options, the entire

forward local volatility surface can be backed out from the prices of options. This surface is the two

dimensional analog of the forward rate curve central to the HJM analysis. Following HJM, these authors

impose a stochastic process on the forward local volatility surface and derive the risk-neutral dynamics of

2

Jagannathan[20] shows that in general options need not be increasing in volatility.

3

For marketing reasons, these contracts are usually written on the standard deviation, despite the focus of the literature

on spanning contracts on variance.

4

This contract is actually a forward contract on realized volatility, but is nonetheless termed a swap.

2

this surface.

The outline of this paper is as follows. The next section looks at trading realized volatility via static

positions in options. The theory of static replication using options is reviewed in order to develop some

new positions for proﬁting from a correct view on volatility. The subsequent section shows how dynamic

trading in the underlying can alternatively be used to create or hedge a volatility exposure. The fourth

section looks at over-the-counter volatility contracts as a further alternative for trading volatility. The

section shows how such contracts can be synthesized by combining static replication using options with

dynamic trading in the underlying asset. A ﬁfth section draws a link between these volatility contracts

and the work on forward local volatility pioneered by Dupire and DKK. The ﬁnal section summarizes and

suggests some avenues for future research.

II Trading Realized Volatility via Static Positions in Options

The classic position for gaining exposure to volatility is to buy an at-the-money

5

straddle. Since at-the-

money options are frequently used to trade volatility, the implied volatility from these options are widely

used as a forecast of subsequent realized volatility. The widespread use of this measure is surprising since

the approach relies on a model which itself assumes that volatility is constant.

This section derives an alternative forecast, which is also calculated from market prices of options. In

contrast to implied volatility, the forecast does not assume constant volatility, or even that the underlying

price process is continuous. In contrast to the implied volatility forecast, our forecast uses the market

prices of options of all strikes. In order to develop the alternative forecast, the next subsection reviews

the theory of static replication using options developed in Ross[26] and Breeden and Litzenberger[3]. The

following subsection applies this theory to determine a model-free forecast of subsequent realized volatility.

5

Note that in the Black model, the sensitivity to volatility of a straddle is actually maximized at slightly below the forward

price.

3

II-A Static Replication with Options

Consider a single period setting in which investments are made at time 0 with all payoﬀs being received at

time T. In contrast to the standard intertemporal model, we assume that there are no trading opportunities

other than at times 0 and T. We assume there exists a futures market in a risky asset (eg. a stock index)

for delivery at some date T

**≥ T. We also assume that markets exist for European-style futures options
**

6

of all strikes. While the assumption of a continuum of strikes is far from standard, it is essentially the

analog of the standard assumption of continuous trading. Just as the latter assumption is frequently made

as a reasonable approximation to an environment where investors can trade frequently, our assumption is

a reasonable approximation when there are a large but ﬁnite number of option strikes (eg. for S&P500

futures options).

It is widely recognized that this market structure allows investors to create any smooth function f(F

T

)

of the terminal futures price by taking a static position at time 0 in options

7

. Appendix 1 shows that any

twice diﬀerentiable payoﬀ can be re-written as:

f(F

T

) = f(κ) +f

(κ)[(F

T

−κ)

+

−(κ −F

T

)

+

]

+

κ

0

f

(K)(K −F

T

)

+

dK +

∞

κ

f

(K)(F

T

−K)

+

dK. (1)

The ﬁrst term can be interpreted as the payoﬀ from a static position in f(κ) pure discount bonds, each

paying one dollar at T. The second term can be interpreted as the payoﬀ from f

**(κ) calls struck at κ less
**

f

(κ) puts, also struck at κ. The third term arises from a static position in f

**(K)dK puts at all strikes
**

less than κ. Similarly, the fourth term arises from a static position in f

**(K)dK calls at all strikes greater
**

than κ.

In the absence of arbitrage, a decomposition similar to (1) must prevail among the initial values. Let

V

f

0

and B

0

denote the initial values of the payoﬀ and the pure discount bond respectively. Similarly, let

6

Note that listed futures options are generally American-style. However, by setting T

**= T, the underlying futures will
**

converge to the spot at T and so the assumption is that there exists European-style spot options in this special case.

7

This observation was ﬁrst noted in Breeden and Litzenberger[3] and established formally in Green and Jarrow[17] and

Nachman[23].

4

P

0

(K) and C

0

(K) denote the initial prices of the put and the call struck at K respectively. Then the no

arbitrage condition requires that:

V

f

0

= f(κ)B

0

+f

(κ)[C

0

(κ) −P

0

(κ)]

+

κ

0

f

(K)P

0

(K)dK +

∞

κ

f

(K)C

0

(K)dK. (2)

Thus, the value of an arbitary payoﬀ can be obtained from bond and option prices. Note that no assumption

was made regarding the stochastic process governing the futures price.

II-B An Alternative Forecast of Variance

Consider the problem of forecasting the variance of the log futures price relative ln

F

T

F

0

. For simplicity,

we refer to the log futures price relative as a return, even though no investment is required in a futures

contract. The variance of the return over some interval [0, T] is of course given by the expectation of the

squared deviation of the return from its mean:

Var

0

ln

F

T

F

0

= E

0

ln

F

T

F

0

−E

0

¸

ln

F

T

F

0

2

. (3)

It is well-known that futures prices are martingales under the appropriate risk-neutral measure. When

the futures contract marks to market continuously, then futures prices are martingales under the measure

induced by taking the money market account as numeraire. When the futures contract marks to market

daily, then futures prices are martingales under the measure induced by taking a daily rollover strategy

as numeraire, where this strategy involves rolling over pure discount bonds with maturities of one day.

Thus, given a mark-to-market frequency, futures prices are martingales under the measure induced by the

rollover strategy with the same rollover frequency.

If the variance in (3) is calculated using this measure, then E

0

ln

F

T

F

0

**can be interpreted as the
**

futures

8

price of a portfolio of options which pays oﬀ f

m

(F) ≡ ln

F

T

F

0

**at T. The spot value of this payoﬀ
**

8

Options do trade futures-style in Hong Kong. However, when only spot option prices are available, one can set T

= T

and calculate the mean and variance of the terminal spot under the forward measure. The variance is then expressed in

terms of the forward prices of options, which can be obtained from the spot price by dividing by the bond price.

5

is given by (2) with κ arbitrary and f

m

(K) =

−1

K

2

. Setting κ = F

0

, the futures price of the payoﬀ is given

by:

F ≡ E

0

¸

ln

F

T

F

0

= −

F

0

0

1

K

2

ˆ

P

0

(K, T)dK −

∞

F

0

1

K

2

ˆ

C

0

(K, T)dK,

where

ˆ

P

0

(K, T) and

ˆ

C

0

(K, T) denote the initial futures price of the put and the call respectively, both for

delivery at T. This futures price is initially negative

9

due to the concavity (negative time value) of the

payoﬀ.

Similarly, the variance of returns is just the futures price of the portfolio of options which pays

oﬀ f

v

(F) =

ln

F

T

F

0

−F

¸

2

at T (see Figure 1): The second derivative of this payoﬀ is f

v

(K) =

0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Futures Price

P

a

y

o

f

f

Payoff for Variance of Return

Figure 1: Payoﬀ for Variance of Return (F

0

= 1; F = −.09).

2

K

2

1 −ln

K

F

0

+F

**. This payoﬀ has zero value and slope at F
**

0

e

F

. Thus, setting κ = F

0

e

F

, the fu-

tures price of the payoﬀ is given by:

Var

0

ln

F

T

F

0

=

F

0

e

F

0

2

K

2

¸

1 −ln

K

F

0

+F

ˆ

P

0

(K, T)dK

+

∞

F

0

e

F

2

K

2

¸

1 −ln

K

F

0

+F

ˆ

C

0

(K, T)dK. (4)

At time 0, this futures price is an interesting alternative to implied or historical volatility as a forecast

of subsequent realized volatility. However, in common with any futures price, this forecast is a reﬂection

9

If the futures price process is a continuous semi-martingale, then Itˆ o’s lemma implies that E

0

ln

FT

F0

= −E

0

1

2

T

0

σ

2

t

dt,

where σ

t

is the volatility at time t.

6

of both statistical expected value and risk aversion. Consequently, by comparing this forecast with the

ex-post outcome, the market price of variance risk can be inferred. We will derive a simpler forecast of

variance in section IV under more restrictive assumptions, principally price continuity.

When compared to an at-the-money straddle, the static position in options used to create f

v

has

the advantage of maintaining sensitivity to volatility as the underlying moves away from its initial level.

Unfortunately, like straddles, these contracts can take on signiﬁcant price exposure once the underlying

moves away from its initial level. An obvious solution to this problem is to delta-hedge with the underlying.

The next section considers this alternative.

III Trading Realized Volatility by Delta-Hedging Options

The static replication results of the last section made no assumption whatsoever about the price process

or volatility process. In order to apply delta-hedging with the underlying futures, we now assume that

investors can trade continuously, that interest rates are constant, and that the underlying futures price

process is a continuous semi-martingale. Note that we maintain our previous assumption that the volatility

of the futures follows an arbitrary unknown stochastic process. While one could specify a stochastic process

and develop the correct delta-hedge in such a model, such an approach is subject to signiﬁcant model risk

since one is unlikely to guess the correct volatility process. Furthermore, such models generally require

dynamic trading in options which is costly in practice. Consequently, in what follows we leave the volatility

process unspeciﬁed and restrict dynamic strategies to the underlying alone. Speciﬁcally, we assume that

an investor follows the classic replication strategy speciﬁed by the Black model, with the delta calculated

using a constant volatility σ

h

. Since the volatility is actually stochastic

10

,, the replication will be imperfect

and the error results in either a proﬁt or a loss realized at the expiration of the hedge.

To uncover the magnitude of this P&L, let V (F, t; σ) denote the Black model value of a European-style

claim given that the current futures price is F and the current time is t. Note that the last argument of

10

In an interesting paper, Cherian and Jarrow[9] show the existence of an equilibrium in an incomplete economy where

investors believe the Black Scholes formula is valid even though volatility is stochastic.

7

V is the volatility used in the calculation of the value. In what follows, it will be convenient to have the

attempted replication occur over an arbitrary future period (T, T

**) rather than over (0, T). Consequently,
**

we assume that the underlying futures matures at some date T

≥ T

.

We suppose that an investor sells a European-style claim at T for the Black model value V (F

T

, T; σ

h

)

and holds

∂V

∂F

(F

t

, t; σ

h

) futures contracts over (T, T

**). Applying Itˆ o’s lemma to V (F, t; σ
**

h

)e

r(T

−t)

gives:

V (F

T

, T

; σ

h

) = V (F

T

, T; σ

h

)e

r(T

−T)

+

T

T

e

r(T

−t)

∂V

∂F

(F

t

, t; σ

h

)dF

t

(5)

+

T

T

e

r(T

−t)

¸

−rV (F

t

, t; σ

h

) +

∂V

∂t

(F

t

, t; σ

h

)

¸

dt +

T

T

e

r(T

−t)

∂

2

V

∂F

2

(F

t

, t; σ

h

)

F

2

t

2

σ

2

t

dt.

Now, by deﬁnition, V (F, t; σ

h

) solves the Black partial diﬀerential equation subject to a terminal condition:

−rV (F, t; σ

h

) +

∂V

∂t

(F, t; σ

h

) = −

σ

2

h

F

2

2

∂

2

V

∂F

2

(F, t; σ

h

), (6)

V (F, T

; σ

h

) = f(F). (7)

Substituting (6) and (7) in (5) and re-arranging gives:

f(F

T

)+

T

T

e

r(T

−t)

F

2

t

2

∂

2

V

∂F

2

(F

t

, t; σ

h

)(σ

2

h

−σ

2

t

)dt = V (F

T

, T; σ

h

)e

r(T

−T)

+

T

T

e

r(T

−t)

∂V

∂F

(F

t

, t; σ

h

)dF

t

. (8)

The right hand side is clearly the terminal value of a dynamic strategy comprising an investment at T of

V (F

T

, T; σ

h

) dollars in the riskless asset and a dynamic position in

∂V

∂F

(F

t

, t; σ

h

) futures contracts over the

time interval (T, T

**). Thus, the left hand side must also be the terminal value of this strategy, indicating
**

that the strategy misses its target f(F

T

) by:

P&L ≡

T

T

e

r(T

−t)

F

2

t

2

∂

2

V

∂F

2

(F

t

, t; σ

h

)(σ

2

h

−σ

2

t

)dt. (9)

Thus, when a claim is sold for the implied volatility σ

h

at T, the instantaneous P&L from delta-hedging

it over (T, T

) is

F

2

t

2

∂

2

V

∂F

2

(F

t

, t; σ

h

)(σ

2

h

− σ

2

t

), which is the diﬀerence between the hedge variance rate and

the realized variance rate, weighted by half the dollar gamma. Note that the P&L (hedging error) will be

zero if the realized instantaneous volatility σ

t

is constant at σ

h

. It is well known that claims with convex

payoﬀs have nonnegative gammas (

∂

2

V

∂F

2

(F

t

, t; σ

h

) ≥ 0) in the Black model. For such claims (eg. options),

if the hedge volatility is always less that the true volatility (σ

h

< σ

t

for all t ∈ [T, T

**]), then a loss results,
**

8

regardless of the path. Conversely, if the claim with a convex payoﬀ is sold for an implied volatility σ

h

which dominates

11

the subsequent realized volatility at all times, then delta-hedging at σ

h

using the Black

model delta guarantees a positive P&L.

When compared with static options positions, delta hedging appears to have the advantage of being

insensitive to the price of the underlying. However, (9) indicates that the P&L at T

does depend on

the ﬁnal price as well as on the price path. An investor with a view on volatility alone would like to

immunize the exposure to this path. One solution is to use a stochastic volatility model to conduct the

replication of the desired volatility dependent payoﬀ. However, as mentioned previously, this requires

specifying a volatility process and employing dynamic replication with options. A better solution is to

choose the payoﬀ function f(·), so that the path dependence can be removed or managed. For example,

Neuberger[24] recognized that if f(F) = 2 ln F, then

∂

2

V

∂F

2

(F

t

, t; σ

h

) = e

−r(T

−t) −2

F

2

t

and thus from (9), the

P&L at T

**is the payoﬀ of a variance swap
**

T

T

(σ

2

t

− σ

2

h

)dt. This volatility contract and others related to

it are explored in the next section.

IV Trading Realized Volatility by Using Volatility Contracts

This section shows that several interesting volatility contracts can be manufactured by taking options

positions and then delta-hedging them at zero volatility. Accordingly, suppose we set σ

h

= 0 in (8) and

negate both sides:

T

T

F

2

t

2

f

(F

t

)σ

2

t

dt = f(F

T

) −f(F

T

) −

T

T

f

(F

t

)dF

t

. (10)

The left hand side is a payoﬀ at T

**based on both the realized instantaneous volatility σ
**

2

t

and the price

path. The dependence of this payoﬀ on f arises only through f

**, and accordingly, we will henceforth only
**

consider payoﬀ functions f which have zero value and slope at a given point κ. The right hand side of (10)

depends only on the price path and results from adding the following three payoﬀs:

11

See El Karoui, Jeanblanc-Picque, and Shreve[13] for the extension of this result to the case when the hedger uses a

delta-hedging strategy assuming that volatility is a function of stock price and time. Also see Avellaneda et. al.[1][2] and

Lyons[22] for similar results.

9

1. The payoﬀ from a static position in options maturing at T

paying f(F

T

) at T

.

2. The payoﬀ from a static position in options maturing at T paying −e

−r(T

−T)

f(F

T

) and future-valued

to T

**3. The payoﬀ from maintaining a dynamic position in −e
**

−r(T

−t)

f

(F

t

) futures contracts over the time

interval (T, T

**) (assuming continuous marking-to-market and that the margin account balance earns
**

interest at the riskfree rate).

Thus, the payoﬀ on the left-hand side can be achieved by combining a static position in options as discussed

in section II, with a dynamic strategy in futures as discussed in section III. The dynamic strategy can be

interpreted as an attempt to create the payoﬀ −f(F

T

) at T

**, conducted under the false assumption of zero
**

volatility. Since realized volatility will be positive, an error arises, and the magnitude of this error is given

by

T

T

F

2

t

2

f

(F

t

)σ

2

t

dt, which is the left side of (10). The payoﬀ f(·) can be chosen so that when its second

derivative is substituted into this expression, the dependence on the path is consistent with the investor’s

joint view on volatility and price. In this section, we consider the following 3 second derivatives of payoﬀs

at T

**and work out the f(·) which leads to them:
**

Description of Payoﬀ f

(F

t

) Payoﬀ at T

**Variance over Future Period
**

2

F

2

t

T

T

σ

2

t

dt

Future Corridor Variance

2

F

2

t

1[F

t

∈ (κ −κ, κ +κ)]

T

T

1[F

t

∈ (κ −κ, κ +κ)]σ

2

t

dt

Future Variance Along Strike

2

κ

2

δ(F

t

−κ)

T

T

δ(F

t

−κ)σ

2

t

dt.

IV-A Contract Paying Future Variance

Consider the following payoﬀ function φ(F) (see Figure 2):

φ(F) ≡ 2

¸

ln

κ

F

+

F

κ

−1

, (11)

where κ is an arbitrary ﬁnite positive number. The ﬁrst derivative is given by:

φ

(F) = 2

¸

1

κ

−

1

F

. (12)

10

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0

0.5

1

1.5

2

2.5

3

3.5

4

Futures Price

P

a

y

o

f

f

Payoff to Delta Hedge to Create Variance

Figure 2: Payoﬀ to Delta-Hedge to Create Contract Paying Variance (κ = 1).

Thus, the value and slope both vanish at F = κ. The second derivative of φ is simply:

φ

(F) =

2

F

2

. (13)

Substituting (11) to (13) into (10) results in a relationship between a contract paying the realized variance

over the time interval (T, T

) and three payoﬀs based on price:

T

T

σ

2

t

dt = 2

¸

ln

κ

F

T

+

F

T

κ

−1

−2

¸

ln

κ

F

T

+

F

T

κ

−1

−2

T

T

¸

1

κ

−

1

F

t

dF

t

. (14)

The ﬁrst two terms on the right hand side arise from static positions in options. Substituting (13) into (2)

implies that for each term, the required position is given by:

2

¸

ln

κ

F

+

F

κ

−1

=

κ

0

2

K

2

(K −F)

+

dK +

∞

κ

2

K

2

(F −K)

+

dK, (15)

Thus, to create the contract paying

T

T

σ

2

t

dt at T

**, at t = 0, the investor should buy options at the
**

longer maturity T

and sell options at the nearer maturity T. The inital cost of this position is given by:

κ

0

2

K

2

P

0

(K, T

)dK +

∞

κ

2

K

2

C

0

(K, T

)dK

−e

−r(T

−T)

[

κ

0

2

K

2

P

0

(K, T)dK +

∞

κ

2

K

2

C

0

(K, T)dK]. (16)

When the nearer maturity options expire, the investor should borrow to ﬁnance the payout of

2e

−r(T

−T)

ln

κ

F

T

+

F

T

κ

−1

**. At this time, the investor should also start a dynamic strategy in futures,
**

11

holding −2e

−r(T

−t)

1

κ

−

1

Ft

futures contracts for each t ∈ [T, T

]. The net payoﬀ at T

is:

2

¸

ln

κ

F

T

+

F

T

κ

−1

−2

¸

ln

κ

F

T

+

F

T

κ

−1

−2

T

T

¸

1

κ

−

1

F

t

dF

t

=

T

T

σ

2

t

dt,

as required. Since the initial cost of achieving this payoﬀ is given by (16), an interesting forecast ˆ σ

2

T,T

of

the variance between T and T

**is given by the future value of this cost:
**

ˆ σ

2

T,T

= e

rT

κ

0

2

K

2

P

0

(K, T

)dK +

∞

κ

2

K

2

C

0

(K, T

)dK

−e

rT

¸

κ

0

2

K

2

P

0

(K, T)dK +

∞

κ

2

K

2

C

0

(K, T)dK

.

In contrast to implied volatility, this forecast does not use a model in which volatility is assumed to be

constant. However, in common with any forward price, this forecast is a reﬂection of both statistical

expected value and risk aversion. Consequently, by comparing this forecast with the ex-post outcome, the

market price of volatility risk can be inferred.

IV-B Contract Paying Future Corridor Variance

In this subsection, we generalize to a contract which pays the “corridor variance”, deﬁned as the variance

calculated using only the returns at times for which the futures price is within a speciﬁed corridor. In

particular, consider a corridor (κ − κ, κ +κ) centered at some arbitrary level κ and with width 2κ.

Suppose that we wish to generate a payoﬀ at T

of

T

T

1[F

t

∈ (κ − κ, κ + κ)]σ

2

t

dt. Thus, the variance

calculation is based only on returns at times in which the futures price is inside the cooridot.

Consider the following payoﬀ φ

κ

(·):

φ

κ

(F) ≡ 2

¸

ln

κ

¯

F

+F

1

κ

−

1

¯

F

, (17)

where:

¯

F

t

≡ max[κ −κ, min(F

t

, κ +κ)]

is the futures price ﬂoored at κ −κ and capped at κ +κ (see Figure 3):

12

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

Futures Price

P

a

y

o

f

f

Capped and Floored Futures Price

Figure 3: Futures Price Capped and Floored(κ = 1, κ = 0.5).

From inspection, the payoﬀ φ

κ

(·) is the same as φ deﬁned in (11), but with F replaced by

¯

F. The new

payoﬀ is graphed in Figure 4: This payoﬀ is actually a generalization of (11) since lim

κ↑∞

¯

F = F. For a ﬁnite

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Futures Price

P

a

y

o

f

f

Payoff to Delta Hedge to Create Corridor Variance

Figure 4: Trimming the Log Payoﬀ (κ = 1, κ = 0.5).

corridor width, the payoﬀ φ

κ

(F) matches φ(F) for futures prices within the corridor. Consequently, like

φ(F), φ

κ

(F) has zero value and slope at F = κ. However, in contrast to φ(F), φ

κ

(F) is linear outside

the corridor with the lines chosen so that the payoﬀ is continuous and diﬀerentiable at κ ±κ. The ﬁrst

derivative of (17) is given by:

φ

κ

(F) = 2

¸

1

κ

−

1

¯

F

, (18)

13

while the second derivative is simply:

φ

κ

(F) =

2

F

2

1[F ∈ (κ −κ, κ +κ)]. (19)

Substituting (17) to (19) into (10) implies that the volatility-based payoﬀ decomposes as:

T

T

σ

2

t

1[F

t

∈ (κ −κ, κ +κ)]dt = 2

¸

ln

κ

¯

F

T

+F

T

1

κ

−

1

¯

F

T

¸

−2

¸

ln

κ

¯

F

T

+F

T

1

κ

−

1

¯

F

T

¸

−2

T

T

¸

1

κ

−

1

¯

F

t

¸

dF

t

.

The payoﬀ function φ

κ

(·) has no curvature outside the corridor and consequently, the static positions

in options needed to create the ﬁrst two terms will not require strikes set outside the corridor. Thus, to

create the contract paying the future corridor variance,

T

T

σ

2

t

1[F

t

∈ (κ−κ, κ+κ)]dt at T

, the investor

should initially only buy and sell options struck within the corridor, for an initial cost of:

κ

κ−κ

2

K

2

P

0

(K, T

)dK +

κ+κ

κ

2

K

2

C

0

(K, T

)dK

−e

−r(T

−T)

[

κ

κ−κ

2

K

2

P

0

(K, T)dK +

κ+κ

κ

2

K

2

C

0

(K, T)dK].

At t = T, the investor should borrow to ﬁnance the payout of 2e

−r(T

−T)

ln

κ

¯

F

T

+F

T

1

κ

−

1

¯

F

T

from

having initially written the T maturity options. The investor should also start a dynamic strategy in

futures, holding −2e

−r(T

−t)

1

κ

−

1

¯

F

t

futures contracts for each t ∈ [T, T

**]. This strategy is semi-static in
**

that no trading is required when the futures price is outside the corridor. The net payoﬀ at T

is:

2

¸

ln

κ

¯

F

T

+F

T

1

κ

−

1

¯

F

T

¸

−2

¸

ln

κ

¯

F

T

+F

T

1

κ

−

1

¯

F

T

¸

−2

T

T

¸

1

κ

−

1

¯

F

t

¸

dF

t

=

T

T

σ

2

t

1[F

t

∈ (κ −κ, κ +κ)]dt,

as desired.

IV-C Contract Paying Future Variance Along a Strike

In the last subsection, only options struck within the corridor were used in the static options position, and

dynamic trading in the underlying futures was required ony when the futures price was in the corridor.

14

In this subsection, we shrink the width of the corridor of the last subsection down to a single point and

examine the impact on the volatility based payoﬀ and its replicating strategy. In order that this payoﬀ

have a non-negligible value, all asset positions in subsection IV-B must be re-scaled by

1

2κ

. Thus, the

volatility-based payoﬀ at T

would instead be

T

T

1[Ft∈(κ−κ,κ+κ)]

2κ

σ

2

t

dt. By letting κ ↓ 0, the variance

received can be completely localized in the spatial dimension to

T

T

δ(F

t

− κ)σ

2

t

dt, where δ(·) denotes a

Dirac delta function

12

. Recalling that only options struck within the corridor are used to create the corridor

variance, the initial cost of creating this localized cash ﬂow is given by the following ratioed calendar spread

of straddles:

1

κ

2

[V

0

(κ, T

) −e

−r(T

−T)

V

0

(κ, T)],

where V

0

(κ, T) is the initial cost of a straddle struck at κ and maturing at T:

V

0

(κ, T) ≡ P

0

(κ, T) +C

0

(κ, T).

As usual, at t = T, the investor should borrow to ﬁnance the payout of

|F

T

−κ|

κ

2

from having initially written

the T maturity straddle. The appendix proves that the dynamic strategy in futures initiated at T involves

holding −

e

−r(T

−t)

κ

2

sgn(F

t

−κ) futures contracts, where sgn(x) is the sign function:

sgn(x) ≡

−1 if x < 0;

0 if x = 0;

1 if x > 0.

When T = 0, this strategy reduces to the initial purchase of a straddle maturing at T

, initially borrowing

e

−rT

|F

0

− κ| dollars and holding −

e

−r(T

−t)

κ

2

sgn(F

t

− κ) futures contracts for t ∈ (0, T

). The component

of this strategy involving borrowing and futures is known as the stop-loss start-gain strategy, previously

investigated by Carr and Jarrow[7]. By the Tanaka-Meyer formula

13

, the diﬀerence between the payoﬀ

from the straddles and this dynamic strategy is known as the local time of the futures price process. Local

time is a fundamental concept in the study of one dimensional stochastic processes. Fortunately, a straddle

12

The Dirac delta function is a generalized function characterized by two properties:

1. δ(x) =

0 if x = 0

∞ if x = 0

2.

∞

−∞

δ(x)dx = 1.

. See Richards and Youn[25] for an accessible introduction to such generalized functions.

13

See Karatzas and Shreve[21], pg. 220.

15

combined with a stop-loss start-gain strategy in the underlying provides a mechanism for synthesizing a

contract paying oﬀ this fundamental concept. The initial time value of the straddle is the market’s (risk-

neutral) expectation of the local time. By comparing this time value with the ex-post outcome, the market

price of local time risk can be inferred.

V Connection to Recent Work on Stochastic Volatility

The last contract examined in the last section represents the limit of a localization in the futures price.

When a continuum of option maturities is also available, we may additionally localize in the time dimension

as has been done in some recent work by Dupire[12] and DKK[10]. Accordingly, suppose we further re-scale

all the asset positions described in subsection IV-C by

1

T

where T ≡ T

− T. The payoﬀ at T

would

instead be:

T

T

δ(F

t

−κ)

T

σ

2

t

dt.

The cost of creating this position would be:

1

κ

2

¸

V

0

(κ, T

) −e

−r(T

−T)

V

0

(κ, T)

T

¸

.

By letting T ↓ 0, one gets the beautiful result of Dupire[12] that

1

κ

2

∂V

0

∂T

(κ, T) +rV

0

(κ, T)

is the cost of

creating the payment δ(F

T

−κ)σ

2

T

at T. As shown in Dupire, the forward local variance can be deﬁned as

the number of butterﬂy spreads paying δ(F

T

−κ) at T one must sell in order to ﬁnance the above option

position initially. A discretized version of this result can be found in DKK[10]. One can go on to impose

a stochastic process on the forward local variance as in Dupire[12] and in DKK[10]. These authors derive

conditions on the risk-neutral drift of the forward local variance, allowing replication of price or volatility-

based payoﬀs using dynamic trading in only the underlying asset and a single option

14

. In contrast to

earlier work on stochastic volatility, the form of the market price of volatility risk need not be speciﬁed.

14

When two Brownian motions drive the price and the forward local volatility surface, any two assets whose payoﬀs are

not co-linear can be used to span.

16

Summary and Suggestions for Future Research

We reviewed three approaches for trading volatility. While static positions in options do generate exposure

to volatility, they also generate exposure to price. Similarly, a dynamic strategy in futures alone can yield

a volatility exposure, but always has a price exposure as well. By combining static positions in options

with dynamic trading in futures, payoﬀs related to realized volatility can be achieved which have either no

exposure to price, or which have an exposure contingent on certain price levels being achieved in speciﬁed

time intervals.

Under certain assumptions, we were able to price and hedge certain volatility contracts without spec-

ifying the process for volatility. The principle assumption made was that of price continuity. Under this

assumption, a calendar spread of options emerges as a simple tool for trading the local volatility (or local

time) between the two maturities. It would be interesting to see if this insight survives the relaxation

of the critical assumption of price continuity. It would also be interesting to consider contracts which

pay nonlinear functions of realized variance or local variance. Finally, it would be interesting to develop

contracts on other statistics of the sample path such as the Sharpe ratio, skewness, covariance, correlation,

etc. In the interests of brevity, such inquiries are best left for future research.

References

[1] Avellaneda, M., A. Levy, and A. Paras, 1995, “Pricing and Hedging Derivative Securities in Markets

with Uncertain Volatilities”, Applied Mathematical Finance, 2, 73–88.

[2] Avellaneda, M., A. Levy, and A. Paras, 1996, “Managing the Volatility Risk of Portfolios of Derivative

Securities: The Lagrangian Uncertain Volatility Model”, Applied Mathematical Finance, 3, 21–52.

[3] Breeden, D. and R. Litzenberger, 1978, “Prices of State Contingent Claims Implicit in Option Prices,”

Journal of Business, 51, 621-651.

17

[4] Brenner, M., and D. Galai, 1989, “New Financial Instruments for Hedging Changes in Volatility”,

Financial Analyst’s Journal, July-August 1989, 61–65.

[5] Brenner, M., and D. Galai, 1993, “Hedging Volatility in Foreign Currencies”, The Journal of Deriva-

tives, Fall 1993, 53–9.

[6] Brenner, M., and D. Galai, 1996, “Options on Volatility”, Chapter 13 of Option Embedded Bonds, I.

Nelken, ed. 273–286.

[7] Carr P. and R. Jarrow, 1990, “The Stop-Loss Start-Gain Strategy and Option Valuation: A New

Decomposition into Intrinsic and Time Value”, Review of Financial Studies, 3, 469–492.

[8] Carr P. and D. Madan, 1997, “Optimal Positioning in Derivative Securities”, Morgan Stanley working

paper.

[9] Cherian, J., and R. Jarrow, 1997, “Options Markets, Self-fulﬁlling Prophecies and Implied Volatilities”,

Boston University working paper, forthcoming in Review of Derivatives Research.

[10] Derman E., I. Kani, and M. Kamal, 1997, “Trading and Hedging Local Volatility” Journal of Financial

Engineering, 6, 3, 233-68.

[11] Dupire B., 1993, “Model Art”, Risk. Sept. 1993, pgs. 118 and 120.

[12] Dupire B., 1996, “A Uniﬁed Theory of Volatility”, Paribas working paper.

[13] El Karoui, N., M. Jeanblanc-Picque, and S. Shreve, 1996, “Robustness of the Black and Scholes

Formula”, Carnegie Mellon University working paper.

[14] Fleming, J., B. Ostdiek, and R. Whaley, 1993, “Predicting Stock Market Volatility: A New Measure”,

Duke University working paper.

[15] Galai, D., 1979, “A Proposal for Indexes for Traded Call Options”, Journal of Finance, XXXIV, 5,

1157–72.

18

[16] Gastineau, G., 1977, “An Index of Listed Option Premiums”, Financial Analyst’s Journal, May-June

1977.

[17] Green, R.C. and R.A. Jarrow, 1987, “Spanning and Completeness in markets with Contingent Claims,”

Journal of Economic Theory, 41, 202-210.

[18] Grunbichler A., and F. Longstaﬀ, 1993, “Valuing Options on Volatility”, UCLA working paper.

[19] Heath, D., R. Jarrow, and A. Morton, 1992, “Bond Pricing and the Term Structure of Interest Rates:

A New Methodology for Contingent Claim Valuation”, Econometrica, 66 77–105.

[20] Jagannathan R., 1984, “Call Options and the Risk of Underlying Securities”, Journal of Financial

Economics, 13, 3, 425–434.

[21] Karatzas, I., and S. Shreve, 1988, Brownian Motion and Stochastic Calculus, Springer Verlag, NY.

[22] Lyons, T., 1995, “Uncertain Volatility and the Risk-free Synthesis of Derivatives”, Applied Mathemat-

ical Finance, 2, 117-33.

[23] Nachman, D., 1988, “Spanning and Completeness with Options,” Review of Financial Studies, 3, 31,

311-328.

[24] Neuberger, A. 1990, “Volatility Trading”, London Business School working paper.

[25] Richards, J.I., and H.K. Youn, 1990 Theory of Distributions: A Non-technical Introduction, Cam-

bridge University Press, 1990.

[26] Ross, S., 1976, “Options and Eﬃciency”, Quarterly Journal of Economics, 90 Feb. 75–89.

[27] Whaley, R., 1993, “Derivatives on Market Volatility: Hedging Tools Long Overdue”, The Journal of

Derivatives, Fall 1993, 71–84.

19

Appendix 1: Spanning with Bonds and Options

For any payoﬀ f(F), the sifting property of a Dirac delta function implies:

f(F) =

∞

0

f(K)δ(F −K)dK

=

κ

0

f(K)δ(F −K)dK +

κ

0

f(K)δ(F −K)dK,

for any nonnegative κ. Integrating each integral by parts implies:

f(F) = f(K)1(F < K)

κ

0

−

κ

0

f

(K)1(F < K)dK

+f(K)1(F ≥ K)

∞

κ

+

∞

κ

f

(K)1(F ≥ K)dK.

Integrating each integral by parts once more implies:

f(F) = f(κ)1(F < κ) −f

(K)(K −F)

+

κ

0

+

κ

0

f

(K)(K −F)

+

dK

+f(κ)1(F ≥ κ) −f

(K)(F −K)

+

∞

κ

+

∞

κ

f

(K)(F −K)

+

dK

= f(κ) +f

(κ)[(F −κ)

+

−(κ −F)

+

]

+

κ

0

f

(K)(K −F)

+

dK +

∞

κ

f

(K)(F −K)

+

dK.

20

Appendix 2: Derivation of Futures Position When Synthesizing

Contract Paying Future Variance Along a Strike

Recall from section IV-C, that all asset positions in section IV-B were normalized by multiplying by

1

2κ

.

Thus in particular, the futures position of −2e

−r(T

−t)

1

κ

−

1

¯

F

t

**contracts in subsection IV-B is changed to
**

−

e

−r(T

−t)

κ

1

κ

−

1

¯

F

t

contracts in subsection IV-C. More explicitly, the number of contracts held is given by

−

e

−r(T

−t)

κ

1

κ

−

1

κ−κ

if F

t

≤ κ −κ;

−

e

−r(T

−t)

κ

1

κ

−

1

Ft

if F

t

∈ (κ −κ, κ +κ);

−

e

−r(T

−t)

κ

1

κ

−

1

κ+κ

if F

t

≥ κ +κ.

Now, by Taylor’s series:

1

κ −κ

=

1

κ

+

1

κ

2

κ +O(κ

2

)

and:

1

κ +κ

=

1

κ

−

1

κ

2

κ +O(κ

2

).

Substitution implies that the number of futures contracts held is given by:

−

e

−r(T

−t)

κ

−

1

κ

2

κ +O(κ

2

)

if F

t

≤ κ −κ;

−

e

−r(T

−t)

κ

1

κ

−

1

Ft

if F

t

∈ (κ −κ, κ +κ);

−

e

−r(T

−t)

κ

1

κ

2

κ +O(κ

2

)

if F

t

≥ κ +κ.

Thus, as κ ↓ 0, the number of futures contracts held converges to −

e

−r(T

−t)

κ

2

sgn(F

t

−κ), where sgn(x)

is the sign function:

sgn(x) ≡

−1 if x < 0;

0 if x = 0;

1 if x > 0.

21

I

Introduction

Much research has been directed towards forecasting the volatility1 of various macroeconomic variables such as stock indices, interest rates and exchange rates. However, comparatively little research has been directed towards the optimal way to invest given a view on volatility. This absence is probably due to the belief that volatility is diﬃcult to trade. For this reason, a small literature has emerged which advocates the development of volatility indices and the listing of ﬁnancial products whose payoﬀ is tied to these indices. For example, Gastineau[16] and Galai[15] propose the development of option indices similar in concept to stock indices. Brenner and Galai[4] propose the development of realized volatility indices and the development of futures and options contracts on these indices. Similarly, Fleming, Ostdiek and Whaley[14] describe the construction of an implied volatility index (the VIX), while Whaley[27] proposes derivative contracts written on this index. Brenner and Galai[5, 6] develop a valuation model for options on volatility using a binomial process, while Grunbichler and Longstaﬀ[18] instead assume a mean reverting process in continuous time. In response to this hue and cry, some volatility contracts have been listed. For example, the OMLX, which is the London based subsidiary of the Swedish exchange OM, has launched volatility futures at the beginning of 1997. At this writing, the Deutsche Terminborse (DTB) recently launched its own futures based on its already established implied volatility index. Thus far, the volume in these contracts has been disappointing. One possible explanation for this outcome is that volatility can already be traded by combining static positions in options on price with dynamic trading in the underlying. Neuberger[24] showed that by delta-hedging a contract paying the log of the price, the hedging error accumulates to the diﬀerence between the realized variance and the ﬁxed variance used in the delta-hedge. The contract paying the log of the price can be created with a static position in options as shown in Breeden and Litzenberger[3]. Independently of Neuberger, Dupire[11] showed that a calendar spread of two such log contracts pays

1

In this article, the term “volatility” refers to either the variance or the standard deviation of the return on an investment.

1

thereby developing the ﬁrst stochastic volatility model in which the market price of volatility risk does not require speciﬁcation. even though volatility is imperfectly correlated with the price of the underlying.the variance between the 2 maturities. 3 2 2 . These authors develop the notion of forward local volatility. This contract pays the buyer the diﬀerence betweeen the realized volatility3 and the ﬁxed swap rate determined at the outset of the contract4 . Dupire modelled the evolution of the term structure of this forward variance. Following HJM. since the value usually2 increases with a rise in volatility. The classic example is that of a long position in a straddle. one can synthesize a contract which pays oﬀ the “local volatility”. A secondary purpose of this article is to uncover the link between volatility contracts and some recent path-breaking work by Dupire[12] and by Derman.e. Jarrow and Morton[19](HJM). Following Heath. For marketing reasons. By restricting the set of times and price levels for which returns are used in the volatility calculation. the volatility which will be experienced should the underlying be at a speciﬁed price level at a speciﬁed future date. despite the focus of the literature on spanning contracts on variance. and developed the notion of forward variance. The second method reviewed involves delta-hedging an option position. but is nonetheless termed a swap. Given a complete term and strike structure of options. The ﬁnal method reviewed for trading realized volatility involves buying or selling an over-the-counter contract whose payoﬀ is an explicit function of volatility. and Kamal[10](henceforth DKK). The ﬁrst method reviewed involves taking static positions in options. which is the ﬁxed rate the buyer of the local vol swap pays at maturity in the event the speciﬁed price level is reached. 4 This contract is actually a forward contract on realized volatility. these authors impose a stochastic process on the forward local volatility surface and derive the risk-neutral dynamics of Jagannathan[20] shows that in general options need not be increasing in volatility. i. these contracts are usually written on the standard deviation. The primary purpose of this article is to review three methods which have emerged for trading realized volatility. This surface is the two dimensional analog of the forward rate curve central to the HJM analysis. Kani. the entire forward local volatility surface can be backed out from the prices of options. The simplest example of such a volatility contract is a vol swap. If the investor is successful in hedging away the price risk. then a prime determinant of the proﬁt or loss from this strategy is the diﬀerence between the realized volatility and the anticipated volatility used in pricing and hedging the option.

5 3 . the next subsection reviews the theory of static replication using options developed in Ross[26] and Breeden and Litzenberger[3]. The theory of static replication using options is reviewed in order to develop some new positions for proﬁting from a correct view on volatility. The section shows how such contracts can be synthesized by combining static replication using options with dynamic trading in the underlying asset. Since at-themoney options are frequently used to trade volatility. The fourth section looks at over-the-counter volatility contracts as a further alternative for trading volatility. The widespread use of this measure is surprising since the approach relies on a model which itself assumes that volatility is constant. the forecast does not assume constant volatility. In order to develop the alternative forecast. which is also calculated from market prices of options. our forecast uses the market prices of options of all strikes. The following subsection applies this theory to determine a model-free forecast of subsequent realized volatility. or even that the underlying price process is continuous. In contrast to implied volatility. the implied volatility from these options are widely used as a forecast of subsequent realized volatility. The next section looks at trading realized volatility via static positions in options. The subsequent section shows how dynamic trading in the underlying can alternatively be used to create or hedge a volatility exposure.this surface. A ﬁfth section draws a link between these volatility contracts and the work on forward local volatility pioneered by Dupire and DKK. The ﬁnal section summarizes and suggests some avenues for future research. In contrast to the implied volatility forecast. The outline of this paper is as follows. Note that in the Black model. the sensitivity to volatility of a straddle is actually maximized at slightly below the forward price. II Trading Realized Volatility via Static Positions in Options The classic position for gaining exposure to volatility is to buy an at-the-money5 straddle. This section derives an alternative forecast.

let Note that listed futures options are generally American-style. 6 4 .II-A Static Replication with Options Consider a single period setting in which investments are made at time 0 with all payoﬀs being received at time T . 7 This observation was ﬁrst noted in Breeden and Litzenberger[3] and established formally in Green and Jarrow[17] and Nachman[23]. a stock index) for delivery at some date T ≥ T . We assume there exists a futures market in a risky asset (eg. for S&P500 futures options). it is essentially the analog of the standard assumption of continuous trading. Just as the latter assumption is frequently made as a reasonable approximation to an environment where investors can trade frequently. the fourth term arises from a static position in f (K)dK calls at all strikes greater than κ. In contrast to the standard intertemporal model. Appendix 1 shows that any twice diﬀerentiable payoﬀ can be re-written as: f (FT ) = f (κ) + f (κ)[(FT − κ)+ − (κ − FT )+ ] + κ 0 f (K)(K − FT )+ dK + ∞ κ f (K)(FT − K)+ dK. However. we assume that there are no trading opportunities other than at times 0 and T . (1) The ﬁrst term can be interpreted as the payoﬀ from a static position in f (κ) pure discount bonds. The second term can be interpreted as the payoﬀ from f (κ) calls struck at κ less f (κ) puts. It is widely recognized that this market structure allows investors to create any smooth function f (FT ) of the terminal futures price by taking a static position at time 0 in options7 . by setting T = T . The third term arises from a static position in f (K)dK puts at all strikes less than κ. our assumption is a reasonable approximation when there are a large but ﬁnite number of option strikes (eg. a decomposition similar to (1) must prevail among the initial values. Similarly. In the absence of arbitrage. While the assumption of a continuum of strikes is far from standard. the underlying futures will converge to the spot at T and so the assumption is that there exists European-style spot options in this special case. Similarly. We also assume that markets exist for European-style futures options6 of all strikes. each paying one dollar at T . also struck at κ. Let V0f and B0 denote the initial values of the payoﬀ and the pure discount bond respectively.

For simplicity. then futures prices are martingales under the measure induced by taking the money market account as numeraire. then E0 ln futures8 price of a portfolio of options which pays oﬀ fm (F ) ≡ ln 8 FT F0 can be interpreted as the FT F0 at T . when only spot option prices are available. then futures prices are martingales under the measure induced by taking a daily rollover strategy as numeraire. When the futures contract marks to market continuously. Thus.P0 (K) and C0 (K) denote the initial prices of the put and the call struck at K respectively. T ] is of course given by the expectation of the squared deviation of the return from its mean: Var0 ln FT F0 = E0 ln FT F0 − E0 ln FT F0 2 . When the futures contract marks to market daily. even though no investment is required in a futures contract. However. futures prices are martingales under the measure induced by the rollover strategy with the same rollover frequency. given a mark-to-market frequency. II-B An Alternative Forecast of Variance FT F0 Consider the problem of forecasting the variance of the log futures price relative ln . (2) Thus. The spot value of this payoﬀ Options do trade futures-style in Hong Kong. Note that no assumption was made regarding the stochastic process governing the futures price. we refer to the log futures price relative as a return. one can set T = T and calculate the mean and variance of the terminal spot under the forward measure. which can be obtained from the spot price by dividing by the bond price. The variance of the return over some interval [0. If the variance in (3) is calculated using this measure. where this strategy involves rolling over pure discount bonds with maturities of one day. (3) It is well-known that futures prices are martingales under the appropriate risk-neutral measure. 5 . The variance is then expressed in terms of the forward prices of options. Then the no arbitrage condition requires that: V0f = f (κ)B0 + f (κ)[C0 (κ) − P0 (κ)] + κ 0 f (K)P0 (K)dK + ∞ κ f (K)C0 (K)dK. the value of an arbitary payoﬀ can be obtained from bond and option prices.

Similarly. in common with any futures price. K2 ˆ ˆ where P0 (K.5 Figure 1: Payoﬀ for Variance of Return (F0 = 1. both for delivery at T . the variance of returns is just the futures price of the portfolio of options which pays oﬀ fv (F ) = ln FT F0 −F 2 at T (see Figure 1): The second derivative of this payoﬀ is fv (K) = Payoff for Variance of Return 0.15 0.6 0. T )dK F0 K ˆ + F C0 (K. T ) denote the initial futures price of the put and the call respectively. Thus.25 Payoff 0. This payoﬀ has zero value and slope at F0 eF .3 1.4 1.05 0 0.2 1. T ) and C0 (K. the fu- tures price of the payoﬀ is given by: Var0 FT ln F0 = 2 1 − ln K2 0 ∞ 2 + 1 − ln F 0 eF K 2 F 0 eF K ˆ + F P0 (K.5 0. F = −. then Itˆ’s lemma implies that E0 ln o FT F0 = −E0 1 2 T 0 2 σt dt.is given by (2) with κ arbitrary and fm (K) = by: F ≡ E0 ln FT F0 =− F0 0 −1 . this forecast is a reﬂection 9 If the futures price process is a continuous semi-martingale. However.09).4 0. T )dK.3 0. This futures price is initially negative9 due to the concavity (negative time value) of the payoﬀ.1 0. T )dK − K2 1 ˆ C0 (K.7 0.2 0.35 0.8 0.1 Futures Price 1. this futures price is an interesting alternative to implied or historical volatility as a forecast of subsequent realized volatility.9 1 1. 6 . the futures price of the payoﬀ is given ∞ F0 1 ˆ P0 (K. K2 Setting κ = F0 . 2 K2 1 − ln K F0 + F . where σt is the volatility at time t. T )dK. F0 (4) At time 0. setting κ = F0 eF .

principally price continuity. To uncover the magnitude of this P &L. these contracts can take on signiﬁcant price exposure once the underlying moves away from its initial level. While one could specify a stochastic process and develop the correct delta-hedge in such a model. Unfortunately. in what follows we leave the volatility process unspeciﬁed and restrict dynamic strategies to the underlying alone. such an approach is subject to signiﬁcant model risk since one is unlikely to guess the correct volatility process. with the delta calculated using a constant volatility σh . When compared to an at-the-money straddle. Consequently. σ) denote the Black model value of a European-style claim given that the current futures price is F and the current time is t. Note that we maintain our previous assumption that the volatility of the futures follows an arbitrary unknown stochastic process. We will derive a simpler forecast of variance in section IV under more restrictive assumptions. An obvious solution to this problem is to delta-hedge with the underlying. the replication will be imperfect and the error results in either a proﬁt or a loss realized at the expiration of the hedge. Since the volatility is actually stochastic10 . Consequently. let V (F. such models generally require dynamic trading in options which is costly in practice. Cherian and Jarrow[9] show the existence of an equilibrium in an incomplete economy where investors believe the Black Scholes formula is valid even though volatility is stochastic. the market price of variance risk can be inferred. Speciﬁcally. In order to apply delta-hedging with the underlying futures.of both statistical expected value and risk aversion. like straddles. and that the underlying futures price process is a continuous semi-martingale. by comparing this forecast with the ex-post outcome. The next section considers this alternative. t. we now assume that investors can trade continuously.. 7 . that interest rates are constant. Note that the last argument of 10 In an interesting paper. Furthermore. III Trading Realized Volatility by Delta-Hedging Options The static replication results of the last section made no assumption whatsoever about the price process or volatility process. we assume that an investor follows the classic replication strategy speciﬁed by the Black model. the static position in options used to create fv has the advantage of maintaining sensitivity to volatility as the underlying moves away from its initial level.

T . t. σh )er(T 2 ∂F 2 −T ) + T T er(T −t) ∂V ∂F (Ft . t. (8) The right hand side is clearly the terminal value of a dynamic strategy comprising an investment at T of V (FT . Substituting (6) and (7) in (5) and re-arranging gives: f (FT )+ T T (6) (7) er(T 2 −t) Ft ∂2V 2 2 (Ft . σh )dFt . t. the instantaneous P&L from delta-hedging it over (T. 2 ∂F 2 (9) Thus. 8 2 . T . t. t. t. then a loss results. T . t. Applying Itˆ’s lemma to V (F. σh ) ≥ 0) in the Black model. Consequently. We suppose that an investor sells a European-style claim at T for the Black model value V (FT . σh )er(T + T T + T er(T −t) ∂V er(T −t) T T er(T −t) ∂ V F2 2 (Ft . σh ) = f (F ). which is the diﬀerence between the hedge variance rate and the realized variance rate. σh ) solves the Black partial diﬀerential equation subject to a terminal condition: −rV (F. Thus. σh )er(T o (Ft . ∂F 2 if the hedge volatility is always less that the true volatility (σh < σt for all t ∈ [T. For such claims (eg. t. weighted by half the dollar gamma. σh ) t σt dt. σh ) dollars in the riskless asset and a dynamic position in ∂V ∂F (Ft . the left hand side must also be the terminal value of this strategy. it will be convenient to have the attempted replication occur over an arbitrary future period (T. ∂F 2 2 2 Now. σh )dFt ∂F ∂V (Ft . indicating that the strategy misses its target f (FT ) by: P &L ≡ T T er(T 2 −t) Ft ∂2V 2 2 (Ft . t. It is well known that claims with convex payoﬀs have nonnegative gammas ( ∂ V (Ft . V (F. σh ) futures contracts over (T. t. we assume that the underlying futures matures at some date T ≥ T . T ). σh ) dt + −rV (Ft . t. T . σh )(σh − σt )dt. σh )(σh 2 ∂F 2 2 − σt ). σh ) and holds ∂V ∂F (Ft . Note that the P&L (hedging error) will be zero if the realized instantaneous volatility σt is constant at σh . T ) is 2 Ft ∂ 2 V 2 (Ft .V is the volatility used in the calculation of the value. T ). T ). T ) rather than over (0. σh ) = V (FT . T ]). (F. In what follows. options). σh ) = − h ∂t 2 ∂F 2 V (F. t. when a claim is sold for the implied volatility σh at T . t. σh ) + ∂t T −T ) −t) gives: (5) V (FT . σh ) + ∂V σ2 F 2 ∂2 V (F. σh )(σh −σt )dt = V (FT . t. by deﬁnition. σh ) futures contracts over the time interval (T. t. σh ). t. T . T .

if the claim with a convex payoﬀ is sold for an implied volatility σh which dominates11 the subsequent realized volatility at all times. σh ) ∂F 2 = e−r(T −t) −2 2 Ft and thus from (9). However. suppose we set σh = 0 in (8) and negate both sides: T T Ft2 2 f (Ft )σt dt = f (FT ) − f (FT ) − 2 T T f (Ft )dFt . One solution is to use a stochastic volatility model to conduct the replication of the desired volatility dependent payoﬀ. so that the path dependence can be removed or managed. For example. then delta-hedging at σh using the Black model delta guarantees a positive P&L. we will henceforth only consider payoﬀ functions f which have zero value and slope at a given point κ. t. (10) 2 The left hand side is a payoﬀ at T based on both the realized instantaneous volatility σt and the price path. This volatility contract and others related to IV Trading Realized Volatility by Using Volatility Contracts This section shows that several interesting volatility contracts can be manufactured by taking options positions and then delta-hedging them at zero volatility. this requires specifying a volatility process and employing dynamic replication with options. delta hedging appears to have the advantage of being insensitive to the price of the underlying. Neuberger[24] recognized that if f (F ) = 2 ln F .[1][2] and Lyons[22] for similar results. Conversely. The right hand side of (10) depends only on the price path and results from adding the following three payoﬀs: See El Karoui. T T ∂2V (Ft . as mentioned previously. An investor with a view on volatility alone would like to immunize the exposure to this path. (9) indicates that the P&L at T does depend on the ﬁnal price as well as on the price path. Accordingly. The dependence of this payoﬀ on f arises only through f . and Shreve[13] for the extension of this result to the case when the hedger uses a delta-hedging strategy assuming that volatility is a function of stock price and time. the 2 2 (σt − σh )dt.regardless of the path. Also see Avellaneda et. However. and accordingly. Jeanblanc-Picque. A better solution is to choose the payoﬀ function f (·). 11 9 . al. When compared with static options positions. then P&L at T is the payoﬀ of a variance swap it are explored in the next section.

The ﬁrst derivative is given by: φ (F ) = 2 10 1 1 − .1. T ) (assuming continuous marking-to-market and that the margin account balance earns interest at the riskfree rate). The payoﬀ from a static position in options maturing at T paying −e−r(T to T 3. In this section. which is the left side of (10). The dynamic strategy can be interpreted as an attempt to create the payoﬀ −f (FT ) at T . κ + 2 κ)]σt dt 2 δ(Ft − κ)σt dt. κ + κ)] T T T T 2 2 Ft 2 2 1[Ft Ft 2 δ(Ft κ2 1[Ft ∈ (κ − κ. an error arises. 2. κ F (12) . conducted under the false assumption of zero volatility. Thus. κ (11) where κ is an arbitrary ﬁnite positive number. the payoﬀ on the left-hand side can be achieved by combining a static position in options as discussed in section II. and the magnitude of this error is given by 2 T Ft f T 2 2 (Ft )σt dt. The payoﬀ from maintaining a dynamic position in −e−r(T −t) −T ) f (FT ) and future-valued f (Ft ) futures contracts over the time interval (T. Since realized volatility will be positive. we consider the following 3 second derivatives of payoﬀs at T and work out the f (·) which leads to them: Description of Payoﬀ Variance over Future Period Future Corridor Variance Future Variance Along Strike f (Ft ) Payoﬀ at T T 2 T σt dt ∈ (κ − − κ) κ. IV-A Contract Paying Future Variance Consider the following payoﬀ function φ(F ) (see Figure 2): φ(F ) ≡ 2 ln κ F + F −1 . The payoﬀ f (·) can be chosen so that when its second derivative is substituted into this expression. the dependence on the path is consistent with the investor’s joint view on volatility and price. with a dynamic strategy in futures as discussed in section III. The payoﬀ from a static position in options maturing at T paying f (FT ) at T .

the value and slope both vanish at F = κ. the investor should borrow to ﬁnance the payout of 2e−r(T −T ) ln κ FT + FT κ − 1 . κ Ft (14) The ﬁrst two terms on the right hand side arise from static positions in options. Substituting (13) into (2) implies that for each term. the required position is given by: 2 ln κ F + F −1 = κ T T κ 0 2 (K − F )+ dK + K2 ∞ κ 2 (F − K)+ dK.4 0. When the nearer maturity options expire.5 1 0. The inital cost of this position is given by: κ 2 0 K 2 P0 (K. K2 (15) Thus.5 3 2. T )dK + ∞ 2 κ K 2 C0 (K. T )dK (16) −e−r(T −T ) [ κ 2 0 K 2 P0 (K.4 1. At this time.6 1.5 0 0 0. T )dK + ∞ 2 κ K 2 C0 (K. the investor should also start a dynamic strategy in futures. 11 . the investor should buy options at the longer maturity T and sell options at the nearer maturity T . T )dK].8 1 1. T ) and three payoﬀs based on price: T T 2 σt dt κ = 2 ln FT κ FT − 1 − 2 ln + κ FT FT −1 −2 + κ T T 1 1 − dFt . Thus.2 Futures Price 1.Payoff to Delta Hedge to Create Variance 4 3. The second derivative of φ is simply: φ (F ) = 2 . at t = 0.6 0.5 Payoff 2 1.8 2 Figure 2: Payoﬀ to Delta-Hedge to Create Contract Paying Variance (κ = 1).2 0. to create the contract paying 2 σt dt at T . F2 (13) Substituting (11) to (13) into (10) results in a relationship between a contract paying the realized variance over the time interval (T.

T )dK + C0 (K. T )dK . T ]. consider a corridor (κ − κ. κ + κ) centered at some arbitrary level κ and with width 2 κ. this forecast is a reﬂection of both statistical expected value and risk aversion. deﬁned as the variance calculated using only the returns at times for which the futures price is within a speciﬁed corridor. IV-B Contract Paying Future Corridor Variance In this subsection. Since the initial cost of achieving this payoﬀ is given by (16). the variance calculation is based only on returns at times in which the futures price is inside the cooridot. the market price of volatility risk can be inferred. Consequently. by comparing this forecast with the ex-post outcome. κ + κ)] κ (·): κ (F ) ≡ 2 ln κ ¯ +F F 1 1 − ¯ κ F .T ˆ2 = erT −erT κ 0 ∞ 2 2 P0 (K. Consider the following payoﬀ φ φ where: ¯ F t ≡ max[κ − is the futures price ﬂoored at κ − κ. min(Ft .T of ˆ2 the variance between T and T is given by the future value of this cost: σT. Thus. in common with any forward price. T T Suppose that we wish to generate a payoﬀ at T of 1[Ft ∈ (κ − κ. T as required. 0 K2 κ K2 In contrast to implied volatility. However. κ + 2 κ)]σt dt. T )dK + C0 (K. we generalize to a contract which pays the “corridor variance”. (17) κ and capped at κ + κ (see Figure 3): 12 . T )dK K2 κ K2 κ 2 ∞ 2 P0 (K. an interesting forecast σT. In particular. this forecast does not use a model in which volatility is assumed to be constant. The net payoﬀ at T is: FT + −1 −2 κ T κ 2 ln FT κ FT + − 1 − 2 ln κ FT T 1 1 dFt = − κ Ft T 2 σt dt.holding −2e−r(T −t) 1 κ − 1 Ft futures contracts for each t ∈ [T.

5).6 0. the payoﬀ φ κ (·) κ = 0.4 0.8 2 Figure 3: Futures Price Capped and Floored(κ = 1.2 Futures Price 1.4 1.4 0.Capped and Floored Futures Price 2 1.8 1.1 0 0 0.4 Payoff 0. the payoﬀ φ φ(F ). like κ (F ) has zero value and slope at F = κ.4 1.3 0.2 0.6 0.2 0 0 0.2 0.8 2 Figure 4: Trimming the Log Payoﬀ (κ = 1.7 0.6 1.5 0. φ κ (F ) κ (F ) κ = 0.5). ¯ is the same as φ deﬁned in (11).4 0.6 0. φ is linear outside κ. From inspection. κ F (18) 13 . in contrast to φ(F ). matches φ(F ) for futures prices within the corridor.4 1.8 1 1.2 Futures Price 1.6 1.6 1.6 0.2 Payoff 1 0.8 0. However. Consequently. The ﬁrst the corridor with the lines chosen so that the payoﬀ is continuous and diﬀerentiable at κ ± derivative of (17) is given by: φ κ (F ) =2 1 1 − ¯ . For a ﬁnite Payoff to Delta Hedge to Create Corridor Variance 0.2 0. The new κ↑∞ ¯ payoﬀ is graphed in Figure 4: This payoﬀ is actually a generalization of (11) since lim F = F .8 1 1. corridor width. but with F replaced by F .

T κ K2 0 )dK −e−r(T −T ) [ κ 2 κ− κ K 2 P0 (K. The investor should also start a dynamic strategy in futures. to create the contract paying the future corridor variance. κ K2 0 −T ) At t = T . T )dK + κ+ κ 2 C (K. κ + κ)]dt at T . The net payoﬀ at T is: 2 ln −2 as desired. T T κ ¯ FT + FT 1 1 − ¯ κ FT T T − 2 ln κ ¯ FT + FT κ. This strategy is semi-static in that no trading is required when the futures price is outside the corridor. κ + κ)]. the investor should initially only buy and sell options struck within the corridor. the investor should borrow to ﬁnance the payout of 2e−r(T ln κ ¯ FT + FT 1 κ − 1 ¯ FT from having initially written the T maturity options. only options struck within the corridor were used in the static options position. κ + κ)]dt = 2 ln −2 T T κ ¯ FT + FT 1 1 − ¯ κ FT − 2 ln κ ¯ FT + FT 1 1 − ¯ κ FT 1 1 − ¯ dFt . T ]. T )dK]. and dynamic trading in the underlying futures was required ony when the futures price was in the corridor. T )dK + κ+ κ 2 C (K. the static positions in options needed to create the ﬁrst two terms will not require strikes set outside the corridor.while the second derivative is simply: φ κ (F ) = 2 1[F ∈ (κ − F2 κ. for an initial cost of: κ 2 κ− κ K 2 P0 (K. T T 2 σt 1[Ft ∈ (κ − κ. holding −2e−r(T −t) 1 κ − 1 ¯ Ft futures contracts for each t ∈ [T. κ Ft The payoﬀ function φ κ (·) has no curvature outside the corridor and consequently. κ + 1 1 − ¯ κ FT κ)]dt. Thus. 14 . (19) Substituting (17) to (19) into (10) implies that the volatility-based payoﬀ decomposes as: T T 2 σt 1[Ft ∈ (κ − κ. 1 1 − ¯ dFt = κ Ft 2 σt 1[Ft ∈ (κ − IV-C Contract Paying Future Variance Along a Strike In the last subsection.

T )]. initially borrowing e−rT |F0 − κ| dollars and holding − e −r(T −t) κ2 sgn(Ft − κ) futures contracts for t ∈ (0. When T = 0. the By letting κ ↓ 0. previously investigated by Carr and Jarrow[7]. As usual. T ) is the initial cost of a straddle struck at κ and maturing at T : V0 (κ. T ) − e−r(T 2 0 κ −T ) V0 (κ. all asset positions in subsection IV-B must be re-scaled by volatility-based payoﬀ at T would instead be T 1[Ft ∈(κ− κ. The component of this strategy involving borrowing and futures is known as the stop-loss start-gain strategy. The appendix proves that the dynamic strategy in futures initiated at T involves holding − e −r(T −t) κ2 sgn(Ft − κ) futures contracts. where δ(·) denotes a Dirac delta function12 . In order that this payoﬀ have a non-negligible value. the diﬀerence between the payoﬀ from the straddles and this dynamic strategy is known as the local time of the futures price process.κ+ κ)] 2 σt dt. 15 . T ). if x = 0. Fortunately. 13 See Karatzas and Shreve[21]. Recalling that only options struck within the corridor are used to create the corridor variance. where V0 (κ. . T ) + C0 (κ. ∞ −∞ δ(x)dx = 1. By the Tanaka-Meyer formula13 . 2 κ Thus. δ(x) = 2. Local time is a fundamental concept in the study of one dimensional stochastic processes. if x > 0.In this subsection. at t = T . pg. we shrink the width of the corridor of the last subsection down to a single point and examine the impact on the volatility based payoﬀ and its replicating strategy. the initial cost of creating this localized cash ﬂow is given by the following ratioed calendar spread of straddles: 1 [V (κ. 220. where sgn(x) is the sign function: sgn(x) ≡ −1 0 1 if x < 0. this strategy reduces to the initial purchase of a straddle maturing at T . T ) ≡ P0 (κ. T ). the variance received can be completely localized in the spatial dimension to 2 δ(Ft − κ)σt dt. the investor should borrow to ﬁnance the payout of |FT −κ| κ2 from having initially written the T maturity straddle. a straddle 12 The Dirac delta function is a generalized function characterized by two properties: 0 if x = 0 ∞ if x = 0 1. See Richards and Youn[25] for an accessible introduction to such generalized functions. T 2 κ T T 1 .

By comparing this time value with the ex-post outcome. any two assets whose payoﬀs are not co-linear can be used to span. In contrast to earlier work on stochastic volatility. we may additionally localize in the time dimension as has been done in some recent work by Dupire[12] and DKK[10]. T ) 1 κ2 . the market price of local time risk can be inferred. the forward local variance can be deﬁned as the number of butterﬂy spreads paying δ(FT − κ) at T one must sell in order to ﬁnance the above option position initially. ∂V0 (κ. 16 . V Connection to Recent Work on Stochastic Volatility The last contract examined in the last section represents the limit of a localization in the futures price. The initial time value of the straddle is the market’s (riskneutral) expectation of the local time. 14 When two Brownian motions drive the price and the forward local volatility surface. T ) ∂T T ↓ 0. allowing replication of price or volatilitybased payoﬀs using dynamic trading in only the underlying asset and a single option14 .combined with a stop-loss start-gain strategy in the underlying provides a mechanism for synthesizing a contract paying oﬀ this fundamental concept. One can go on to impose a stochastic process on the forward local variance as in Dupire[12] and in DKK[10]. T ) − e−r(T κ2 T By letting −T ) V0 (κ. suppose we further re-scale all the asset positions described in subsection IV-C by instead be: T T 1 T where T ≡ T − T . one gets the beautiful result of Dupire[12] that + rV0 (κ. When a continuum of option maturities is also available. These authors derive conditions on the risk-neutral drift of the forward local variance. the form of the market price of volatility risk need not be speciﬁed. As shown in Dupire. The payoﬀ at T would δ(Ft − κ) 2 σt dt. T ) is the cost of 2 creating the payment δ(FT − κ)σT at T . A discretized version of this result can be found in DKK[10]. T The cost of creating this position would be: 1 V0 (κ. Accordingly.

skewness. Levy. and A. 2. Applied Mathematical Finance. Under certain assumptions. such inquiries are best left for future research. D. [3] Breeden. Paras. [2] Avellaneda.” Journal of Business.. Levy.. 621-651. they also generate exposure to price. and R. A. 73–88. “Prices of State Contingent Claims Implicit in Option Prices. etc. Paras. References [1] Avellaneda. It would also be interesting to consider contracts which pay nonlinear functions of realized variance or local variance. 1996. “Managing the Volatility Risk of Portfolios of Derivative Securities: The Lagrangian Uncertain Volatility Model”. 3. 51. 1995. M. 21–52. we were able to price and hedge certain volatility contracts without specifying the process for volatility. correlation. The principle assumption made was that of price continuity. or which have an exposure contingent on certain price levels being achieved in speciﬁed time intervals. covariance. Litzenberger. By combining static positions in options with dynamic trading in futures. A. In the interests of brevity. it would be interesting to develop contracts on other statistics of the sample path such as the Sharpe ratio. 1978. M. 17 . and A. Similarly. a calendar spread of options emerges as a simple tool for trading the local volatility (or local time) between the two maturities. payoﬀs related to realized volatility can be achieved which have either no exposure to price. While static positions in options do generate exposure to volatility. It would be interesting to see if this insight survives the relaxation of the critical assumption of price continuity. Under this assumption. Finally.Summary and Suggestions for Future Research We reviewed three approaches for trading volatility. Applied Mathematical Finance. but always has a price exposure as well. a dynamic strategy in futures alone can yield a volatility exposure. “Pricing and Hedging Derivative Securities in Markets with Uncertain Volatilities”.

“Options Markets. I. “Model Art”. [15] Galai. 3. [14] Fleming. M.. 118 and 120. Whaley. 1993. and D. Fall 1993. and R. 273–286. and M. 1157–72. Journal of Finance. 1993. Morgan Stanley working paper. 1990. [7] Carr P. 1997. “Options on Volatility”. [5] Brenner. and D.. “The Stop-Loss Start-Gain Strategy and Option Valuation: A New Decomposition into Intrinsic and Time Value”. 1979. Madan. Sept. 1996.. “Predicting Stock Market Volatility: A New Measure”. M. “Trading and Hedging Local Volatility” Journal of Financial Engineering. 1993. Kamal. Boston University working paper.. [12] Dupire B. “Optimal Positioning in Derivative Securities”. 5. 18 . and R. Nelken. Carnegie Mellon University working paper. M. 1996. 233-68. Paribas working paper. 469–492. Chapter 13 of Option Embedded Bonds. M. ed. and D. 1996. 1997.[4] Brenner. [8] Carr P. Self-fulﬁlling Prophecies and Implied Volatilities”. Review of Financial Studies. Jarrow. Galai. “Hedging Volatility in Foreign Currencies”.. N. 3. [13] El Karoui. 1997. [11] Dupire B. Duke University working paper.. Shreve. Kani. Financial Analyst’s Journal. The Journal of Derivatives. July-August 1989. 61–65. [10] Derman E. XXXIV. pgs. J. and D. 1989. and R. Risk.. Galai. “A Uniﬁed Theory of Volatility”. [9] Cherian. “Robustness of the Black and Scholes Formula”.. and S. I. J. B. Jeanblanc-Picque. Ostdiek. Galai. “New Financial Instruments for Hedging Changes in Volatility”. 1993. “A Proposal for Indexes for Traded Call Options”. 53–9. D. forthcoming in Review of Derivatives Research. Jarrow. 6... [6] Brenner.

41. “Spanning and Completeness in markets with Contingent Claims.. R.[16] Gastineau.. 71–84. Fall 1993. Applied Mathematical Finance. Quarterly Journal of Economics. 75–89. Morton. “An Index of Listed Option Premiums”. [17] Green. 1988. “Uncertain Volatility and the Risk-free Synthesis of Derivatives”. UCLA working paper. G. “Spanning and Completeness with Options. Econometrica.” Journal of Economic Theory. D. “Options and Eﬃciency”.I. 2. “Bond Pricing and the Term Structure of Interest Rates: A New Methodology for Contingent Claim Valuation”. 90 Feb. Cambridge University Press. Journal of Financial Economics. 31. I. May-June 1977. 1976. J. NY.. T. Longstaﬀ. [20] Jagannathan R.” Review of Financial Studies. and A. [27] Whaley. R. and R. and H.. Shreve. Youn. Jarrow. [18] Grunbichler A. Jarrow.C.. and S. 13. 1995. 1987. 3. “Valuing Options on Volatility”. 1990 Theory of Distributions: A Non-technical Introduction. [26] Ross. 1990. R. 66 77–105. 1988. The Journal of Derivatives. Brownian Motion and Stochastic Calculus. “Call Options and the Risk of Underlying Securities”. [21] Karatzas. 117-33. [22] Lyons. Springer Verlag. 19 . 425–434. 202-210.K. 1992. and F. 1990... 311-328. Financial Analyst’s Journal. [23] Nachman. 3. “Derivatives on Market Volatility: Hedging Tools Long Overdue”. 1984. [25] Richards. S. A. 1977.A. London Business School working paper. 1993.. [24] Neuberger.. [19] Heath. 1993.. “Volatility Trading”. D.

the sifting property of a Dirac delta function implies: f (F ) = = ∞ 0 0 κ f (K)δ(F − K)dK κ 0 f (K)δ(F − K)dK + f (K)δ(F − K)dK.Appendix 1: Spanning with Bonds and Options For any payoﬀ f (F ). for any nonnegative κ. Integrating each integral by parts implies: f (F ) = f (K)1(F < K) − 0 κ κ 0 f (K)1(F < K)dK ∞ κ +f (K)1(F ≥ K) Integrating each integral by parts once more implies: ∞ κ + f (K)1(F ≥ K)dK. f (F ) = f (κ)1(F < κ) − f (K)(K − F )+ + 0 κ κ 0 f (K)(K − F )+ dK ∞ κ +f (κ)1(F ≥ κ) − f (K)(F − K) + + + ∞ κ + f (K)(F − K)+ dK = f (κ) + f (κ)[(F − κ) − (κ − F ) ] + κ 0 f (K)(K − F )+ dK + ∞ κ f (K)(F − K)+ dK. 20 .

κ κ − κ κ e−r(T −t) 1 − κ κ e−r(T −t) −r(T −t) 1 . the number of contracts held is given by 1 − − 1 Ft if Ft ∈ (κ − if Ft ≥ κ + κ. the futures position of −2e−r(T −e − F contracts in subsection ¯ κ κ t −r(T −t) 1 −e − κ−1 κ if Ft ≤ κ − κ. κ + κ. More explicitly. 2 κ −t) 1 κ − 1 ¯ Ft contracts in subsection IV-B is changed to 1 1 IV-C. 21 . κ + O( κ2) Substitution implies that the number of futures contracts held is given by: − e−r(T −t) − 1 κ + O( κ2 ) if Ft ≤ κ − κ. as κ ↓ 0. Thus. κ). κ + κ. 1 κ+ κ Now. the number of futures contracts held converges to − e −1 −r(T −t) κ2 sgn(Ft − κ).Appendix 2: Derivation of Futures Position When Synthesizing Contract Paying Future Variance Along a Strike Recall from section IV-C. if x > 0. if x = 0. that all asset positions in section IV-B were normalized by multiplying by Thus in particular. by Taylor’s series: 1 1 1 = + 2 κ− κ κ κ and: 1 1 1 = − 2 κ+ κ κ κ κ + O( κ2). κ κ2 − Ft − κ κ e−r(T −t) 1 − κ + O( 2 κ κ e−r(T −t) 1 1 if Ft ∈ (κ − κ2) if Ft ≥ κ + κ. where sgn(x) is the sign function: sgn(x) ≡ 0 1 if x < 0. κ).

Sign up to vote on this title

UsefulNot useful- MS VIX Futs Opts Primer
- Energy derivativesCW.docx
- Dynamic Hedging
- 4 Powerful Rules to Successful Options Trading by McMillan
- US Federal Reserve
- Zhang_J
- Local Vol Delta-Hedging
- Implied Volatility from Options on Gold Futures
- Bruno Dupire - Breakeven Volatility
- Which Free Lunch Would You Like Today, Sir - Delta Hedging, Volatility Arbitrage and Optimal Portfolios
- Wilmott_WhichFreeLunch
- FMWJP1
- Option Approximation
- 2_Trading Guides - OptionsCourse-301_Volatility_Trading
- CasePacket2013_v4
- Veg A
- Reconstructing the unknown local volatility function
- Dupire Bruno
- Ebooksclub.org Volatility and Correlation the Perfect Hedger and the Fox Wiley Finance
- Indian Market Future & Options Full Tutorial
- Indian Market Future & Options Full Tutorial
- Eurodollar Futures Convexity Adjustments in Stochastic Volatility Models
- 67190_1985-1989
- ED Convexity Adjustment
- 890 IV Surface
- poon_granger_jel
- VOLATILITY SMILES, SURFACES AND OPTION PRICES
- 02_04
- All Greeks
- No Small Probabilities Are Not Attractive to Sell NNT Paper OTM Options
- Towards a Theory of Vol Trading

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd