126 Risk July 2007

CUTTING EDGE CLASSIC FROM JANUARY 1994. VOLATILITY

model
1
gives options
prices as a function of
volatility. If an option price is given by the market we can invert
this relationship to get the implied volatility.
If the model were perfect, this implied value would be the
same for all option market prices, but reality shows this is not
the case. Implied Black-Scholes volatilities strongly depend on
the maturity and the strike of the European option under scru-
tiny. If the implied volatilities of at-the-money options on the
Nikkei 225 index are 20% for a maturity of six months and
18% for a maturity of one year, we are in the uncomfortable
position of assuming that the Nikkei oscillates with a constant
volatility of 20% for six months but also oscillates with a con-
stant volatility of 18% for one year.
It is easy to solve this paradox by allowing volatility to be time-
dependent, as Merton did.
2
Te Nikkei would rst exhibit an
instantaneous volatility of 20% and subsequently a lower one,
computed by a forward relationship to accommodate the one-year
volatility. We now have a single process, compatible with the two
option prices. From the term structure of implied volatilities we
can infer a time-dependent instantaneous volatility, because the
former is the quadratic mean of the latter. Te spot process S is
then governed by the following stochastic dierential equation:
dS
S
= r t ( )dt + t ( )dW
where r(t) is the instantaneous forward rate of maturity t implied
from the yield curve.
Some Wall Street houses incorporate this temporal informa-
tion in their discretisation schemes to price American or path-
dependent options.
However, the dependence of implied volatility on the strike,
for a given maturity (known as the smile eect) is trickier.
Researchers have attempted to enrich the BlackScholes model
to compute a theoretical smile. Unfortunately, they have to
introduce a non-traded source of risk such as jumps, stochastic
volatility or transaction costs, thus losing the completeness
(ability to hedge options with the underlying asset) of the
model.
3
Completeness is of the highest value: it allows for arbi-
trage pricing and hedging.
Terefore, we must ask whether it is possible to build a spot
process that:
n is compatible with the observed smiles at all maturities, and
n keeps the model complete.
More precisely, given the arbitrage-free prices C(K, T) of Euro-
pean calls of all strikes K and maturities T, is it possible to nd a
risk-neutral process for the spot in the form of a diusion:
dS
S
= r t ( )dt + S, t ( )dW
where the instantaneous volatility s is a deterministic function of
the spot and of the time?
Tis would extend the Black-Scholes model to make full use of
its diusion setting without increasing the dimension of the
uncertainty. We would have the features of a one-factor model
(hence easily discretisable) to explain all European option prices.
We could then price and hedge any American or path-dependent
options (even for European options, the knowledge of the whole
process is necessary for hedging). We would also know which
volatility to use to price a barrier option and how to hedge a com-
pound option. It is quite simple to work on a discretised version of
the spot, as we show later, but here we also give an analytical
treatment, which is more revealing.
If the spot price follows a one-dimensional diusion process,
then the model is complete and option prices can be computed by
discounting an expectation with respect to a risk-neutral proba-
bility under which the discounted spot has no drift (but retains
the same diusion coecient).
More precisely, path-dependent options are priced as the dis-
counted expected value of their terminal payo over all possible
paths. In the case of European options, this boils down to an
expectation about the terminal values of the spot (which can be
seen as bundling the paths that end at a same point).
It follows that knowledge of the prices of all path-dependent
options is equivalent to knowledge of the full (risk-neutral) diu-
sion process of the spot; knowing all European option prices
merely amounts to knowing the probability densities of the spot
at dierent times, conditional on its current value.
Te full diusion contains much more information than the
conditional laws, as distinct diusions may generate identical
conditional laws. For instance, a Gaussian process with mean
reversion can generate the same conditional laws as another Gaus-
Pricing
with
a smile
In the January 1994 issue of Risk, Bruno Dupire
showed how the Black-Scholes model can be extended
to make it compatible with observed market volatility
smiles, allowing consistent pricing and hedging of
exotic options
The Black-Scholes
1
Black, F, and M Scholes, 1973, Te pricing of options and corporate liabilities, Journal of Political
Economy 81, pages 637654
2
Merton, R, 1973, Te theory of rational option pricing, Bell Journal of Economics and Management
Science 4, pages 141183
3
For an account on completeness for stochastic volatility, see Dupire, Arbitrage pricing with stochastic
volatility, Proceedings of AFFI Conference, Paris, June 1992, and Model art, Risk, September 1993,
pages 118124
dupire.indd 2 2/7/07 16:56:30
risk.net 127
sian process with volatility decreasing over time. However, as we
shall see, if we restrict ourselves to risk-neutral diusions, the
ambiguity is removed and we can retrieve from the conditional
laws the unique risk-neutral diusion from which they come.
Tis result is interesting in itself but we will also exploit its conse-
quences in hedging terms.
A difusion from prices
We can gain considerable clarity without losing much in general-
ity by assuming that the interest rate is zero. For a given maturity
T, the collection of option prices of dierent strikes C(K, T)
which in practice requires a smooth interpolation from a few
points yields the risk-neutral density function T of the spot at
time T through the relationship:
C K,T ( ) = max S K, 0 ( )
T
S ( )dS
0

which we dierentiate twice with respect to K to obtain:

T
K ( ) =

2
C
K
2
K,T ( )
which is the risk-neutral probability density of the spot being
equal to K at time T. We recall that European option prices are
equivalent to the densities
T
and that path-dependent option
prices are equivalent to the diusion process. We are then left
with an interesting stochastic problem with the notation (x, t)
instead of (K, T): knowing all the densities conditional on an ini-
tial xed (x
0
, t
0
), is there a unique diusion process dx = a(x, t)dt
+ b(x, t)dW which generates these densities?
Te solution in general is not unique; however, if we restrict
ourselves to risk-neutral diusions, we can recover, under some
technical assumptions, a unique diusion process from the
T

(see box 1). Te interest rate being zero, we pay attention only to
martingale (ie, driftless) diusions dx = b(x, t)dW.
Tanks to the Fokker-Planck equation, we can after some
maths
4
write:

b
2
K,T ( )
2

2
C
K
2
=
C
T

(E)
where C(S, t, K, T) denotes the premium at time t for a spot S of
a European call of strike K and maturity T.
Both derivatives are positive by arbitrage (buttery for the con-
vexity and conversion for the maturity). Equation (E) can be used
to determine b, as:

2
C
K
2
and
C
T
are known from the market smiles. We can infer the instanta-
neous volatility at time T for a spot equal to K from the knowl-
edge of the option prices of maturities and strikes around T
If we restrict ourselves to difusions, there is a unique risk-neutral (drift equal
to the short-term rate) process for the spot which is compatible with Euro-
pean option prices:
Diffusions
Risk-neutral
processes
Processes
compatible
with market
smiles
Unique
sought
diffusion
This means that, if we assume the spot is following a difusion process, we
can obtain exotic option prices from European option prices through the
scheme:
European
prices
Path-dependent
prices
Risk-neutral
densities
Risk-neutral
diffusion
1. A unique diffusion process
4
See Dupire, Pricing and hedging with smiles, Proceedings of AFFI Conference, La Baule, June 1993
(also presented at IAFE meeting, New York, December 1993)
Performance
Measurement in
Financial Institutions
in an ERM framework
By Ashish Dev and Vandana Rao
Price 85 / US$153 / 128
Format Hardback,198pp
ISBN 1 904339 72 7
Focusing on internal business unit performance, this in-
depth practitioner guide addresses both traditional and
risk-adjusted performance measures in financial
institutions, providing you with the tools to implement a
balanced scorecard approach to performance
measurement.
In the spirit of enterprise risk management, you will benefit
from an integral treatment of the three pillars of internal
performance measurement in a financial institution:
Funds transfer pricing
Economic capital
Expense allocation
Performance Measurement in Financial Institutions in an
ERM Framework will enable you to successfully
measure and control performance. It is recommended
for chief financial officers, treasurers, financial analysts,
accountants, chief risk officers, risk managers, auditors
and management consultants engaged in financial
consulting to financial institutions.
Online
riskbooks.com
Email
[email protected]
Fax
+44 (0)20 7484 9800
Telephone
+44 (0)870 240 8859
RBID07_PMAD65x198.indd 1 2/7/07 14:30:30
dupire.indd 3 2/7/07 16:56:32
128 Risk July 2007
CUTTING EDGE CLASSIC FROM JANUARY 1994. VOLATILITY
and K, which is our primary purpose. Going back to the spot
process dS/S = s(S, t)dw, we indeed obtain the instantaneous
volatility by:
S, t ( ) =
b S, t ( )
S
A new way to compute prices
Equation (E) can also be interpreted in another fashion. If b is
known, it establishes a relationship between the price as of
today of call options of varying maturities and strikes.
Equation (E) has the same avour as, but is distinct from, the
classical Black-Scholes partial dierential equation which
involves, for a xed option (ie, K and T xed), derivatives with
respect to the current time and value of the spot. With zero inter-
est rates, the Black-Scholes equation takes the form:

b
2
S, t ( )
2

2
C
S
2
=
C
t

(BS)
Equations (E) and (BS) can be thought as being dual to each
other. However, the relationship is not so universal, as (BS)
applies to any contingent claim, though (E) holds only because
the intrinsic value of a call happens to be the second integral of a
Dirac function. It is very fortunate that the market trades this
particular payo!
It also provides an algorithm to compute an option price
through a forward tree and even the price of many dierent
options in a single sweep of the tree!
To price the (K, T) call, we build a tree with its root at (K, T),
expanding backward in time up to the current date where it is fed
by an intrinsic value which is the value today of an option of
immediate maturity. Pricing is performed forward in time by tak-
ing the discounted expectation at each node until the root (K, T)
is reached and the premium can be collected (see box 2).
An internal node of the tree will be labelled with todays value
of a European call where strike and maturity correspond to this
node, as opposed to a standard tree where each node carries the
premium of a xed option at a future time and spot associated to
that node.
It is indeed possible to compute b numerically from the relation
(E) obtained from the continuous time and price analysis, and to
discretise the associated spot process with explicit recombining
binomial
5
or trinomial
6
schemes. We prefer however to present a
construction that makes use of a new technique widely used for
interest rate model tting: forward induction,
7
as it can be under-
stood without any stochastic machinery.
It is worth stressing that it is quite easy to nd a set of coe-
cients that price options correctly, since degrees of freedom are in
superabundance compared with the constraints. Te situation is
analogous to the one encountered in the continuous case, where
various diusions could generate the same densities. However,
imposing the martingale condition (risk-neutrality) in the dis-
crete time setting at each node gives additional constraints. Tis
extra structure is a key point in our pricing/hedging approach but
existence and uniqueness are in general not achieved by a simplis-
tic discretisation. As we shall see, a trinomial one does ensure
existence and uniqueness of the discretised process, through a
parsimonious use of its degrees of freedom (the weights carried by
the connections).
We build a trinomial tree with equally spaced time-steps. Te
ratio of price-step over time-step, which determines the opening
of the tree, has to be large enough to cater for the local variance of
the process. Tis condition is equivalent to the one that guaran-
tees the stability of explicit discretisations of a partial dierential
equation. If the market Black-Scholes smiles are not too pro-
nounced, equal steps on the logarithm of the spot are best. If the
Spot follows dx
t
= b(x, t)dW
t
(interest rates are 0).
Two ways to compute C(S
0
, 0, K
0
, T
0
):
Black-Scholes PDE (BS)
K, T fxed
C
t
=
b
2
S, t ( )
2

2
C
S
2
computes C(S, t, K
0
, T
0
)
S
0
0 T
0
C(S, t, K
0
, T
0
)
K
0
(S K
0
)+
(S, t)
Fokker-Planck (E)
S, t fxed
C
T
=
b
2
K,T ( )
2

2
C
K
2
computes C(S
0
, 0, K, T)
S
0
0 T
0
C(S
0
, 0, K, t)
K
0
(S
0
K)
+
(K, t)
In both cases, C(S
0
, 0, K
0
, T
0
) collected at the root of the tree.
2. A new way to price options
5
Nelson, D, and K Ramaswamy, 1990, Simple binomial processes as diusion approximations in
nancial models, Te Review of Financial Studies 3, pages 393430
6
Hull, J, and A White, 1990, Valuing derivative securities using the explicit nite dierence method,
Journal of Financial and Quantitative Analysis 25, pages 87100
7
Jamshidian, F, 1991, Forward induction and construction of yield curve diusion models, Journal of
Fixed Income 1; Hull, J, and A White, 1992, One factor interest-rate models and the valuation of
interest-rate contingent claims, working paper, University of Toronto
dupire.indd 4 2/7/07 16:56:32
risk.net 129
initial guess of the opening is not high enough, it should be
increased to ensure that the procedure described below can be
carried out. Weights will be assigned to the connections, which
will allow us to compute the discounted probability of each path
and hence to value any path-dependent option. It is possible to
reduce the complexity of the computation in many cases.
At each discrete date, all proles consisting of continuous
piecewise linear functions with break points located at inner
nodes of the tree are required to be correctly priced by the tree. At
the nth step, any such prole is uniquely characterised by the
value it takes on the 2n + 1 nodes of that step, thus the space of all
proles is of dimension 2n + 1. Tis contains the zero coupon, the
asset itself and all calls (and puts) whose strikes are the inner
nodes. With each node we associate an Arrow-Debreu prole
whose value is 1 on this node and 0 on the others.
A node is labelled (n, i) with n denoting the time-step and i the
price-step. Its associated Arrow-Debreu price is denoted A(n, i)
and the weight of the connection between nodes (n, i) and (n + 1,
j), j = i 1, i, or i + 1 is denoted w(n, i, j). Arrow-Debreu prices
are computed from market prices, as prices of portfolios of Euro-
pean calls, spot and cash positions. Te weights are computed
through the tree in a forward fashion.
We can exploit two types of relations:
n forward relations, which relate the Arrow-Debreu price of a
node to the Arrow-Debreu prices of its immediate predecessors;
n standard backward relations, which link the value of a contin-
gent claim at a node to its value at the immediate successors. We
apply this relation to two simple claims: a unit of cash and a unit
of the spot, both to be received one time-step later (see box 3).
Te generic step of the algorithm is:
Compute w(n, i, i 1) from A(n + 1, i 1), A(n, i), A(n, i 1),
A(n, i 2), w(n, i 1, i 1) and w(n, i 2, i 1).
Compute w(n, i, i) and w(n, i, i + 1) from the forward discount
factors of the cash and spot.
Hedging
Knowledge of the whole process allows for the pricing of path-
dependent options (by Monte-Carlo methods) and American
options (by dynamic programming). It also allows for hedging
through an equivalent spot position because the sensitivity of the
options with respect to the spot can be computed. Knowing the
full process, it is possible to shift the initial value and to infer the
process that starts from this new value and the new price it incurs.
Delta hedging can then be achieved, which will be eective
throughout the life of the option if the spot behaves according to
the inferred process.
It probably will not, which leads us to a more sophisticated
method of hedging. We can build a robust hedge that will be
ecient even if the spot does not behave according to the instan-
taneous inferred volatilities of the diusion process. Te idea is
to associate with every contingent claim X a portfolio of Euro-
pean options (which should be rebalanced periodically) that
will be tangential to it in the sense that it will change in value
identically up to the rst order for changes in the volatility
manifold s(K, T)
K,T
.
We proceed as follows. A local move of the volatility manifold
around (K
0
, T
0
) will lead to a new diusion process, hence to a
new value of X. We can then compute the sensitivity of X to a
change of volatility s(K
0
, T
0
) and the equivalent (K
0
, T
0
) call posi-
tion. Repeating for all (K, T), we obtain a spectrum of sensitivi-
ties Vega(K, T) and the associated (continuous) portfolio of (K,
T) calls, which can be seen as a projection of X on the calls. Tis
portfolio will behave up to the rst order as X, even if the market
evolves by transgressing the induced forward volatilities com-
puted above.
Conclusion
Under certain conditions, it is possible to recover from the condi-
tional laws a full diusion process whose drift is imposed. Tis
means that from option prices observed in the market we can
induce a unique diusion process. Clearly it would be excessive to
pretend that the spot will follow this diusion. What we can say
is that the market prices European options as if the process was
this diusion.
In practice, this shows how a sound pricing for path-dependent
and American options can be elaborated. Moreover, it nely
assesses the risk of such options by performing a risk analysis
along both strikes and maturities. Tis enables these options to be
fully integrated into a book of standard European options, which
is clearly a key point for many nancial institutions. n
Bruno Dupire was head of options research at Paribas Capital Markets in
London. He is happy to mention fruitful conversations with Nicole El Karoui
and colleagues from the SORT (Swaps and Options Research Team) at Paribas
We assume the connections have been computed over the frst time-step
and pay attention to the second one:
A
D
I
H
G
F
E
6
6
1
5 4
4
3
2
2
B
C
1 is computed forward through the Arrow-Debreu prices of B and E
2 are computed backward through 1 and the zero coupon and the spot at
period 2
3 is computed forward through the Arrow-Debreu prices of B, C, F and 2
4 are computed backward through 3 and the zero coupon and the spot at
period 2
5 is computed forward through the Arrow-Debreu prices of B, C, D, G, 2
and 4
6 are computed backward through 5 and the zero coupon and the spot at
period 2
Arrow-Debreu profles of H and I need not be exploited, as they are neces-
sarily correctly priced by the tree. In efect, they are spanned by the Arrow-
Debreu profles of E, F and G, the zero coupon and S, which are correctly
priced.
3. Building the tree
dupire.indd 5 2/7/07 16:56:33