6
A Unified Theory of Volatility
Bruno Dupire
Bloomberg
Modern finance can be seen as the art of establishing links
between liquid and illiquid assets. For a long time pricing has
consisted of assessing the fair value of let us say, a stock, from
some information such as a balance sheet or economic fundamen-
tals. The advent of options displaced this focus by establishing a
link between the stock Price, taken as given, and the option price.
This has been rendered Possible thanks to the Black-Scholes-
Merton model which reduced the problem to merely one of volatil-
ity estimation.
We now have to make a further step forward. The proliferation
of more complex options, whether it be in variety or volume, has
Spurred the necessity of. establishing a third link, between European
options and exotic options. European options prices are taken as
given and a model has to be developed which respects those prices
and exhibits a reasonable degree of plausibility.
Respecting European option prices is feasible with a simple
extension of Black-Scholes, which considers local volatilities as
a deterministic function of time and spot level. This model has
the considerable merit of preserving the main features of Black-
Scholes model (only one Brownian Motion, theoretical ability to
hedge any exotic with the asset itself). It even seems to emerge as
the compulsory benchmark model to price exotics consistently
with Europeans. Unfortunately, it has the considerable drawback
Of assuming that the local volatilities computed today will be
ae 185DERIVATIVES PRICING: THE CLASSIC COLLECTION
186
realised with certainty in the future. In that respect it lacks plau-
sibility, as the real world is not amenable to such simplicity and
tends to bewilder us with the unexpected.
What is mentioned here as unified is a model which combines
two essential features: first, compatibility with European option
prices, second random local volatilities. The derivation is along the
lines of Dupire (1992). It parallels the logic of Heath-Jarrow-Morton,
transposing to volatility the analysis which they conducted on
interest rates. We start by extracting some instantaneous forward
volatilities (thus ensuring compatibility with current implied
volatilities), then we make some assumptions about their evolution
and finally obtain the risk neutral dynamics of the current instan-
taneous volatility, which is exactly what is needed to price and
hedge contingent claims.
In the second section, we show how to synthesise instantaneous
forward variances conditional to spot values, make assumptions
on their dynamics and obtain the risk neutral process of the instan-
taneous volatility in the third section. The fourth section makes use
of the Martingale Representation Theorem to price contingent
claims and the fifth section addresses hedging issues. Applications
are considered in the sixth section, extensions investigated in the
seventh section and conclusions presented in the eighth section.
PRIME: MARKET EVIDENCE
Market smiles
Market participants have unanimously adhered to the Black-Scholes
model since its unveiling in 1973. In the Black-Scholes model, the
stock follows a geometric Brownian motion.
ds
S=pdt+odw
S
and generates option Prices as a function of the volatility ¢, assumed
tobe constant. This relationship can be inverted, which allows one to
infer for any option the only volatility input compatible with its mar-
ket price. Its called the implied volatility.
Were the Black-Scholes model a perfect description of real price
moves, all implied volatilities would be equal across all strikes and
maturities. In recent years, markets have
e been displaying volatility
Patterns strongly differing from this fla noes
it world.A UNIFIED THEORY OF VOLATIUTY
Why smiles?
Volatility smiles are more easily observed than explained. However,
one can put forward tentative justifications for their presence.
The most blatant one is simply demand and supply. Investors
are structurally long the stock and will tend either to buy Put
options (low strike) to protect the downside or to sell Call options
(high strike) to enhance their returns, hence unbalancing the
market. This behaviour yields to the classical downwards smile
(expensive Puts, cheap Calls) that is observed on the stock market,
aside from consideration of the dynamics the spot will display in
the future. The crash aversion generated by October 1987 fosters
deeply pronounced smiles on stock indices, which dramatically
challenge Black-Scholes assumptions. It should be noted that on
other markets, like Gold, a reverse pattern is experienced as an
expression of sudden burn-up fears.
Another stream of explanation finds its roots in the examination
of spot dynamics. This means that we could in turn retrieve (or
guess) them through inspection of current market prices.
Examples of dynamics that produce non-flat smiles are, amongst
many other possible choices, jump processes or non-constant
instantaneous (will indifferently be called local) volatilities, which
may either be path-dependent, deterministic function of time and
spot (Dupire, 1993) or stochastic, which brings another dimension
of uncertainty into the picture.
A parallel, mapping local volatility to short-term interest rate
and implied volatility surface to yield curve, will be conducted
throughout the chapter.
A most important point is that even if the current yield curve
shape is mainly determined by demand and supply, arbitrage
strategies enable the locking of forward parameters.
Impact of smiles ;
These widely differing implied volatilities substantially impact
exotic (path dependent) options. Barrier options clearly illustrate
this: their prices do not only depend on the implied volatility
Of their European counterpart but also indeed on the likelihood
of hitting the barrier, which in tum depends on what happens at
any time before maturity and how the spot behaves at different
levels,
eeeDERIVATIVES PRICING: THE CLASSIC COLLECTION
188
WHICH MODEL?
We are looking for a model which
a) respects option prices, ie, fits the current implied volatility surface
b) reflects the randomness of volatility, as displayed by the real world
c) is complete, to ensure preference-free pricing and ability to hedge
d) indeed is arbitrage-free
‘The general logic that will be followed is
O compute by arbitrage from European options the Instantaneous
Forward Variances (IFV) conditional to forward values of the
spot
OQ make a (Real World) assumption on the way the IFV’s evolve
through time
Q tisk neutralise these IFV dynamics
Q obtain Risk Neutral dynamics of instantaneous variances
O apply the Martingale Representation Theorem to price and hedge
contingent claims
The first point is quite thomy, requiring the use of options of differ-
ent maturities and strikes and then taking limits in the right order.
To lighten the analysis and make sure that the emphasis is put on
the concepts and not on the notations, we assume throughout the
chapter that rates are nil and dividends or repos are absent.
ARBITRAGE PORTFOLIO.
Ithas been shown in my previous work (Dupire, 1992) how to com-
Pute the value today of a forward contract that will pay the instan-
taneous variance (square of instantaneous volatility) at time T. In
other terms, the contract binds the two parties to exchange at time
T the realised instantaneous variance against a fixed sum which is
agreed at time ty, The chapter shows the unique possible price for
the fixed leg of the contract thanks to arbitrage agreements there
developed.
The exchange of the floating leg for the fixed leg was carried
Over irrespective of the value of the Spot. As we are here interested
in the Strike dimension as well, we will need a refinement of this
concept, that is to say a contract that will exchange the floating leg
(instantaneous Varia
nce) against the fixed leg, only if the spot at T is‘A UNIFIED THEORY OF VOLATILITY
around a certain value K. We will need a continuum of such con-
tracts, for any value K.
To set things more formally, we define Vgz a forward contract
agreed at fy that will exchange at time T, v(Sz,T) against Vir(So, to)
for a Dirac amount, if and only if S; = K. If Sp # K, no exchange
takes place. In this chapter, we deal with absolute variance as
opposed to proportional variance (in Dupire, 1992).
This parallel with computing an instantancous forward rate gives
us some hints on how to proceed. To compute an IFR, between T and
T + 8T we lend over T + 8T and borrow over T. The transcription in
option terms is a Calendar Spread: buy one Cx r+s7/ sell on Cx7-
Call spread/butterfly arbitrage
Asa first step in the direction of understanding the concept of IFV
and how to lock it, let us try to gain some trading intuition. If we
look at a portfolio comprised of a long position on Cx, 7, and a short
Position on Cx 7, with T; < T, (known as a “Calendar Spread” in
the market) we can compute, in a Black-Scholes world, its value at
time T, as a function of S;,. We get a graph that captures the time
value of Cy, 7, at Ty:
Ky
This Portfolio at time T, can be approximated by a simple combi-
nation of Calls maturity Ty.
@ (Cy-e — 2Cx + Cre)
known as a “Butterfly” and whose profile is
__£\__
K-e Kee Sy
_-‘DERIVATIVES PRICING: THE CLASSIC COLLECTION
190
A perfect match can virtually be achieved through the use of a con-
tinuum of strikes, as any profile can be decomposed as an infinite
sum of Calls: J5 a(K’)Cy.pdK’.
We have thus linked the two dimensions of our implied volatility
surface: the maturity one (Calendar Spread) and the strike one
(Butterfly).
The value of the Calendar Spread at T, will depend on the then
prevailing volatilities in addition to the spot S;,, as the portfolio of
options maturity T, is insensitive to the volatility. We would have
the ingredients to synthesise the contract V7 had the “width” of
the calendar spread at T; been independent of the volatility.
AN
This difficulty can be turned around by the introduction of other
forward contracts. We define Uk,x,,7 2S a contract that exchanges at
time T the local variance u(S,T) against a fixed payment Ux, 1650 to)
determined at fy if K, = S <= K (no exchange otherwise).
Let us define Pary 7, contract that gives Max (Sy — K,0)? at time
Tand the Parabola Calendar Spread
PCS im Paar Paty y
whose value at T is either 0, if S < K, or (5,1) if $ = K.
Now PCSx,1 — PCS,,7 is an instrument that gives u(S,7) if and
only if K, a SS K,. It appears to be the variable leg of the contract
Ux,x,r which delivers o(8,1) if Ky < $< K, at T.
n oe = Teg pays Ux,x,7(So,t,) if and only if K, < § < Kyand is
nen actly replicated by Uk.x1 (Sorto)(Dx,r — Dx,z) where Der
sa digital option, whose payoff at Tis 1 if S;= K, Ootherwise
Equating the values at fy of the two legs gives
Un... (Soto Dy, (Spt) — Dx,1(Soto))
= PCSx,1(So,fo) — PCSx,7(So, fo)
Dividing both sides by K, - Ky and letting K; go to‘A UNIFIED THEORY OF VOLATILITY
ADx.1(Sorto) _ APCSp r(Spto)
0 Vex(Sr fg) reno) = PCSe1(Sorto)
aK
As Vx7(So,to) is the limit as K, goes to K, of Ux,x,r (So-fo) we can
tewrite the two derivatives as follows:
Dg.rlSyito) _ ~PCx Sof)
aK aK?
and
OPCSq.r(Sovto) _ a | emf)
aK ~ aK aT
3 sPia ut)
” aT aK
a
a — 5p Cx Sorted)
which after substitution in (*) gives
9 dCx (5-4)
aT
FC 1(Spsf0)
aK?
Vk.r (Soto) =
Which is the local variance of the local volatility model (Dupire,
1993),
Arrow Debreu Numeraire .
The previous section provided a means to obtain (*) with the arbi-
tage portfolio. Another way to get it is the Fokker-Planck equation
(Dupire, 1993), :
We present here a third and hopefully enlightening way this
time in a generalised form.
We fix a given (K, T) and define CS and But as:
pe Cacteat TOKE
a ag
er’DERIVATIVES PRICING: THE CLASSIC COLLECTION
and
Cuneat 7 2Ck.t +}Cr-est
eee eens
But = lim
0 &
CS and But are financial instruments, which can be valued at any
time t for any spot values. Both happen to behave at time T like
Dirac masses at K. More precisely, we have:
C(S,T) -[78] But(S,T)
Which we can read as: the price of CS expressed in terms of But at
time T is:
5,7)
2
Keeping in mind that the price of an asset expressed in terms of a
second one is a martingale under the measure associated with the
second asset. Hence:
Seu (Soy ta) = E* [CS py (Sp, T)] = EO" [He 2)
Where Qk, is the martingale measure associated with But which is
in tum the Arrow-Debreu security linked with (K,T). Qxr is not
equivalent to Q but is a limit of martingales equivalent to Q.
Re-expressing in $ terms we get:
C5(S, 1) = ER [S| But(Sy, t)
From the definition of CS and But we have:
2c,
CStSr) = EL (6, 1)
and
FC,
ae
Buty to)= 5
192‘A UNIFIED THEORY OF VOLATIUTY
On top of that, for any random variable X.
rx) = E2| x Ser
EX] = [x Ser]
[ae
=E2[X|S,=K]
So taking an expectation under Qx7 is no more than taking a con-
ditional expectation under Q. Combining the preceding results, we
get the fundamental PDE.
ac us, ay #Cer
Big py a pre | MD c JE EKT Gt)
57 Soro) Ee [ > |e OK? Gorty
To obtain a stochastic volatility model compatible with the market
smile, we need to have:
2°EE Gf)
ER*"[0(S, T)] = Vi. (Sp»f0) = se ———
For example, a lognormal model would give:
By
(Syrto)e ?
¥
— Sith,
zh,
Bile
Where the denominator has to be numerically computed.
REAL WORLD SDE FOR Ver
Wejust saw how to extract the (Vir(So telex from the (Cx(S04))k
For a given (K,T), the value of Vir computed tomorrow from the
new smile surface will likely offer from the one computed today and
will reflect market uncertainty regarding volatility. As Vir is a Va"
ance, it must remain positive and we hence model the proportional
193DERIVATIVES
194
ICING: THE CLASSIC COLLECTION
evolution of Vxr, through the SDE
V; f
ies) ) =< aat+ Sb, aw;
Ve r(S,1)
where is a drift term and the b;’s account for the sensitivity to the
Brownian Motion W, affecting the whole volatility surface through-
out all K’s and T's. Both « and the bs may be of very general form.
For the sake of notational simplicity, we concentrate on the one
Brownian Motion case. W is a Brownian motion under real world
measure P.
dv, 7(S, t)
Ver) =adt+bdw,
Under Qk, Vir is a martingale:
We r(S,t) _
or
Verso
and under Q
dV, 7(S,t)
Sealer) _ 2
Ver(St) adi+bdwy
where a is obtained by the change of measure from Qx1r to Q.
Indeed, if we investigate the deterministic variance case (b = 0) we
get for alll S, t, K, T,
Var (S,t) = Ver (Sorte)
Thus,
2 2Gax( Soto)
Ct ee
9Cx.r(Sp, fo)
ak
This result was initial
ly proved by double i i
Fokker-Planck equation y double integration of the
(see Dupire, 1992),A UNIFIED THEORY OF VOLATILITY
PRICING
‘We now pay attention to a contingent claim X that will pay at time T
a payoff that may depend on the full path of S (and possibly of v)
and T.
Once the Risk Neutral dynamics of S and v are obtained, pricing
becomes almost an easy matter. We can apply the Martingale
Representation Theorem to ht defined as
1(S,0,t) = E°[X;|F,]
which is the expectation of the final payoff of X conditioned by the
current value of the spot and of the variance at f. Indeed, hr isa Q
martingale, thus we can apply the MRT which gives the existence
of a and B such that
' '
MS, v,t)=I(5,,09, to) + f a, dW? +B, dW,
which gives us two things:
1. The initial premium of the claim: lM(Sp,0),to)=E°%CXr|Fol
2. The hedging strategy (c.,, f,), which allows us to transform this
initial premium into the final payoff.
The expectation in 1 can be computed by Monte Carlo simulation
or PDE methods.
REFERENCES
Blac, F, and M. Scholes, 1973, “The Pricing of Options and Corporate Lihiities own!
Political Economy, 81, pp. 637-54.
Boyle, R, 1977, “Options: A Monte Carlo Approach”, Joural of Financial Economics,
Pp.323-38,
Breeden, D,, and R. Litzenberger, 1978, “Prices of State-Contingent Claims Implicit in
Option Prices”, journal of Business, SL, pp. 621-51
Cor, J.C. J. E.Ingersoll, and S. A. Ross, 1985, A Theory ofthe Term Structure of Interest
Rates", Econometrica, $3, p. 385-07.
Duffie, D,, 1988, Security Markets, Stoclastic Matels (San Diego: Academic Press).
— eeeDERIVATIVES PRICIN
196
HE CLASSIC COLLECTION
Dupire, B, 1992, “Arbitrage Pricing with Stochastic Volatility”, Proceedings of AFFI
Conference in Paris, June.
Dupire, B,, 1993, “Pricing and Hedging with Smiles”, Proceedings of AFFI conference in
La Baule, June.
EI Karoui, N,, R. Myneni, and R. Viswanathan, 1992, “Arbitrage Pricing and Hedging of
Interest Rates Claims with State Variables”, Working Paper.
Hanzison, J.M. and D. Kreps, 1979, “Martingales and Arbitrage in Multiperiod Securities
Markers", Journal of Economic Theory, 20, pp. 381-408.
Harsison, J. M, and S. Pliska, 1981, “Martingales and Stochastic Integrals in the Theory
of Continuous Trading”, Stochastic Processes and their Applications, 11, p. 215-60.
Heath, D., A. Jarrow, and A. Morton, 1987, “Bond Pricing and the Term Structure of
Interest Rates: A New Methodology”, Preprint, Comell University.
Ho, T. S,, and S. B. Lee, 1986, “Term Structure Movements and Pricing Interest Rate
Contingent Claims”, Journal of Finance, 41, p. 1011-29.
Hull, J., and A. White, 1987, “The Pricing of Options on Assets with Stochastic
Volatlities”, The Journal of Finance, 3, pp. 281-300.
Hull, J,, and A. White, 1990, “Valuing Derivative Securities Using the Explicit Finite
Difference Method”, Journal of Financial and Quantitative Analysis, 25, pp. 87-100.
Hall, J. and A. White, 1992, “One Factor Interest-Rate Models and the Valuation of
Interest-Rate Contingent Claims”, Working Paper, University of Toronto.
Jamshidian, F, 1991, “Forward Induction and Construction of Yield Curve Diffusion
Models”, Journal of Fixed Income, 1.
Johnson, H, and D. Shanno, 1987, “Option Pricing when the Variance is Changing”,
Journal of Financial and Quantitative Analysis, 2, pp. 143-51.
Karatzas, I, and S. E. Shreve,
Springer-Verlag).
Merton, R, 1973, “The The
‘Management Science,
1988, Brownian Motion and Stochastic Calculus, (New York:
'eory of Rational Option Pricing”, Bell Journal of Economics and
L pp. 141-83,