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

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

:

Theory, Rules of Thumb, and Empirical Evidence

Toby Daglish

School of Economics and Finance

Victoria University of Wellington

Email: toby.daglish@vuw.ac.nz

John Hull

Rotman School of Management

University of Toronto

Email: hull@rotman.utoronto.ca

Wulin Suo

School of Business, Queen’s University

Email: wsuo@business.queensu.ca

First version: March 2001

This version: August 2006

Abstract

Implied volatilities are frequently used to quote the prices of options. The implied volatility

of a European option on a particular asset as a function of strike price and time to maturity is

known as the asset’s volatility surface. Traders monitor movements in volatility surfaces closely.

In this paper we develop a no-arbitrage condition for the evolution of a volatility surface. We

examine a number of rules of thumb used by traders to manage the volatility surface and test

whether they are consistent with the no-arbitrage condition and with data on the trading of

options on the S&P 500 taken from the over-the-counter market. Finally we estimate the factors

driving the volatility surface in a way that is consistent with the no-arbitrage condition.

1 Introduction

Option traders and brokers in over-the-counter markets frequently quote option prices using implied

volatilities calculated from Black and Scholes (1973) and other similar models. The reason is that

implied volatilities usually have to be updated less frequently than the prices themselves. Put–call

parity implies that, in the absence of arbitrage, the implied volatility for a European call option is

the same as that for a European put option when the two options have the same strike price and

time to maturity. This is convenient: when quoting an implied volatility for a European option with

a particular strike price and maturity date, a trader does not need to specify whether a call or a put

is being considered.

The implied volatility of European options on a particular asset as a function of strike price and

time to maturity is known as the volatility surface. Every day traders and brokers estimate volatility

surfaces for a range of diﬀerent underlying assets from the market prices of options. Some points

on a volatility surface for a particular asset can be estimated directly because they correspond to

actively traded options. The rest of the volatility surface is typically determined by interpolating

between these points.

If the assumptions underlying Black–Scholes held for an asset, its volatility surface would be ﬂat

and unchanging. In practice the volatility surfaces for most assets are not ﬂat and change stochas-

tically. Consider for example equities and foreign currencies. Rubinstein (1994) and Jackwerth and

Rubinstein (1996), among others, show that the implied volatilities of stock and stock index options

exhibit a pronounced “skew” (that is, the implied volatility is a decreasing function of strike price).

For foreign currencies this skew becomes a “smile” (that is, the implied volatility is a U-shaped

function of strike price). For both types of assets, the implied volatility can be an increasing or

decreasing function of the time to maturity. The volatility surface changes through time, but the

general shape of the relationship between volatility and strike price tends to be preserved.

Traders use a volatility surface as a tool to value a European option when its price is not directly

observable in the market. Provided there are a reasonable number of actively traded European

options and these span the full range of the strike prices and times to maturity that are encountered,

this approach ensures that traders price all European options consistently with the market. However,

as pointed out by Hull and Suo (2002), there is no easy way to extend the approach to price path-

dependent exotic options such as barrier options, compound options, and Asian options. As a result

there is liable to be some model risk when these options are priced.

Traders also use the volatility surface in an ad hoc way for hedging. They attempt to hedge

against potential changes in the volatility surface as well as against changes in the asset price.

Derman (1999) discusses alternative approaches to hedging against asset price movements. The

“volatility-by-strike” or “sticky strike” rule assumes that the implied volatility for an option with a

given strike price and maturity will be unaﬀected by changes in the underlying asset price. Another

1

popular approach is the “volatility-by-moneyness” or “sticky delta” rule. This assumes that the

volatility for a particular maturity depends only on the moneyness (that is, the ratio of the price of

the underlying asset to the strike price).

The ﬁrst attempts to model the volatility surface were by Rubinstein (1994), Derman and Kani

(1994), and Dupire (1994). These authors show how a one-factor model for an asset price, known as

the implied volatility function (IVF) model, can be developed so that it is exactly consistent with the

current volatility surface. Unfortunately, the evolution of the volatility surface under the IVF model

can be unrealistic. The volatility surface given by the model at a future time is liable to be quite

diﬀerent from the initial volatility surface. For example, in the case of a foreign currency the initial

U-shaped relationship between implied volatility and strike price is liable to evolve to one where

the volatility is a monotonic increasing or decreasing function of strike price. Dumas, Fleming, and

Whaley (1997) have shown that the IVF model does not capture the dynamics of market prices well.

Hull and Suo (2002) have shown that it can be dangerous to use the model for the relative pricing

of barrier options and plain vanilla options.

This paper extends existing research in a number of ways. Similarly to Ledoit and Santa Clara

(1998), Sch¨onbucher (1999), Brace et al (2001), and Britten–Jones and Neuberger (2000), we develop

no arbitrage conditions for the evolution of the volatility surface. We consider a general model where

there are a number of factors driving the volatility surface and these may be correlated with stock

price movements. We then investigate the implications of the no-arbitrage condition for the shapes

of the volatility surfaces likely to be observed in diﬀerent situations and examine whether the various

rules of thumb that have been put forward by traders are consistent with the no-arbitrage condition.

Finally we carry out two empirical studies. The ﬁrst empirical study extends the work of Derman

(1999) to investigate whether movements in the volatility surface are consistent with diﬀerent rules of

thumb. The second extends work of Kamal and Derman (1997), Skiadopoulos, Hodges, and Clewlow

(2000), Cont and Fonseca (2002), and Cont, Fonseca and Durrleman (2002) to estimate the factors

driving movements in the volatility surface in a way that reﬂects the theoretical drifts of points on

the surface. One distinguishing feature of our work is that we use over-the-counter consensus data

for our empirical studies. To the best of our knowledge we are the ﬁrst derivatives researchers to use

this type of data. Another distinguishing feature of our research is that, in addition to the sticky

strike and sticky delta rules we consider a “square root of time” rule that is often used by traders and

documented by, for example, Natenberg (1994), where the implied volatility is assumed to depend

on the moneyness and the square root of the time to maturity through a very speciﬁc relationship.

The rest of the paper is organized as follows. Section 2 explains the rules of thumb. Section

3 develops a general model for the evolution of a volatility surface and derives the no-arbitrage

condition. Section 4 discusses the implications of the no-arbitrage condition. Section 5 examines a

number of special cases of the model. Section 6 considers whether the rules of thumb are consistent

2

with the no-arbitrage condition. Section 7 tests whether there is empirical support for the rules of

thumb. Section 8 derives factors driving the volatility surface in a way that is consistent with the

no-arbitrage condition. Conclusions are in Section 9.

2 Rules of Thumb

A number of rules of thumb have been proposed about volatility surfaces. These rules of thumb fall

into two categories. In the ﬁrst category are rules concerned with the way in which the volatility

surface changes through time. They are useful in the calculation of the Greek letters such as delta

and gamma. In the second category are rules concerned with the relationship between the volatility

smiles for diﬀerent option maturities at a point in time. They are useful in creating a complete

volatility surface when market prices are available for a relatively small number of options. In this

section we explain three diﬀerent rules of thumb: the sticky strike rule, the sticky delta rule and the

square root of time rule. The ﬁrst two of these rules are in the ﬁrst category and provide a basis for

calculating Greek letters. The third rule is in the second category and assists with the mechanics of

“ﬁlling in the blanks” when a complete volatility surface is being produced.

We consider a European call option with maturity T, strike price K and implied volatility σ

TK

.

We deﬁne the call price as c, the underlying asset price as S, and the delta, ∆ of the option as the

total rate of change of its price with respect to the asset price:

∆ =

dc

dS

.

2.1 The Sticky Strike Rule

The sticky strike rule assumes that σ

TK

is independent of S. This is an appealing assumption

because it implies that the sensitivity of the price of an option to S is

∂c

∂S

,

where for the purposes of calculating the partial derivative the option price, c, is considered to be a

function of the asset price, S, σ

TK

, and time, t. The assumption enables the Black–Scholes formula

to be used to calculate delta with the volatility parameter set equal to the option’s implied volatility.

The same is true of gamma, the rate of change of delta with respect to the asset price.

In what we will refer to as the basic sticky strike model, σ

TK

is a function only of K and T. In

the generalized sticky strike model it is independent of S, but possibly dependent on other stochastic

variables.

3

2.2 The Sticky Delta Rule

An alternative to the sticky strike rule is the sticky delta rule. This assumes that the implied

volatility of an option depends on the S and K, through its dependence on the moneyness variable,

K/S. The delta of a European option in a stochastic volatility model is

∆ =

∂c

∂S

+

∂c

∂σ

TK

∂σ

TK

∂S

.

Again, for the purposes of calculating partial derivatives the option price c is considered to be a

function of S, σ

TK

, and t. The ﬁrst term in this expression is the delta calculated using Black-

Scholes with the volatility parameter set equal to implied volatility. In the second term, ∂c/∂σ

TK

is positive. It follows that, if σ

TK

is a declining (increasing) function of the strike price, it is an

increasing (declining) function of S and ∆ is greater than (less than) that given by Black-Scholes.

For equities σ

TK

is a declining function of K and so the Black-Scholes delta understates the true

delta. For an asset with a U-shaped volatility smile Black-Scholes understates delta for low strike

prices and overstates it for high strike prices.

In the most basic form of the sticky delta rule the implied volatility is assumed to be a deter-

ministic function of K/S and T − t. We will refer to this as the basic sticky delta model. A more

general version of the sticky delta rule is where the process for V

TK

depends on K, S, T, and t only

through its dependence on K/S and T −t. We will refer to this as the generalized sticky delta model.

The implied volatility changes stochastically but parameters describing the process followed by the

volatility (for example, the volatility of the volatility) are functions only of K/S and T −t.

Many traders argue that a better measure of moneyness than K/S is K/F where F is the forward

value of S for a contract maturing at time T. A version on the sticky delta rule sometimes used by

traders is that σ

TK

(S, t) −σ

TF

(S, t) is a function only of K/F and T −t. Here it is the excess of the

volatility over the at-the-money volatility, rather than the volatility itself, which is assumed to be a

deterministic function of the moneyness variable, K/F. This form of the sticky delta rule allows the

overall level of volatility to change through time and the shape of the volatility term structure to

change, but when measured relative to the at-the-money volatility, the volatility is dependent only

on K/S and T −t. We will refer to this model as the relative sticky delta model. If the at-the-money

volatility is stochastic, but independent of S, the model is a particular case of the generalized sticky

delta model just considered.

2.3 The Square Root of Time Rule

Another rule that is sometimes used by traders is what we will refer to as the “square root of time

rule”. This is described in Natenberg (1994) and Hull (2006). It provides a speciﬁc relationship

between the volatilities of options with diﬀerent strike prices and times to maturity at a particular

4

time. One version of this rule is

σ

TK

(S, t)

σ

TF

(S, t)

= Φ

_

ln(K/F)

√

T −t

_

,

where Φ is a function. An alternative that we will use is

σ

TK

(S, t) −σ

TF

(S, t) = Φ

_

ln(K/F)

√

T −t

_

. (1)

We will refer to this version of the rule where Φ does not change through time as the stationary

square root of time model and the version of the rule where the form of the function Φ changes

stochastically as the stochastic square root of time model.

The square root of time model (whether stationary or stochastic) simpliﬁes the speciﬁcation of

the volatility surface. If we know

1. The volatility smile for options that mature at one particular time T

∗

, and

2. At-the-money volatilities for other maturities,

we can compute the complete volatility surface. Suppose that F

∗

is the forward price of the asset

for a contract maturing at T

∗

. We can compute the volatility smile at time T from that at time T

∗

using the result that

σ

TK

(S, t) −σ

TF

(S, t) = σ

T

∗

K

∗(S, t) −σ

T

∗

F

∗(S, t),

where

ln(K/F)

√

T −t

=

ln(K

∗

/F

∗

)

√

T

∗

−t

,

or

K

∗

= F

∗

_

K

F

_

√

(T

∗

−t)/(T−t)

.

If the at-the-money volatility is assumed to be stochastic, but independent of S, the stationary

square root of time model is a particular case of the relative sticky strike model and the stochastic

square root of time model is a particular case of the generalized sticky strike model.

3 The Dynamics of the Implied Volatility

We suppose that the risk-neutral process followed by the price of an asset, S, is

dS

S

= [r(t) −q(t)] dt + σ dz, (2)

where r(t) is the risk-free rate, q(t) is the yield provided by the asset, σ is the asset’s volatility, and

z is a Wiener process. We suppose that r(t) and q(t) are deterministic functions of time and that σ

follows a diﬀusion process. Our model includes the IVF model and stochastic volatility models such

as Hull and White (1987), Stein and Stein (1991) and Heston (1993) as special cases.

5

Most stochastic volatility models specify the process for σ directly. We instead specify the

processes for all implied volatilities. As in the previous section we deﬁne σ

TK

(t, S) as the implied

volatility at time t of an option with strike price K and maturity T when the asset price is S. For

convenience we deﬁne

V

TK

(t, S) = [σ

TK

(t, S)]

2

as the implied variance of the option. Suppose that the process followed by V

TK

in a risk-neutral

world is

dV

TK

= α

TK

dt + V

TK

N

i=1

θ

TKi

dz

i

, (3)

where z

1

, · · · , z

N

are Wiener processes driving the volatility surface. The z

i

may be correlated with

the Wiener process, z, driving the asset price in equation (2). We deﬁne ρ

i

as the correlation

between z and z

i

. Without loss of generality we assume that the z

i

are correlated with each other

only through their correlation with z. This means that the correlation between z

i

and z

j

is ρ

i

ρ

j

.

The initial volatility surface is σ

TK

(0, S

0

) where S

0

is the initial asset price. This volatility surface

can be estimated from the current (t = 0) prices of European call or put options and is assumed to

be known.

The family of processes in equation (3) deﬁnes the multi-factor dynamics of the volatility surface.

The parameter θ

TKi

measures the sensitivity of V

TK

to the Wiener process z

i

. In the most general

form of the model the parameters α

TK

and θ

TKi

(1 ≤ i ≤ N) may depend on past and present

values of S, past and present values of V

TK

, and time.

There is clearly a relationship between the instantaneous volatility σ(t) and the volatility surface

σ

TK

(t, S). The appendix shows that the instantaneous volatility is the limit of the implied volatility

of an at-the-money option as its time to maturity approaches zero. For this purpose an at-the-money

option is deﬁned as an option where the strike price equals the forward asset price.

1

Formally:

lim

T→t

σ

TF

(t, S) = σ(t), (4)

where F is the forward price of the asset at time t for a contract maturing at time T. In general

the process for σ is non-Markov. This is true even when the processes in equation (3) deﬁning the

volatility surface are Markov.

2

Deﬁne c(S, V

TK

, t; K, T) as the price of a European call option with strike price K and maturity

T when the asset price S follows the process in equations (2) and (3). From the deﬁnition of implied

1

Note that the result is not necessarily true if we deﬁne an at-the-money option as an option where the strike price

equals the asset price. For example, when

σ

TK

(t, S) = a + b

1

T −t

ln(F/K)

with a and b constants, lim

T→t

σ

TF

(t, S) is not the same as lim

T→t

σ

TS

(t, S). For more on the relationship between

the implied and spot volatilities, see Durrleman (2004).

2

There is an analogy here to the Heath, Jarrow, and Morton (1992) model. When each forward rate follows a

Markov process the instantaneous short rate does not in general do so.

6

volatility and the results in Black and Scholes (1973) and Merton (1973) it follows that:

c(S, V

TK

, t; K, T) = e

−

_

T

t

q(τ) dτ

SN(d

1

) −e

−

_

T

t

r(τ) dτ

KN(d

2

),

where

d

1

=

ln(S/K) +

_

T

t

[r(τ) −q(τ)]dτ

_

V

TK

(T −t)

+

1

2

_

V

TK

(T −t),

d

2

=

ln(S/K) +

_

T

t

[r(τ) −q(τ)]dτ

_

V

TK

(T −t)

−

1

2

_

V

TK

(T −t).

Using Ito’s lemma equations (2) and (3) imply that the drift of c in a risk-neutral world is:

∂c

∂t

+ (r −q)S

∂c

∂S

+

1

2

σ

2

S

2

∂

2

c

∂S

2

+ α

TK

∂c

∂V

TK

+

1

2

V

2

TK

∂

2

c

∂V

2

TK

_

N

i=1

(θ

TKi

)

2

+

i=j

θ

TKi

θ

TKj

ρ

i

ρ

j

_

+SV

TK

σ

∂

2

c

∂S∂V

TK

N

i=1

θ

TKi

ρ

i

.

In the most general form of the model ρ

i

, the correlation between z and z

i

, is a function of past

and present values of S, past and present values of V

TK

and time. For there to be no arbitrage the

process followed by c must provide an expected return of r in a risk-neutral world. It follows that

∂c

∂t

+ (r −q)S

∂c

∂S

+

1

2

σ

2

S

2

∂

2

c

∂S

2

+ α

TK

∂c

∂V

TK

+

1

2

V

2

TK

∂

2

c

∂V

2

TK

_

N

i=1

(θ

TKi

)

2

+

i=j

θ

TKi

θ

TKj

ρ

i

ρ

j

_

+SV

TK

σ

∂

2

c

∂S∂V

TK

N

i=1

θ

TKi

ρ

i

= rc.

When V

TK

is held constant, c satisﬁes the Black-Scholes (1973) and Merton (1973) diﬀerential

equation. As a result

∂c

∂t

+ (r −q)S

∂c

∂S

= rc −

1

2

V

TK

S

2

∂

2

c

∂S

2

.

It follows that

1

2

S

2

∂

2

c

∂S

2

(σ

2

−V

TK

) + α

TK

∂c

∂V

TK

+

1

2

V

2

TK

∂

2

c

∂V

2

TK

_

N

i=1

(θ

TKi

)

2

+

i=j

θ

TKi

θ

TKj

ρ

i

ρ

j

_

+SV

TK

σ

∂

2

c

∂S∂V

TK

N

i=1

θ

TKi

ρ

i

= 0,

or

α

TK

= −

1

2∂c/∂V

TK

_

S

2

∂

2

c

∂S

2

(σ

2

−V

TK

) +

∂

2

c

∂V

2

TK

V

2

TK

N

i=1

(θ

TKi

)

2

+

∂

2

c

∂V

2

TK

V

2

TK

i=j

θ

TKi

θ

TKj

ρ

i

ρ

j

+ 2SV

TK

σ

∂

2

c

∂S∂V

TK

N

i=1

θ

TKi

ρ

i

_

. (5)

7

The partial derivatives of c with respect to S and V

TK

are the same as those for the Black–Scholes

model:

∂c

∂S

= e

−

_

T

t

q(τ) dτ

N(d

1

),

∂

2

c

∂S

2

=

φ(d

1

)e

−

_

T

t

q(τ) dτ

S

_

V

TK

(T −t)

,

∂c

∂V

TK

=

Se

−

_

T

t

q(τ) dτ

φ(d

1

)

√

T −t

2

√

V

TK

,

∂

2

c

∂V

2

TK

=

Se

−

_

T

t

q(τ) dτ

φ(d

1

)

√

T −t

4V

3/2

TK

(d

1

d

2

−1),

∂

2

c

∂S∂V

TK

= −

e

−

_

T

t

q(τ) dτ

φ(d

1

)d

2

2V

TK

,

where φ is the density function of the standard normal distribution:

φ(x) =

1

2π

exp

_

−

x

2

2

_

, −∞< x < ∞.

Substituting these relationships into equation (5) and simplifying we obtain

α

TK

=

1

T −t

(V

TK

−σ

2

) −

V

TK

(d

1

d

2

−1)

4

_

_

N

i=1

(θ

TKi

)

2

+

i=j

θ

TKi

θ

TKj

ρ

i

ρ

j

_

_

+σd

2

_

V

TK

T −t

N

i=1

θ

TKi

ρ

i

. (6)

Equation (6) provides an expression for the risk-neutral drift of an implied variance in terms of

its volatility. The ﬁrst term on the right hand side is the drift arising from the diﬀerence between the

implied variance and the instantaneous variance. It can be understood by considering the situation

where the instantaneous variance, σ

2

, is a deterministic function of time. The variable V

TK

is then

also a function of time and

V

TK

=

1

T −t

_

T

t

σ(τ)

2

dτ

Diﬀerentiating with respect to time we get

dV

TK

dt

=

1

T −t

[V

TK

−σ(t)

2

],

which is the ﬁrst term.

The analysis can be simpliﬁed slightly by considering the variable

ˆ

V

TK

instead of V where

ˆ

V

TK

= (T −t)V

TK

. Because

d

ˆ

V

TK

= −V

TK

dt + (T −t)dV

TK

,

it follows that

d

ˆ

V

TK

=

_

−σ

2

−

ˆ

V

TK

(d

1

d

2

−1)

4

_

_

N

i=1

(θ

TKi

)

2

+

i=j

θ

TKi

θ

TKj

ρ

i

ρ

j

_

_

8

+σd

2

_

ˆ

V

TK

N

i=1

θ

TKi

ρ

i

_

dt +

ˆ

V

TK

N

i=1

θ

TKi

dz

i

.

4 Implications of the No-Arbitrage Condition

Equation (6) provides a no-arbitrage condition for the drift of the implied variance as a function

of its volatility. In this section we examine the implications of this no-arbitrage condition. In the

general case where V

TK

is nondeterministic, the ﬁrst term in equation (6) is mean ﬂeeing; that

is, it provides negative mean reversion. This negative mean reversion becomes more pronounced

as the option approaches maturity. For a viable model the θ

TKi

must be complex functions that

in some way oﬀset this negative mean reversion. Determining the nature of these functions is not

easy. However, it is possible to make some general statements about the volatility smiles that are

consistent with stable models.

4.1 The Zero Correlation Case

Consider ﬁrst the case where all the ρ

i

are zero so that equation (6) reduces to

α

TK

=

1

T −t

(V

TK

−σ

2

) −

V

TK

(d

1

d

2

−1)

4

N

i=1

(θ

TKi

)

2

As before, we deﬁne F as the forward value of the asset for a contract maturing at time T so

that

F = Se

_

T

t

[r(τ)−q(τ)]dτ

.

The drift of V

TK

−V

TF

is

1

T −t

(V

TK

−V

TF

) −

V

TK

(d

1

d

2

−1)

4

N

i=1

(θ

TKi

)

2

−

V

TF

[1 + V

TF

(T −t)/4]

4

N

i=1

(θ

TFi

)

2

.

Suppose that d

1

d

2

− 1 > 0 and V

TK

< V

TF

. In this case each term in the drift of V

TK

− V

TF

is

negative. As a result V

TK

−V

TF

tends to get progressively more negative and the model is unstable

with negative values of V

TK

being possible. We deduce from this that V

TK

must be greater than

V

TF

when d

1

d

2

> 1. The condition d

1

d

2

> 1 is satisﬁed for very large and very small values of K.

It follows that the case where the ρ

i

are zero can be consistent with the U-shaped volatility smile.

It cannot be consistent with a upward or downward sloping smile because in these cases V

TK

< V

TF

for either very high or very low values of K.

Our ﬁnding is consistent with a result in Hull and White (1987). These authors show that when

the instantaneous volatility is independent of the asset price, the price of a European option is the

Black–Scholes price integrated over the distribution of the average variance. They demonstrate that

when d

1

d

2

> 1 a stochastic volatility tends to increase an option’s price and therefore its implied

volatility.

9

A U-shaped volatility smile is commonly observed for options on a foreign currency. Our analysis

shows that this is consistent with the empirical result that the correlation between implied volatilities

and the exchange rate is close to zero (see, for example, Bates (1996)).

4.2 Volatility Skews

We have shown that a zero correlation between volatility and asset price can only be consistent

with a U-shaped volatility smile. We now consider the types of correlations that are consistent with

volatility skews.

Consider ﬁrst the situation where the volatility is a declining function of the strike price. The

variance rate V

TK

is greater than V

TF

when K is very small and less than V

TF

when K is very

large. The drift of V

TK

−V

TF

is

1

T −t

(V

TK

−V

TF

) −

V

TK

(d

1

d

2

−1)

4

N

i=1

(θ

TKi

)

2

−

V

TF

(1 + V

TF

(T −t)/4)

4

N

i=1

(θ

TFi

)

2

+σd

2

_

V

TK

T −t

N

i=1

θ

TKi

ρ

i

+ σ

V

TF

2

N

i=1

θ

TFi

ρ

i

−

V

TK

(d

1

d

2

−1)

4

i=j

θ

TKi

θ

TKj

ρ

i

ρ

j

−

V

TF

(1 + V

TF

(T −t)/4)

4

N

i=1

θ

TFi

θ

TFj

ρ

i

ρ

j

When K > F, V

TK

−V

TF

is negative and the eﬀect of the ﬁrst three terms is to provide a negative

drift. For a stable model we require the last four terms to give a positive drift. As K increases and

we approach option maturity the ﬁrst of the last four terms dominates the other three. Because

d

2

< 0 we must have

N

i=1

θ

TKi

ρ

i

< 0 (7)

when K is very large. The instantaneous covariance of the asset price and its variance is

σ

N

i=1

θ

TKi

ρ

i

.

Because σ > 0 it follows that when K is large the asset price must be negatively correlated with its

variance.

Equities provide an example of a situation where there is a volatility skew of the sort we are

considering. As has been well documented by authors such as Christie (1982), the volatility of an

equity price tends to be negatively correlated with the equity price. This is consistent with the result

we have just presented.

Consider next the situation where volatility is an increasing function of the strike price. (This

is the case for options on some commodity futures.) A similar argument to that just given shows

that the no-arbitrage relationship implies that the volatility of the underlying must be positively

correlated with the value of the underlying.

10

4.3 Inverted U-Shaped Smiles

An inverted U-shaped volatility smile is much less common than a U-shaped volatility smile or a

volatility skew. An analysis similar to that in Section 4.2 shows why this is so. For a stable model

N

i=1

θ

TKi

ρ

i

must be less than zero when K is large and greater than zero when K is small. It is diﬃcult to see

how this can be so without the stochastic terms in the processes for the V

TK

having a form that

quickly destroys the inverted U-shaped pattern.

5 Special Cases

In this section we consider a number of special cases of the model developed in Section 3. This

will enable us to reach conclusions about whether the rules of thumb in Section 2 can satisfy the

no-arbitrage condition in Section 3.

Case 1: V

TK

is a deterministic function only of t, T, and K

In this situation θ

TKi

= 0 for all i ≥ 1 so that from equation (3)

dV

TK

= α

TK

dt.

Also from equation (6)

α

TK

=

1

T −t

_

V

TK

−σ

2

¸

,

so that

dV

TK

=

1

T −t

_

V

TK

−σ

2

¸

dt,

which can be written as

(T −t)dV

TK

−V

TK

dt = −σ

2

dt,

or

σ

2

= −

d [(T −t)V

TK

]

dt

. (8)

This shows that σ is a deterministic function of time. The only model that is consistent with

V

TK

being a function only of t, T, and K is therefore the model where the instantaneous volatility

(σ) is a function only of time. This is Merton’s (1973) model.

In the particular case where V

TK

depends only on T and K, equation (8) shows that V

TK

= σ

2

and we get the Black-Scholes constant-volatility model.

11

Case 2: V

TK

is independent of the asset price, S .

In this situation ρ

i

is zero and equation (6) becomes

α

TK

=

1

T −t

(V

TK

−σ

2

) −

V

TK

(d

1

d

2

−1)

4

N

i=1

(θ

TKi

)

2

.

Both α

TK

and the θ

TKi

must be independent of S. Because d

1

and d

2

depend on S we must have

θ

TKi

equal zero for all i. Case 2 therefore reduces to Case 1. The only model that is consistent with

V

TK

being independent of S is therefore Merton’s (1973) model where the instantaneous volatility

is a function only of time.

Case 3: V

TK

is a deterministic function of t, T, and K/S.

In this situation

V

TK

= G

_

T, t,

K

F

_

,

where G is a deterministic function and as before F is the forward price of S. From equation (4)

the spot instantaneous volatility, σ, is given by

σ

2

= G(t, t, 1) .

This is a deterministic function of time. It follows that, yet again, the model reduces to Merton’s

(1973) deterministic volatility model.

Case 4: V

TK

is a deterministic function of t, T, S and K.

In this situation we can write

V

TK

= G(T, t, F, K), (9)

where G is a deterministic function. From equation (4), the instantaneous volatility, σ is given by

σ

2

= lim

T→t

G(T, t, F, F).

This shows that the instantaneous volatility is a deterministic function of F and t. Equivalently

it is a deterministic function of underlying asset price, S, and t. It follows that the model reduces

to the implied volatility function model. Writing σ as σ(S, t), Dupire (1994) and Andersen and

Brotherton–Ratcliﬀe (1998) show that

[σ(K, T)]

2

= 2

∂c/∂T + q(T)c + K[r(T) −q(T)]∂c/∂K

K

2

(∂

2

c/∂K

2

)

,

where c is here regarded as a function of S, K and T for the purposes of taking partial derivatives.

6 Theoretical Basis for Rules of Thumb

We are now in a position to consider whether the rules of thumb considered in Section 2 can be

consistent with the no-arbitrage condition in equation (6). Similar results to ours on the sticky strike

and sticky delta model are provided by Balland (2002).

12

In the basic sticky strike rule of Section 2.1 the variance V

TK

is a function only of K and T.

The analysis in Case 1 of the previous section shows that the only version of this model that is

internally consistent is the model where the volatilities of all options are the same and constant.

This is the original Black and Scholes (1973) model. In the generalized sticky strike model V

TK

is

independent of S, but possibly dependent on other stochastic variables. As shown in Case 2 of the

previous section, the only version of this model that is internally consistent is the model where the

instantaneous volatility of the asset price is a function only of time. This is Merton’s (1973) model.

When the instantaneous volatility of the asset price is a function only of time, all European

options with the same maturity have the same implied volatility. We conclude that all versions of

the sticky strike rule are inconsistent with any type of volatility smile or volatility skew. If a trader

prices options using diﬀerent implied volatilities and the volatilities are independent of the asset

price, there must be arbitrage opportunities.

Consider next the sticky delta rule. In the basic sticky delta model, as deﬁned in Section 2.2, the

implied volatility is a deterministic function of K/S and T −t. Case 3 in the previous section shows

that the only version of this model that is internally consistent is Merton’s (1973) model where the

instantaneous volatility of the asset price is a function only of time. As in the case of the sticky

strike model it is inconsistent with any type of volatility smile or volatility skew.

The generalized sticky delta model, where V

TK

is stochastic and depends on K, S, T and t only

through the variable K/S and T − t, can be consistent with the no-arbitrage condition. This is

because equation (6) shows that if each θ

TKi

depends on K, S, T, and t only through a dependence

on K/S and T −t, the same is true of α

TK

.

The relative sticky delta rule and the square root of time rule can be at least approximately

consistent with the no-arbitrage condition. For example, for a mean-reverting stochastic volatility

model, the implied volatility approximately satisﬁes the square root of time rule when the reversion

coeﬃcient is large (Andersson, 2003).

7 Tests of the Rules of Thumb

As pointed out by Derman (1999) apocryphal rules of thumb for describing the behavior of volatility

smiles and skews may not be conﬁrmed by the data. Derman’s research looks at exchange-traded

options on the S&P 500 during the period September 1997 to October 1998 and considers the sticky

strike and sticky delta rules as well as a more complicated rule based on the IVF model. He ﬁnds

subperiods during which each of the rules appears to explain the data best.

Derman’s results are based on relatively short maturity S&P 500 option data. We use monthly

volatility surfaces from the over-the-counter market for 47 months (June 1998 to April 2002). The

data for June 1998 is shown in Table 1. Six maturities are considered ranging from six months to

13

ﬁve years. Seven values of K/S are considered, ranging from 0.8 to 1.2. A total of 42 points on

the volatility surface are therefore provided each month and the total number of volatilities in our

data set is 42 ×47 = 1974. As illustrated in Table 1, implied volatilities for the S&P 500 exhibit a

volatility skew with σ

TK

(t, S) being a decreasing function of K.

[Table 1 about here.]

The data was supplied to us by Totem Market Valuations Limited with the kind permission of a

selection of Totem’s major bank clients.

3

Totem collects implied volatility data in the form shown

in Table 1 from a large number of dealers each month. These dealers are market makers in the

over-the-counter market. Totem uses the data in conjunction with appropriate averaging procedures

to produce an estimate of the mid-market implied volatility for each cell of the table and returns

these estimates to the dealers. This enables dealers to check whether their valuations are in line

with the market. Our data consist of the estimated mid-market volatilities returned to dealers. The

data produced by Totem are considered by market participants to be more accurate than either the

volatility surfaces produced by brokers or those produced by any one individual bank.

Consider ﬁrst the sticky strike rule. Sticky strike rules where the implied volatility is a function

only of K and T may be plausible in the exchange-traded market where the exchange deﬁnes a

handful of options that trade and traders anchor on the volatility they ﬁrst use for any one of these

options. Our data comes from the over-the-counter market. It is diﬃcult to see how this form of the

sticky strike rule can apply in that market because there is continual trading in options with many

diﬀerent strike prices and times to maturity. A more plausible version of the rule is that the implied

volatility is a function only of K and T −t. We tested this version of the rule using

σ

TK

(t, S) = a

0

+ a

1

K + a

2

K

2

+ a

3

(T −t) + a

4

(T −t)

2

+ a

5

K(T −t) + , (10)

where the a

i

are constant and is a normally distributed error term. The terms on the right hand

side of this equation can be thought of as the ﬁrst few terms in a Taylor series expansion of a general

function of K and T − t. Here our analysis is similar in spirit to the “ad-hoc model” of Dumas,

Fleming and Whaley (1998). As shown in Table 2, the model is supported by the data, but has an

R

2

of only 27%. As a test of the model’s viability out-of-sample, we ﬁtted the model using the ﬁrst

two years of data and then tested it for the surfaces for the remaining 23 months. We obtained a

root mean square error (RMSE) of 0.0525, representing a quite sizable error.

[Table 2 about here.]

We now move on to test the sticky delta rule. The version of the rule we consider is the relative

sticky delta model where σ

TK

(t, S) −σ

TF

(t, S) is a function of K/F and T −t. The model we test

3

Totem was acquired by Mark-it Partners in May 2004

14

is

σ

TK

(t, S) −σ

TF

(t, S) = b

0

+ b

1

ln

_

K

F

_

+ b

2

_

ln

_

K

F

__

2

+ b

3

(T −t)

+b

4

(T −t)

2

+ b

5

ln

_

K

F

_

(T −t) + (11)

where the b

i

are constants and is a normally distributed error term. The model is analogous to

the one used to test the sticky strike rule. The terms on the right hand side of this equation can be

thought of as the ﬁrst few terms in a Taylor series expansion of a general function of ln(K/F) and

T −t. The results are shown in Table 3. In this case the R

2

is much higher at 94.93%.

[Table 3 about here.]

Here we repeat our out-of-sample exercise, in this case with more satisfactory results: a RMSE

of 0.0073 is obtained, representing a fairly close ﬁt. In Table 3 it should be noted that the estimates

obtained from using only the ﬁrst half of the sample are quite close to those obtained using the

whole data set.

If two models have equal explanatory power, then the observed ratio of the two models’ squared

errors should be distributed F(N

1

, N

2

) where N

1

and N

2

are the number of degrees of freedom in the

two models. When comparing the sticky strike and relative sticky delta models using a two tailed

test, this statistic must be greater than 1.12 or less than 0.89 for signiﬁcance at the 1% level. The

value of the statistic is 32.71 indicating that we can overwhelmingly reject the hypothesis that the

models have equal explanatory power. The relative sticky delta model in equation (11) can explain

the volatility surfaces in our data much better than the sticky strike model in equation (10).

The third model we test is the version of the stationary square root of time rule where the

function Φ in equation (1) does not change through time so that σ

TK

(t, S) − σ

TF

(t, S) is a known

function of ln(K/S)/

√

T −t. Using a similar Taylor Series expansion to the other models we test

σ

TK

(t, S) −σ

TF

(t, S) = c

1

ln(K/F)

√

T −t

+ c

2

_

ln(K/F)

√

T −t

_

2

+c

3

_

[ln(K/F)]

√

T −t

_

3

+ c

4

_

ln(K/F)

√

T −t

_

4

+ (12)

where c

1

, c

2

, c

3

and c

4

are constants and is a normally distributed error term. The results for this

model are shown in Table 4. We consider more restricted forms of the model where progressively c

4

and c

3

are set to zero, to render a more parsimonious model for the volatility smile. In this case the

R

2

is 95.62%. This is somewhat better than the R

2

for the sticky delta model in (11) even though

the model in equation (12) involves two parameters and the the one in equation (11) involves six

parameters.

[Table 4 about here.]

15

Testing the out of sample performance of the three possible forms of the square-root model gives

RMSEs of 0.0069, 0.0069 and 0.0073 for the four term, three term and two term forms of the model.

We conclude that for both in-sample and out of sample performance, including a cubic term leads to

improvement in performance while the addition of a quartic term has negligible eﬀect. Accordingly,

in the following analysis, when working with the square-root rule, we restrict our attention to the

cubic version.

We can calculate a ratio of sums of squared errors to compare the stationary square root of time

model in equation (12) to the relative sticky delta model in equation (11). In this case, the ratio is

1.373.

4

When a two tailed test is used it is possible to reject the hypothesis that the two models

have equal explanatory power at the 1% level. We conclude that equation (12) is an improvement

over equation (11).

In the stochastic square root of time rule, the functional relationship between σ

TK

(t, S) −

σ

TF

(t, S) and

ln(K/F)

√

T −t

changes stochastically through time. A model capturing this is

σ

TK

(t, S) −σ

TF

(t, S) = c

1

(t)

ln(K/F)

√

T −t

+ c

2

(t)

_

ln(K/F)

√

T −t

_

2

+ c

3

(t)

_

ln(K/F)

√

T −t

_

3

+ (13)

where c

1

(t), c

2

(t) and c

3

(t) are stochastic. To provide a test of this model we ﬁtted the model in

equation (12) to the data on a month by month basis. When the model in equation (13) is compared

to the model in equation (12) the ratio of sums of squared errors statistic is 3.11 indicating that the

stochastic square root of time model does provide a statistically signiﬁcantly better ﬁt to the data

that the stationary square root of time model at the 1% level.

5

The coeﬃcient, c

1

, in the monthly tests of the square root of time rule is always signiﬁcantly

diﬀerent from zero with a very high level of conﬁdence. In almost all cases, the coeﬃcients c

1

, c

2

and c

3

are all signiﬁcantly diﬀerent from zero at the 5% level. We note, however, that in spite of

this statistical signiﬁcance, the most economically signiﬁcant of the three terms is the linear term.

6

Figure 1 shows the level of c

1

, c

2

and c

3

plotted against the S&P 500 for the period covered by

our data. We note that there is fairly weak correlation between c

1

and the level of the S&P 500

index. However, there is reasonably high correlation between c

2

and c

3

and the index, during the

earlier part of our sample.

[Figure 1 about here.]

4

The ratios for comparing the quadratic model and quartic model to (11) are 1.157 and 1.373 respectively

5

Although since this model eﬀectively has many more parameters than its stationary counterpart, this result should

not be surprising.

6

The approximate linearity of the volatility skew for S&P 500 options has been mentioned by a number of re-

searchers including Derman (1999).

16

To provide an alternative benchmark for the square root of time rule, we now consider the

possibility that volatility is a function of some other power of time to maturity. We consider both

the stationary power rule model:

σ

TK

(t, S) −σ

TF

(t, S) = d

1

_

ln(K/F)

(T −t)

d

0

_

+ d

2

_

ln(K/F)

(T −t)

d

0

_

2

+d

3

_

ln(K/F)

(T −t)

d

0

_

3

+ d

4

_

ln(K/F)

(T −t)

d

0

_

4

+ (14)

and the stochastic power rule model:

σ

TK

(t, S) −σ

TF

(t, S) = d

1

(t)

_

ln(K/F)

(T −t)

d

0

(t)

_

+ d

2

(t)

_

ln(K/F)

(T −t)

d

0

(t)

_

2

+d

3

(t)

_

ln(K/F)

(T −t)

d

0

(t)

_

3

+ d

4

(t)

_

ln(K/F)

(T −t)

d

0

(t)

_

4

+ (15)

Estimating this model gives the results in Table 5. We note that the estimates of d

0

(the power

of time to maturity which best explains volatility changes) are approximately equal to 0.44, and is

statistically signiﬁcantly diﬀerent from 0.5. Performing an out of sample calculation of RMSE for

the stationary power rule, we obtain 0.0070 for each model. All three seem to display reasonable

parameter stability over time.

[Table 5 about here.]

Testing for signiﬁcance between the performance of this power rule and the square-root rule, we

obtain a ratio of squared errors of 1.04, insuﬃcient to reject the hypothesis of equal performance.

When the stochastic variants of each model are compared, the ratio becomes 5.39 showing that a

considerable improvement in ﬁt is obtained when the power relationship is allowed to vary with

time. All our model comparison results are summarized in Table 6.

[Table 6 about here.]

For the model in equation (15) there is little or no relationship between the level of the index

and the parameters of the rule. We do, however, note that in the later part of the data, there does

seem to be evidence of the power rule being very close to the square root rule (d

0

= 0.5).

[Figure 2 about here.]

We conclude that there is reasonably strong support for the square-root rule as a parsimonious

parametrisation of the volatility surface. There is some evidence that the rule could be marginally

improved upon by considering powers other than 0.5.

17

8 Estimation of Volatility Factors

We also used the data described in the previous section to estimate the factors underlying the model

in equation (3). We focus on implied volatility surfaces where moneyness is described as F/K rather

than S/K, as originally reported. We create this data, and the attendant changes in log volatility,

using interpolation.

The conventional analysis involves the use of principal components analysis applied to the

variance-covariance matrix of changes in implied volatilities. The results of this are shown in Table

7 and Figure 3. This is a similar analysis to that undertaken by Kamal and Derman (1997), Ski-

adopoulos, Hodges and Clewlow (2000), Cont and Fonseca (2002) and Cont, Foseca and Durrleman

(2002). Our study is diﬀerent from these in that it focuses on over-the-counter options rather than

exchange-traded options. It also uses options with much longer maturities.

8.1 Taking Account of the No-Arbitrage Drift

We now carry out a more sophisticated analysis that takes account of the no-arbitrage drift derived

in Section 3. The traditional principal components analysis does not use any information about the

drift of volatilities. In estimating the variance-covariance matrix, a constant drift for each volatility

is assumed. From our results in Section 3 we know that in a risk-neutral world the drifts of implied

volatilities depend on the factors. We assume that the market price of volatility risk is zero so that

volatilities have the same drift in the real world as in the risk-neutral world. We also assume a form

of the generalized sticky delta rule where each θ

TKi

is a function K/F and T − t. Consistent with

this assumption we write V

TK

(t) and θ

TKi

(t) as V

(T−t),F/K,t

and θ

(T−t),F/K,i

.

Deﬁne ˆ α as the drift of ln V

(T−t),F/K,t

. Applying Ito’s lemma to equation (6) the process for

ln V

(T−t),F/K,t

is

d ln V

(T−t),F/K,t

= ˆ αdt +

N

i=1

θ

(T−t),F/K,i

dz

i

,

where

ˆ α

(T−t),F/K

=

1

T −t

(1 −

σ

2

V

(T−t),F/K

) + σd

2

1

_

V

(T−t),F/K

(T −t)

N

i=1

θ

(T−t),F/K,i

ρ

i

−

(d

1

d

2

+ 1)

4

N

i=1

_

_

θ

2

(T−t),F/K,i

+

j=i

ρ

i

ρ

j

θ

(T−t),F/K,i

θ

(T−t),F/K,j

_

_

. (16)

We can write the discretised process for ln V

(T−t),F/K,t

as

∆ln(V

(T−t),F/K,t

) = ˆ α

(T−t),F/K

∆t +

(T−t),F/K,t

. (17)

Here, the term

(T−t),F/K,t

(which has expectation zero) represents the combined eﬀects of the

factors on the movement of the volatility of an option with time to maturity T − t and relative

moneyness F/K, observed at time t.

18

The correlations between the ’s arise from a) the possibility that the same factor aﬀects more

than one option and b) the correlation between the factors arising from their correlation with the

asset price. The covariance between two ’s is given by

E(

(T

1

−t),(F/K

1

),t

(T

2

−t),(F/K

2

),t

) = ∆t

N

k=1

_

θ

(T

1

−t),(F/K

1

),k

θ

(T

2

−t),(F/K

2

),k

+

l=k

θ

(T

1

−t),(F/K

1

),k

θ

(T

2

−t),(F/K

2

),l

ρ

k

ρ

l

_

+

_

_

_

σ

2

if T

1

= T

2

and K

1

= K

2

.

0 otherwise.

(18)

The analysis assumes no correlation between any volatility movements at diﬀerent times. The

σ

**captures further eﬀects over and above those given by the ﬁnite set of factors considered.
**

We carry out a maximum likelihood analysis to determine the values of the θ’s, ρ

s, and σ

that

best describe the observed changes in option implied volatilities. As these parameters are changed,

ˆ α changes and there is a diﬀerent covariance matrix for the errors.

For any set of parameters we back out an for each observation in our data set, using (17). Since

we know that these ’s are jointly normally distributed, it is a standard result in econometrics

7

that

the log likelihood function for this set of parameters is given by:

L =

M

i=1

−

t

i

(Σ

t

i

)

−1

t

i

−ln det(Σ

t

i

) (19)

where

t

i

is the vector of ’s for time t

i

and Σ

t

i

is the variance-covariance matrix described by (18)

at time t

i

.

8

To estimate the factors, we can thus use an optimization routine to maximize (19)

by varying θ

(T−t),F/K,j

(where T − t and F/K are taken to be the varying levels of maturity and

moneyness we observe in our volatility matrices), ρ

j

and σ

2

**. For an N factor model, describing a
**

volatility matrix of M volatilities, this results in N ×M + N + 1 parameters. We used MATLAB’s

fmincon optimization routine (constraining the ρ’s to lie between -1 and 1) and found that the

routine is eﬃcient for solving the problem, even for this large number of parameters.

Fitting a four factor model (N = 4) we ﬁnd the factors given in Figure 4 and Table 8. To ease

the comparison with the results from Figure 3 and Table 7, we have converted our factors from

a collection of factors each correlated with stock price movements (and therefore each other) to a

collection of ﬁve orthonormal factors, one of which (factor zero) is driven by the same Brownian

Motion as in the underlying stock price. The normalization results in the factors having to be

7

See, for example Greene (1997), page 89.

8

It is worth observing that our model’s intertemporally independent residuals allows us to consider the likelihood

ratio as the sum of individual time steps’ log likelihood ratios. This considerably speeds up the calculation of both

the inner product term (

t

i

(Σ

t

i

)

−1

t

i

) and ln det(Σ

t

i

), which (for a large sample) could involve quite large matrices.

19

multiplied by a scaling factor, which allows the familiar analysis of considering the proportion of the

covariance matrix of the surface which is explained by each factor.

Focusing on factor zero (the factor which is perfectly correlated with stock prices) we see that

an increase in stock prices leads to an increase in the slope of implied volatilities. In the case of the

typical (negative) skew observed, this would result in a decrease in stock prices causing the skew to

steepen, rendering out of the money puts more valuable. The eﬀect is most pronounced for short

maturity options, and tapers oﬀ for longer maturity options.

The most important of all the factors is the ﬁrst one. This factor shows evidence of hyperbolic

behaviour in the time dimension, and works to oﬀset the explosive behaviour we noted in section

3. It is roughly level in the moneyness dimension, and can therefore be seen as representing moves

up and down of the existing smile. Factors two and three show evidence of a square root eﬀect,

having roughly a square-root shape in the time dimension, but both showing evidence of increasing

sensitivity for higher strike options. Factor four is less easy to interpret, seeming to be a factor

which aﬀects low moneyness, long maturity options.

With four factors, we can explain 99.6% of the variation in implied volatilities, with both the

last two factors making a signiﬁcant contribution.

Contrasting the results of our consistent factor analysis with the simpler approach originally

followed, we note that we obtain a richer factor structure. The principal components approach

gives two dominant factors which explain 98.8% of the variation. The ﬁrst factor is similar to

that uncovered by the drift consistent approach, while the second factor shows some evidence of a

square-root behaviour in the maturity dimension.

[Figure 3 about here.]

[Figure 4 about here.]

[Table 7 about here.]

[Table 8 about here.]

9 Summary

It is a common practice in the over-the-counter markets to quote option prices using their Black–

Scholes implied volatilities. In this paper we have developed a model of the evolution of implied

volatilities and produced a no-arbitrage condition that must be satisﬁed by the volatilities. Our

model is exactly consistent with the initial volatility surface, but more general than the IVF model

of Rubinstein (1994), Derman and Kani (1994), and Dupire (1994). The no-arbitrage condition

leads to the conclusions that a) when the volatility is independent of the asset price there must be

a U-shaped volatility smile and b) when the implied volatility is a decreasing (increasing) function

20

of the asset price there must be a negative (positive) correlation between the volatility and the

asset price. We outline how these no-arbitrage conditions can be incorporated into a factor analysis,

resulting in no-arbitrage consistent factors. Our empirical implementation of this analysis suggests

that more factors may in fact govern volatility movements than would be thought otherwise.

A number of rules of thumb have been proposed for how traders manage the volatility surface.

These are the sticky strike, sticky delta, and square root of time rules. Some versions of these

rules are clearly inconsistent with the no-arbitrage condition; for other versions of the rules the

no-arbitrage condition can in principle be satisﬁed.

Our empirical tests of the rules of thumb using 47 months of volatility surfaces for the S&P 500

show that the relative sticky delta model (where the excess of the implied volatility of an option

over the corresponding at-the-money volatility is a function of moneyness) outperforms the sticky

strike rule. Also, the stochastic square root of time model outperforms the relative sticky delta rule.

A Proof of Equation (4)

For simplicity of notation, we assume that r and q are constants. Deﬁne F(τ) as the forward price

at time τ for a contract maturing at time t + ∆t so that

F(t) = S(t)e

(r−q)∆t

.

The price at time t of a call option with strike price F(t) and maturity t + ∆t is given by

c(S(t), V

t+∆t,F(t)

, t; F(t), t + ∆t) = e

−q∆t

S(t)[N(d

1

) −N(d

2

)],

where in this case

d

1

= −d

2

=

√

∆t

2

σ

t+∆t,F(t)

(t, S(t)).

The call price can be written

c(S(t), V

t+∆t,F(t)

, t; F(t), t + ∆t) = e

−q∆t

S(t)

_

d

1

−d

1

φ(x) dx,

where

φ(x) =

1

√

2π

exp

_

−

x

2

2

_

.

For some ¯ x

_

d

1

−d

1

φ(x) dx = 2d

1

φ(¯ x)

and the call price is therefore given by

c(S(t), V

t+∆t,F(t)

, t; F(t), t + ∆t) = 2e

−q∆t

S(t)d

1

φ(¯ x),

or

c(S, V

t+∆t,F(t)

, t; F(t), t + ∆t) = e

−q∆t

S(t)φ(¯ x)σ

t+∆t,F(t)

(t, S(t))

√

∆t. (20)

21

The process followed by S is

dS

S

= (r −q) dt + σ dz.

Using Ito’s lemma

dF = σF dz.

When terms of order higher that (∆t)

1/2

are ignored

F(t + ∆t) −F(t) = σ(t)F(t)[z(t + ∆t) −z(t)].

It follows that

lim

∆t→0

1

√

∆t

E[F(t + ∆t) −F(t)]

+

= lim

∆t→0

1

√

∆t

σ(t)F(t)E[z(t + ∆t) −z(t)]

+

,

where E denotes expectations under the risk-neutral measure.

Because

E[F(t + ∆t) −F(t)]

+

= e

r∆t

c(S(t), V

t+∆t,F(t)

, t; F(t), t + ∆t),

and

lim

∆t→0

1

√

∆t

E[z(t + δt) −z(t)]

+

=

1

√

2π

,

it follows that

lim

∆t→0

1

√

∆t

e

r∆t

c(S(t), V

t+∆t,F(t)

, t; F(t), t + ∆t) =

σ(t)F(t)

√

2π

.

Substituting from equation (20)

lim

∆t→0

e

r∆t

e

−q∆t

S(t)φ(¯ x)σ

t+∆t,F(t)

(t, S(t)) =

σ(t)F(t)

√

2π

.

As ∆t tends to zero, φ(¯ x) tends to 1/

√

2π so that

lim

∆t→0

σ

t+∆t,F(t)

(t, S(t)) = σ(t).

This is the required result.

References

Andersson, K. (2003) “Stochastic volatility,” U.U.D.M. Project Report, Department of Mathematics,

Uppsala University

Balland, P. (2002) “Deterministic implied volatility models,” Quantitative Finance, 2, 31–44

Black, F. and M.S. Scholes (1973), “The pricing of options and corporate liabilities,” Journal of

Political Economy, 81, 637–659

Bates, D. S. (1996), “Jumps and Stochastic Volatility: Exchange Rate Process Implicit in Deutsche

Mark Options,” Review of Financial Studies, 9, 1, 69–107

22

Brace, A., B. Goldys, F. Klebaner, and R. Womersley (2001), “Market model of stochastic volatility

with applications to the BGM model,” Working paper S01-1, Dept of Statistics, University of New

South Wales

Britten–Jones, M., and A. Neuberger (2000) “Option Prices, Implied Price Processes, and Stochastic

Volatility,” Journal of Finance, 55, 2 (2000), 839–866

Cont, R., J. Fonseca, and V. Durrleman (2002) “Stochastic Models of Implied Volatility Surfaces,”

Economic Notes, 31, 2 (2002), 361–377

Cont, R., and J. Fonseca (2002), “Dynamics of implied volatility surfaces,” Quantitative Finance, 2,

45–60

Christie, A. A. (1982), “The stochastic behavior of common stock variances: Value, leverage, and

interest rate eﬀects,” Journal of Financial Economics, 10, 4, 407–432

Derman, E. (1999), “Regimes of volatility,” Quantitative Strategies Research Notes, Goldman Sachs,

New York, NY. Also in Risk Magazine, April, 1999

Derman, E. and I. Kani (1994a), “The volatility smile and its implied tree,” Quantitative Strategies

Research Notes, Goldman Sachs, New York, NY

Dumas, B., J. Fleming and R.E. Whaley (1997), “Implied volatility functions: empirical tests,”

Journal of Finance 6, 2059–2106

Dupire, B. (1994), “Pricing with a smile,” Risk, 7, 18–20

Durrleman, V. (2004), “From implied to spot volatilities,” Ph.D Dissertation, Department of Oper-

ational Research and Financial Engineering, Princeton University

Greene, W. (1997), Econometric Analysis, 3rd Edition, Prentice Hall, Upper Saddle River, NJ

Heath, D., R. Jarrow, and A. Morton (1992), “Bond Pricing and the Term Structure of the Interest

Rates: A New Methodology,” Econometrica, 60, 1, 77–105

Heston, S.L. (1993), “A closed-form solution for options with stochastic volatility applications to

bond and currency options,” Review of Financial Studies, 6, 327–343

Hull, J. (2002) Options, Futures, and Other Derivatives, 5th Edition, Prentice Hall, Englewood

Cliﬀs, NJ

Hull, J. and W. Suo (2002), “A methodology for assessing model risk and its application to the

implied volatility function model,” Journal of Financial and Quantitative Analysis, 37, 2, 297–

318

23

Hull, J. and A. White (1987), “The pricing of options with stochastic volatility,” Journal of Finance,

42, 281–300

Jackwerth, J.C. and M. Rubinstein (1996), “Recovering probabilities from option prices”, Journal

of Finance, 51, 1611–1631

Kamal, M. and E. Derman, “The patterns of change in implied index volatilities,” Quantitative

Strategies Research Notes, Goldman Sachs, New York, NY

Ledoit O. and P. Santa Clara (1998) “Relative pricing of options with stochastic volatility” Working

Paper, Anderson Graduate School of Management, University of California, Los Angeles

Merton, R. C. (1973), “Theory of rational option pricing,” Bell Journal of Economics and Manage-

ment Science, 4, 141–183

Natenberg S. (1994) Option Pricing and Volatility: Advanced Trading Strategies and Techniques,

2nd ed. McGraw-Hill

Rubinstein, M. (1994), “Implied binomial trees,” Journal of Finance, 49, 771–818

Sch¨onbucher, P.J. (1999), “A market model of stochastic implied volatility,” Philosophical Transac-

tions of the Royal Society, Series A, 357, 2071–2092

Skiadopolous, G., S. Hodges and L. Clewlow (2000), “Dynamics of the S&P500 implied volatility

surface,” Review of Derivatives Research, 3, 263–82

Stein, E. and C. Stein (1991), “Stock price distributions with stochastic volatilities: An analytical

approach,” Review of Financial Studies, 4, 727–752

24

List of Figures

1 Plot of the level of the S&P 500 index (solid line, with scale on the right axis) and the

estimates of c

1

, c

2

and c

3

(dashed line, with scale on the left axes). Note the positive

correlations between c

2

and c

3

and the index during the earlier part of the sample. . 26

2 Plot of the level of the S&P 500 index (solid line, with scale on the right axis) and

the estimates of d

0

, d

1

, d

2

and d

3

(dashed line, with scale on the left axes). Note

the positive correlations between c

3

and the index. The dotted line on the ﬁrst graph

shows the level of d

0

(0.5) which would be consistent with the square-root rule. . . . 27

3 Principal Component factors for the Totem data. The height of the surface measures

the value of the normalised factor, for each maturity and moneyness combination. . 28

4 Estimated factors for the Totem data, accounting for no-arbitrage drift in volatilities.

The height of the surface measures the normalised factor for each maturity and mon-

eyness combination. Factors 1,2,3 and 4 are independent of each other and the stock

price, while factor 0 is perfectly correlated with the stock price. . . . . . . . . . . . . 29

25

1999 2000 2001 2002

−0.3

−0.15

c1

1999 2000 2001 2002

1000

1200

1400

1600

1999 2000 2001 2002

−0.15

0.3

c2

1999 2000 2001 2002

1000

1200

1400

1600

1999 2000 2001 2002

−0.8

1.6

c3

1999 2000 2001 2002

1000

1200

1400

1600

Figure 1: Plot of the level of the S&P 500 index (solid line, with scale on the right axis) and the

estimates of c

1

, c

2

and c

3

(dashed line, with scale on the left axes). Note the positive correlations

between c

2

and c

3

and the index during the earlier part of the sample.

26

1999 2000 2001 2002

0.2

0.5

0.8

d0

1999 2000 2001 2002

1000

1200

1400

1600

1999 2000 2001 2002

−0.35

−0.15

d1

1999 2000 2001 2002

1000

1200

1400

1600

1999 2000 2001 2002

0

0.15

d2

1999 2000 2001 2002

1000

1200

1400

1600

1999 2000 2001 2002

−0.1

0.5

d3

1999 2000 2001 2002

1000

1200

1400

1600

Figure 2: Plot of the level of the S&P 500 index (solid line, with scale on the right axis) and the

estimates of d

0

, d

1

, d

2

and d

3

(dashed line, with scale on the left axes). Note the positive correlations

between c

3

and the index. The dotted line on the ﬁrst graph shows the level of d

0

(0.5) which would

be consistent with the square-root rule.

27

0

5

80

100

120

0

0.2

0.4

T−t

Factor 1: Variance 2.823 Portion 0.926

K/F 0

5

80

100

120

−0.5

0

0.5

T−t

Factor 2: Variance 0.188 Portion 0.062

K/F

0

5

80

100

120

−0.5

0

0.5

T−t

Factor 3: Variance 0.016 Portion 0.005

K/F 0

5

80

100

120

−1

0

1

T−t

Factor 4: Variance 0.011 Portion 0.004

K/F

Figure 3: Principal Component factors for the Totem data. The height of the surface measures the

normalised factor, for each maturity and moneyness combination.

28

0

5

80

100

120

0

0.2

0.4

T−t

Factor 1: Variance 2.043, Portion 0.720

K/F 0

5

80

100

120

−0.3

−0.2

−0.1

T−t

Factor 2: Variance 0.284, Portion 0.100

K/F

0

5

80

100

120

−1

0

1

T−t

Factor 3: Variance 0.147, Portion 0.052

K/F 0

5

80

100

120

−0.5

0

0.5

T−t

Factor 4: Variance 0.125, Portion 0.044

K/F

0

5

80

100

120

−0.5

0

0.5

T−t

Factor 0: Variance 0.229, Portion 0.081

K/F

Figure 4: Estimated factors for the Totem data, accounting for no-arbitrage drift in volatilities. The

height of the surface measures the value of the normalised factor for each maturity and moneyness

combination. Factors 1,2,3 and 4 are independent of each other and the stock price, while factor 0

is perfectly correlated with the stock price.

29

List of Tables

1 Volatility matrix for June 1998. Time to maturity is measured in months while strike

is in percentage terms, relative to the level of the S&P 500 index. . . . . . . . . . . . 31

2 Estimates for the version of the sticky strike model in equation (10) where volatility

depends on strike price and time to maturity. . . . . . . . . . . . . . . . . . . . . . . 32

3 Estimates for the version of the relative sticky delta model in equation (11). . . . . . 33

4 Estimates for the stationary square root of time rule in equation (12). . . . . . . . . 34

5 Estimates of the power rule from equation (14). . . . . . . . . . . . . . . . . . . . . . 35

6 Comparison of Models using the ratio of sums of squared errors statistic. In a two-

tailed test the statistic must be greater than 1.123 (1.111) to reach the conclusion

that the ﬁrst equation to provide a better explanation of the data than the second

equation at the 1% (5%) level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

7 Factors obtained by Principal Components analysis applied to changes in the volatility

surface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

8 Factors obtained by using maximum likelihood estimation, incorporating the no-

arbitrage consistent drift. The factors are all orthogonal, with factors 1, 2, 3 and

4 being uncorrelated with the stock price Brownian Motion, and factor 0 being per-

fectly correlated with the stock price. . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

30

Time to Maturity (months)

Strike 6 12 24 36 48 60

120 15.91% 18.49% 20.20% 21.03% 21.44% 21.64%

110 18.13% 20.33% 21.48% 21.98% 22.18% 22.30%

105 19.50% 21.26% 22.21% 22.52% 22.59% 22.70%

100 20.94% 22.38% 23.06% 23.14% 23.11% 23.07%

95 22.73% 23.71% 23.92% 23.73% 23.58% 23.53%

90 24.63% 24.99% 24.78% 24.40% 24.10% 23.96%

80 28.41% 27.71% 26.66% 25.83% 25.23% 24.83%

Table 1: Volatility matrix for June 1998. Time to maturity is measured in months while strike is in

percentage terms, relative to the level of the S&P 500 index.

31

Variable Estimate Standard Error t-statistic Estimate from ﬁrst 2 years

a

0

0.4438616 0.0238337 18.62 0.5405505

a

1

-0.0001944 0.000036 -5.40 -0.0002511

a

2

0.0000000 0.0000000 1.44 0.0000000

a

3

-0.0262681 0.0033577 -7.82 -0.0335188

a

4

-0.0006589 0.0003454 -1.91 -0.0011062

a

5

0.0000290 0.0000022 13.30 0.0000355

R

2

0.2672

Standard error

of residuals 0.0327

Table 2: Estimates for the version of the sticky strike model in equation (10) where volatility depends

on strike price and time to maturity.

32

Variable Estimate Standard Error t-statistic Estimates from ﬁrst 2 years

b

0

0.0058480 0.0003801 15.39 0.0067000

b

1

-0.2884075 0.0019565 -147.41 -0.3345461

b

2

0.0322727 0.0067576 4.78 0.0112730

b

3

-0.0075740 0.0003487 -21.72 -0.0091289

b

4

0.0015705 0.0000701 22.42 0.0018404

b

5

0.0414902 0.0009180 45.20 0.0496937

R

2

0.9493

Standard error

of residuals 0.0057

Table 3: Estimates for the version of the relative sticky delta model in equation (11).

33

Variable Estimate Standard Error t-statistic Estimates from ﬁrst 2 years

Unrestricted

c

1

-0.2529421 0.0022643 -111.7102965 -0.2702221

c

2

0.1423444 0.0124579 11.4260104 0.0675920

c

3

0.5101281 0.0572173 8.9156322 0.2231858

c

4

-0.2067651 0.2106077 -0.9817545 -0.1157022

R

2

0.9631

Standard error

of residuals 0.0049

c

4

= 0

c

1

-0.2547240 0.0013537 -188.1644600 -0.2712324

c

2

0.1319096 0.0064984 20.2987718 0.0620709

c

3

0.5595131 0.0272658 20.5206944 0.2522889

R

2

0.9631

Standard error

of residuals 0.0049

c

3

= c

4

= 0

c

1

-0.2402662 0.0012591 -190.8294624 -0.2654298

c

2

0.0565335 0.0058389 9.6821751 0.0240899

R

2

0.9562

Standard error

of residuals 0.0052

Table 4: Estimates for the stationary square root of time rule in equation (12).

34

Parameter Estimate Standard Error t-statistic Estimates from ﬁrst two years

Unrestricted

d

0

0.4392204 0.0052036 84.4076018 0.5114474

d

1

-0.2416622 0.0023361 -103.4488377 -0.2714956

d

2

0.0725670 0.0152348 4.7632480 0.0864143

d

3

0.3307833 0.0617382 5.3578373 0.2373512

d

4

0.3775721 0.2574542 1.4665602 -0.3039707

R

2

0.9652

Standard error

of residuals 0.0047

d

4

= 0

d

0

0.4419942 0.0048206 91.6878322 0.5083725

d

1

-0.2394926 0.0018407 -130.1086545 -0.2733074

d

2

0.0912235 0.0076917 11.8600018 0.0674926

d

3

0.2581095 0.0390363 6.6120391 0.2901229

R

2

0.9651

Standard error

of residuals 0.0047

d

4

= d

3

= 0

d

0

0.4220134 0.0034726 121.5254560 0.4825231

d

1

-0.2308670 0.0012548 -183.9941367 -0.2632187

d

2

0.0540911 0.0055657 9.7185927 0.0238527

R

2

0.9644

Standard error

of residuals 0.0048

Table 5: Estimates of the power rule from equation (14).

35

Test Statistic

Sticky Strike (10) versus Sticky delta (11) 32.71

Sticky Delta (11) versus (cubic) Stationary Square-root rule (12) 1.37

Stationary Square-root rule (12) versus Stochastic Square-root rule (13) 3.11

Stationary Square-root rule (12) versus Stationary power rule (14) 1.04

Stochastic Square-root rule (13) versus Stochastic power rule (15) 5.39

Table 6: Comparison of Models using the ratio of sums of squared errors statistic. In a two-tailed

test the statistic must be greater than 1.123 (1.111) to reach the conclusion that the ﬁrst equation

to provide a better explanation of the data than the second equation at the 1% (5%) level

36

Factor 1 Variance=2.823464 Factor 2 Variance=0.187797

Portion=0.926429 Portion=0.061620

Time to Maturity (months) Time to Maturity (months)

Strike 6 12 24 36 48 60 Strike 6 12 24 36 48 60

0.80 0.17 0.15 0.12 0.11 0.10 0.10 0.80 0.15 0.03 -0.06 -0.10 -0.14 -0.15

0.90 0.19 0.16 0.13 0.12 0.11 0.10 0.90 0.18 0.06 -0.06 -0.11 -0.15 -0.16

0.95 0.20 0.16 0.13 0.12 0.11 0.10 0.95 0.20 0.06 -0.06 -0.11 -0.15 -0.17

1.00 0.21 0.17 0.14 0.12 0.11 0.10 1.00 0.22 0.06 -0.06 -0.11 -0.16 -0.18

1.00 0.21 0.17 0.14 0.12 0.11 0.10 1.00 0.22 0.06 -0.06 -0.11 -0.16 -0.18

1.05 0.21 0.17 0.14 0.12 0.11 0.10 1.05 0.24 0.06 -0.06 -0.11 -0.16 -0.18

1.10 0.22 0.18 0.14 0.12 0.11 0.10 1.10 0.26 0.07 -0.05 -0.11 -0.17 -0.19

1.20 0.22 0.19 0.15 0.13 0.11 0.11 1.20 0.29 0.10 -0.05 -0.11 -0.18 -0.20

Factor 3 Variance=0.015721 Factor 4 Variance=0.010684

Portion=0.005158 Portion=0.003505

Time to Maturity (months) Time to Maturity (months)

Strike 6 12 24 36 48 60 Strike 6 12 24 36 48 60

0.80 0.41 0.10 -0.02 0.04 0.10 0.14 0.80 -0.15 -0.27 -0.13 -0.07 0.07 0.08

0.90 0.31 0.03 -0.07 0.02 0.09 0.12 0.90 -0.07 -0.21 -0.11 -0.05 0.08 0.09

0.95 0.24 -0.02 -0.09 0.01 0.07 0.10 0.95 -0.02 -0.20 -0.11 -0.05 0.08 0.10

1.00 0.17 -0.08 -0.12 -0.01 0.05 0.08 1.00 0.05 -0.18 -0.10 -0.05 0.08 0.12

1.00 0.17 -0.08 -0.12 -0.01 0.05 0.08 1.00 0.05 -0.18 -0.10 -0.05 0.08 0.12

1.05 0.08 -0.14 -0.15 -0.03 0.03 0.07 1.05 0.15 -0.13 -0.09 -0.06 0.09 0.13

1.10 -0.02 -0.22 -0.20 -0.04 0.01 0.05 1.10 0.27 -0.08 -0.07 -0.07 0.09 0.15

1.20 -0.28 -0.39 -0.28 -0.07 -0.03 0.02 1.20 0.58 -0.00 -0.05 -0.08 0.10 0.18

Table 7: Factors obtained by Principal Components analysis applied to changes in the volatility

surface.

37

Factor 1 Variance=2.043035 Factor 2 Variance=0.283968

Portion=0.719536 Portion=0.100011

Time to Maturity (months) Time to Maturity (months)

Strike 6 12 24 36 48 60 Strike 6 12 24 36 48 60

0.80 0.16 0.15 0.12 0.11 0.10 0.09 0.80 -0.12 -0.11 -0.11 -0.11 -0.12 -0.12

0.90 0.18 0.16 0.13 0.12 0.10 0.10 0.90 -0.15 -0.13 -0.12 -0.12 -0.13 -0.13

0.95 0.19 0.17 0.14 0.12 0.11 0.10 0.95 -0.17 -0.13 -0.12 -0.12 -0.13 -0.13

1.00 0.20 0.17 0.14 0.12 0.11 0.10 1.00 -0.19 -0.14 -0.13 -0.13 -0.13 -0.13

1.00 0.20 0.17 0.14 0.12 0.11 0.10 1.00 -0.19 -0.14 -0.13 -0.13 -0.13 -0.13

1.05 0.21 0.18 0.15 0.13 0.11 0.10 1.05 -0.21 -0.15 -0.13 -0.13 -0.13 -0.13

1.10 0.21 0.18 0.15 0.13 0.11 0.10 1.10 -0.24 -0.16 -0.13 -0.13 -0.13 -0.13

1.20 0.22 0.20 0.16 0.13 0.12 0.11 1.20 -0.28 -0.17 -0.13 -0.13 -0.14 -0.14

Factor 3 Variance=0.146822 Factor 4 Variance=0.125169

Portion=0.051709 Portion=0.044083

Time to Maturity (months) Time to Maturity (months)

Strike 6 12 24 36 48 60 Strike 6 12 24 36 48 60

0.80 -0.07 -0.00 0.05 0.07 0.07 0.08 0.80 -0.08 -0.07 -0.11 -0.15 -0.19 -0.20

0.90 -0.16 -0.05 0.03 0.06 0.07 0.08 0.90 -0.05 -0.06 -0.11 -0.15 -0.20 -0.21

0.95 -0.21 -0.07 0.03 0.06 0.08 0.09 0.95 -0.04 -0.05 -0.11 -0.15 -0.20 -0.22

1.00 -0.27 -0.08 0.02 0.06 0.08 0.09 1.00 -0.02 -0.04 -0.10 -0.15 -0.20 -0.22

1.00 -0.27 -0.08 0.02 0.06 0.08 0.09 1.00 -0.02 -0.04 -0.10 -0.15 -0.20 -0.22

1.05 -0.33 -0.10 0.02 0.06 0.09 0.10 1.05 -0.01 -0.04 -0.10 -0.15 -0.20 -0.22

1.10 -0.40 -0.12 0.01 0.06 0.09 0.10 1.10 0.00 -0.02 -0.09 -0.15 -0.20 -0.23

1.20 -0.52 -0.17 -0.00 0.06 0.10 0.11 1.20 0.03 0.01 -0.08 -0.14 -0.21 -0.24

Factor 0 Variance=0.229168

Portion=0.080711

Time to Maturity (months)

Strike 6 12 24 36 48 60

0.80 0.49 0.26 0.12 0.08 0.05 0.04

0.90 0.39 0.22 0.10 0.07 0.05 0.04

0.95 0.32 0.19 0.09 0.06 0.04 0.03

1.00 0.25 0.16 0.08 0.06 0.04 0.03

1.00 0.25 0.16 0.08 0.06 0.04 0.03

1.05 0.17 0.12 0.07 0.05 0.03 0.02

1.10 0.08 0.09 0.05 0.05 0.03 0.02

1.20 -0.13 0.01 0.03 0.04 0.02 0.01

Table 8: Factors obtained by using maximum likelihood estimation, incorporating the no-arbitrage

consistent drift. The factors are all orthogonal, with factors 1, 2, 3 and 4 being uncorrelated with

the stock price Brownian Motion, and factor 0 being perfectly correlated with the stock price.

38

1

Introduction

Option traders and brokers in over-the-counter markets frequently quote option prices using implied volatilities calculated from Black and Scholes (1973) and other similar models. The reason is that implied volatilities usually have to be updated less frequently than the prices themselves. Put–call parity implies that, in the absence of arbitrage, the implied volatility for a European call option is the same as that for a European put option when the two options have the same strike price and time to maturity. This is convenient: when quoting an implied volatility for a European option with a particular strike price and maturity date, a trader does not need to specify whether a call or a put is being considered. The implied volatility of European options on a particular asset as a function of strike price and time to maturity is known as the volatility surface. Every day traders and brokers estimate volatility surfaces for a range of diﬀerent underlying assets from the market prices of options. Some points on a volatility surface for a particular asset can be estimated directly because they correspond to actively traded options. The rest of the volatility surface is typically determined by interpolating between these points. If the assumptions underlying Black–Scholes held for an asset, its volatility surface would be ﬂat and unchanging. In practice the volatility surfaces for most assets are not ﬂat and change stochastically. Consider for example equities and foreign currencies. Rubinstein (1994) and Jackwerth and Rubinstein (1996), among others, show that the implied volatilities of stock and stock index options exhibit a pronounced “skew” (that is, the implied volatility is a decreasing function of strike price). For foreign currencies this skew becomes a “smile” (that is, the implied volatility is a U-shaped function of strike price). For both types of assets, the implied volatility can be an increasing or decreasing function of the time to maturity. The volatility surface changes through time, but the general shape of the relationship between volatility and strike price tends to be preserved. Traders use a volatility surface as a tool to value a European option when its price is not directly observable in the market. Provided there are a reasonable number of actively traded European options and these span the full range of the strike prices and times to maturity that are encountered, this approach ensures that traders price all European options consistently with the market. However, as pointed out by Hull and Suo (2002), there is no easy way to extend the approach to price pathdependent exotic options such as barrier options, compound options, and Asian options. As a result there is liable to be some model risk when these options are priced. Traders also use the volatility surface in an ad hoc way for hedging. They attempt to hedge against potential changes in the volatility surface as well as against changes in the asset price. Derman (1999) discusses alternative approaches to hedging against asset price movements. The “volatility-by-strike” or “sticky strike” rule assumes that the implied volatility for an option with a given strike price and maturity will be unaﬀected by changes in the underlying asset price. Another 1

popular approach is the “volatility-by-moneyness” or “sticky delta” rule. This assumes that the volatility for a particular maturity depends only on the moneyness (that is, the ratio of the price of the underlying asset to the strike price). The ﬁrst attempts to model the volatility surface were by Rubinstein (1994), Derman and Kani (1994), and Dupire (1994). These authors show how a one-factor model for an asset price, known as the implied volatility function (IVF) model, can be developed so that it is exactly consistent with the current volatility surface. Unfortunately, the evolution of the volatility surface under the IVF model can be unrealistic. The volatility surface given by the model at a future time is liable to be quite diﬀerent from the initial volatility surface. For example, in the case of a foreign currency the initial U-shaped relationship between implied volatility and strike price is liable to evolve to one where the volatility is a monotonic increasing or decreasing function of strike price. Dumas, Fleming, and Whaley (1997) have shown that the IVF model does not capture the dynamics of market prices well. Hull and Suo (2002) have shown that it can be dangerous to use the model for the relative pricing of barrier options and plain vanilla options. This paper extends existing research in a number of ways. Similarly to Ledoit and Santa Clara (1998), Sch¨nbucher (1999), Brace et al (2001), and Britten–Jones and Neuberger (2000), we develop o no arbitrage conditions for the evolution of the volatility surface. We consider a general model where there are a number of factors driving the volatility surface and these may be correlated with stock price movements. We then investigate the implications of the no-arbitrage condition for the shapes of the volatility surfaces likely to be observed in diﬀerent situations and examine whether the various rules of thumb that have been put forward by traders are consistent with the no-arbitrage condition. Finally we carry out two empirical studies. The ﬁrst empirical study extends the work of Derman (1999) to investigate whether movements in the volatility surface are consistent with diﬀerent rules of thumb. The second extends work of Kamal and Derman (1997), Skiadopoulos, Hodges, and Clewlow (2000), Cont and Fonseca (2002), and Cont, Fonseca and Durrleman (2002) to estimate the factors driving movements in the volatility surface in a way that reﬂects the theoretical drifts of points on the surface. One distinguishing feature of our work is that we use over-the-counter consensus data for our empirical studies. To the best of our knowledge we are the ﬁrst derivatives researchers to use this type of data. Another distinguishing feature of our research is that, in addition to the sticky strike and sticky delta rules we consider a “square root of time” rule that is often used by traders and documented by, for example, Natenberg (1994), where the implied volatility is assumed to depend on the moneyness and the square root of the time to maturity through a very speciﬁc relationship. The rest of the paper is organized as follows. Section 2 explains the rules of thumb. Section 3 develops a general model for the evolution of a volatility surface and derives the no-arbitrage condition. Section 4 discusses the implications of the no-arbitrage condition. Section 5 examines a number of special cases of the model. Section 6 considers whether the rules of thumb are consistent

2

In this section we explain three diﬀerent rules of thumb: the sticky strike rule. and the delta. In the generalized sticky strike model it is independent of S. Conclusions are in Section 9. In the ﬁrst category are rules concerned with the way in which the volatility surface changes through time. σT K . t. ∂S where for the purposes of calculating the partial derivative the option price. is considered to be a function of the asset price. In what we will refer to as the basic sticky strike model. Section 8 derives factors driving the volatility surface in a way that is consistent with the no-arbitrage condition. σT K is a function only of K and T . In the second category are rules concerned with the relationship between the volatility smiles for diﬀerent option maturities at a point in time. 3 . The third rule is in the second category and assists with the mechanics of “ﬁlling in the blanks” when a complete volatility surface is being produced. 2 Rules of Thumb A number of rules of thumb have been proposed about volatility surfaces. They are useful in creating a complete volatility surface when market prices are available for a relatively small number of options. They are useful in the calculation of the Greek letters such as delta and gamma. We deﬁne the call price as c. Section 7 tests whether there is empirical support for the rules of thumb. dS 2. S. The ﬁrst two of these rules are in the ﬁrst category and provide a basis for calculating Greek letters. ∆ of the option as the total rate of change of its price with respect to the asset price: ∆= dc . strike price K and implied volatility σT K . This is an appealing assumption because it implies that the sensitivity of the price of an option to S is ∂c . c. the underlying asset price as S. We consider a European call option with maturity T . The same is true of gamma. and time. the rate of change of delta with respect to the asset price. the sticky delta rule and the square root of time rule.with the no-arbitrage condition. The assumption enables the Black–Scholes formula to be used to calculate delta with the volatility parameter set equal to the option’s implied volatility. but possibly dependent on other stochastic variables.1 The Sticky Strike Rule The sticky strike rule assumes that σT K is independent of S. These rules of thumb fall into two categories.

∂S ∂σT K ∂S Again. In the most basic form of the sticky delta rule the implied volatility is assumed to be a deterministic function of K/S and T − t. This assumes that the implied volatility of an option depends on the S and K. σT K .2 The Sticky Delta Rule An alternative to the sticky strike rule is the sticky delta rule. through its dependence on the moneyness variable. if σT K is a declining (increasing) function of the strike price. In the second term. T . We will refer to this as the generalized sticky delta model. S. A version on the sticky delta rule sometimes used by traders is that σT K (S. it is an increasing (declining) function of S and ∆ is greater than (less than) that given by Black-Scholes. We will refer to this as the basic sticky delta model. The implied volatility changes stochastically but parameters describing the process followed by the volatility (for example. It provides a speciﬁc relationship between the volatilities of options with diﬀerent strike prices and times to maturity at a particular 4 . the model is a particular case of the generalized sticky delta model just considered. This is described in Natenberg (1994) and Hull (2006).2. which is assumed to be a deterministic function of the moneyness variable. for the purposes of calculating partial derivatives the option price c is considered to be a function of S. 2. but when measured relative to the at-the-money volatility.3 The Square Root of Time Rule Another rule that is sometimes used by traders is what we will refer to as the “square root of time rule”. If the at-the-money volatility is stochastic. the volatility of the volatility) are functions only of K/S and T − t. The delta of a European option in a stochastic volatility model is ∆= ∂c ∂c ∂σT K + . but independent of S. The ﬁrst term in this expression is the delta calculated using BlackScholes with the volatility parameter set equal to implied volatility. A more general version of the sticky delta rule is where the process for VT K depends on K. We will refer to this model as the relative sticky delta model. and t only through its dependence on K/S and T − t. K/S. Many traders argue that a better measure of moneyness than K/S is K/F where F is the forward value of S for a contract maturing at time T . For equities σT K is a declining function of K and so the Black-Scholes delta understates the true delta. This form of the sticky delta rule allows the overall level of volatility to change through time and the shape of the volatility term structure to change. rather than the volatility itself. It follows that. Here it is the excess of the volatility over the at-the-money volatility. the volatility is dependent only on K/S and T − t. and t. t) − σT F (S. ∂c/∂σT K is positive. t) is a function only of K/F and T − t. For an asset with a U-shaped volatility smile Black-Scholes understates delta for low strike prices and overstates it for high strike prices. K/F .

S (2) where r(t) is the risk-free rate. but independent of S. 5 . One version of this rule is σT K (S. Stein and Stein (1991) and Heston (1993) as special cases. t) − σT ∗ F ∗ (S. t) − σT F (S. 3 The Dynamics of the Implied Volatility We suppose that the risk-neutral process followed by the price of an asset. t) = σT ∗ K ∗ (S. and 2. where Φ is a function. t) − σT F (S.time. q(t) is the yield provided by the asset. If we know 1. We can compute the volatility smile at time T from that at time T ∗ using the result that σT K (S. We suppose that r(t) and q(t) are deterministic functions of time and that σ follows a diﬀusion process. Suppose that F ∗ is the forward price of the asset for a contract maturing at T ∗ . At-the-money volatilities for other maturities. t) = Φ ln(K/F ) √ T −t . the stationary square root of time model is a particular case of the relative sticky strike model and the stochastic square root of time model is a particular case of the generalized sticky strike model. If the at-the-money volatility is assumed to be stochastic. (1) We will refer to this version of the rule where Φ does not change through time as the stationary square root of time model and the version of the rule where the form of the function Φ changes stochastically as the stochastic square root of time model. where ln(K ∗ /F ∗ ) ln(K/F ) √ = √ ∗ . is dS = [r(t) − q(t)] dt + σ dz. An alternative that we will use is σT K (S. t) ln(K/F ) √ T −t . we can compute the complete volatility surface. T −t T −t √ K =F ∗ ∗ or K F (T ∗ −t)/(T −t) . The volatility smile for options that mature at one particular time T ∗ . t). Our model includes the IVF model and stochastic volatility models such as Hull and White (1987). The square root of time model (whether stationary or stochastic) simpliﬁes the speciﬁcation of the volatility surface. S. t) =Φ σT F (S. and z is a Wiener process. σ is the asset’s volatility.

Jarrow.2 Deﬁne c(S. see Durrleman (2004). From the deﬁnition of implied 1 Note that the result is not necessarily true if we deﬁne an at-the-money option as an option where the strike price 1 ln(F/K) T −t equals the asset price. For convenience we deﬁne VT K (t. This is true even when the processes in equation (3) deﬁning the volatility surface are Markov. In the most general form of the model the parameters αT K and θT Ki (1 ≤ i ≤ N ) may depend on past and present values of S. limT →t σT F (t. The initial volatility surface is σT K (0. S) is not the same as limT →t σT S (t. and time. 2 There is an analogy here to the Heath. S) = [σT K (t. S) as the implied volatility at time t of an option with strike price K and maturity T when the asset price is S. When each forward rate follows a Markov process the instantaneous short rate does not in general do so. when σT K (t. S) = σ(t). T ) as the price of a European call option with strike price K and maturity T when the asset price S follows the process in equations (2) and (3). S). zN are Wiener processes driving the volatility surface. This means that the correlation between zi and zj is ρi ρj . VT K . S) = a + b with a and b constants. In general the process for σ is non-Markov. S)]2 as the implied variance of the option. For example. For more on the relationship between the implied and spot volatilities. K. 6 .1 Formally: T →t lim σT F (t. There is clearly a relationship between the instantaneous volatility σ(t) and the volatility surface σT K (t.Most stochastic volatility models specify the process for σ directly. We instead specify the processes for all implied volatilities. t. For this purpose an at-the-money option is deﬁned as an option where the strike price equals the forward asset price. (3) where z1 . Without loss of generality we assume that the zi are correlated with each other only through their correlation with z. As in the previous section we deﬁne σT K (t. · · · . We deﬁne ρi as the correlation between z and zi . S). The parameter θT Ki measures the sensitivity of VT K to the Wiener process zi . driving the asset price in equation (2). S0 ) where S0 is the initial asset price. The family of processes in equation (3) deﬁnes the multi-factor dynamics of the volatility surface. Suppose that the process followed by VT K in a risk-neutral world is dVT K = αT K dt + VT K i=1 N θT Ki dzi . The zi may be correlated with the Wiener process. (4) where F is the forward price of the asset at time t for a contract maturing at time T . z. past and present values of VT K . This volatility surface can be estimated from the current (t = 0) prices of European call or put options and is assumed to be known. The appendix shows that the instantaneous volatility is the limit of the implied volatility of an at-the-money option as its time to maturity approaches zero. and Morton (1992) model.

VT K . t. the correlation between z and zi . is a function of past and present values of S. K. c satisﬁes the Black-Scholes (1973) and Merton (1973) diﬀerential equation. For there to be no arbitrage the process followed by c must provide an expected return of r in a risk-neutral world. It follows that ∂c ∂c 1 ∂2c ∂c 1 2 ∂2c + (r − q)S + σ 2 S 2 2 + αT K + VT K 2 ∂t ∂S 2 ∂S ∂VT K 2 ∂VT K +SVT K σ ∂2c ∂S∂VT K N N (θT Ki )2 + i=1 i=j θT Ki θT Kj ρi ρj θT Ki ρi = rc. i=1 (5) 7 . VT K (T − t). T ) = e where d1 d2 = = ln(S/K) + ln(S/K) + T [r(τ ) t T [r(τ ) t − T t q(τ ) dτ SN (d1 ) − e − T t r(τ ) dτ KN (d2 ). i=1 In the most general form of the model ρi . i=1 When VT K is held constant.volatility and the results in Black and Scholes (1973) and Merton (1973) it follows that: c(S. ∂t ∂S 2 ∂S It follows that 1 2 ∂2c 2 ∂c 1 2 ∂2c S (σ − VT K ) + αT K + VT K 2 2 2 ∂S ∂VT K 2 ∂VT K +SVT K σ or αT K = − + 1 ∂2c ∂2c 2 S 2 2 (σ 2 − VT K ) + 2 V 2∂c/∂VT K ∂S ∂VT K T K ∂ c 2 2 V ∂VT K T K 2 N N (θT Ki )2 + i=1 i=j θT Ki θT Kj ρi ρj ∂2c ∂S∂VT K N θT Ki ρi = 0. As a result ∂c ∂c 1 ∂2c + (r − q)S = rc − VT K S 2 2 . i=1 (θT Ki )2 i=1 2 N θT Ki θT Kj ρi ρj + 2SVT K σ i=j ∂ c ∂S∂VT K θT Ki ρi . past and present values of VT K and time. − q(τ )]dτ − q(τ )]dτ VT K (T − t) VT K (T − t) + − 1 2 1 2 VT K (T − t). Using Ito’s lemma equations (2) and (3) imply that the drift of c in a risk-neutral world is: ∂c ∂c 1 ∂2c ∂c 1 2 ∂2c + (r − q)S + σ 2 S 2 2 + αT K + VT K 2 ∂t ∂S 2 ∂S ∂VT K 2 ∂VT K +SVT K σ ∂2c ∂S∂VT K N N (θT Ki )2 + i=1 i=j θT Ki θT Kj ρi ρj θT Ki ρi .

The partial derivatives of c with respect to S and VT K are the same as those for the Black–Scholes model: ∂c ∂S ∂2c ∂S 2 ∂c ∂VT K ∂2c 2 ∂VT K ∂2c ∂S∂VT K = e = − T t q(τ ) dτ − T t N (d1 ). Substituting these relationships into equation (5) and simplifying we obtain N 1 VT K (d1 d2 − 1) θT Ki θT Kj ρi ρj αT K = (θT Ki )2 + (VT K − σ 2 ) − T −t 4 i=1 i=j +σd2 VT K T −t N θT Ki ρi . 2 VT K T √ q(τ ) dτ t φ(d1 ) T − t (d1 d2 − 1). is a deterministic function of time. The variable VT K is then also a function of time and VT K = Diﬀerentiating with respect to time we get 1 dVT K = [VT K − σ(t)2 ]. it follows that ˆ dVT K = ˆT K (d1 d2 − 1) N V (θT Ki )2 + −σ 2 − 4 i=1 8 θT Ki θT Kj ρi ρj i=j 1 T −t T t σ(τ )2 dτ . 2VT K where φ is the density function of the standard normal distribution: φ(x) = 1 x2 exp − 2π 2 . = = − √ φ(d1 ) T − t √ . −∞ < x < ∞. ˆ The analysis can be simpliﬁed slightly by considering the variable VT K instead of V where ˆ VT K = (T − t)VT K . dt T −t which is the ﬁrst term. It can be understood by considering the situation where the instantaneous variance. 3/2 4VT K q(τ ) dτ T t = − e − q(τ ) dτ φ(d1 )d2 . σ 2 . q(τ ) dτ φ(d1 )e S Se Se − VT K (T − t) T t . i=1 (6) Equation (6) provides an expression for the risk-neutral drift of an implied variance in terms of its volatility. The ﬁrst term on the right hand side is the drift arising from the diﬀerence between the implied variance and the instantaneous variance. Because ˆ dVT K = −VT K dt + (T − t)dVT K .

In this case each term in the drift of VT K − VT F is negative. Our ﬁnding is consistent with a result in Hull and White (1987). They demonstrate that when d1 d2 > 1 a stochastic volatility tends to increase an option’s price and therefore its implied volatility. i=1 Suppose that d1 d2 − 1 > 0 and VT K < VT F . 4. Determining the nature of these functions is not easy. the ﬁrst term in equation (6) is mean ﬂeeing. This negative mean reversion becomes more pronounced as the option approaches maturity. As a result VT K − VT F tends to get progressively more negative and the model is unstable with negative values of VT K being possible. the price of a European option is the Black–Scholes price integrated over the distribution of the average variance. (θT Ki )2 − i=1 VT F [1 + VT F (T − t)/4] 4 N (θT F i )2 .1 The Zero Correlation Case Consider ﬁrst the case where all the ρi are zero so that equation (6) reduces to αT K = 1 VT K (d1 d2 − 1) (VT K − σ 2 ) − T −t 4 N (θT Ki )2 i=1 As before.N N +σd2 ˆ VT K i=1 θT Ki ρi ˆ dt + VT K i=1 θT Ki dzi . These authors show that when the instantaneous volatility is independent of the asset price. It cannot be consistent with a upward or downward sloping smile because in these cases VT K < VT F for either very high or very low values of K. However. that is. In the general case where VT K is nondeterministic. 9 . It follows that the case where the ρi are zero can be consistent with the U-shaped volatility smile. In this section we examine the implications of this no-arbitrage condition. it is possible to make some general statements about the volatility smiles that are consistent with stable models. The condition d1 d2 > 1 is satisﬁed for very large and very small values of K. 4 Implications of the No-Arbitrage Condition Equation (6) provides a no-arbitrage condition for the drift of the implied variance as a function of its volatility. For a viable model the θT Ki must be complex functions that in some way oﬀset this negative mean reversion. it provides negative mean reversion. we deﬁne F as the forward value of the asset for a contract maturing at time T so that T F = Se The drift of VT K − VT F is 1 VT K (d1 d2 − 1) (VT K − VT F ) − T −t 4 N t [r(τ )−q(τ )]dτ . We deduce from this that VT K must be greater than VT F when d1 d2 > 1.

4. The drift of VT K − VT F is 1 VT K (d1 d2 − 1) (VT K − VT F ) − T −t 4 +σd2 VT K T −t N N N VT F (1 + VT F (T − t)/4) (θT Ki ) − 4 i=1 2 N (θT F i )2 i=1 θT Ki ρi + σ i=1 VT F 2 θ T F i ρi − i=1 N VT K (d1 d2 − 1) 4 θT F i θT F j ρi ρj θT Ki θT Kj ρi ρj i=j VT F (1 + VT F (T − t)/4) − 4 i=1 When K > F . The instantaneous covariance of the asset price and its variance is N σ i=1 θT Ki ρi . Bates (1996)). Our analysis shows that this is consistent with the empirical result that the correlation between implied volatilities and the exchange rate is close to zero (see. Because d2 < 0 we must have N θT Ki ρi < 0 i=1 (7) when K is very large.A U-shaped volatility smile is commonly observed for options on a foreign currency. VT K − VT F is negative and the eﬀect of the ﬁrst three terms is to provide a negative drift. We now consider the types of correlations that are consistent with volatility skews. This is consistent with the result we have just presented. Consider ﬁrst the situation where the volatility is a declining function of the strike price. (This is the case for options on some commodity futures. Equities provide an example of a situation where there is a volatility skew of the sort we are considering. The variance rate VT K is greater than VT F when K is very small and less than VT F when K is very large. Consider next the situation where volatility is an increasing function of the strike price.2 Volatility Skews We have shown that a zero correlation between volatility and asset price can only be consistent with a U-shaped volatility smile.) A similar argument to that just given shows that the no-arbitrage relationship implies that the volatility of the underlying must be positively correlated with the value of the underlying. Because σ > 0 it follows that when K is large the asset price must be negatively correlated with its variance. For a stable model we require the last four terms to give a positive drift. As has been well documented by authors such as Christie (1982). the volatility of an equity price tends to be negatively correlated with the equity price. for example. 10 . As K increases and we approach option maturity the ﬁrst of the last four terms dominates the other three.

and K is therefore the model where the instantaneous volatility (σ) is a function only of time. or σ2 = − d [(T − t)VT K ] . equation (8) shows that VT K = σ 2 and we get the Black-Scholes constant-volatility model. It is diﬃcult to see how this can be so without the stochastic terms in the processes for the VT K having a form that quickly destroys the inverted U-shaped pattern. Case 1: VT K is a deterministic function only of t. T .3 Inverted U-Shaped Smiles An inverted U-shaped volatility smile is much less common than a U-shaped volatility smile or a volatility skew. T −t 1 VT K − σ 2 . An analysis similar to that in Section 4. In the particular case where VT K depends only on T and K. Also from equation (6) αT K = so that dVT K which can be written as (T − t)dVT K − VT K dt = −σ 2 dt. The only model that is consistent with VT K being a function only of t.2 shows why this is so. This is Merton’s (1973) model. T −t This shows that σ is a deterministic function of time. For a stable model N θT Ki ρi i=1 must be less than zero when K is large and greater than zero when K is small. This will enable us to reach conclusions about whether the rules of thumb in Section 2 can satisfy the no-arbitrage condition in Section 3. 11 . 5 Special Cases In this section we consider a number of special cases of the model developed in Section 3.4. T . dt (8) = 1 VT K − σ 2 dt. and K In this situation θT Ki = 0 for all i ≥ 1 so that from equation (3) dVT K = αT K dt.

T )]2 = 2 ∂c/∂T + q(T )c + K[r(T ) − q(T )]∂c/∂K . F. and K/S. t. In this situation we can write VT K = G(T. T →t This shows that the instantaneous volatility is a deterministic function of F and t. From equation (4). F. 6 Theoretical Basis for Rules of Thumb We are now in a position to consider whether the rules of thumb considered in Section 2 can be consistent with the no-arbitrage condition in equation (6). K F . t. σ is given by σ 2 = lim G(T. This is a deterministic function of time.Case 2: VT K is independent of the asset price. The only model that is consistent with VT K being independent of S is therefore Merton’s (1973) model where the instantaneous volatility is a function only of time. It follows that the model reduces to the implied volatility function model. Dupire (1994) and Andersen and Brotherton–Ratcliﬀe (1998) show that [σ(K. Equivalently it is a deterministic function of underlying asset price. the model reduces to Merton’s (1973) deterministic volatility model. T . 1) . Case 2 therefore reduces to Case 1. t. Writing σ as σ(S. Similar results to ours on the sticky strike and sticky delta model are provided by Balland (2002). where G is a deterministic function and as before F is the forward price of S. Because d1 and d2 depend on S we must have θT Ki equal zero for all i. t. It follows that. i=1 Both αT K and the θT Ki must be independent of S. S. S . Case 4: VT K is a deterministic function of t. t). In this situation ρi is zero and equation (6) becomes αT K = 1 VT K (d1 d2 − 1) (VT K − σ 2 ) − T −t 4 N (θT Ki )2 . S and K. F ). T . the instantaneous volatility. is given by σ 2 = G (t. K 2 (∂ 2 c/∂K 2 ) where c is here regarded as a function of S. In this situation VT K = G T. 12 . K). Case 3: VT K is a deterministic function of t. (9) where G is a deterministic function. K and T for the purposes of taking partial derivatives. From equation (4) the spot instantaneous volatility. σ. yet again. and t.

In the basic sticky strike rule of Section 2. T and t only through the variable K/S and T − t. As shown in Case 2 of the previous section. Derman’s research looks at exchange-traded options on the S&P 500 during the period September 1997 to October 1998 and considers the sticky strike and sticky delta rules as well as a more complicated rule based on the IVF model. We use monthly volatility surfaces from the over-the-counter market for 47 months (June 1998 to April 2002). for a mean-reverting stochastic volatility model. He ﬁnds subperiods during which each of the rules appears to explain the data best. S. S. In the generalized sticky strike model VT K is independent of S. The analysis in Case 1 of the previous section shows that the only version of this model that is internally consistent is the model where the volatilities of all options are the same and constant. This is Merton’s (1973) model. When the instantaneous volatility of the asset price is a function only of time.2. If a trader prices options using diﬀerent implied volatilities and the volatilities are independent of the asset price. but possibly dependent on other stochastic variables. 7 Tests of the Rules of Thumb As pointed out by Derman (1999) apocryphal rules of thumb for describing the behavior of volatility smiles and skews may not be conﬁrmed by the data. The generalized sticky delta model. This is because equation (6) shows that if each θT Ki depends on K. Consider next the sticky delta rule. all European options with the same maturity have the same implied volatility. The data for June 1998 is shown in Table 1. The relative sticky delta rule and the square root of time rule can be at least approximately consistent with the no-arbitrage condition. the only version of this model that is internally consistent is the model where the instantaneous volatility of the asset price is a function only of time. there must be arbitrage opportunities. the implied volatility approximately satisﬁes the square root of time rule when the reversion coeﬃcient is large (Andersson. and t only through a dependence on K/S and T − t. T . the implied volatility is a deterministic function of K/S and T − t. For example. where VT K is stochastic and depends on K. can be consistent with the no-arbitrage condition. Case 3 in the previous section shows that the only version of this model that is internally consistent is Merton’s (1973) model where the instantaneous volatility of the asset price is a function only of time. the same is true of αT K . as deﬁned in Section 2. We conclude that all versions of the sticky strike rule are inconsistent with any type of volatility smile or volatility skew. As in the case of the sticky strike model it is inconsistent with any type of volatility smile or volatility skew.1 the variance VT K is a function only of K and T . This is the original Black and Scholes (1973) model. 2003). Derman’s results are based on relatively short maturity S&P 500 option data. In the basic sticky delta model. Six maturities are considered ranging from six months to 13 .

but has an R2 of only 27%.] The data was supplied to us by Totem Market Valuations Limited with the kind permission of a selection of Totem’s major bank clients. We obtained a root mean square error (RMSE) of 0. implied volatilities for the S&P 500 exhibit a volatility skew with σT K (t. S) being a decreasing function of K. The version of the rule we consider is the relative sticky delta model where σT K (t.3 Totem collects implied volatility data in the form shown in Table 1 from a large number of dealers each month. As a test of the model’s viability out-of-sample.ﬁve years. Our data comes from the over-the-counter market. where the ai are constant and (10) is a normally distributed error term. Our data consist of the estimated mid-market volatilities returned to dealers. S) − σT F (t. As shown in Table 2. The data produced by Totem are considered by market participants to be more accurate than either the volatility surfaces produced by brokers or those produced by any one individual bank. Totem uses the data in conjunction with appropriate averaging procedures to produce an estimate of the mid-market implied volatility for each cell of the table and returns these estimates to the dealers. [Table 1 about here. These dealers are market makers in the over-the-counter market. Consider ﬁrst the sticky strike rule. A total of 42 points on the volatility surface are therefore provided each month and the total number of volatilities in our data set is 42 × 47 = 1974.] We now move on to test the sticky delta rule. Fleming and Whaley (1998). the model is supported by the data. This enables dealers to check whether their valuations are in line with the market. ranging from 0.0525. representing a quite sizable error.8 to 1. we ﬁtted the model using the ﬁrst two years of data and then tested it for the surfaces for the remaining 23 months. The model we test 3 Totem was acquired by Mark-it Partners in May 2004 14 . S) is a function of K/F and T − t. The terms on the right hand side of this equation can be thought of as the ﬁrst few terms in a Taylor series expansion of a general function of K and T − t. We tested this version of the rule using σT K (t. Seven values of K/S are considered.2. S) = a0 + a1 K + a2 K 2 + a3 (T − t) + a4 (T − t)2 + a5 K(T − t) + . Here our analysis is similar in spirit to the “ad-hoc model” of Dumas. [Table 2 about here. A more plausible version of the rule is that the implied volatility is a function only of K and T − t. Sticky strike rules where the implied volatility is a function only of K and T may be plausible in the exchange-traded market where the exchange deﬁnes a handful of options that trade and traders anchor on the volatility they ﬁrst use for any one of these options. It is diﬃcult to see how this form of the sticky strike rule can apply in that market because there is continual trading in options with many diﬀerent strike prices and times to maturity. As illustrated in Table 1.

62%. [Table 3 about here.] Here we repeat our out-of-sample exercise.] 15 . We consider more restricted forms of the model where progressively c4 and c3 are set to zero. this statistic must be greater than 1.0073 is obtained. Using a similar Taylor Series expansion to the other models we test σT K (t. then the observed ratio of the two models’ squared errors should be distributed F (N1 . This is somewhat better than the R2 for the sticky delta model in (11) even though the model in equation (12) involves two parameters and the the one in equation (11) involves six parameters.71 indicating that we can overwhelmingly reject the hypothesis that the models have equal explanatory power. S) − σT F (t. [Table 4 about here. S) = b0 + b1 ln K F + b2 ln K F K F 2 + b3 (T − t) (11) +b4 (T − t)2 + b5 ln where the bi are constants and (T − t) + is a normally distributed error term. representing a fairly close ﬁt.93%. In Table 3 it should be noted that the estimates obtained from using only the ﬁrst half of the sample are quite close to those obtained using the whole data set.89 for signiﬁcance at the 1% level. The relative sticky delta model in equation (11) can explain the volatility surfaces in our data much better than the sticky strike model in equation (10).is σT K (t. If two models have equal explanatory power. In this case the R2 is much higher at 94. The results are shown in Table 3. S) = ln(K/F ) + c2 c1 √ T −t +c3 where c1 . S) is a known √ function of ln(K/S)/ T − t. c2 . S) − σT F (t. When comparing the sticky strike and relative sticky delta models using a two tailed test. The model is analogous to the one used to test the sticky strike rule. The terms on the right hand side of this equation can be thought of as the ﬁrst few terms in a Taylor series expansion of a general function of ln(K/F ) and T − t. The value of the statistic is 32. The results for this model are shown in Table 4. In this case the R2 is 95. N2 ) where N1 and N2 are the number of degrees of freedom in the two models. c3 and c4 are constants and [ln(K/F )] √ T −t ln(K/F ) √ T −t 3 2 + c4 ln(K/F ) √ T −t 4 + (12) is a normally distributed error term. The third model we test is the version of the stationary square root of time rule where the function Φ in equation (1) does not change through time so that σT K (t. in this case with more satisfactory results: a RMSE of 0.12 or less than 0. to render a more parsimonious model for the volatility smile. S) − σT F (t.

in the following analysis. the coeﬃcients c1 .373 respectively since this model eﬀectively has many more parameters than its stationary counterpart.Testing the out of sample performance of the three possible forms of the square-root model gives RMSEs of 0.11 indicating that the stochastic square root of time model does provide a statistically signiﬁcantly better ﬁt to the data that the stationary square root of time model at the 1% level. In almost all cases.373. in the monthly tests of the square root of time rule is always signiﬁcantly diﬀerent from zero with a very high level of conﬁdence.6 Figure 1 shows the level of c1 . [Figure 1 about here. 0. the ratio is 1. three term and two term forms of the model. c2 (t) and c3 (t) are stochastic. When the model in equation (13) is compared to the model in equation (12) the ratio of sums of squared errors statistic is 3. there is reasonably high correlation between c2 and c3 and the index. 6 The approximate linearity of the volatility skew for S&P 500 options has been mentioned by a number of researchers including Derman (1999). 16 . c1 . In the stochastic square root of time rule.157 and 1. S) − σT F (t.5 The coeﬃcient. We can calculate a ratio of sums of squared errors to compare the stationary square root of time model in equation (12) to the relative sticky delta model in equation (11). c2 and c3 are all signiﬁcantly diﬀerent from zero at the 5% level.0069. We note.0069 and 0. S) and ln(K/F ) √ T −t changes stochastically through time. when working with the square-root rule.4 When a two tailed test is used it is possible to reject the hypothesis that the two models have equal explanatory power at the 1% level. this result should not be surprising. including a cubic term leads to improvement in performance while the addition of a quartic term has negligible eﬀect. c2 and c3 plotted against the S&P 500 for the period covered by our data. We note that there is fairly weak correlation between c1 and the level of the S&P 500 index. we restrict our attention to the cubic version. the most economically signiﬁcant of the three terms is the linear term. In this case. We conclude that equation (12) is an improvement over equation (11). that in spite of this statistical signiﬁcance. the functional relationship between σT K (t.] 4 The 5 Although ratios for comparing the quadratic model and quartic model to (11) are 1. A model capturing this is ln(K/F ) σT K (t. Accordingly. S) − σT F (t. However. We conclude that for both in-sample and out of sample performance.0073 for the four term. S) = c1 (t) √ + c2 (t) T −t ln(K/F ) √ T −t 2 + c3 (t) ln(K/F ) √ T −t 3 + (13) where c1 (t). To provide a test of this model we ﬁtted the model in equation (12) to the data on a month by month basis. however. during the earlier part of our sample.

[Figure 2 about here. We consider both the stationary power rule model: σT K (t. note that in the later part of the data.44. we obtain 0.04. however. We note that the estimates of d0 (the power of time to maturity which best explains volatility changes) are approximately equal to 0.5).] Testing for signiﬁcance between the performance of this power rule and the square-root rule. Performing an out of sample calculation of RMSE for the stationary power rule.5. S) − σT F (t. S) = d1 (t) +d3 (t) ln(K/F ) ln(K/F ) + d2 (t) (T − t)d0 (t) (T − t)d0 (t) ln(K/F ) (T − t)d0 (t) 3 2 ln(K/F ) ln(K/F ) + d2 d0 (T − t) (T − t)d0 ln(K/F ) (T − t)d0 3 2 + d4 ln(K/F ) (T − t)d0 4 + (14) + d4 (t) ln(K/F ) (T − t)d0 (t) 4 + (15) Estimating this model gives the results in Table 5.To provide an alternative benchmark for the square root of time rule. [Table 6 about here. All our model comparison results are summarized in Table 6. All three seem to display reasonable parameter stability over time. we now consider the possibility that volatility is a function of some other power of time to maturity. we obtain a ratio of squared errors of 1.5. There is some evidence that the rule could be marginally improved upon by considering powers other than 0. the ratio becomes 5. and is statistically signiﬁcantly diﬀerent from 0. When the stochastic variants of each model are compared. insuﬃcient to reject the hypothesis of equal performance. S) − σT F (t.] We conclude that there is reasonably strong support for the square-root rule as a parsimonious parametrisation of the volatility surface.] For the model in equation (15) there is little or no relationship between the level of the index and the parameters of the rule. S) = d1 +d3 and the stochastic power rule model: σT K (t. 17 . [Table 5 about here.0070 for each model. We do. there does seem to be evidence of the power rule being very close to the square root rule (d0 = 0.39 showing that a considerable improvement in ﬁt is obtained when the power relationship is allowed to vary with time.

F/K ∆t + ˆ Here.i ρi T −t V(T −t). Deﬁne α as the drift of ln V(T −t).t (T −t). the term (T −t).F/K. 18 .F/K. a constant drift for each volatility is assumed. 8.F/K. We focus on implied volatility surfaces where moneyness is described as F/K rather than S/K. as originally reported.F/K V(T −t).F/K.F/K. observed at time t. We create this data.F/K ˆ 1 1 σ2 (1 − ) + σd2 θ(T −t). where α(T −t).t as ∆ ln(V(T −t).i + − ρi ρj θ(T −t). (17) (which has expectation zero) represents the combined eﬀects of the factors on the movement of the volatility of an option with time to maturity T − t and relative moneyness F/K. Foseca and Durrleman (2002). The traditional principal components analysis does not use any information about the drift of volatilities.1 Taking Account of the No-Arbitrage Drift We now carry out a more sophisticated analysis that takes account of the no-arbitrage drift derived in Section 3.j .t .i θ(T −t). Our study is diﬀerent from these in that it focuses on over-the-counter options rather than exchange-traded options. We assume that the market price of volatility risk is zero so that volatilities have the same drift in the real world as in the risk-neutral world.i dzi .t and θ(T −t). using interpolation.8 Estimation of Volatility Factors We also used the data described in the previous section to estimate the factors underlying the model in equation (3).t is N d ln V(T −t). From our results in Section 3 we know that in a risk-neutral world the drifts of implied volatilities depend on the factors.F/K. The results of this are shown in Table 7 and Figure 3.i . Hodges and Clewlow (2000).F/K.t = αdt + ˆ i=1 θ(T −t).t ) = α(T −t). It also uses options with much longer maturities. Consistent with this assumption we write VT K (t) and θT Ki (t) as V(T −t). We also assume a form of the generalized sticky delta rule where each θT Ki is a function K/F and T − t. 4 i=1 = j=i N (16) We can write the discretised process for ln V(T −t).F/K. In estimating the variance-covariance matrix.F/K.t . This is a similar analysis to that undertaken by Kamal and Derman (1997). Cont and Fonseca (2002) and Cont. The conventional analysis involves the use of principal components analysis applied to the variance-covariance matrix of changes in implied volatilities.F/K.F/K. and the attendant changes in log volatility. Skiadopoulos.F/K.F/K.F/K (T − t) i=1 N (d1 d2 + 1) 2 θ(T −t).F/K. Applying Ito’s lemma to equation (6) the process for ˆ ln V(T −t).

The σ captures further eﬀects over and above those given by the ﬁnite set of factors considered. we can thus use an optimization routine to maximize (19) by varying θ(T −t). 8 It for example Greene (1997). which (for a large sample) could involve quite large matrices.(F/K2 ). this results in N × M + N + 1 parameters.(F/K2 ).k θ(T2 −t). using (17).l ρk ρl + l=k σ2 + 0 if T1 = T2 and K1 = K2 . ρ s. (18) The analysis assumes no correlation between any volatility movements at diﬀerent times.The correlations between the ’s arise from a) the possibility that the same factor aﬀects more than one option and b) the correlation between the factors arising from their correlation with the asset price.F/K. describing a volatility matrix of M volatilities. 19 .8 To estimate the factors. We carry out a maximum likelihood analysis to determine the values of the θ’s. For an N factor model. As these parameters are changed. one of which (factor zero) is driven by the same Brownian Motion as in the underlying stock price. is worth observing that our model’s intertemporally independent residuals allows us to consider the likelihood −1 ti ) ti (Σti ) ratio as the sum of individual time steps’ log likelihood ratios.k θ(T2 −t). The covariance between two ’s is given by N E( (T1 −t). page 89.j (where T − t and F/K are taken to be the varying levels of maturity and moneyness we observe in our volatility matrices). We used MATLAB’s fmincon optimization routine (constraining the ρ’s to lie between -1 and 1) and found that the routine is eﬃcient for solving the problem. To ease the comparison with the results from Figure 3 and Table 7. This considerably speeds up the calculation of both the inner product term ( and ln det(Σti ). even for this large number of parameters. we have converted our factors from a collection of factors each correlated with stock price movements (and therefore each other) to a collection of ﬁve orthonormal factors. α changes and there is a diﬀerent covariance matrix for the errors. ρj and σ 2 . Fitting a four factor model (N = 4) we ﬁnd the factors given in Figure 4 and Table 8. Since we know that these ’s are jointly normally distributed.(F/K2 ).(F/K1 ).k θ(T1 −t).t (T2 −t). ˆ For any set of parameters we back out an for each observation in our data set.t ) = ∆t k=1 θ(T1 −t). otherwise. it is a standard result in econometrics7 that the log likelihood function for this set of parameters is given by: M L= i=1 − −1 ti ti (Σti ) − ln det(Σti ) (19) where ti is the vector of ’s for time ti and Σti is the variance-covariance matrix described by (18) at time ti .(F/K1 ). The normalization results in the factors having to be 7 See.(F/K1 ). and σ that best describe the observed changes in option implied volatilities.

Factor four is less easy to interpret.6% of the variation in implied volatilities.multiplied by a scaling factor. Factors two and three show evidence of a square root eﬀect. but both showing evidence of increasing sensitivity for higher strike options.] [Table 7 about here. with both the last two factors making a signiﬁcant contribution. while the second factor shows some evidence of a square-root behaviour in the maturity dimension. Our model is exactly consistent with the initial volatility surface. long maturity options. Contrasting the results of our consistent factor analysis with the simpler approach originally followed. This factor shows evidence of hyperbolic behaviour in the time dimension. rendering out of the money puts more valuable. and Dupire (1994). With four factors. The most important of all the factors is the ﬁrst one. we can explain 99.] [Table 8 about here. and can therefore be seen as representing moves up and down of the existing smile. Derman and Kani (1994). and works to oﬀset the explosive behaviour we noted in section 3. seeming to be a factor which aﬀects low moneyness. Focusing on factor zero (the factor which is perfectly correlated with stock prices) we see that an increase in stock prices leads to an increase in the slope of implied volatilities.] 9 Summary It is a common practice in the over-the-counter markets to quote option prices using their Black– Scholes implied volatilities. In this paper we have developed a model of the evolution of implied volatilities and produced a no-arbitrage condition that must be satisﬁed by the volatilities. and tapers oﬀ for longer maturity options. which allows the familiar analysis of considering the proportion of the covariance matrix of the surface which is explained by each factor. but more general than the IVF model of Rubinstein (1994). The principal components approach gives two dominant factors which explain 98. this would result in a decrease in stock prices causing the skew to steepen. It is roughly level in the moneyness dimension. The no-arbitrage condition leads to the conclusions that a) when the volatility is independent of the asset price there must be a U-shaped volatility smile and b) when the implied volatility is a decreasing (increasing) function 20 . The eﬀect is most pronounced for short maturity options. In the case of the typical (negative) skew observed. we note that we obtain a richer factor structure. having roughly a square-root shape in the time dimension.] [Figure 4 about here. [Figure 3 about here.8% of the variation. The ﬁrst factor is similar to that uncovered by the drift consistent approach.

t. Some versions of these rules are clearly inconsistent with the no-arbitrage condition. where in this case d1 = −d2 = The call price can be written c(S(t).F (t) . x or √ c(S. t. For some x ¯ φ(x) dx = 2d1 φ(¯) x −d1 and the call price is therefore given by c(S(t). t + ∆t) = e−q∆t S(t)φ(¯)σt+∆t. S(t)). t.F (t) (t. t. x 21 (20) . Our empirical tests of the rules of thumb using 47 months of volatility surfaces for the S&P 500 show that the relative sticky delta model (where the excess of the implied volatility of an option over the corresponding at-the-money volatility is a function of moneyness) outperforms the sticky strike rule.F (t) . t + ∆t) = e−q∆t S(t)[N (d1 ) − N (d2 )]. F (t). for other versions of the rules the no-arbitrage condition can in principle be satisﬁed. 2 φ(x) dx. Vt+∆t.F (t) (t. A number of rules of thumb have been proposed for how traders manage the volatility surface.of the asset price there must be a negative (positive) correlation between the volatility and the asset price. we assume that r and q are constants. sticky delta. Vt+∆t. and square root of time rules.F (t) . Vt+∆t. the stochastic square root of time model outperforms the relative sticky delta rule. −d1 x2 1 φ(x) = √ exp − 2 2π d1 . F (t). We outline how these no-arbitrage conditions can be incorporated into a factor analysis. Our empirical implementation of this analysis suggests that more factors may in fact govern volatility movements than would be thought otherwise. F (t). Deﬁne F (τ ) as the forward price at time τ for a contract maturing at time t + ∆t so that F (t) = S(t)e(r−q)∆t . A Proof of Equation (4) For simplicity of notation. resulting in no-arbitrage consistent factors. These are the sticky strike. t + ∆t) = 2e−q∆t S(t)d1 φ(¯). t + ∆t) = e−q∆t S(t) where d1 √ ∆t σt+∆t. Also. F (t). The price at time t of a call option with strike price F (t) and maturity t + ∆t is given by c(S(t).F (t) . S(t)) ∆t. Vt+∆t.

and M. ∆t→0 ∆t→0 ∆t ∆t where E denotes expectations under the risk-neutral measure. φ(¯) tends to 1/ 2π so that x lim σt+∆t. ∆t→0 2π ∆t it follows that ∆t→0 1 σ(t)F (t) lim √ er∆t c(S(t).F (t) (t. K.M. Department of Mathematics.F (t) . t. t + ∆t) = √ . 9.U. and 1 1 lim √ E[z(t + δt) − z(t)]+ = √ . ∆t→0 This is the required result.” Quantitative Finance. S Using Ito’s lemma dF = σF dz. F (t).S. (2002) “Deterministic implied volatility models. (1996). References Andersson. (2003) “Stochastic volatility. Vt+∆t. 69–107 22 . Project Report. F (t). 81.D. Vt+∆t.F (t) . 1. D.” U. S(t)) = x ∆t→0 √ As ∆t tends to zero. Scholes (1973). Uppsala University Balland. It follows that 1 1 lim √ E[F (t + ∆t) − F (t)]+ = lim √ σ(t)F (t)E[z(t + ∆t) − z(t)]+ . S. 2. P.” Review of Financial Studies. 2π ∆t σ(t)F (t) √ . “Jumps and Stochastic Volatility: Exchange Rate Process Implicit in Deutsche Mark Options. 31–44 Black.The process followed by S is dS = (r − q) dt + σ dz. When terms of order higher that (∆t)1/2 are ignored F (t + ∆t) − F (t) = σ(t)F (t)[z(t + ∆t) − z(t)]. 637–659 Bates. t + ∆t).” Journal of Political Economy. t.F (t) (t. “The pricing of options and corporate liabilities. F. Because E[F (t + ∆t) − F (t)]+ = er∆t c(S(t). 2π Substituting from equation (20) lim er∆t e−q∆t S(t)φ(¯)σt+∆t. S(t)) = σ(t).

and I. “The volatility smile and its implied tree. Econometric Analysis.Brace.” Quantitative Finance. Englewood Cliﬀs. 31.” Working paper S01-1. E. M.” Quantitative Strategies Research Notes. J. and A. Futures. 1999 Derman. Prentice Hall. F. “Regimes of volatility. Fleming and R. A. Dept of Statistics. 18–20 Durrleman. 2059–2106 Dupire. leverage.” Journal of Finance.” Journal of Finance 6. Fonseca. J. 3rd Edition. New York.” Quantitative Strategies Research Notes. E.” Journal of Financial and Quantitative Analysis. 55. and J. 361–377 Cont. “Implied volatility functions: empirical tests.” Review of Financial Studies. (2002) Options. 327–343 Hull. 6. S. Princeton University Greene. “A methodology for assessing model risk and its application to the implied volatility function model. Fonseca (2002). 2. D. (1993). April. Also in Risk Magazine. 7. Womersley (2001). 1. Jarrow.” Journal of Financial Economics. Implied Price Processes. J. NJ Heath. R. “Pricing with a smile. R. (1997). Durrleman (2002) “Stochastic Models of Implied Volatility Surfaces. “Bond Pricing and the Term Structure of the Interest Rates: A New Methodology. and W. and A... Department of Operational Research and Financial Engineering. and Stochastic Volatility. 839–866 Cont. (1982). B. (1994). Goldys. 37. B. Goldman Sachs.. and R. A. V.L. Klebaner. B. “The stochastic behavior of common stock variances: Value. W.. J. NJ Hull. “Dynamics of implied volatility surfaces. Morton (1992). 4.” Risk.” Econometrica. 2 (2002). University of New South Wales Britten–Jones. Prentice Hall.” Economic Notes. 10. Suo (2002). 45–60 Christie. 77–105 Heston. “From implied to spot volatilities. Whaley (1997). NY. 2. Kani (1994a). Goldman Sachs. (1999). R. 2 (2000). “A closed-form solution for options with stochastic volatility applications to bond and currency options. 297– 318 23 .” Ph. A. NY Dumas. “Market model of stochastic volatility with applications to the BGM model. and V... Neuberger (2000) “Option Prices. New York. Upper Saddle River. and interest rate eﬀects. 407–432 Derman. 60.D Dissertation. and Other Derivatives.E. (2004). 5th Edition.

C. 4. “Recovering probabilities from option prices”. “Dynamics of the S&P500 implied volatility surface. 49.” Philosophical Transaco tions of the Royal Society. 1611–1631 Kamal. (1999). and A. C. “A market model of stochastic implied volatility. 727–752 24 . Goldman Sachs. Anderson Graduate School of Management. “The patterns of change in implied index volatilities. P. 357. 263–82 Stein. 42. and M. and C. J. NY Ledoit O.. Clewlow (2000).Hull.” Journal of Finance. M. Santa Clara (1998) “Relative pricing of options with stochastic volatility” Working Paper. 2071–2092 Skiadopolous. (1994). M. (1973). J.” Review of Financial Studies. Hodges and L. 281–300 Jackwerth. R. 4. 51. G.J. “Stock price distributions with stochastic volatilities: An analytical approach. White (1987). “Implied binomial trees. “Theory of rational option pricing. University of California. (1994) Option Pricing and Volatility: Advanced Trading Strategies and Techniques.” Quantitative Strategies Research Notes. and E. McGraw-Hill Rubinstein. Rubinstein (1996). Derman. Journal of Finance. “The pricing of options with stochastic volatility. 3.” Journal of Finance. Los Angeles Merton. E. and P. S. Series A. 771–818 Sch¨nbucher. New York.” Review of Derivatives Research. Stein (1991). 2nd ed. 141–183 Natenberg S.” Bell Journal of Economics and Management Science.

accounting for no-arbitrage drift in volatilities. d1 . .List of Figures 1 Plot of the level of the S&P 500 index (solid line. Plot of the level of the S&P 500 index (solid line. . The dotted line on the ﬁrst graph shows the level of d0 (0. d2 and d3 (dashed line. . c2 and c3 (dashed line. with scale on the left axes). Note the positive correlations between c3 and the index. . .3 and 4 are independent of each other and the stock price. 26 2 27 28 3 4 29 25 . Estimated factors for the Totem data. . Factors 1. with scale on the right axis) and the estimates of d0 . . with scale on the right axis) and the estimates of c1 . Principal Component factors for the Totem data. . . with scale on the left axes).5) which would be consistent with the square-root rule. while factor 0 is perfectly correlated with the stock price. for each maturity and moneyness combination. Note the positive correlations between c2 and c3 and the index during the earlier part of the sample. . The height of the surface measures the value of the normalised factor. . . .2. . . . . The height of the surface measures the normalised factor for each maturity and moneyness combination.

with scale on the left axes). c2 and c3 (dashed line.15 1.3 1999 2000 2001 2002 1000 1600 1400 1200 −0. with scale on the right axis) and the estimates of c1 . Note the positive correlations between c2 and c3 and the index during the earlier part of the sample.c1 −0.6 1999 2000 2001 2002 1000 1600 1400 1200 −0.8 1999 2000 2001 2002 1000 c2 c3 Figure 1: Plot of the level of the S&P 500 index (solid line.15 1600 1400 1200 −0.3 0. 26 .

Note the positive correlations between c3 and the index.2 1999 2000 d1 −0. d2 and d3 (dashed line.5 1600 1400 1200 −0. d1 .1 1000 2001 2002 1000 2001 2002 1000 2001 2002 1000 1999 2000 2001 2002 Figure 2: Plot of the level of the S&P 500 index (solid line. 27 . with scale on the right axis) and the estimates of d0 .15 1600 1400 1200 0 1999 2000 d3 0.5 1200 0.35 1999 2000 d2 0.d0 0. The dotted line on the ﬁrst graph shows the level of d0 (0.8 1600 1400 0. with scale on the left axes).5) which would be consistent with the square-root rule.15 1600 1400 1200 −0.

The height of the surface measures the normalised factor.823 Portion 0.5 120 100 K/F 80 0 T−t 5 5 Factor 3: Variance 0.Factor 1: Variance 2.005 0.5 0 −0.5 0 −0.188 Portion 0. for each maturity and moneyness combination.5 120 100 K/F 80 0 T−t Factor 4: Variance 0.4 0.011 Portion 0.2 0 120 100 K/F 80 0 T−t Factor 2: Variance 0.926 0. 28 .062 0.016 Portion 0.004 1 0 −1 120 100 K/F 80 0 T−t 5 5 Figure 3: Principal Component factors for the Totem data.

720 0. Portion 0.Factor 1: Variance 2.100 −0.5 0 −0.2 0 120 100 K/F 80 0 T−t 5 −0. Portion 0.229. Factors 1.4 Factor 2: Variance 0.043.5 120 100 K/F 80 0 T−t 5 Factor 0: Variance 0.052 1 Factor 4: Variance 0. while factor 0 is perfectly correlated with the stock price.3 and 4 are independent of each other and the stock price.2 −0.125. 29 . Portion 0. Portion 0.284.147.3 120 100 K/F 80 0 T−t 5 Factor 3: Variance 0.081 0. Portion 0.5 0 0 −1 120 100 K/F 80 0 T−t 5 −0. accounting for no-arbitrage drift in volatilities.1 0.044 0. The height of the surface measures the value of the normalised factor for each maturity and moneyness combination.2.5 120 100 K/F 80 0 T−t 5 Figure 4: Estimated factors for the Totem data.

. . . . . . . . . . . . . . . . . . . . with factors 1. . . . . . . . . . . . . . relative to the level of the S&P 500 index. . . The factors are all orthogonal. . . . . . . . . . . . . . . . Estimates for the stationary square root of time rule in equation (12). . 31 32 33 34 35 36 37 7 8 38 30 . . . . . . . . . . . . . . . . . . . . . . . . . and factor 0 being perfectly correlated with the stock price. . . . . . . . . . . . . . . 2. Estimates for the version of the relative sticky delta model in equation (11). . . . . Factors obtained by Principal Components analysis applied to changes in the volatility surface. . . . . . . . . Factors obtained by using maximum likelihood estimation. .List of Tables 1 2 3 4 5 6 Volatility matrix for June 1998. . . . . . incorporating the noarbitrage consistent drift. . . . 3 and 4 being uncorrelated with the stock price Brownian Motion. . . .123 (1. . . . . . . . Estimates for the version of the sticky strike model in equation (10) where volatility depends on strike price and time to maturity. . . . . . . . . . In a twotailed test the statistic must be greater than 1. . . . . . .111) to reach the conclusion that the ﬁrst equation to provide a better explanation of the data than the second equation at the 1% (5%) level . Time to maturity is measured in months while strike is in percentage terms. . Estimates of the power rule from equation (14). . . . . . . . . . . . . . . . . . . . . . Comparison of Models using the ratio of sums of squared errors statistic. . . . . . . . . . .

83% 25.20% 21.18% 21.94% 22.30% 22. Time to maturity is measured in months while strike is in percentage terms.91% 18.78% 24.71% 26.99% 24. 31 .53% 23.59% 22.73% 23.10% 27.71% 23.48% 21.58% 24.44% 20.13% 19. relative to the level of the S&P 500 index.50% 20.92% 23.07% 23.14% 23.52% 22.26% 22.83% Table 1: Volatility matrix for June 1998.23% 60 21.06% 23.49% 20.63% 28.64% 22.96% 24.Strike 120 110 105 100 95 90 80 6 15.41% Time to Maturity (months) 12 24 36 48 18.70% 23.21% 22.38% 23.03% 21.40% 24.66% 25.98% 22.33% 21.11% 23.73% 24.

5405505 -0.000036 0.82 -1.0002511 0.0327 Standard Error 0.0000000 -0.4438616 -0.0238337 0.0003454 0.0000022 t-statistic 18.40 1. 32 .0000000 0.0335188 -0.0000355 Table 2: Estimates for the version of the sticky strike model in equation (10) where volatility depends on strike price and time to maturity.0033577 0.2672 0.0000290 0.0001944 0.Variable a0 a1 a2 a3 a4 a5 R2 Standard error of residuals Estimate 0.91 13.0262681 -0.0006589 0.62 -5.0011062 0.30 Estimate from ﬁrst 2 years 0.0000000 -0.44 -7.

0015705 0.0009180 t-statistic 15.39 -147.20 Estimates from ﬁrst 2 years 0.78 -21.41 4.0112730 -0.0496937 Table 3: Estimates for the version of the relative sticky delta model in equation (11).0057 Standard Error 0.72 22.0003801 0.0322727 -0.9493 0.0414902 0.0018404 0.0000701 0.0058480 -0.0075740 0.0091289 0.0019565 0. 33 .0067000 -0.0067576 0.3345461 0.Variable b0 b1 b2 b3 b4 b5 R2 Standard error of residuals Estimate 0.0003487 0.42 45.2884075 0.

2529421 0.0012591 -190.1157022 c4 = 0 0.0572173 8.7102965 0.2654298 0.0052 Standard Error t-statistic Unrestricted 0.2712324 0.2067651 0.2987718 0.9631 0.5206944 -0.5595131 0.9817545 Estimates from ﬁrst 2 years -0.0620709 0.9562 0.Variable c1 c2 c3 c4 R2 Standard error of residuals c1 c2 c3 R2 Standard error of residuals c1 c2 R2 Standard error of residuals Estimate -0.0565335 0.6821751 -0.2547240 0.4260104 0.9156322 0.0013537 -188. 34 .5101281 -0.0124579 11.8294624 0.0058389 9.9631 0.0049 -0.2402662 0.0022643 -111.0049 -0.0272658 20.0240899 Table 4: Estimates for the stationary square root of time rule in equation (12).0675920 0.0064984 20.2106077 -0.1423444 0.2231858 -0.2702221 0.1644600 0.2522889 c3 = c4 = 0 0.1319096 0.

4488377 0.6878322 0.4220134 -0.0725670 0.0052036 84.8600018 0.0864143 0.0674926 0.0048206 91.Parameter d0 d1 d2 d3 d4 R2 Standard error of residuals d0 d1 d2 d3 R2 Standard error of residuals d0 d1 d2 R2 Standard error of residuals Estimate 0.7632480 0.4076018 0.3578373 0.0047 0.4825231 -0.2416622 0.0152348 4.4392204 -0.5114474 -0.4419942 -0.1086545 0.0055657 9.6120391 0. 35 .2901229 d4 = d3 = 0 0.2373512 -0.0047 0.0617382 5.0012548 -183.3775721 0.9652 0.0912235 0.0390363 6.9651 0.7185927 0.9644 0.2581095 0.0076917 11.4665602 Estimates from ﬁrst two years 0.2308670 0.2632187 0.3307833 0.5254560 0.9941367 0.0023361 -103.0540911 0.0048 Standard Error t-statistic Unrestricted 0.2733074 0.5083725 -0.2394926 0.2574542 1.0238527 Table 5: Estimates of the power rule from equation (14).0034726 121.0018407 -130.2714956 0.3039707 d4 = 0 0.

11 1.Test Sticky Strike (10) versus Sticky delta (11) Sticky Delta (11) versus (cubic) Stationary Square-root rule (12) Stationary Square-root rule (12) versus Stochastic Square-root rule (13) Stationary Square-root rule (12) versus Stationary power rule (14) Stochastic Square-root rule (13) versus Stochastic power rule (15) Statistic 32.123 (1. In a two-tailed test the statistic must be greater than 1.37 3.71 1.39 Table 6: Comparison of Models using the ratio of sums of squared errors statistic.111) to reach the conclusion that the ﬁrst equation to provide a better explanation of the data than the second equation at the 1% (5%) level 36 .04 5.

15 -0.15 0.24 0.16 0.18 -0.12 -0.28 -0.17 0.18 -0.07 0.05 -0.12 0.06 -0.21 0.09 0.19 0.15 0.00 1.01 0.00 1.20 0.07 -0.08 -0.05 -0.14 0.27 -0.80 0.20 6 -0.17 0.20 60 0.00 1.15 0.12 0.11 -0.16 0.13 -0.12 0.05 0.03 -0.95 1.12 0.17 0.03 -0.10 -0.15 -0.14 0.14 0.01 -0.11 0.00 1.11 0.05 0.05 1.20 Factor 6 0.22 3 Strike 0.10 0.00 1.21 0.823464 Portion=0.15 -0.03 -0.926429 Time to Maturity (months) 12 24 36 48 0.04 0.10 1.05 0.18 -0.20 Factor 6 0.10 0.061620 Time to Maturity (months) 12 24 36 48 0.13 0.06 -0.17 -0.11 0.29 4 60 0.31 0.10 0.10 0.08 0.06 -0.14 0.10 -0.17 0.06 -0.11 -0.08 0.05 1.05 -0.58 Variance=0.80 0.10 0.17 0.05 -0.18 Variance=0.22 0.05 0.06 -0.05 1.19 0.05 1.02 Strike 0.09 -0.010684 Portion=0.22 0.95 1.09 0.11 Strike 0.95 1.15 0.11 -0.80 0.16 0.18 -0.20 0.02 0.11 0.05 0.41 0.18 0.08 -0.10 0.09 -0.06 -0.08 0.015721 Portion=0.06 0.11 0.07 -0.07 0.95 1.13 0.10 1.11 -0.07 -0.07 -0.10 -0.12 0.14 0.12 0.00 1.39 -0.06 -0.05 -0.90 0.19 -0.15 0.15 0.10 0.22 -0.22 0.13 -0.18 0.10 1.20 6 0.03 0.10 0.05 0.08 -0.11 -0.12 -0.11 -0.27 0.13 0.12 0.14 0.21 0.08 0. 37 .06 -0.16 -0.07 -0.003505 Time to Maturity (months) 12 24 36 48 -0.17 0.10 0.80 0.09 -0.08 -0.10 -0.18 -0.06 -0.10 1.20 -0.00 1.17 0.10 -0.11 -0.04 0.21 -0.01 0.10 60 -0.26 0.187797 Portion=0.10 0.02 0.10 0.90 0.90 0.02 -0.18 Table 7: Factors obtained by Principal Components analysis applied to changes in the volatility surface.90 0.24 0.005158 Time to Maturity (months) 12 24 36 48 0.09 -0.06 -0.16 0.12 0.12 0.11 -0.02 -0.03 Factor 2 60 0.08 -0.12 0.11 Variance=0.13 0.00 -0.28 Variance=2.08 -0.11 0.11 0.07 0.20 -0.14 -0.Factor 1 Strike 0.15 0.08 -0.16 0.07 0.07 -0.11 -0.02 0.01 0.05 0.00 1.08 -0.06 -0.

02 0.02 60 0.15 -0.13 -0.00 1.90 0.04 1.16 0.95 -0.13 Strike 6 0.80 -0.18 0.20 Factor 6 0.13 -0.15 -0.03 Variance=0.11 0.06 0.13 -0.95 1.09 -0.15 0.15 -0.05 0.11 -0.00 1.20 6 -0.28 Factor 4 Strike 0.11 0.11 Strike 6 0.11 0.04 0.10 -0.19 1.23 -0.02 0.21 -0.02 1.19 0.06 0.11 Variance=0.04 0.05 1.13 -0.00 1.09 0.12 0. incorporating the no-arbitrage consistent drift.04 0.13 -0.13 -0.13 -0.13 -0.16 -0.11 -0.11 -0.10 0.20 0.14 0.09 0.10 1.10 0.17 0.20 0.10 -0.05 -0.10 0.229168 Portion=0.10 1.07 -0.15 -0.17 0.13 -0.06 0.10 0.13 -0.03 0.12 0. and factor 0 being perfectly correlated with the stock price.12 0.10 1.19 0.09 0.05 -0.39 0.12 -0.11 0. 2.00 1.02 -0.14 -0.07 -0.03 0.25 0.00 -0.13 -0.22 3 Strike 0.12 -0.10 -0.22 0.283968 Portion=0.80 0.07 0. with factors 1.49 0.95 1.02 0.13 -0.13 -0.10 0.15 -0.04 0.16 0.06 0. The factors are all orthogonal.80 0.20 -0.19 -0.95 1.13 -0.05 0.20 0.08 0.14 0.14 Variance=0.719536 Time to Maturity (months) 12 24 36 48 0.90 0.13 0.11 -0.08 -0.20 -0.00 0.02 0.12 0.07 -0.13 -0.11 -0.12 0.13 -0.10 Factor 0 Factor 2 60 0.05 0.03 0.04 -0.13 -0.08 -0.90 -0.24 60 0.09 -0.02 1.06 0.00 1.32 0.16 0.18 0.15 -0.17 -0.15 0.17 0.08 0.00 -0.05 0. 38 .08 -0.01 1.10 0.20 0.10 0.07 0.04 0.27 -0.20 6 0.146822 Portion=0.03 0.044083 Time to Maturity (months) 12 24 36 48 -0.06 0.08 0.10 0.10 -0.05 0.05 0.21 0.16 0.21 -0.05 0.12 0.10 0.20 -0.90 0.08 -0.25 0.40 -0.05 1.13 -0.080711 Time to Maturity (months) 12 24 36 48 0.00 1.03 0.03 0.04 -0.100011 Time to Maturity (months) 12 24 36 48 -0.80 0.21 60 -0.09 0.13 0.10 0.16 0.17 0.13 0.05 1.12 0.22 -0.06 0.11 0.Factor 1 Strike 0.21 1.22 -0.15 0.03 0.20 -0.06 0.14 0.12 -0.00 -0.12 0. 3 and 4 being uncorrelated with the stock price Brownian Motion.11 -0.20 0.07 0.14 -0.15 0.06 0.12 0.21 0.00 0.17 1.043035 Portion=0.12 -0.15 -0.27 -0.06 -0.18 0.12 Variance=0.26 0.07 -0.05 -0.051709 Time to Maturity (months) 12 24 36 48 -0.80 -0.12 -0.16 -0.15 -0.125169 Portion=0.33 -0.14 -0.20 -0.13 -0.08 0.13 0.01 0.07 0.95 -0.24 1.01 -0.08 0.03 0.52 Variance=2.09 0.90 -0.09 -0.08 0.13 -0.08 0.19 1.10 0.13 -0.20 -0.06 0.08 -0.20 -0.05 0.13 -0.12 -0.10 0.00 1.01 0.13 -0.10 0.00 -0.02 0.08 0.11 0.04 -0.14 60 -0.22 -0.09 0.17 -0.01 Table 8: Factors obtained by using maximum likelihood estimation.04 0.13 -0.22 -0.

- An Empirical Analysis of Reactions to Nadac
- Regularization of Portfolio Allocation - Lyxor White Paper,
- Far 4 Project Paper
- Capital Investment Decision, Project Planning and Control
- Microsoft PowerPoint - PP07final-MarkowitzOptimization
- Avramov Doron Financial Econometrics
- The network of global corporate control
- Target date funds
- Sales Maximization and Oligopoly
- Computational Finance
- Asset Pricing
- Alex Lipton and Artur Sepp- Stochastic Volatility Models and Kelvin waves
- FRM Formulas
- The Complete Investment Banker Extract
- Analysis of Zynga's Profits-Revenues Data
- Implied Bankruptcy Prob - Merton Based
- Derivatives - Session 09
- ConferencePaper-ArbitrageOpportunitiesInIntradayTrading
- The Importance of Accounting Information in Portfolio Optimization
- QBA assignment.docx
- PQM PPT
- Boyle Broadie Glasserman Mc Overview Jedc
- Untitled
- Simple optimized technical trading
- SSRN-id2478751.pdf
- Lecture 10 Beta2 Intro Stat of Port
- Advance Financial Instruments and MarketsLecture No. 02_ Risk and Financial Crisi
- Technical Analysis of Indian Stock Market BSE Sensex Index
- SSRN-id1633027

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 listening from where you left off, or restart the preview.

scribd