2004 Option Pricing Valuation Models and Applications

MANAGEMENT SCIENCE informs ®
Vol. 50, No. 9, September 2004, pp. 1145–1177 doi 10.1287/mnsc.1040.0275

issn 0025-1909 eissn 1526-5501 04 5009 1145 © 2004 INFORMS
50th Anniversary Article

Option Pricing: Valuation Models and Applications
Mark Broadie
Graduate School of Business, Columbia University, 3022 Broadway, New York, New York 10027-6902, mnb2@columbia.edu
Jerome B. Detemple
School of Management, Boston University, 595 Commonwealth Avenue, Boston, Massachusetts 02215, detemple@bu.edu
T his paper surveys the literature on option pricing from its origins to the present. An extensive review of
valuation methods for European- and American-style claims is provided. Applications to complex securities
and numerical methods are surveyed. Emphasis is placed on recent trends and developments in methodology
and modeling.
Key words: option pricing; American options; risk-neutral valuation; jump and stochastic volatility models
1. Derivatives Markets: maturity date T , for a prespecified price K, called the

Introduction and Definitions strike or exercise price. A call (put) option is a right to
Management Science has a long tradition of publish- buy (sell). Because exercise is a right and not an obli-
ing important research in the finance area, includ- gation, the exercise payoff is S − K+ ≡ maxS − K 0
ing significant contributions to portfolio optimization; for a call option and K − S+ ≡ maxK − S 0 for a
asset-liability management; utility theory and stochas- put option, where S denotes the price of the underly-
tic dominance; and empirical finance and derivative ing asset. Options can be European style, which can
securities. Within the derivatives area, contributions only be exercised at the maturity date, or American
in the journal have advanced our understanding of style, where exercise is at the discretion of the holder,
the pricing, hedging, and risk management of deriva- at any time before or at the maturity date.
tive securities in a wide variety of financial mar- Plain vanilla options, such as those described above,
kets, including equity, fixed income, commodity, and were introduced on organized option exchanges such
credit markets. In addition to theoretical advances, as the Chicago Board of Options Exchange (CBOE) in
several articles have focused on practical applications 1973. Since then, innovation has led to the creation
through the design of efficient numerical procedures of numerous products designed to fill the needs of
for valuing and hedging derivative securities. In this various types of investors. Path-dependent options,
paper we survey the option pricing literature over such as barrier options, Asian options, and lookbacks
the last four decades, including many articles that are examples of contractual forms that have emerged
have appeared in the pages of Management Science. We since and are now routinely traded in markets or
begin with a description of derivative securities and quoted by financial institutions, or both. Even more
their properties. exotic types of contracts, whose payoffs depend on
A derivative security is a financial asset whose pay- multiple underlying assets or on occupation times of
off depends on the value of some underlying variable. predetermined regions, have emerged in recent years
The underlying variable can be a traded asset, such and have drawn interest.
as a stock; an index portfolio; a futures price; a cur- This paper surveys past contributions in the field
rency; or some measurable state variable, such as the and provides an overview of recent trends. We
temperature at some location or the volatility of an first review the fundamental valuation principles for
index. The payoff can involve various patterns of cash European-style (§2) and American-style (§3) deriva-
flows. Payments can be spread evenly through time, tives. Next, we outline extensions of the basic
occur at specific dates, or a combination of the two. model (§4) and survey numerical methods (§5). An
Derivatives are also referred to as contingent claims. assessment of the future and comments on open prob-
An option is a derivative security that gives the right lems are formulated in the concluding section (§6).
to buy or sell the underlying asset, at or before some The appendix treats derivatives in the fixed income
1145
Broadie and Detemple: Option Pricing: Valuation Models and Applications
1146 Management Science 50(9), pp. 1145–1177, © 2004 INFORMS
and credit markets, real options, path-dependent con- on . Letting V t S and V t S be the mean and
tracts, and derivatives written on multiple underlying volatility of the derivative’s return enables us to write
assets. dVt = Vt V t St dt + V t St dWt .
A self-financing portfolio of the risk-free asset, the
2. European-Style Derivatives underlying asset, and the derivative contract, how-
The focus of this section is on the fundamentals ever, has value X evolving according to
of derivatives’ valuation and applications to plain dXt = rXt dt + Xt t − r dt + dWt
vanilla European options. We first review the no-
arbitrage (§2.1) and the risk-neutral (§2.2) approaches + Xt tV V t St − r dt + V t St dWt
to valuation for European-style derivatives. We then
with initial value X0 = x the cost of the portfolio
turn our attention to valuation in the presence of
at initiation. Here X represents the (dollar) amount
unhedgeable risks (§2.3).
invested in the underlying asset, X V the amount in
2.1. No-Arbitrage Valuation the derivative, and X1 − − V the balance in the
The no-arbitrage principle is central to the valuation riskless asset ( V and 1 − − V are the fractions
of derivative securities. One of the most important of the portfolio’s value invested in the three assets).
insights of the seminal papers by Black and Scholes With the choices Xt tV = V St t and Xt t = −St ×
(1972) and Merton (1973) was to show how the prin- V St t/S we obtain

ciple can be used to characterize the price of an arbi- V V V
trary derivative asset. dXt = rXt − S − r + + S −
S t t S t
2.1.1. The Black-Scholes-Merton Framework. The
1 2V 2 2
Black-Scholes-Merton (BSM) analysis starts from the + S − rV dt
premise that the underlying asset price S follows a 2 S 2 t
geometric Brownian motion process Note that this portfolio is locally riskless. Because the
dSt initial capital could also have been invested risklessly,
= − dt + dWt (1) to preclude the existence of arbitrage opportunities
St the portfolio return must be equal to the risk-free rate.
where , and are constants, which represent the Equivalently, the derivative’s price V must satisfy
expected (total) return on the asset, the dividend rate,
V V 1 2V 2 2
and the return volatility, respectively. The process W + St r − + S − rV = 0 (2)
is a standard Brownian motion, under the physical t S 2 S 2 t
(or statistical) probability measure P , that captures the on . This partial differential equation, along with the
underlying uncertainty in this market. Trading in this boundary conditions
asset is unrestricted, i.e., no taxes, transactions costs, 
 V ST T = gST on +
constraints, or other frictions. Likewise, investors can 

invest without restrictions, at the constant risk-free V 0 t = g0e−rT −t on 0 T (3)
rate r.1 The asset paying interest at the riskless rate 

 lim V S t = g e−rT −t on 0 T
and the underlying risky asset are sometimes called S→
the primary assets.
characterizes the derivative’s price. Equation (2) is
In the sequel we refer to this model as the BSM
known as the fundamental valuation equation for
framework or setting.
derivatives’ prices. The equation is called fundamen-
2.1.2. The Fundamental Valuation Equation. Sup- tal because it applies to any derivative security, inde-
pose that we seek to value a European-style derivative pendently of its payoff structure. What changes across
that pays off gST at a given maturity date T . Assum- securities are the relevant Boundary Conditions (3).
ing that the current price Vt ≡ V S t of the security One can also introduce the new variables x and
has suitable differentiability properties (specifically the function ux such that
V S t ∈ 2 1 on the domain ≡ + × 0 T ) one
can apply Itô’s lemma (see Karatzas and Shreve 1988, S = K expx t = T − 1 2 V St = e−x− ux
p. 149) to show that 2

where ≡ r − − 12 2 / 2 , ≡ 2 + 2r/ 2 , and where
V V 1 2V 2 2 V
dVt = + St −+ St dt + S dWt K is an arbitrary positive constant (in the special
t S 2 S 2 S t
case of options, K represents the strike). The function
1
ux solves the modified valuation equation
Trading strategies are subject to a mild regularity condition. To
ensure a well-defined value process the quadratic variation of a u 2 u
portfolio value must be finite P -a.s. = 2 (4)
x
Management Science 50(9), pp. 1145–1177, © 2004 INFORMS 1147
subject to the boundary conditions for all t ∈ 0 T . The condition reflects the fact that
 −x the portfolios on the left- and right-hand sides of
 e ux0 = gK expx on the equation have the same payoff ST − K+ + K =


 2 K − ST + + ST at the maturity date: Because terminal
lim e−x− ux = g0e−2r/ on ∈ 0 12 2 T


x→− payoffs are identical, their prices must also be iden-
 lim e−x− ux = g e−2r/ 2 on ∈ 0 1 2 T
 tical, at all times prior to maturity, to preclude arbi-
x→ 2
trages. Solving for the put price from (9), substituting
(5) the call Formula (6), and simplifying establishes (8).
Extensions of the model to settings with trad-
Condition (4) is known as the heat equation. This ing costs, short-sales constraints, and other frictions
partial differential equation, which characterizes the have been developed. See, for example, Leland (1985),
propagation of heat in a continuous medium, has Hodges and Neuberger (1989), Bensaid et al. (1992),
been extensively studied in physics. Its fundamen- Boyle and Vorst (1992), Karatzas and Kou (1996), and
tal solution (subject to the boundary condition Broadie et al. (1998).
limx→± ux√ = 0) is the Gaussian density function
ux t = 2 t√−1
exp−x2 /4t with mean 0 and stan- 2.1.4. Delta Hedging. An important function of
dard deviation 2t (see Wilmott et al. 1993, p. 81). options and, more generally, of derivatives, is to pro-
vide the means to hedge given exposures to the
2.1.3. The Black-Scholes Formula. When the pay- underlying sources of risk or given positions in
off is specialized to a call option gS = S − K+ the underlying assets. Conversely, firms dealing in
the valuation equation and boundary conditions options often wish to hedge against the fluctuations
admit the solution inherent in their derivatives positions. The key to
cSt t! K = St e−T −t N dSt ! K T − t immunizing these positions is the delta of the option:
√ The delta is the sensitivity of the option price with
−Ke−rT −t N dSt ! K T − t − T − t respect to the underlying asset price. Simple differen-
(6) tiation shows that the delta of a call is
CSt t
where N · is the cumulative standard normal distri- %c St t ≡ = e−T −t N dSt ! K T − t
S
bution and
Thus, immunizing a call position against fluctuations
dSt ! K T − t in the value of the underlying asset entails shorting

1 S 1 2 %c St t units of the underlying asset. Because delta

= √ log t + r − + T − t depends on the asset price and time to maturity, delta
T −t K 2
hedging is inherently a dynamic process that requires
(7) continuous monitoring and rebalancing of the hedge
This expression is the celebrated Black-Scholes for- over time.
mula for the price of a European call option with Similar derivations establish the formula for the
strike K and maturity date T . That the Black-Scholes delta of a put: %p St t = −e−T −t N −dSt ! K T − t.
Formula (6)–(7) satisfies all the pricing conditions can Immunizing a put position entails a long posi-
be verified by computing the relevant derivatives and tion consisting of %p St t units of the underlying
limits, and substituting in (2)–(3). Alternatively, one asset.
can work with the fundamental solution of the heat Comparison with the formulas of §2.1.2 shows that
equation to deduce CS t. the delta of an option is closely related to the self-
Similar arguments lead to the value of a European financing portfolio used to synthesize a riskless posi-
put option, tion. Simple analysis reveals that delta also represents
√ the number of shares to be held in order to replicate
pSt t! K = Ke−rT −t N −dSt ! K T − t + T − t the option, i.e., the number of shares in the replicating
portfolio. Further perspective on the relation between
−St e−T −t N −dSt ! K T − t (8) delta hedging and the self-financing condition can be
with dSt ! K T − t as defined in (7). The put for- found in Ingersoll (1987, p. 363). Carr and Jarrow
mula can also be derived by a straightforward appli- (1990) provide a nice analysis of a trading strategy,
cation of the put-call parity relationship for European the stop-loss start-gain strategy; it appears to replicate
options on dividend-paying assets. This no-arbitrage the option’s payoff but in fact does not.
condition, which relates call and put options with The sensitivities of option prices with respect to
identical maturities and strike prices, states that other parameters of the model are also of interest
to traders. In particular, the second derivative of the
cSt t! K + Ke−rT −t = pSt t! K + St e−T −t (9) option’s price with respect to the underlying price
(gamma) and the derivatives with respect to time price implied by (11), or equivalently over the distri-
(theta), volatility (vega), and the interest rate (rho) bution of the Brownian motion W ∗ .
have received attention. Formulas for these Greeks are The risk-neutral Formula (11) seems to emerge as
easy to derive from (6) and (8) and can be found in an alternative to the standard present value rule,
standard textbooks such as Hull (2003). which states that the expected payoff of a security
ought to be discounted at a risk-adjusted rate. Risk
2.2. Risk-Neutral Valuation adjustment follows from equilibrium considerations
The fundamental valuation equations (2)–(3) reveal linking the riskiness of an asset to its expected return
that the only market parameters relevant for pricing (or risk premium). Somewhat surprisingly, Formula
derivatives are the interest rate r, the volatility , and (11) appears to obviate the need for adjusting discount
the dividend yield of the underlying asset. Surpris- rates for the purpose of valuing derivatives.
ingly, the expected return does not matter. This 2.2.2. The Equivalent Martingale Measure: Some
property lies at the heart of the second approach to Definitions. Resolution of this puzzle follows by tak-
option valuation, the so-called risk-neutralization pro- ing a closer look at the relationship between the no-
cedure. This method was discovered by Cox and Ross arbitrage and risk-neutralization approaches.
(1976). Its formalization and the identification of some Consider again the price dynamics of the underly-
fundamental underlying principles can be found in ing security (1) in our risk-averse economy and note
Harrison and Kreps (1979) and Harrison and Pliska that it can be expressed in the form
(1981). dSt
= r − dt + dWt + ' dt
2.2.1. A Risk-Neutral Pricing Formula. Suppose St
that the financial market behaves in a risk-neutral where ' ≡ −1 − r. The variable ' is known as
manner with underlying asset price evolving accord- the market price of risk or the Sharpe ratio. It mea-
ing to sures the reward (i.e., the risk premium) per unit
dSt risk. Defining dWt∗ ≡ dWt + ' dt then enables us to
= r − dt + dWt∗ (10)
St write the price evolution in precisely the form (10). Of
where W ∗ is a Brownian motion process. Risk neutral- course, W ∗ is a Brownian motion process with drift
ity implies that the expected total return on the asset ' in this original economy with physical probability
equals the risk-free rate: Et∗ dSt /St + dt = r dt, where measure P .
Et∗ · is the conditional expectation at t with respect At this point, however, one can appeal to the
to the Brownian motion W ∗ . This property is reflected Girsanov change of measure (see Karatzas and Shreve
1998, p. 191) to construct a new measure Q under
in the dynamics (10) of the asset price. Also assume
which W ∗ has the (standard) Brownian motion prop-
that, in other respects, the market structure remains
erty. Specifically, define the process
the same (securities are freely traded and a riskless
asset is available). )t ≡ exp − 12 ' 2 t − 'Wt t ∈ 0 T
In this environment, the no-arbitrage analysis of
and construct the measure dQ ≡ )T dP . An applica-
§2.1 applies and, as can be verified, leads to the same tion of Itô’s lemma shows that ) is a P -martingale
valuation equation. Derivative security prices in our with initial value )0 = 1.2 Combining this with the
risk-neutral economy and in the original risk-averse fact that )T > 0 (P -as) enables us to conclude that
economy will therefore coincide! Moreover, an appli- Q is a probability measure, and that Q is equivalent
cation of the Feynman-Kac formula (see Karatzas and to P .3 It is referred to as the equivalent martingale
Shreve 1988, p. 366) shows that measure (EMM). The random variable )T represents
the Radon-Nikodym derivative of Q with respect to P
V St t = e−rT −t Et∗ gST (11) (i.e., )T = dQ/dP ). Because )T depends on the mar-
ket price of risk, the measure Q can be interpreted
solves (2)–(3). It follows that the derivative’s price is as a risk-adjusted probability measure. Girsanov’s
simply the expected payoff discounted at the risk-free theorem can then be invoked to assert that W ∗ is a
rate, as should be the case with risk neutrality. Q-Brownian motion.4
These observations motivate the second approach
to the valuation of derivatives, the risk-neutralization 2
The process ) is a P -martingale if and only if Et )v = )t for all
procedure, which involves two steps. The first step t ≤ v, where Et · is the expectation with respect to P .
consists of risk neutralizing the underlying price pro- 3
The measures Q and P are equivalent if they have the same null
cess, as in (10), by replacing the expected return sets.
with the risk-free rate r. The second step con- 4
Alternatively, this can also be verified by computing the charac-
sists of computing the price from (10) by taking the teristic function of W ∗ under Q, given by
∗ ∗ ∗
expectation over the distribution of the underlying E ∗ ei/Wt = Eei/Wt )T = Eei/Wt )t
2.2.3. The Equivalent Martingale Measure: Asset Finally, it bears pointing out that the risk-neutral
Price Representations. The developments above valuation Formula (12) for the underlying asset can
show that W ∗ is a Brownian motion under Q. Let us also be restated in terms of expectations with respect
now describe the behavior of prices under the new to P . Indeed, changing the measure yields the alter-
measure. Another straightforward application of Itô’s native representation
lemma shows that T
T T St = Et 0t T ST + 0t v Sv dv (13)
e−rT −t ST + e−rv−t Sv dv = St + e−rv−t Sv dWv∗ t
t t
−rT −t
where 0t T ≡ e )T /)t . The quantity 0v ≡ 00 v is
for all t ∈ 0 T . Taking expectations with respect known as the state price density. The Arrow-Debreu
to Q, on both sides of this equality, establishes the price at date 0 of a dollar received at date v in state 1
formula equals 0v dP 1. Conditional Arrow-Debreu prices at
T date t for cash flows received at v are given by
St = Et∗ e−rT −t ST + e−rv−t Sv dv (12) 0t v dPt 1. Arrow-Debreu prices are also known as
t
state prices (see Debreu 1959). The present value of
This expression shows that the asset price equals the underlying asset (13) is the sum of cash flows mul-
the expected value of the discounted dividends aug- tiplied by state prices.
mented by the expected value of the discounted ter- 2.2.4. Martingale Representation and No-Arbitrage
minal price. Expectations are computed under the Pricing. Let · be the filtration generated by the
EMM Q. The discount factor is evaluated using the Brownian motion W .5 The martingale representation
risk-free rate. The formula therefore suggests that theorem—hereafter MRT—(see Karatzas and Shreve
the asset is priced, under Q, as if the market were risk 1988, p. 188) states that any random variable B that
neutral. This is why the EMM is often called the risk- is measurable with respect to T and is almost surely
neutral measure. This label should be used with some finite can be written as a sum of Brownian increments.
caution, however, because (as pointed out above) the More formally,
EMM is effectively adjusted for risk. Recall that the
T
standard present-value formula computes expected
B = EB + 3s dWs (14)
cash flows under the statistical (i.e., real-world) mea- 0
sure P and discounts those at a risk-adjusted discount
for some process 3 that is adapted and square inte-
rate. Our alternative Formula (12) proceeds the other
grable (P -a.s.).6 If the stochastic integral in (14) is a
way: It first corrects probabilities for risk and cal-
martingale we can also take conditional expectations
culates a risk-adjusted expected cash flow, then dis- t
and get Et B = EB + 0 3s dWs for all t ∈ 0 T . This
counts those at the risk-free rate.
holds, in particular, if B has a finite second moment,
It is also important to note that the discounted stock
EB 2 < (i.e., B ∈ L2 ). In that instance the stochastic
price augmented by the discounted value of divi-
integral also has finite second moment and
dends is a martingale under the EMM. This can easily
be seen by rewriting (12) in the following manner: t 2 t
E 3s dWs =E 32s ds < !
t T 0 0
e−rt St + e−rv Sv dv = Et∗ e−rT ST + e−rv Sv dv
0 0 in addition, 3 is unique (see Karatzas and Shreve
1988, p. 185 and p. 189).
This property explains why Q is referred to as an When applied to the deflated payoff of a derivative
equivalent martingale measure. security such that 0T gST ∈ L2 the MRT gives
1 2
t
= Eei/−'Wt +i/'− 2 ' t Et 0T gST = E0T gST + 3s dWs (15)
1 2 t+i/'− 1 ' 2 t 1 2t 0
= e 2 i/−' 2 = e− 2 /
t
for all t ∈ 0 T . In this sequence of equalities the first follows for some adapted process 3 such that E 0 32s ds < .
from the passage to the measure P , the second uses the martin- Suppose now that we invest an initial amount x ≡
gale property of ), and the fourth the distributional properties of E0T gST and follow the portfolio policy consisting
the Brownian motion W under P and the moment-generating func-
tion of a normal distribution. Other equalities involve standard
5
manipulations. The final expression corresponds to the characteris- The filtration generated by W is the increasing collection of sigma-
tic function of a normal distribution with mean zero and variance t. algebras t 6 t ∈ 0 T where t = − Ws 6 s ∈ 0 t, i.e., t is the
Combining this distributional property with the continuity of the collection of possible trajectories of the Brownian motion at t.
6
sample path of W ∗ , the initial condition W0∗ = 0, and the indepen- The process 3 is said to be adapted to the filtration · if 3t is
dence of increments establishes the Brownian motion behavior of t -measurable
T for all t ∈ 0 T . The process 3 is square integrable
W ∗ under the risk-adjusted probability measure Q. (P -a.s.) if 0 32s ds < , P -as
of an investment Xt t = Xt −1 ' +0t−1 −1 3t in the risky where are progressively measurable pro-
asset and the complement Xt 1 − t in the riskless cesses and is invertible.8 In this setting W is a
asset, for t ∈ 0 T . Then, by Itô’s lemma, we obtain7 d-dimensional Brownian motion, is a d × 1 vec-
tor of appreciation rates (expected total returns), is
d0t Xt = 0t Xt t − ' dWt = 3t dWt ! a d × 1 vector of dividend rates, and is a d × d
subject to 00 X0 = X0 = x matrix of volatility coefficients. This class of models
includes Markovian diffusion models with stochastic
Writing this differential equation in integral form interest rates, stochastic dividends, and or stochastic
yields volatilities. It includes deterministic volatility func-
t tion models proposed in Dupire (1994), Derman and
0t Xt = x + 3s dWs = Et 0T gST Kani (1994), and Rubinstein (1994). An important lim-
0 itation is that all processes are driven by the hedge-
where the second equality follows from the defini- able multidimensional Brownian motion W ; therefore,
tion x ≡ E0T gST and the martingale representation models with nonhedgeable factors are excluded.
of the discounted payoff (15). In particular, for t = T Under the assumption that the market price of risk
we see that 0T XT = 0T gST , which means XT = gST : 't ≡ t−1 t − rt 1 is bounded, the density process
Our selected portfolio duplicates the derivative’s pay-
1 t t
off at maturity. The no-arbitrage principle can then be )t ≡ exp − 'v 'v dv − 'v dWv t ∈ 0 T
invoked to conclude that the claim’s price must be the 2 0 0
same as the portfolio value. That is, (17)
V St t = Xt = 0t−1 Et 0T gST = Et 0t T gST is a P -martingale with initial value )0 = 1. The mea-
sure Q, such that dQ ≡ )T dP , is well defined and
for all t ∈ 0 T . Any breakdown in this relation, at has the same properties as in §2.2.3. It represents the
any point in time, means that two traded assets with EMM or risk-neutral measure in our more general
identical payoffs have different prices, which implies market. The process
the existence of an arbitrage opportunity.
These arguments establish that derivative securi- t
ties, in the BSM market, satisfy the same type of valu- Wt∗ ≡ Wt + 'v dv t ∈ 0 T
0
ation formulas as those characterizing the underlying
asset. In effect, we can write is a d-dimensional Q-Brownian motion and the asset
price S has the representations
V St t = Et 0t T gST = e−T −t Et∗ gST
T
St = Et∗ bt T ST + bt v Sv v dv
where we recall that Et∗ ·
is the expectation under the t
risk-neutral measure Q. As before, we see that dis- T
counted prices are Q-martingales, e−rt V St t = e−rT · = Et 0t T ST + 0t v Sv v dv (18)
t
Et∗ gST . We also observe that deflated prices are
P -martingales, 0t V St t = Et 0T gST . under the measures v Q and P , respectively. The quan-
2.2.5. Risk-Neutral Pricing with Itô Price Pro- tity bt v = exp− t ru du is the discount factor at the
cesses: Some Generalizations. The principles out- locally riskless rate, and 0t = b0 t )t ≡ bt )t is the appli-
lined above, namely the change of measure and the cable state price density.
representations of underlying and derivatives’ prices, A multidimensional version of the arguments in
remain valid in the context of more general markets §2.2.4 can again be invoked to yield no-arbitrage rep-
with stochastic interest rate and Itô price processes. resentations of derivatives prices. For a derivative
Specifically, suppose that the interest rate r is a pro- security generating a continuous payment ft , per unit
gressively measurable process and that the market is time, at date t and a terminal cash flow YT we obtain
comprised of d risky securities whose evolution is the valuation formula
governed by the stochastic differential equation (SDE) T
Vt f Y = Et∗ bt T YT + bt v fv dv
dSt t
= t − t dt + t dWt (16) T
St = Et 0t T YT + 0t v fv dv (19)
t
7
Itô’s lemma gives d0t Xt = 0t dXt + Xt d0t + d0 Xt where d0 Xt
8
is the cross-variation of the processes 0 and X. Substituting dX d0 A process x is said to be progressively measurable if xt is measur-
and d0 Xt = −0t Xt t ' dt, and rearranging gives the result able with respect to the product sigma-algebra t ⊗ 0 t for all
stated. t ∈ 0 T , where 0 t is the Borel sigma-field on 0 t.
where the expectations are taken with respect to Q on that interval. The fundamental quantity, aggregate
and P , respectively. consumption, is assumed to follow an Itô process
The principles presented in this section and §2.2.4
are either special cases or bear relationship to the fun- dCt = Ct ct dt + tc dWt
damental theorem of asset pricing (FTAP). In essence,
where W is a d-dimensional Brownian motion, c
the FTAP states that a market is arbitrage free if and
is a progressively measurable process representing
only if there exists a probability measure Q under the expected consumption growth rate, and c is a
which (discounted) security prices are martingales. 1 × d vector of progressively measurable processes
The probability measure Q is the equivalent martin- capturing the volatility exposures of the consumption
gale measure. Harrison and Kreps (1979) show that growth rate with respect to the various sources of risk.
market completeness, as defined there (see also §2.3), Financial markets are composed of primary and
is equivalent to the existence of a unique EMM. The derivative assets. Primary assets include n risky
EMM clearly relates to a change of numeraire. When stocks and one riskless asset. The riskless asset is
the EMM as defined above exists, prices expressed in zero net supply and pays interest at the rate r,
in units of the bond (i.e., in the bond numeraire) are a progressively measurable process. Each stock is in
Q-martingales. As discussed in more detail in §3.6, unit supply (one share outstanding) and pays divi-
prices can also be expressed in other numeraires and dends. Stock j pays Dj per unit time, where dDt =
j
will retain the martingale property relative to a suit- j j j j j

Dt t dt + t dWt with progressively measur-
ably adjusted probability measure. able. Assuming that dividends are the only source
Original versions of the theorem were established of consumption mandates the consistency condition
by Harrison and Kreps (1979), Harrison and Pliska n j
j=1 Dt = Ct . Derivative assets are in zero net supply
(1981), and Kreps (1981). Versions of the theorem, as j
and consist of k securities with payoffs f j YT 6 j =
well as extensions to more-general contingent claims j
1 k, where f is a continuous, progressively mea-
spaces or to more-general models with frictions can j
surable payment and YT a terminal, measurable cash
be found in a variety of papers. References include flow. No specific restrictions are placed on n d, and k.
Dybvig and Ross (1987), Delbaen and Schachermayer Hence n + k < d is an admissible market structure.
(1994), Jacod and Shiryaev (1998), Pham and Touzi Stock prices S j 6 j = 1 n and derivative prices
(1999), and Kabanov and Stricker (2001). V j 6 j = 1 k are conjectured to satisfy Itô pro-
cesses whose evolution is described by
2.3. Incomplete Markets
j j j s j s j
A critical aspect of the models described in prior sec- dSt + Dt dt = St t dt + t dWt j = 1 n (20)
tions is the ability to hedge all risks with the existing j j j d j d j
dVt + ft dt = Vt t dt + t dWt
menu of assets. This property is called market com-
j
pleteness. When the underlying source of risk consists VT = YT ! j = 1 k (21)
of a d-dimensional Brownian motion, market com-
pleteness can be ensured if d + 1 primary securities, with progressively measurable coefficients s j
namely d risky assets and one locally riskless asset, s j j=1n and d j d j j=1k . These coefficients, as
are freely traded (see Duffie 1986). In §2.3 and its well as the rate of interest, are endogenous in equilib-
subsections we examine valuation in market settings rium. All assets, primary and derivatives, are freely
where not all risks can be eliminated. traded at the relevant prices.
The economy has a representative agent who max-
2.3.1. Effectively Complete Markets. A simple imizes welfare by consuming and investing in the
approach to valuation, when risks cannot all be assets available. The consumption space is the space
hedged, is to resort to a general equilibrium analysis. of progressively measurable, nonnegative processes.
Pioneered by Lucas (1978) and Cox et al. (1985) this Preferences over consumption processes, denoted
approach rests on the notion that the economy, in the by c ≡ ct 6 t ∈ 0 T , are represented by the von
aggregate, behaves like a single, well-defined agent. Neumann–Morgenstern utility
Analysis of this representative agent’s behavior com- T
bined with market-clearing conditions lead to equilib- U c ≡ E uct t dt (22)
rium values for the interest rate, the market price of 0
risk, and the prices of primary and derivative assets. where uct t is the utility of consumption at time t.
For exposition purposes we sketch a continuous The utility function is assumed to be strictly increas-
time version of the Lucas pure exchange economy ing and concave in consumption and to satisfy
model. The economy has a finite time horizon 0 T : the limiting (Inada) conditions limc→0 u c t = ,
All processes described below are understood to live limc→ u c t = 0, for all t ∈ 0 T , where u c t is
the derivative with respect to consumption. Invest- Itô’s lemma to both sides of (24 ). This yields
ment policies are progressively measurable processes
u Ct t/t
denoted by ≡ s d ≡ ts td 6 t ∈ 0 T rt = − + RCt tct
where s represents the vector of fractions of wealth u Ct t
invested in stocks and d the vector of fractions in 1
derivatives. The complement 1 − s + d 1 is the − RCt tP Ct t tc tc (27)
2
fraction in the risk-free asset (1 is a vector of ones 't = RCt t tc (28)
with suitable dimension). For a given consumption-
portfolio policy c , wealth X evolves according to where RC t = −u C tC/u C t is the relative
the dynamic budget constraint risk-aversion coefficient and P C t = −u Ct tCt /
dXt = rXt − ct dt + Xt ts st − rt 1 dt + ts dWt u Ct t the relative-prudence coefficient.
The valuation Formulas (25)–(26) issued from this
+ Xt td dt − rt 1 dt + td dWt (23) general equilibrium analysis have the same structure
subject to some initial condition X0 = x. The represen- as those prevailing in the complete market settings of
tative agent maximizes (22) over policies c satis- earlier sections. In light of this structural property, the
fying the budget constraint (23) and the nonnegativity representative agent model is said to have effectively
constraint Xt ≥ 0, for all t ∈ 0 T . Policies that solve complete markets.
this maximization problem are said to be optimal for Formulas (24)–(28) and their extensions to settings
U at the given price processes S V r. with discontinuous price processes (see, e.g., Naik
To close this model we need to specify a notion of and Lee 1990), have been used in a variety of con-
equilibrium. A competitive rational expectations equi- texts involving the valuation of primitive as well as
librium is a collection of price processes S s s derivative assets. The formulas are especially useful
V d d r and consumption-portfolio policies in settings with stochastic volatility, stochastic inter-
c such that c is optimal for U at the price est rates, and jump risk, where all the risks cannot be
processes S V r and markets clear. Market clearing hedged away. In those instances pricing based on the
mandates c = C (consumption good market), X s = representative agent paradigm provides a simple and
S (stock market), X d = 0 (derivatives market), and often attractive approach to valuation.
X1 − s + d 1 = 0 (riskless asset market). 2.3.2. Incomplete Markets: Private Values and
Under suitable conditions, standard arguments can Certainty-Equivalents. There are situations, how-
be invoked (see, for instance, Karatzas and Shreve ever, in which a representative agent analysis might
1998) to show that optimal consumption satisfies the not be adequate. A good example is the valuation of
first-order condition u ct t = y0t , where y is a con- executive stock options (ESOs). These securities are
stant and 0 is the relevant state price density implied granted to executives as part of their compensation
by the given price processes S V r (recall that packages, to provide incentives for them to act in the
0t = bt )t with ) as in (17)). Equilibrium in the goods best interest of shareholders. ESOs cannot be traded in
market then gives the condition u Ct t = y0t . Since organized markets and carry restrictions on the abil-
00 = 1 it must be that y = u C0 0 and therefore ity to trade the underlying asset (grantees, typically,
u Ct t cannot short the stock of the firm). In essence they
0t = (24) represent nontraded assets that are in positive supply
u C0 0
and cannot be fully hedged.
The equilibrium state price density is therefore equal
For those types of situations valuation becomes an
to the marginal rate of substitution between consump-
individual matter. An ESO, for instance, has value for
tions at time t and time 0. Moreover, the same argu-
the executive holding it because it affects his or her
ments that were used in prior sections can be applied
private consumption in the future. To the extent that
in this context to establish the price representations
it does not materially impact the rest of the market, it
T T
j
St = Et∗ bt v Dvj dv = Et 0t v Dvj dv (25) will have little relevance for prices in general and for
t t the policies pursued by other investors. The execu-
T tive’s trading strategy and consumption choices, how-
j j
Vt f Y = Et∗ bt T YT + bt v fvj dv ever, could be significantly affected.
t
T In order to value securities that are privately held
j
= Et 0t T YT + 0t v fvj dv (26) one can resort to the notion of certainty-equivalent
t (or private value) discussed by Pratt (1964). The
v j
where bt v = exp− t ru du (by convention ST = 0 certainty-equivalent of a claim f Y is the sure cash
because the economy lives on 0 T and ceases to pay amount V f Y that leaves the individual indifferent
dividends at T ). The equilibrium interest rate r and between holding the claim to maturity or receiving
the market price of risk ' are obtained by applying the said cash amount.
Before proceeding with the continuous time market J x + V f Y 0 0 = J x f Y . If the value function
setting, it is useful to visualize the notion in a sim- J · 0 0 is invertible the solution is
ple one-period context. Accordingly, suppose that the
investor cares about terminal wealth, which consists f Y = J −1 J x f Y 0 0 − x
V
exclusively of the risky payoff Y . By definition the
certainty-equivalent of the risky cash flow Y is the The certainty-equivalent value is a private value to

sure amount V Y such that the extent that it depends on the preferences of the
individual holding the asset and exposed to the non-
Y = EuY
uV traded risks. It represents the ask price for this partic-
ular individual.10
Equivalently, V . Simple properties
Y = u−1 EuY The practical computation of the certainty equiv-
of this certainty-equivalent can be readily established. alent is arduous due to the difficulty in resolving
Under the standard assumption of a concave and the optimization problem with the nontraded asset
increasing utility function, V Y is increasing in the (a constrained optimization problem). Problems of
expected cash flow and decreasing in the riskiness of this nature are analyzed by Duffie et al. (1997),
the cash flow. Cuoco (1997), Cvitanić et al. (2001), Karatzas and
In a continuous time setting, the investor holding Zitkovic (2003), and Hugonnier and Kramkov (2004),
the nontraded asset consumes and invests in traded in continuous time settings. Implementation of the
securities. Assume the Itô market of §2.2.5 and sup- certainty-equivalent method is easier in some discrete
pose that the investor derives utility from consump- time settings. For results in the binomial framework
tion over time as well as from terminal wealth. Also and an application to ESO valuation, see Detemple
suppose that the nontraded asset produces cash flows and Sundaresan (1999).
f Y adapted to the filtration generated by W Z, Several other approaches have been proposed to
where Z is a Brownian motion orthogonal to W . deal with incomplete markets and unspanned risks.
The individual’s optimization problem becomes A popular notion, which appears in several con-
T texts, is to use the minimal martingale measure 0 o
max E uct t dt + BXT for pricing purposes. In essence, this measure assigns
c 0 null market price to risks that cannot be hedged
subject to a dynamic budget constraint with the menu of traded assets. Derivative prices
satisfy the usual present value Formula (19), substi-
dXt = rXt + ft − ct dt + Xt t t − rt dt + t dWt tuting 0 o for 0. The method is consistent with the
dynamic consumption-portfolio choice problem of a
t ∈ 0 T ! X0 = x XT = XT − + YT
myopic individual, with logarithmic utility function.
where u· t is utility of consumption at time t The resulting price can be interpreted as a price at the
and B· is the utility of terminal wealth (bequest margin calculated using the marginal rate of substitu-
function). As usual, this optimization is subject to tion of that particular individual. Originally proposed
the liquidity constraint X ≥ 0. This constraint man- by Föllmer and Sondermann (1986) and Föllmer and
dates a nonnegative portfolio value, i.e., the indi- Schweizer (1991), the method has since been gen-
vidual cannot borrow against the future cash-flows eralized or applied in a number of papers (e.g.,
generated by the nontraded asset.9 Let J x f Y be Schweizer 1992, 1995a). Other pricing measures, such
the value function associated with this optimization as the minimal variance measure (Schweizer 1995b,
problem. Delbaen and Schachermayer 1996) and the mini-
Alternatively, suppose that f Y is exchanged, at mal entropy measure (Fritelli 2000) have also been
date zero, against a sure cash amount V f Y . The studied.
new optimization problem faced by the agent has the
same structure as the one just described, but with 3. American Options
f Y = 0 0 and X0 = x + Vf Y . The associated
American-style derivatives can be exercised at any
f Y 0 0.
value function is J x + V point in time before maturity. As a result, part of the
The certainty-equivalent V f Y is now easily valuation problem consists of identifying the optimal
defined. It is the cash amount that leaves the indi- exercise policy, i.e., the exercise time that maximizes
vidual indifferent between holding on to the claim value for the holder of the security.
or giving it up, i.e., the solution of the problem
10
This definition assumes that the claim is held to maturity. Varia-
9
One could weaken this restriction by allowing for borrowings tions of the concept could allow for liquidation at a selected set of
against the hedgeable components of these cash flows. dates and for divisibility of the claim in a finite number of lots.
This section reviews the various approaches to Itô’s lemma applies and the arguments leading to (2)
value American-style contracts. We survey the free- can be invoked. The derivative’s price satisfies (2)
boundary approach (§3.1), the variational inequalities in .
method (§3.2), and the risk-neutralization procedure Characterizing the immediate exercise boundary is
(§3.3). Useful representation formulas for the values more difficult. Of course, for S t ∈ B immediate
of those derivatives are reviewed last (§3.4). exercise is optimal, so V Bt t = gBt. This con-
dition, however, is not enough to fully determine
3.1. The Free-Boundary Approach B. Indeed, one could specify any arbitrary function
The free-boundary method goes back to Samuelson Bt and simply solve the valuation partial differen-
(1965), McKean (1965), Taylor (1967), and Merton tial equation (PDE) subject to that boundary condi-
(1973). It applies to Markovian settings such as the tion. This would only produce different functions V
Black-Scholes framework of §2.1. An important ele- corresponding to different contractual specifications
ment is the observation that the arguments of §2.1.2 (i.e., exercise policies). To solve the pricing problem
apply, even when the claim under consideration is under consideration one must therefore add a condi-
American style. The fundamental valuation equation tion reflecting the optimality of exercise along B.
therefore characterizes its price, provided the contract No-arbitrage arguments can be used for that pur-
is alive. In the complementary event, where the claim pose. Suppose that the derivative of the option price,
is exercised, its value must equal the exercise payoff. at Bt, does not coincide with that of the payoff,
Let V S t be the price of an American deriva- i.e., V Bt t/S = gBt. If so, one can show that
tive with payoff gS at exercise. Because exercise an arbitrage opportunity exists. Indeed, consider a
is at the option of the holder we know that it is sequence S n ∈ that converges to Bt and such that
suboptimal to exercise if gS < 0. Attention can limn→ V S n t/S > g Bt. Then, for n large
therefore be restricted to securities with nonnegative enough (i.e., S n ∈ and close to Bt), it must be
payoffs, gS ≥ 0. Assume that the payoff function that V S n t < gS, in contradiction with the defi-
g· is continuous, continuously differentiable almost nition of (equivalently, an arbitrage opportunity).
everywhere, and twice continuously differentiable on The existence of an arbitrage opportunity can also
S6 gS > 0.11 The continuation region, ≡ S t ∈ be established if g Bt > limn→ V S n t/S. An
+ × 0 T 6 V S t > gS, is the set of point S t informal argument is provided by Wilmott et al.
at which the derivative is worth more alive. Its com- (1993).12 The combination of these two results gives
plement is the exercise region, ≡ S t ∈ + × the condition V S t/S = g S at the point S = Bt.
0 T 6 V S t = gS. Under the weak assumption This condition, mandating that the price and pay-
that the derivative’s price is continuous with respect off derivatives match along the immediate exercise
to its arguments, the continuation region is an open boundary, is known as the high contact or smooth
set and the exercise region a closed set. The boundary pasting condition (see, e.g., Ingersoll 1987, p. 374 or
of , denoted by B, is then part of the exercise region Dixit and Pindyck 1994, p. 130).
(i.e., B ∈ ) and serves to distinguish points where In summary, the price of the derivative solves the
exercise is optimal from those where continuation is fundamental valuation Equation (2) subject to the
best. This set is called the immediate exercise bound- boundary conditions
ary. Its t-section, Bt ≡ S ∈ + 6 S t ∈ B, is the set

of boundary points at time t. Note that Bt, for gen-  V S T = gS on +


eral payoff function g·, might not be single valued: 


 V 0 t = g0 for t ∈ 0 T
The exercise region can, in principle, have lower and
upper boundaries, or can even be the union of discon- V S t = gS for S t = Bt t (29)


nected intervals. The immediate exercise boundary is 


the collection of its t-sections: B = Bt6 t ∈ 0 T .  V S t = g S for S t = Bt t

The identification of B is key to the resolution of the S
valuation problem. Because B is not known a priori it The system (2), (29) is the free-boundary characteriza-
is often referred to as a free boundary. The American tion of the pricing function.
claim valuation problem is a free-boundary problem.
When the option is alive, in , it makes intu- 3.2. Variational Inequalities
itive sense to assume that the price is smooth, pre- Free-boundary problems are often difficult to solve
cisely that V S t ∈ 2 1 . The same assumption was because of the need to identify the unknown bound-
invoked for European claims. Under this assumption ary in the resolution process. An alternative is to
11 12
These are weak assumptions, satisfied by standard option con- Another informal proof, based on the notion that Bt is optimal,
tracts such as puts and calls. is found in Merton (1992).
recast the problem in terms of a variational inequality. The variational inequality formulation is obtained
This transformation is attractive because the inequal- by integrating and relaxing the previous conditions
ity does not explicitly depend on the boundary. The (30)–(31). Specifically, let be a set of test functions
pricing function can be obtained directly from this consisting of functions vx that are continuous,
characterization. The boundary can then be identified continuously differentiable in , differentiable almost
from this solution. everywhere in x, and that satisfy the boundary con-
To understand the variational inequality formula- ditions (32) and the inequality constraint vx ≥
tion it might be useful to first express the pricing ĝx for all x ∈ ×0 12 2 T . Note that the solu-
problem in linear complementarity form. Recall that tion of our problem, the function u, belongs to .
the pricing function can be recovered from the solu- Moreover, for any v ∈ we have vx ≥ ĝx and
tion, denoted by u, of the heat Equation (4) and its therefore ux vx − ĝx ≥ 0, with equality
corresponding boundary condition (5). To simplify if vx = ĝx . Integrating over x yields13
notation, write
ux vx − ĝx dx ≥ 0 (33)
u 2 u −
ux ≡ −
x2
ux ux − ĝx dx = 0 (34)
−
and note that is the Itô differential operator
associated with a time-reversed Brownian motion Taking the difference between (33) and (34) eliminates
x ≡ x 6 ∈ 0 12 2 T : It represents the drift of the the function ĝx and gives
function ux as time moves backward (i.e., as
the rescaled time-to-maturity measure = 12 2 T − t ux vx − ux dx ≥ 0
increases). In the continuation region where −
ux > ex+ gK expx ≡ ĝx the price satisfies Integration by parts now enables us to get rid of sec-
ux = 0. In the exercise region ux = ĝx , ond derivatives14
and therefore u
0≤ vx − ux dx
ux = ĝx −

u
Moreover, because ĝx represents the local gains − vx − ux
x
from exercising, it cannot be negative when exer- −
u
cise is optimal: This gives ux = ĝx ≥ 0. In
+ v x − u x dx
summary, − x
ux ux − ĝx = 0 (30) v u ∈ , thus they satisfy the same boundary condi-
tions at x = ± and therefore, after collecting terms,
ux − ĝx ≥ 0 and ux ≥ 0 (31) u

0≤ vx − ux dx
The problem of finding a continuous and almost
−
everywhere differentiable function ux that satis- u
fies (30)–(31) subject to the boundary conditions + v x − u x dx (35)
− x


 ux 0 = ex gK expx for x ∈ The variational inequality problem is to find u ∈

 such that (35) holds for all t ∈ 0 T , for all test func-
lim e−x− ux = g0 for ∈ 0 21 2 T
 x→− tions v ∈ . One could also go an extra step by inte-


 lim e−x− ux = g for ∈ 0 1 2 T grating over time, to produce a global variational
2
x→ inequality.
(32) The variational inequality approach for pricing
American claims is developed in Jaillet et al. (1990).
is called the linear complementarity problem (LCP). Mathematical treatises on the method can be found
Note that the unknown exercise boundary corre- in Bensoussan and Lions (1978) and Kinderlehrer and
sponding to the problem expressed in terms of x does Stampacchia (1980).
not appear in (30)–(32). The LCP can be interpreted as
an obstacle problem where a function, whose uncon- 13
Note (30)–(31) holds if and only if (33)–(34) holds for u ∈ and
strained evolution is governed by the fundamental for all v ∈ .
valuation equation, must exceed an obstacle given by 14
A consequence is that the existence of a second derivative is not
the exercise payoff and meet conditions at the bound- needed. The variational inequality approach applies even if the
aries of its domain. payoff is only required to be differentiable almost everywhere.
3.3. The Risk-Neutralization Approach for all t ∈ 0 0 and for some progressively mea-
The risk-neutralization principle described in §2.2 also surable process 3. The arguments in §2.2.4 can then
applies to American claims. To develop this point let be used to show that the portfolio X ≡ Xt t =
us adopt the general framework with Itô processes of Xt −1 ' + 0t−1 −1 3t 6 t ∈ 0 0 has value Xt =
§2.2.5 and suppose that the claim under consideration Et 0t 0 Y0 . X0 = Y0 , therefore this policy repli-
has an exercise payoff given by a progressively mea- cates the claim’s payoff at the exercise time 0. Its
surable process Y . In this context, let t T be the initial cost is X0 = E00 Y0 . A claim paying off Y0
set of stopping times of the filtration with values in at 0 must therefore be worth
t T and define the process
V0 = E00 Y0
Zt ≡ sup Et∗ b Y (36)
∈ t T at inception, to preclude arbitrage opportunities. But
by definition 0 maximizes E0 Y (recall that
where we recall that b ≡ exp− 0 rv dv is the dis- Z0 ≡ sup∈ 0T E ∗ b Y = sup∈ 0T E0 Y ). It can
count factor at the locally risk-free rate and Et∗ · is therefore not be improved upon: 0 represents the
the conditional expectation at date t, under the risk- optimal exercise policy for the American claim issued
neutral measure Q defined earlier. The process Z is at t = 0 and V0 is its market (no-arbitrage) value.
known as the Snell envelope of the discounted payoff Similarly, t can be interpreted as the optimal exer-
b Y . It is the result of the maximization described on cise time for the American claim issued at time t; its
the right-hand side of ( 36). The solution of this maxi- market value is Vt = Et 0t t Yt , which can also be
mization is a stopping time, indexed by time, denoted expressed as Vt = Et∗ bt t Yt under the risk-neutral
by t. The Snell envelope is the value function of measure.
this maximization. These developments show that valuation principles
The Snell envelope has several attractive properties for American claims are fully consistent with those
(see El Karoui 1981 and Karatzas and Shreve 1998 for established for European-style contracts. The value of
details). For one, it is a Q-supermartingale because15 a claim is simply the present value of its exercise pay-
off, where present value is calculated by discount-
Et∗ Zs = Et∗ sup Es∗ b Y ing cash flows at the locally risk-free rate and taking
∈ s T
expectations under the risk-neutral measure. In this
= Et∗ bs Ys ≤ sup Et∗ b Y = Zt calculation the exercise payoff is at the optimal exer-
∈ t T
cise time, i.e., the stopping time that maximizes value.
where the first equality on the second line follows Alternatively, the price can be written as the expected
from the definition of s, i.e., the solution of (36) deflated value of the exercise payoff where the expec-
at time s, and the law of iterated expectations, and tation is under the original measure and the (stochas-
the inequality on the same line from the inclusion tic) deflator is the state price density.
s T ⊂ t T , for t ≤ s. The last equality is The theory of optimal stopping has a long history.
simply the definition of Z. The Markovian case is considered by several authors,
It also has paths that are right continuous with left including Fakeev (1971), Bismut and Skalli (1977),
limits RCLL and is nonnegative because exercise is and Shiryaev (1978). An extensive treatment for gen-
at the option of the holder. Immediate exercise at t eral stochastic processes can be found in El Karoui
is always feasible, thus it must also be that Zt ≥ bt Yt : (1981). Karatzas and Shreve (1998, Appendix D) con-
The Snell envelope majorizes the discounted payoff. tains detailed results for continuous time processes.
In fact, it is the smallest supermartingale majorant of The connection between the Snell envelope and no-
the discounted payoff. arbitrage pricing is drawn by Bensoussan (1984) and
To see the relation with our pricing problem it suf- Karatzas (1988).
fices to invoke a no-arbitrage argument. Let 0 be
the solution at t = 0. An application of the martingale 3.4. Exercise Premium Representations
representation theorem (assuming 00 Y0 is almost The arguments outlined in §3.3 provide a theoreti-
surely finite) gives cal resolution of the valuation problem for American
claims. Several transformations of the pricing for-
t mula, however, prove useful for interpretation pur-
Et 00 Y0 = E00 Y0 + 3v dWv poses and practical implementation.
0
The first transformation leads to a formula known
15
A process Z ≡ Zt 6 t ∈ 0 T is a supermartingale if Et Zs ≤ Zt ,
as the early exercise premium (EEP) representation of
P -a.s., for all s ≥ t and t ∈ 0 T . It is a Q-supermartingale if the the claim’s price. The key to this formula is the behav-
expectation is taken with respect to the measure Q. ior of the Snell envelope Z. In the event v = v,
where stopping is optimal, the Snell envelope is the The first term is the immediate exercise payoff. The
discounted payoff Zv = bv Yv . Therefore, by Itô’s second is the present value of the gains from delaying
lemma, exercise, the negative of the gains from exercising (see
the structure of the EEP above). The second compo-
dZv = bv dYv − rv Yv dv for v = v
nent can also be interpreted as the time value of the
On the complementary event v > v, the formula American claim; the immediate exercise payoff is the
Zv ≡ Ev∗ bv Yv shows that Z is a Q-martingale, i.e., intrinsic value.
dZv = 3∗v dWv∗ for some progressively measurable pro- The EEP representation corresponds to the Riesz
cess 3∗ . Combining these two elements shows that decomposition of the Snell envelope. This decompo-
T sition states that any supermartingale is the sum of a
Z T = Z0 + dZv martingale and a potential.17 The EEP representation
0 essentially identifies the structure of the processes in
T
this decomposition in terms of the underlying param-
= Z0 + 1v=v bv dYv − rv Yv dv
0 eters of the model and the payoff process. The Riesz
T decomposition of the Snell envelope was established
+ 1v>v 3∗v dWv∗ by El Karoui and Karatzas (1991) for a setting with
0
Brownian motion. The EEP formula for American
Taking expectations (under Q) and rearranging them options, in the Black-Scholes framework with constant
yields coefficients, is due to Kim (1990), Jacka (1991), and
T Carr et al. (1992). An extension to payoffs adapted
Z0 = E ∗ ZT + E ∗ 1v=v bv rv Yv dv − dYv to a general filtration is given in Rutkowski (1994)
0
and to jump-diffusion processes in Gukhal (2001). The
where ZT ≡ bT YT is the discounted terminal payoff.
DEP formula above extends the corresponding for-
Now note that the left-hand side of this expression is
mula in Carr et al. (1992) derived for the Black-Scholes
the initial price V0 of the American claim. The first
framework.
term on the right-hand side is the initial price of its
European counterpart. The second term on the right 3.5. Integral Equation for American Options
is called the EEP. It captures the present value of the For American call options immediate exercise will
gains associated with exercise prior to the maturity only occur if S > K. Yv = Sv − K, thus the local gains
date. These gains are interest-rate gains, realized by from exercise, in the exercise region, equal rv Yv dv −
investing the exercise payoff at the risk-free rate, net Ev∗ dYv = v Sv − rv K dv and the EEP formula for the
of the opportunity cost associated with the forgone call price is given by
appreciation of the payoff.
The same argument yields, for any time t ∈ 0 T , Ct = Et∗ bt T ST − K+
T T
Vt = Et∗ bt T YT + Et∗ 1v=v bt v rv Yv dv − dYv + Et∗ 1v=v bt v v Sv − rv K dv
t t
(37)
The net local benefits from exercise are now seen to
The price of a newly issued American claim (or an consist of the dividends paid on the underlying asset
unexercised claim, if t < 0) is the same as the net of the interest forgone by paying the strike.
European claim value plus an EEP given by the In the Black-Scholes setting, with constant coeffi-
present value of the gains to be realized by optimally cients, the formula can be further refined. First, it is
exercising prior to maturity.16 easy to verify that immediate exercise takes place, at
The second transformation, called the delayed exer- date v, provided Sv > Bv where Bv is the value at
cise premium (DEP) representation, emphasizes the v of a function of time B·. Expectations can then be
gains realized by deferring exercise. This formula fol- computed more explicitly to give
lows almost immediately from the pricing formula for
the claim. Indeed, CSt t!B·K
t T
Vt = Et∗ bt t Yt = Yt + Et∗ dbt v Yv = cSt t!K+ St e−v−t N dSt !Bvv −t dv
t t
T √
t
= Yt + Et∗ bt v dYv − rv Yv dv − rKe−rv−t N dSt !Bvv −t− v −t dv
t
t
(38)
16
Note that only the drift of dY under the risk-neutral measure mat-
17
ters in (37). Martingale components vanish because of the expecta- A process x is a potential if it is a right-continuous nonnegative
tion in front. supermartingale such that limt→ Ext = 0.
where cSt t! K is the Black-Scholes Formula (6) and where d W

v ≡ −dW ∗ + v dv is a Q-Brownian motion.
v
dSt ! B v − t is defined in (7). Passage to the measure Q is akin to using the stock
In order to implement (38) one still needs to as a new numeraire. The right-hand side of (40) is the
identify the unknown exercise boundary B. A few value of a call, exercised at , with strike St , written
additional steps provide the characterization needed. on an asset with price S and paying dividends at the
Indeed, recall that immediate exercise is optimal rate r, and in a financial market with interest rate .
when S = B. It follows that
Although derived for the optimal put exercise
Bt − K = cBt t! K time , the relation (40) holds for any stopping time
T of the filtration. It follows that optimal stopping times
+ Bte−v−t N dBt! Bv v − t dv must be identical across the two markets, i.e., the opti-
t
mal exercise time of the American put in the original
T
− rKe−rv−t N dBt! Bv v − t market is the same as the optimal exercise time of the
t American call option, identified by the formula, in the
√
− v − t dv (39) modified financial market. The call option in the mod-
ified market is therefore symmetric to the put in the
Moreover, because t → T , it can be verified that original market.
limt→T Bt = maxK r/K. Equation (39) is the Early versions of the symmetry relation, in
integral equation for the exercise boundary. Solving restricted settings, can be found in Grabbe (1983),
the integral equation subject to the boundary condi- Bjerksund and Stensland (1993), Chesney and Gibson
tion indicated identifies the optimal exercise bound- (1995), McDonald and Schroder (1998). For general
ary. Substituting in (38) provides the American call markets with semimartingale prices the symmetry
option price.
relation is established by Schroder (1999); see also
This characterization of the exercise boundary and
Kholodnyi and Price (1998), for an approach using
the resulting two-step valuation procedure have come
group theory. This general result relies on a change
to be known as the integral equation method. This
of numeraire method introduced in Jamshidian (1989)
approach to American option pricing can be traced to
and Geman et al. (1995). A review of these results and
Kim (1990), Jacka (1991), and Carr et al. (1992). Vari-
ations of the method have been proposed in subse- some extensions appear in Detemple (2001).
quent work. Little et al. (2000), in particular, point to
an interesting reduction in dimensionality leading to
an equation for the boundary with a single integral.
4. Beyond the BSM Model
The BSM Equation (6) is one of the most successful
3.6. Put-Call Symmetry and widely used in financial economics. The under-
The valuation of put options is a simple matter once lying Model (1) and its related assumptions are sim-
call option prices are known. Indeed, a straightfor- ple and elegant. But as a tool for real-world use the
ward change of variables shows that a put is identical model must be compared to financial data. Devia-
to a call in a suitably modified financial market. This tions between the model and empirical evidence offer
result is known as put-call symmetry. opportunities for the development of more-realistic
For exposition purposes we adopt the Itô finan- models, which in turn create new computational chal-
cial market of §2.2.5 assuming a single asset and lenges for the pricing and hedging of derivative
Brownian motion (d = 1) and focus on American-style securities. Section 4 and its subsections present brief
claims (the symmetry relation for European claims is empirical results (§4.1), as well as extensions of the
a special case). In this context the price of a put is BSM model to include jumps in returns (§4.2) and
given by Pt = Et∗ bt K − S + where stands for the stochastic volatility (§4.3).
optimal exercise time. Simple transformations yield
Pt = Et∗ bt St K/St − St + 4.1. Empirical Evidence
The BSM model can be tested using data from the
=Et exp − v dv S − St + (40) underlying asset (or underlying variable) or options
t data. The first approach checks for consistency
where the first equality uses St ≡ S /St and the sec- between the data and the assumed geometric
ond one a change of measure to Q such that d Q= Brownian motion process. The second approach exam-
1 T 2 T ∗
exp− 2 0 v dv + 0 v dWv dQ. The expectation Et · ines the data for consistency between observed option
and
is under Q S ≡ K/St is a new price whose evo- prices and those implied by the model, and therefore
is described by
lution (under Q) jointly tests the model for the asset price process and
additional assumptions required for deriving pricing
d
Sv = v
Sv v − rv dv + v d W
St = K (41) equations (e.g., frictionless trading).
The BSM model in Equation (1) is an exactly solv- Figure 1 Autocorrelation of S&P 500 Squared Log-Returns
able SDE with solution 18%
√
−− 2 /2h+ hZ
St+h = St e (42) 16%
14%
where Z is a standard normal random variable. 12%
Correlation
Equation (42) implies that log-returns lnSt+h /St are
10%
normally distributed.
A quick examination of financial time series for 8%
almost any underlying asset (equities, commodi- 6%

ties, bonds) indicates violations of the normality 4%
assumption for log-returns. To give an example, on 2%
October 19, 1987, the S&P 500 index closed at 224.84,
0%
which represented a one-day log-return of −229% 0 5 10 15 20
from the previous business day’s close of 282.70. Lag
Using reasonable values for the parameters , and Note. Raw results are given by the jagged line; the other line shows a fitted
, the probability of a market drop of this magni- smooth curve.
tude or smaller is approximately 10−97 under the nor-
mality assumption of the BSM model. The expected volatility clustering is one motivation for the addi-
time to observe an event as extreme as this is approx- tion of a stochastic volatility process to the BSM
imately 1084 billion years. Repeating the calculation model.
for the less extreme one-day log return of −63% Although a diffusion process for volatility can gen-
on October 13, 1989, gives a probability of approxi- erate autocorrelations in returns that are consistent
mately 10−9 . The expected time to observe an event with the data, such a process cannot easily generate
with this probability is approximately 1 million years. the extreme returns (i.e., jumps) found in the data.
The data are clearly not consistent with the normal- These two departures from the BSM model are in a
ity assumption of log-returns. Statistical tests that are sense complementary.
more formal show that the kurtosis (fourth moment) It is useful to measure option prices in units
of log-returns is significantly larger than implied by of volatility rather than dollars. The BSM implied
normality, i.e., financial data have fatter tails than volatility (or simply the implied volatility) of an
does the normal distribution. Extremely large returns option is the volatility parameter that equates the
that occur more frequently in financial data is one BSM model price with the market price. Suppose a
motivation for the addition of a jump process to the European call option with a strike K and maturity T
BSM diffusion model for asset prices. is traded in the market at the dollar price ĉ. Define
The BSM model assumes that price changes over the option’s implied volatility ˆ to be the solution of
nonoverlapping time intervals are independent. The the equation
quantity E ≡ ElnSt+h /St 2 lnSt+u+h /St+u 2 rep- ĉ = cSt t! K T
ˆ (43)
resents the autocorrelation of lagged squared log-
returns. The independent increments assumption in where cSt t! K T
ˆ is computed using the BSM
the BSM model implies that E should be zero. Choos- Formula (6). Because the right-hand side of (6) is
ing the return interval h and time lag u both equal a monotonically increasing function of , ˆ a unique
to one business day, and computing E using clos- solution exists as long as the market price does not
ing prices of the S&P 500 index from January 1997 violate the no-arbitrage condition. Different options
through December 2002 gives an estimated autocor- will have different implied volatilities, so ˆ can be
relation value of Ê = 165%. If the assumptions of viewed as a function of the option parameters, i.e.,
the BSM model hold, the standard error of the esti- ˆ t t! K T . Quoting an option price in terms
ˆ = S
mate is 2.5. In other words, the observed autocorrela- of implied volatility is analogous to quoting a bond
tion estimate cannot be explained by statistical error price in terms of yield to maturity. For a bond the
because the estimate is more than six standard errors price-yield equation allows a dollar bond price to be
from zero. Figure 1 shows that autocorrelations are converted to a yield; for an option the BSM equation
also statistically significant at much longer time lags. allows a dollar option price to be converted to an
Squared log-returns are a measure of volatility, thus implied volatility.
the positive autocorrelation value indicates that high- Whether or not the BSM model is correct, option
volatility days (i.e., large positive or negative returns) prices can be, and often are, quoted in units of implied
are more likely to be followed by high-volatility days volatility. In the BSM model, the constant volatility
than by low-volatility days. This phenomenon of is a property of the underlying asset, and does not
Figure 2 BSM-Implied Volatility Curves, February 4, 1985 (Top Panel) Figure 3 Time-Series of One Month at-the-Money BSM-Implied
and September 16, 1999 (Bottom Panel) Volatility
60% Historical Volatility
80
Bursting of
50% 70
Dot–Com Bubble
Crash of 1987
Gulf War II
BSM Implied Volatility
60 Russian 9–11
40% Default/LTCM
Volatility
50
Gulf War I
30% 40 Asian Currency

Crises
30
days = 39, points = 346
20%
20
10% 10
0.6 0.7 0.8 0.9 1.0 1.1
Strike / Spot 1988 1990 1992 1994 1996 1998 2000 2002
60%
the implied volatility range of 16% to 53%. Restricted
to the single shortest maturity, the implied volatili-
50% ties ranged from 23% to 53%. For a fixed maturity,
the implied-volatility curve S ˆ t t! K T as a func-
tion of K is often called the implied volatility skew

40%
or the implied volatility smile (see, e.g., the curve for
the shortest maturity on September 16, 1999). Simi-
30%
lar departures from the BSM model are obtained for
many other financial markets.
days = 1, points = 306 The systematic dependence of implied volatility
20% days = 29, points = 307 on strike and maturity is further dramatic evidence
days = 64, points = 56 of the inconsistency of the BSM model with mar-
ket data. Higher implied volatilities for low-strike
10% options means that these options are relatively more
0.6 0.7 0.8 0.9 1.0 1.1 1.2
Strike / Spot
expensive than predicted by the BSM model. Expres-
sion (11) relates the option price to the expected dis-
Note. Days indicate the number of days to maturity, and points indicate the
number of trades used to create the curves.
counted payoff under the risk-neutral measure. For a
fixed-option payoff, e.g., a put payoff, a higher put
price is generated by a higher risk-neutral probabil-
depend on an option’s strike or maturity. So if the ity of ST < K. In other words, a downward slop-
BSM model is correct, then options of all strikes and ing implied volatility curve (which associates higher
maturities traded on the same underlying asset will volatilities with lower strikes) corresponds to a fat-
have identical implied volatilities. In practice, option- ter left-tail of the risk-neutral distribution of ST than
implied volatilities depend in a systematic way on K would be implied by the BSM model. Thus market
and T and vary through time t. option data suggest the need for an asset price model
Figure 2 shows BSM-implied volatilities for options where log-returns have fatter tails than occurs with
on the S&P 500 (futures) on February 4, 1985, and the normal distribution.
on September 16, 1999. These options began trad- Not only do implied volatility curves depend in
ing on the Chicago Mercantile Exchange (CME) in a systematic way on K and T , but this dependence
1985. For the two maturities shown on February 4, changes in an unpredictable way with the passage
1985, options traded in the narrow implied volatility of calendar time t. Figure 3 shows a time series
range of 14% to 17%. Prior to the crash of October of BSM-implied volatility for 30-day maturity at-
1987, option-implied volatilities showed relatively lit- the-money options (i.e., the strike Kt equals St ).18
tle dependence on strike and maturity. After the
crash, however, the situation changed dramatically. 18
An option with this exact strike and maturity cannot be traded,
For example, on September 16, 1999, options traded in thus interpolation might be necessary to define this quantity.
Large changes in volatility are seen to occur in the discussion of the jump-risk premium, see Bates (1988,
data and these changes are often associated with 1996) and Naik and Lee (1990).
significant economic or political events. If the BSM- To price call and put options in this model, note
constant volatility model assumption were correct, that the risk-neutralized model has the same form as
then the correlation of daily changes in implied Equation (44), but with the drift − replaced by
volatility with daily log-returns should be zero. Com- r −−/1 . The SDE under the risk-neutral Q-measure
puting this correlation using S&P 500 futures options has the explicit solution
data from January 1997 through December 2002 gives
NT
a value of −57%, which is again highly statistically 2 /2T +
√
ST = S0 er−−/1 − T Z0
eZn (45)
significant. The negative correlation is interpreted to n=1
mean that options become more expensive when mar-
ket returns are negative and is reflected in a skewed where Z0 is a standard normal variate. Let xt = lnSt .
distribution of asset returns. Then conditional on x0 = lnS0 and NT = j, the distri-
There is a vast literature on empirical tests and bution of XT = lnST is normal with mean x0 + r −
2
econometric estimation of option pricing models. An −/1 − 2 T +js and variance 2 T +j s2 . Analytical
influential early paper is Engle (1982). Recent con- tractability is maintained because ST is a product of a
tributions include Bakshi et al. (1997), Bates (2000), (random number) of lognormal random variables.
Chernov and Ghysels (2000), Pan (2002), Chernov Using the extension of Equation (11) to the setting
et al. (2003) and Eraker et al. (2003). For recent with discontinuous prices under consideration, the
surveys see Garcia et al. (2004) and Bates (2003). time t = 0 price of a call option, cM S0 0, can be writ-
Empirical price data for the underlying asset and ten as
associated options suggest significant departures from
the assumptions of the BSM model. Researchers have e−rT E0∗ ST − K+
proposed many extensions of the BSM model to incor- = e−rT E0∗ ST 1xT >lnK − e−rT KE0∗ 1xT >lnK (46)
porate these and other empirical features. Some of
these extensions are described next. The term E0∗ 1xT >lnK = QxT > lnK can be com-
puted as follows. Let f 3 = E0∗ ei3xT be the charac-
4.2. Jump Models teristic function of xT . Using the Fourier inversion
For the underlying price, Merton (1976) proposed the formula
following jump-diffusion model:
1 1 e−i3y
Nt QxT > y = + Re f 3 d3 (47)
dSt Z 2 0 i3
= − dt + dWt + d e − 1
n
(44)
St − n=1 the characteristic function can be computed explicitly
where Nt is a Poisson process with intensity / and the as
jumps in returns are determined by Zn ∼ N s s2 ,

/T j e−/T
which are independent of W . This model has three f 3 = E0∗ ei3xT = E0∗ ei3xT x0 NT = j
j=0
j!
additional parameters: / determines the arrival rate

of jumps; and s and s determine the mean and /T e
j −/T
32 2 T
variance of the jumps in return. When the nth jump = exp i3m −
j=0
j! 2
occurs at time t the stock price changes from St− to
j
eZn St− . The average jump size is then EeZn − 1 = 32 s2
exps + s2 /2 − 1 ≡ 1 . It is often more convenient to · exp i3s −
2
specify 1 and s rather than s and s as primitive 2 2

3 T
model parameters. = exp i3m − − /T
In this model, option payoffs cannot be replicated 2

by trading in the primitive assets, i.e., the market 32 s2
is not complete, so option prices are not uniquely · exp /T exp i3s − (48)
2
determined from arbitrage considerations (i.e., market
prices of risk, implied by the asset structure, are not where m ≡ x0 + r − − /1 − 2 /2T .19
unique). In order to proceed, Merton (1976) assumed The term E0∗ ST 1xT >lnK in Equation (46) can be
that jump risk was diversifiable, so the market price transformed to a probability computation by using
of jump risk is zero. This model can provide a better
fit to empirical asset price data, it can produce a range 19
The equality on the second line follows from Eei3Z = expi3 −
of implied volatility curves, and maintains almost all 32 2 /2 when Z ∼ N 2 and the equality on the last line uses

of the analytical tractability of the BSM model. For a ey = j
j=0 y /j! for all complex y.
the change of measure outlined in §3.6, adapted to Figure 4 Merton Jump-Diffusion Model Implied Volatility Curves (Top
the present context. Let )t ≡ e−r−t St /S0 and recall the Panel); Heston SV Model Implied Volatility Curves (Bottom
≡ )T dQ∗ . Then E ∗ )T hXT = E 0 hXT Panel)
measure d Q 0
∗
and so E0 ST 1xT >lnK = e r−T
S0 QxT > lnK. The 45%
characteristic function under the new measure can be

40%
computed explicitly as
35%
0 ei3xT = E ∗ )T ei3xT
g3 = E 0
30%
= e−r−T −x0 E0∗ e1+i3xT = e−r−T −x0 f −i + 3
25%
(49)
20%
T>
where f · is defined in (48). The probability Qx
lnK can be computed using the Fourier inversion 15%
Formula (47) with f replaced by g. 70 80 90 100 110 120
Strike
To summarize, the price of a European call in
Merton’s jump-diffusion model is 1-day 30-day 60-day
T > lnK − e−rT KQxT > lnK

cM S0 0 = S0 e−T Qx 45%
(50)
40%
where the two probabilities can be computed using
(47) and the explicit Formulas (48) and (49). The 35%
result is equivalent to Merton’s original formula that
expressed the option price as an infinite weighted 30%
sum of BSM formulas. Although the expressions (48) 25%

and (49) are lengthy, the final Equation (50) involves
two one-dimensional integrals that can be easily com- 20%
puted numerically. Put prices can be derived in a sim-
ilar manner, or computed from call prices using the 15%
70 80 90 100 110 120
put-call parity relation (9). Strike
The form of Equation (50) mirrors the BSM For-
1-day 30-day 60-day
mula (6). Indeed, the two formulas coincide for / = 0,
and so this Fourier inversion approach provides an
alternate derivation and procedure for computing the Other models that feature jumps in returns include
BSM formula. Bates (1996), Madan et al. (1998), Carr et al. (2002),
The Merton jump-diffusion model can provide a and Carr and Wu (2003), Scott (1997), and Kou (2002).
better fit to underlying asset price data and it can gen- For a more complete treatment of jump models, see
erate implied volatility curves that are more consis- Cont and Tankov (2004). Although jump-diffusion
tent with market option prices. To illustrate, Figure 4 and pure-jump models can generate jumps in returns
shows implied volatility curves for the Merton model and implied volatility skews consistent with empirical
with the parameters S0 = 100, = 20%, / = 10%, data, asset returns under these models produce zero
1 = −20%, s = 40%, = 1%, and r = 5%. For exam- autocorrelation of squared log-returns. Furthermore,
ple, for 30-day maturity options (i.e., T = 30/36525 implied volatility curves, for a fixed option matu-
years) with strikes of K = 80 90, and 100, European rity, would remain constant through time. To address
put option prices under the Merton model are these empirical observations, researchers have pro-
0.103, 0.211, and 2.243, respectively. These dollar posed modeling volatility as a random process rather
option prices correspond to BSM-implied volatilities than as a constant, as in the BSM model. An illustra-
of 40.7%, 25.6%, and 21.1%, respectively. The negative tive stochastic volatility model is discussed next.
value of the average jump size parameter, 1 , gener-
ates a fatter left tail compared with the lognormal dis- 4.3. Stochastic Volatility Models
tribution of ST under the BSM model, which in turn Heston (1993) proposed the following two-factor asset
produces the declining implied volatilities mentioned. price model:
For the shortest one-day maturity options, the param- dSt
eter s is large enough to generate positive jumps, so = − dt + Vt dWt (51)
St
the implied volatility curve becomes the smile shown
in Figure 4. dVt = H' − Vt dt + v Vt dWtv (52)
where Wt and Wtv are two Brownian motion pro- The attractive features of jump-diffusion and SV
cesses with EdWt dWtv = E dt, H is the speed of mean models have been combined in the models of Bates
reversion in the variance process, and ' is the long- (1996), Scott (1997), Bakshi et al. (1997), and the recent
run variance mean. Equation (52) models variance model in Carr et al. (2003). A general affine jump-
as a mean-reverting process: When Vt > ' the drift diffusion framework that includes many of these
is negative and variance is more likely to decrease models as special cases is provided in Duffie et al.
than increase and, conversely, when Vt < ' variance is (2000). Affine jump-diffusion interest-rate models are
more likely to increase. Variance cannot become neg- given in Chen and Scott (1992), Duffie and Kan (1996),
ative because the volatility of variance approaches and Dai and Singleton (2000).
zero
as V decreases due to the Vt term. This same
t
Vt term leads to a exponential affine structure in 5. Numerical Methods
the solution, which affords a considerable degree of The pricing of American, path-dependent, and multi-
analytical tractability. Equation (52) generates volatil- asset options in the BSM framework generally requires
the use of numerical methods. More recent models
ity clustering—i.e., high-volatility periods are more
incorporating jumps in returns, SV, jumps in volatility,
likely to be followed by high-volatility periods—and
stochastic interest rates, default risk, etc., pose com-
also generates positive autocorrelation of squared log-
putational challenges for European-style options and
returns. The parameter E allows correlation between
more-exotic structures. Calibration of models to mar-
log-returns and changes in variance to be incorpo-
ket data typically involves an optimization procedure
rated in the model. The distribution of ST can be
to determine a set of parameters that best fit the data,
viewed as an infinite mixture of lognormals with
and each step in the optimization can require many
different volatility parameters. This mixing property price and derivative computations that can result in
generates excess kurtosis (compared with the BSM a tremendous computational challenge. Investment
model), which produces implied volatility curves. banks, mutual funds, and many other financial and
Heston (1993) showed that the price of a nonfinancial institutions hold portfolios containing
European call, in this setting, has the same form large numbers of derivative securities. Determining
as Equation (50). The characteristic functions have the risk of these positions typically requires portfo-
explicit, exponential-affine forms and the Fourier lio revaluations under many potential future market
inversion formula can again be used to numerically outcomes, and this presents another difficult yet prac-
compute the probabilities required for the option tically important computational task.20
price. Alternative inversion formulas are presented In theory, closed-form analytical solutions are quick
and analyzed in Lee (2004). For a discussion of the to evaluate and perfectly accurate. In practice, how-
volatility risk premium and its impact on option ever, even the BSM formula requires either numerical
prices, see Bates (1988, 1996) and Lewis (2000). As an integration or the use of approximations for determin-
illustration, Figure 4 shows implied volatility curves ing the normal probabilities in the formula. Although
for the Heston model with parameters S0 = 100, typically not a major issue for the BSM formula, it
V0 = 0262 , H = 6, ' = 0352 , v = 3, E = −50%, does illustrate the point that almost all derivative
= 1%, r = 5%, and the strikes and maturities indi- pricing problems ultimately require a numerical pro-
cated. Although the parameters were not chosen to cedure, and the choice of methods involves trade-offs
optimally fit any given set of data, it is clear that the between speed and accuracy and between simplicity
stochastic volatility (SV) model can generate implied- and generality.
volatility curves that are similar to those observed in The major numerical approaches can be classified
the data. Furthermore, the curves change randomly in one of four categories: (i) formulas and approx-
through time as Vt evolves, with correlated changes in imations, (ii) lattice and finite difference methods,
volatility and returns. The model has a difficult time (iii) Monte Carlo simulation, or (iv) other specialized
reproducing steep smiles typically observed for very methods.
short maturity options. In the first category, the most significant advances
The Fourier inversion approach was first proposed have been in the application of transform meth-
in Stein and Stein (1991) and Heston (1993). Impor- ods and asymptotic expansion techniques. Fourier,
tant early contributions to the SV literature include Laplace, and generalized transform methods have
Cox (1975, 1996) and Hull and White (1987). Recent
contributions include Duan (1995), Fouque et al. 20
Financial risk management is another important area of theoret-
(2000), Davydov and Linetsky (2001), Detemple and ical and practical interest. A theoretical foundation is provided in
Artzner et al. (1999). A related early paper is Rudd and Schroeder
Tian (2002), and Cvsa and Ritchken (2001). For an (1982). Numerical algorithms are treated in Glasserman et al. (2000).
overview of SV models see Ghysels et al. (1996) and For a comprehensive introduction to this field, see Jorion (2001) and
Lewis (2000). the references therein.
been applied to SV models (Stein and Stein 1991, continuous distribution. For m = 2, the points x1 and
Heston 1993, Duffie et al. 2000), the pricing of Asian x2 and associated probabilities are chosen so that the
options (Reiner 1990, Geman and Yor 1993) and many first and second moments either match exactly or
other options pricing models. Asymptotic expansion match in the limit as h → 0. Additional lattices have
and singular perturbation techniques have proved been proposed with m = 4 or higher, and with those
especially useful in developing analytical formulas it is possible to match higher moments of the dis-
and approximations for SV models (Hull and White tribution (see, e.g., Heston and Zhou 2000, Alford
1987, Hagan and Woodward 1999, Fouque et al. 2000, and Webber 2001). Unfortunately, when applied to the
Andersen et al. 2001). pricing of American options, values of m greater than
two have not resulted in better overall convergence
5.1. Lattice and Finite Difference Methods
when the additional computational time is taken into
Lattice methods use discrete-time and discrete-state
consideration.
approximations to SDEs to compute derivative prices.
Pricing options with the trinomial Lattice (4)
Lattice approaches were first proposed in Parkinson
requires three calculations at each node instead
(1977) and Cox et al. (CRR) (1979). Lattice meth-
of two. Although the additional outcome produces
ods are easy to explain and implement and they are
American option prices that are more accurate, com-
described in virtually every textbook on derivatives.
pared with the binomial method, the additional
The triangular lattice obviates the need for spatial
accuracy is almost exactly balanced by the addi-
or side boundary conditions required for finite dif-
ference methods. These features make lattice meth- tional computational cost, as shown in Broadie and
ods attractive for pedagogical purposes and for the Detemple (1996). An advantage of the Boyle (1986) tri-
computation of derivative prices in simpler models. nomial approach is that the additional stretch param-
Finite difference methods provide numerical solutions eter / ≥ 1 can be used to adjust a particular asset node
to the fundamental pricing PDE, and are often the to a convenient level, e.g., to coincide with a strike or
method of choice for models and securities that are barrier.
more complicated. Lattices (1) and (4) have the advantage of maintain-
The risk-neutral BSM model has an exact solution ing constant asset price levels through time, so the
St+h = St eZ , where Z ∼
r − − 2 /2h 2 h. Over a number of distinct asset price levels ST is linear in
discrete time step, a lattice method replaces Z with a the number of time steps, n. This happens because
discrete random variable X, where X = xi with prob- up and down moves lead back to the same asset
ability pi for i = 1 2 m. Over multiple time steps price level. For example, in Lattice (1), St ex1 ex2 = St but
this leads to a distribution of asset prices: binomial St ex1 ex2 = St in Lattices (2) and (3). In Lattices (1) and
for m = 2, trinomial for m = 3, and multinomial for (4), the drift is captured by the probabilities, whereas
general m. There have been dozens of proposals for in (2) and (3) the drift is captured in the stock price
discrete approximations X; the four widely used lat- levels. When pricing options, the constant level prop-
tices of CRR (1979), Jarrow and Rudd (1983), Amin erty of Lattices (1) and (4) results in computational
(1991), and Boyle (1986) are given in Table 1. savings because On distinct asset price levels can be
The discrete distribution St+h i = St exi , for i = computed once and stored; in the other lattices On2
1 m, is chosen to closely approximate the exact distinct price levels need to be computed.
Table 1 Standard Lattice Methods
Lattice Outcome Probability

√
√ er − h − e− h
CRR (1979) (1) x1 = h p1 = √ √
√ e −e
h − h
x2 = − h p2 = 1 − p1
√
2 h/2
√ e − e− h
2
Jarrow and Rudd (1983) (2) x1 = r − − /2 h + h p1 = √ √
√ e h − e− h
x2 = r − − 2 /2 h − h p2 = 1 − p1
√ √
Amin (1991) (3) x1 = r − − lncosh h h + h p1 = 1/2
√ √
x2 = r − − lncosh h h − h p2 = 1/2
√
√ 1 r − − 2 /2 h
Boyle (1986) (4) x1 = h p1 = +
22 2
x2 = 0 p2 = 1 − 1/2
√
√ 1 r − − 2 /2 h
x3 = − h p3 = 2 −
2 2
Related to lattice methods are numerical solu- derivatives (which includes standard put and call
tions of the PDE (2) with Boundary Conditions (3). payoff structures as well as many others). This lack
The finite difference approach to option pricing was of smoothness typically results in oscillatory con-
first proposed by Brennan and Schwartz (1977, 1978). vergence of the solution as the grid size decreases
A major advantage of this approach is the wealth (i.e., as the number of time steps increases), and
of existing theory, algorithms, and numerical soft- this poses an obstacle to applying extrapolation to
ware that can be brought to bear on the problem. accelerate convergence. For the American option pric-
Issues of numerical consistency, convergence, and sta- ing problem, Broadie and Detemple (1996) propose
bility have been well studied. The generality of finite smoothing the solution by applying the European for-
difference methods is especially important for mod- mula for the continuation value at the penultimate
els that extend beyond the constant coefficients BSM time step in combination with Richardson extrapola-
model. The finite difference method can handle pro- tion. This idea, which applies equally well to lattice
cesses with time-varying coefficients, Itô processes and finite difference methods, can increase the con-
that are more general, jump and SV models, single vergence rate by an order of magnitude (the error
and multifactor interest-rate models, etc. Generaliz- decreases as O1/w) and is also often trivial to imple-
ing lattice methods for these models requires separate ment. Rannacher timestepping (Rannacher 1984) is an
developments, the resulting algorithms are often com- alternative smoothing procedure for finite difference
plicated to implement, and it is difficult to theoret- methods when a closed-form solution or approxima-
ically analyze their convergence properties. In addition is not available. For a general introduction to
tion to widely available numerical libraries of finite extrapolation methods see, e.g., Brezinski and Zaglia
difference methods, a higher-level language for auto- (1991). Richardson extrapolation was introduced in
matic code generation has also been developed and derivatives pricing in Geske and Johnson (1984).
is described in Randall et al. (1997). The language Accelerated convergence for American options in a
enables users to concisely represent financial models finite difference method is given in Forsyth and Vetzal
and solution methods; these specifications are then (2002).
automatically translated into source code in a stan- Lattice and finite difference methods for interest-
dard programming language, e.g., C. rate models are given in Ho and Lee (1986), Black
Lattice methods can be viewed as explicit finite et al. (1990), Hull and White (1994), and Pelsser
difference methods, but the fundamental BSM PDE (2000). Fitting a model to a given initial term structure
can also be solved with a number of other meth- involves numerically solving for a time-dependent
ods, including implicit, Crank-Nicolson, and finite drift function. The forward induction idea of
element methods and alternating direction implicit Jamshidian (1991) has proved to be particularly useful
(ADI) methods for higher-dimensional problems. The in these applications.
finite difference approach offers considerable flexibil- 5.1.1. Path-Dependent Options. Many options
ity in the choices of grids for the time and space have payoff structures that depend not just on the
dimensions, which is useful for dealing with dis- current asset price, but also on the history of the asset
crete dividends, barriers, and other common features. price through time. Pricing options with these general
Higher-order derivative approximations are possible payoff structures require keeping track of the asset
in the finite difference approach, and these can lead price through time, and this leads to nonrecombining
to improved convergence. For a general treatment of lattice or bushy tree algorithms. A nonrecombining
numerical methods for PDEs, see Morton and Mayers lattice algorithm for the pricing of interest-rate deriva-
(1994), and for applications to financial derivatives see tives in the general Heath-Jarrow-Morton (HJM)
Tavella and Randall (2000). framework is proposed in Heath et al. (1990). A
The computation time (or w for work) of lattice nonrecombining lattice with branches determined
methods for pricing options in the BSM model is through Monte Carlo sampling, i.e., a simulated tree
Omn, where m is the number of time steps and n algorithm, is proposed in Broadie and Glasserman
is the number of asset price levels. Typically, m is (1997) for the pricing of path-dependent (and higher-
On, in which case the computation time is On2 . dimensional) American options. The computational
The pricing error decreases as O1/n (see Diener and time of these methods is Omn , where n is the num-
Diener 2004),
√ which implies that the error decreases ber of branch points in the time dimension and m
as O1/ w. Faster convergence for many numer- is the number of branches at each node. The mem-
ical applications can be obtained using extrapola- ory requirement, however, is only Omn (see Broadie
tion techniques, but these techniques usually require et al. 1997 for details). The exponential dependence
a smoothly converging solution. Both lattice and on the number of branch points renders these meth-
finite difference methods are affected by option pay- ods infeasible for all but small values of n. Neverthe-
offs that are discontinuous or have discontinuous less, it is feasible to use n up to 30 or 40 with small
values of m (e.g., two or three) on today’s personal possible using the related dimensionality-reduction
computers (PC). Applications with a small number of idea presented in Hilliard et al. (1995).
exercise opportunities, e.g., some Bermudan options,
5.1.2. Jump Processes and SV Models. When
are natural for this approach. Combined with extrap-
jumps are added to the BSM model as in Equation (44),
olation techniques (see, e.g., Broadie et al. 1997) the
the fundamental pricing PDE (2) becomes a par-
nonrecombining lattice approach can lead to surpris-
tial integro-differential equation (PIDE). Numerical
ingly accurate option price approximations in some
methods to solve the pricing equation are more
applications.
complicated, because the computation of the contin-
Fortunately, most path-dependent payoff structures
uation value at any node now requires an integral
seen in practice do not depend in an arbitrary way on
over the jump distribution. Straightforward exten-
the path of asset prices. Mildly path-dependent pay-
sions of lattice or finite difference methods take an
off structures depend on the current asset price and a
order-of-magnitude more computation time (i.e., the
single sufficient statistic that summarizes the relevant
CPU time is On3 , where n is the number of time
information from the path. For example, for Asian
steps) to achieve comparable accuracy to models
options whose payoff depends on an average asset
without jumps (see, e.g., Amin 1993). Andersen and
price, it suffices to track the current average price (also
Andreasen (2000) develop a finite difference approach
called the running average or the average to date). In
using a combination of fast Fourier transform and
the case of barrier and lookback options, it suffices to
ADI methods; this approach is stable and reduces
track the current maximum or minimum asset price,
the computional time requirement to On2 logn.
or both. Mildly path-dependent options can be priced
Their method applies to a jump-diffusion and local
in an enlarged state space, which includes the cur-
volatility function model that enables exact calibra-
rent price as well information related to the sufficient
tion to a set of market option prices. For the stan-
statistic. For example, in the case of in barrier options,
dard Merton jump-diffusion model in (44), Broadie
the state space includes the current asset price and
and Yamamoto (2003) develop a method based on the
whether or not the option has been knocked in (i.e.,
fast Gauss transform that is easy to implement and
the asset price has breached the knock-in level). In
further reduces the computational time requirement
lattice and finite difference methods, the placement of
to On2 . Hirsa and Madan (2004) develop a finite dif-
the barrier relative to the discrete asset price levels
ference method for the solution of the PIDE that arises
can have a large effect on the accuracy of the method.
in pricing American options in the variance gamma
Boyle and Lau (1994) first noticed and proposed a
model.
solution to the problem. In the case of Asian options,
SV models are treated in detail in Lewis (2000).
the state space can be expanded to include the current
Most of the explicit formulas provided there are
asset price and the current average price. Ingersoll
derived through transform analysis (e.g., gener-
(1987) shows that the Asian option price satisfies a
alized Fourier transforms) and apply to path-
two-dimensional PDE, and finite difference methods
independent European-style options. American and
are given in Zvan et al. (1998). Alternatively, for each
path-dependent options require numerical solutions
node in the lattice or finite difference grid, it is pos-
of the pricing PDE. Examples include Forsyth et al.
sible to keep track of option values corresponding to
(1999), who present a finite element approach to the
each value of the average. Unfortunately, the num-
pricing of lookback options with SV; Kurpiel and
ber of potential average prices is typically exponential
Roncalli (2000), who develop hopscotch methods for
in the number of time steps. So, rather than storing
SV and other two-state models; and Apel et al. (2002),
information for each distinct average price, a range
who develop a finite element method for an SV
of possible averages are grouped into bins or buck-
model.
ets, and interpolation is used in a backwards pricing
procedure. The convergence of lattice and finite dif- 5.1.3. Multiasset Options. A multidimensional
ference methods for pricing mildly path-dependent extension of the BSM model is given in the SDE (16).
options using interpolation is investigated in Forsyth Formulas for the prices of European derivatives usu-
et al. (2002). ally involve multidimensional integrals, so numeri-
In some cases significant computational savings cal methods are needed even in this comparatively
when pricing mildly path-dependent options can simple case. As in the single-asset case, two related
be obtained by a change of measure, also called approaches are multidimensional lattices and multidi-
the change of numeraire technique (see §3.6). Babbs mensional finite difference methods. Suppose that the
(1992) applies this idea to the lattice pricing of look- time dimension is divided in m time steps and that
back options and Rogers and Shi (1995), Andreasen each of d assets is divided into n asset price levels.
(1998), and Vecer (2001) apply it to the pricing of Then a multidimensional lattice or finite difference
Asian options. Significant computational savings are grid will have Omnd nodes and the computational
time will be of the same order. Pricing algorithms do where h represents the discounted derivative payoff
not need to store information at all nodes simulta- that depends on the path of underlying (and possi-
neously, but typically store information at the nodes bly vector-valued) state variables St0 St1 Stm . The
of two adjacent time steps, which leads to an Ond Monte Carlo method consists of three main steps. The
storage (i.e., memory) requirement. Both memory and first step is to generate n random paths of the under-
computation time are exponential in the number of lying state variables. The next step is to compute the
assets d. Given typical processor speeds and mem- corresponding n discounted option payoffs. The final
ory configurations on today’s PCs, it turns out that step is to average the results to estimate the expected
memory is the limiting factor on the size of prob- value; usually a standard error of the estimate is also
lems that can be solved. Problems with three to four computed. In this situation, the convergence rate of
√
assets can be feasibly solved. For example, the stor- the Monte Carlo method is often O1/ n by the
age requirement for a four-asset finite difference grid central limit theorem, independent of the problem
with 50 nodes for each asset and two time steps dimension, which makes it especially attractive for
in memory is 95 megabytes (MB). This assumes an pricing high-dimensional derivatives. This forward
8-byte double precision number is stored for each path generation approach also makes simulation well
of the 504 × 2 nodes, where each double requires suited for pricing complex path-dependent deriva-
8/10242 MB of memory. A similar five-asset problem tives. Much of the focus of recent research has been
would require 4,768 MB of memory, which is not cur- on (i) path simulation methods, especially when there
rently feasible. are nonlinearities in the financial SDEs; (ii) computa-
A lattice for an arbitrary number of assets in the tional improvements through variance reduction tech-
multidimensional BSM model was proposed in Boyle niques; and (iii) extending the Monte Carlo method
et al. (BEG) (1989). At each node in the lattice a to calculate price derivatives and American option
discrete approximation to a multidimensional log- values. For a survey of the literature in this area,
normal distribution is required. Their lattice uses a see Boyle et al. (1997), and for a more extensive
discrete approximation with 2d outcomes, i.e., there treatment see Glasserman (2004), which contains over
are 2d branches at each node. Kamrad and Ritchken 350 references.
(KR) (1991) generalize the BEG lattice by adding an An extensive treatment of the numerical solu-
additional outcome together with a stretch param- tion of nonlinear SDEs is given in Kloeden and
eter (analogous to the parameter in the Boyle 1986 Platen (1999). In the area of efficiency improve-
single-asset trinomial lattice). The additional flexibil- ments, much attention has been given to quasi–Monte
ity in the placement of nodes afforded by the stretch Carlo or low-discrepancy methods (Niederreiter 1992,
parameter is useful, for example, in the pricing of Birge 1994, Paskov and Traub 1995, Joy et al. 1996,
barrier options. An advantage of the BEG and KR Owen 1997). Variance reduction techniques based on
lattices over finite difference approaches is that spa- moment matching are presented and analyzed in Bar-
tial boundary conditions do not need to be specified. raquand (1995), Duan and Simonato (1998), and Duan
Finite difference approaches, however, offer consid- et al. (2001). Algorithms for computing security price
erable flexibility in the choice of grid (i.e., node derivatives using simulation are given in Broadie and
placement) and a variety of approaches for multidi- Glasserman (1996).
mensional PDEs, e.g., ADI methods or improvements The determination of optimal exercise strategies
based on Krylov subspace reduction as presented in in the pricing of American options requires a back-
Druskin et al. (1997). wards dynamic programming algorithm that appears
Relatively little work has been done to date on com- to be incompatible with the forward nature of Monte
putational methods for multiasset options in the pres- Carlo simulation. Much research was focused on the
ence of SV or jumps, or both. development of fast methods to compute approxi-
5.2. Monte Carlo Methods mations to the optimal exercise policy. In this con-
As illustrated in (19), derivative pricing often reduces text, fast refers to a method whose computation
to the computation of an expected value. In some time requirement is a multiple of the time to price
cases this calculation can be done explicitly, e.g., in the corresponding European option using simulation.
the case of the BSM formula, whereas in other cases Notable examples include the functional optimization
lattice or finite difference methods prove useful for approach in Andersen (2000) and the regression-based
numerical evaluation. Monte Carlo simulation is a approaches of Carriere (1996), Longstaff and Schwartz
natural approach for expected value computations. (2001), and Tsitsiklis and Van Roy (1999, 2001).
Consider, for example the calculation of the price of a These methods often incur unknown approxima-
derivative security represented as tion errors and are limited by a lack of error bounds.
Broadie and Glasserman (1997) propose a method
V S0 t0 = E ∗ hSt0 St1 Stm (53) based on simulated trees that generates error bounds
(in the form of confidence intervals) and converges appropriate choice of the martingale M, the inequal-
to the correct value under broadly applicable condi- ity holds with equality, i.e., the duality gap is zero. To
tions. The simulated tree method, while able to han- see this, recall that the American option value process
dle high-dimensional problems, has a computation Vk ≡ V Sk tk is a Q-supermartingale (see §3.3) and
time requirement that is exponential in the number so admits a Doob-Meyer decomposition Vk = Mk − Ak ,
of potential exercise dates, making it practical only where Mk is a martingale, A0 = 0, and Ak is a nonde-
for problems with a small number of potential exer- creasing process. Now set Mk = Mk , i.e., use the mar-
cise dates. The stochastic mesh method of Broadie tingale component of the true American option price
and Glasserman (2004) avoids this exponential depen- to get
dence on the number of exercise opportunities, is a
provably convergent method, and also generates error V S0 t0
bounds. This method has a work requirement that is
≤ M0 + E ∗ max hSk tk − Mk
quadratic in the number of simulated paths; experi- k=1m
mental results suggest a square-root convergence in
the error, leading to an overall complexity in which = V S0 t0 + E ∗ max hSk tk − V Sk tk − Ak
k=1m
the error decreases as the fourth-root of the work,
which is rather slow. Glasserman (2004, §8.6) shows ≤ V S0 t0 (56)
that the regression-based approaches can be viewed
where the last inequality follows from V Sk tk ≥
as special cases of the stochastic mesh method.
hSk tk and Ak ≥ 0.
All of the fast approximation methods mentioned
There are two challenges to developing tight upper
above share the same limitation: the lack of known
bounds with this approach. First, because the true
error bounds. Each method, though, by virtue of
value process V is unknown, the optimal martin-
explicitly or implicitly providing a suboptimal exer-
gale is also unknown. Second, the determination of a
cise strategy, can be used to generate a lower bound
nearly optimal martingale and a corresponding upper
on the true price, because any suboptimal policy triv-
bound must be done in a computationally practi-
ially delivers less value than the value-maximizing
cal manner. The approach taken in Andersen and
optimal stopping rule. A complete bound on the true
Broadie (2004) is to base the upper bound compu-
price would be available if an upper bound could be
tation on any exercise policy by taking M to be the
generated from any given exercise policy.
martingale component of that policy. Their algorithm
A dual formulation that represents the American
is based on a simulation within a simulation. The
option price as the solution to a minimization prob-
inner simulation is used to compute the martingale
lem was independently developed by Haugh and
M based on any approximate exercise policy, and the
Kogan (2004) and Rogers (2002). In addition to its the-
outer simulation is used to compute the expectation
oretical interest, a computationally effective method
E ∗ maxk=1m hXk tk − Mk to determine an upper
to compute upper bounds based on this dual formula-
bound. Although this approach sounds computation-
tion was introduced in Andersen and Broadie (2004).
ally intensive, much of the work in the inner sim-
Let hx tk represent the discounted option payoff
ulation can be avoided by using the properties of
at time tk when the state Stk = x is a d-dimensional
the chosen martingale M. The resulting algorithm can
vector. As in §3.3, the discrete-time American option
generate practically useful lower and upper bounds
price is given by
on many problems of financial importance in nearly
V S0 t0 = sup E ∗ hS (54) real time (e.g., within a few minutes on a current PC)
and further improvements are possible with specially
where is a stopping time taking values in = designed variance reduction techniques.
t1 t2 tm . Let Mtk be any adapted martingale.
Then 6. Conclusions
Valuation methods for derivative securities have
V S0 t0 = sup E ∗ hS + M − M reached a fair degree of maturity. Nevertheless, chal-
∈
lenges remain on several fronts. It is sure to be a
= Mt0 + sup E ∗ hS − M continuing goal to develop asset price models that
∈
better match observed data, while maintaining as
≤ Mt0 + E ∗ max hSk tk − Mtk (55) much parsimony and tractability as possible. Multi-
k=1m
factor SV, stochastic correlation, and regime-switching
where the second equality follows from the optional models are potential avenues for further investiga-
sampling property of martingales. The upper bound tion. Models that are more detailed and that incor-
given above is especially useful because, for an porate information arrival; liquidity; trading volume
and open interest; market incompleteness; and inter- A.1. Fixed Income, Credit, and Other Derivatives
actions between multiple securities or multiple agents The same principles for pricing and hedging derivatives
might prove useful. Models that integrate market and apply to many markets. However, each market has impor-
credit risk in a general equilibrium framework await tant and unique features that need to be modeled, and
development. these in turn raise new analytical and computational issues.
Ever-improving computing technology will expand The rise of equity derivatives in the 1970s was followed
by the development of derivatives in fixed income, foreign
the amounts and types of data that can be ana-
exchange, commodity, and credit markets. More recently,
lyzed; the types of financial models that can be
energy, weather, catastrophe (e.g., earthquake and hurri-
feasibly studied and numerically solved will be simi- cane), and other derivatives have traded. The literature is
larly expanded. Added computing power alone, how- far too vast to do any justice in this brief space, so we
ever, is of fairly limited usefulness. Many problems settle for mentioning a few selected highlights, represen-
are effectively exponential in their memory or com- tative pieces of research, and pointers to more extensive
putation time requirement, and these will require references.
algorithmic advances. For example, it is not currently A central challenge in the fixed income market is the
feasible to calibrate model parameters to a large set consistent modeling of a set of related securities, in this
of securities that are priced by simulation (a prob- case bonds of different maturities that make up a yield
lem of simulation within optimization). Efficient and curve. James and Webber (2000) provide over 500 refer-
convergent methods for pricing high-dimensional and ences in this area. The seminal paper of Heath et al. (1992)
path-dependent American securities depend on the unifes and substantially extends previous interest-rate mod-
development of new algorithms, not faster computers. els, including those of Vasicek (1977), Ho and Lee (1986),
and Black et al. (1990). Working with discretely rather than
New markets and products continue to be intro-
continuously compounded rates, Brace et al. (1997) and Mil-
duced. Recent examples include passport options,
tersen et al. (1997) propose the Libor market model, which
variance swaps, and volatility futures, which began generates prices consistent with the Black (1976) formula
trading on the CBOE in 2004.21 Businesses could run for European options on a market interest rate. Signifi-
markets to better predict the success of a new prod- cant analytical developments were enabled with the change
uct or service, to predict the market share of existing of measure approach proposed in Jamshidian (1989) and
products, or to predict other factors of interest. Mar- Geman et al. (1995). Extensions of interest-rate models to
kets with payoffs tied to political events already exist, include jumps in rates are given in Duffie and Kan (1996)
and trading in the outcomes of other events has been and Glasserman and Kou (2003). An extension to SV is
proposed. Environmental markets and the trading of given in Andersen and Brotherton-Ratcliffe (2001). Empir-
emissions permits might expand on a global scale. ical evidence supporting SV and jumps in interest rates is
New markets might develop to protect an individual’s given in Collin-Dufresne and Goldstein (2002) and Johannes
home value or a country’s gross domestic product, as (2004).
envisioned in Shiller (2003). In each case, new mod- An important application of interest-rate modeling is
to the pricing of mortgage-backed securities, asset-backed
els, valuation theory, and numerical methods might
securities (e.g., securities backed by credit card receivables),
need to be developed. What is certain is that there
and related derivative securities. Relevant papers in this
will be surprising developments and continuing chal- area include Dunn and Spatt (1985), Schwartz and Torous
lenges in this exciting and expanding field. (1989), Kau et al. (1990), McConnell and Singh (1993),
Zipkin (1993), and Akesson and Lehoczky (2000). See the
Acknowledgments textbook by Sundaresan (2002) for more discussion and
The authors thank the editor, David Hsieh, as well as Mike references.
Chernov, and especially Leif Andersen for many comments Credit risk and the pricing of securities in the presence
that improved the paper. This work was supported in part of default risk have been investigated since at least the
by the National Science Foundation under Grant No. DMS-
1970s. A commonly traded credit derivative is the credit
0074637.
default swap, or CDS. This is an agreement between two
Appendix. Securities and Markets parties, where party A makes a fixed periodic payment
Since 1973, when plain vanilla equity options were first to party B, until either the maturity of the CDS, or until
introduced on the CBOE, numerous products have been cre- there is a default by the reference entity (e.g., a publicly
ated to fill various needs of investors. In this appendix we traded company). In the event of a default, party B pays
briefly review the vast literature on fixed income deriva- party A a contractually specified amount, which can depend
tives, credit derivatives, real options (§A.1), path-dependent on the market value of a specific bond, (or in some cases,
contracts (§A.2), and derivatives with multiple underlying party B delivers to party A a specific bond issued by the
assets (§A.3). reference entity). Credit risk models are often classified
as structural or reduced form. In the structural approach
21
For a description and analysis of passport and related options (sometimes called the firm-value approach) default occurs
see Andersen et al. (1998) and Shreve and Vecer (2000). Volatility based on a model of an issuer’s ability to meet its liabil-
derivatives are treated in Neuberger (1994), Detemple and Osakwe ities. In reduced-form models (sometimes called intensity-
(2000), Carr and Madan (1998), and Demeterfi et al. (1999). based or hazard rate models) default is an exogenous
process that is calibrated to observed historical or current A.2.1. Barrier Options. A barrier option is an option
market data. The seminal contributions given in Black and whose payoff depends on the hitting time of some prespeci-
Scholes (1973), Merton (1974), and Black and Cox (1976), fied barrier. Standard barrier options have a constant barrier
and the more recent models of Leland (1994) and Anderson and come into existence (in-options) or expire (out-options)
and Sundaresan (1996) fall in the first category. Important when the barrier is breached. Examples include down-
reduced-form models include Jarrow and Turnbull (1995) and-out, down-and-in, up-and-out, and up-and-in options.
and Duffie and Singleton (1999). In recent years the mar- A down-and-out option, for instance, expires when the asset
kets for convertible bonds and credit derivatives (including price breaches a barrier whose level is below the price of the
default swaps, collateralized debt obligations, multiname underlying asset at inception of the contract. Out-options
credit derivatives, and others) has grown substantially and sometimes involve a rebate, such as a flat payment, in the
there has been a corresponding growth in related research. event of early termination.
Representative papers in this area include Hull and White In- and out-barrier options with constant barriers are easy
(1995), Das and Sundaram (2000), Li (2000), and Andersen to price in the BSM setting. A comprehensive set of formu-
and Buffum (2004). Excellent sources for more information las can be found in Rubinstein and Reiner (1991).
include the books by Bielecki and Rutkowski (2002), Duffie Symmetry relations also tie pairs of barrier options
and Singleton (2003), and Schonbucher (2003). together. For example, an up-and-out put can be priced
Government deregulation is one factor that has fueled from a down-and-out call by using a change of variables.
the expansion of existing markets for energy and the devel- Indeed, suppose that H is the put barrier and that H =
opment of new markets for energy derivatives. Electricity infv ∈ t T 6 Sv = H represents the first hitting time of H.
markets present new challenges primarily because of elec- Performing the transformations described in §3.6 yields
tricity’s limited ability to be stored. Supply and demand where H
H = H, = KSt /H and
must be balanced through time, whereas demand varies
H = KSt /Sv =
= infv ∈ t T 6 H Sv
significantly between day and evening hours and between
warm and cold months. These factors together with recent
with S described in (41). It is then apparent that an up-
instances of enormous price spikes illustrate the need for
and-out put on S, with barrier H, strike K, and matu-
models that are specifically tailored to these markets. Recent
rity date T has the same price as a down-and-out call
work in this area includes Schwartz and Smith (2000),
Lucia and Schwartz (2002), Barlow (2002), and the books by on S, with barrier H = KSt /H, strike St , and maturity
Pilipović (1998) and Clewlow and Strickland (2000). date T in a modified financial market with interest rate
Weather affects the revenues and expenses of many and dividend rate r. Equivalently, puo S t! K H T r =
T r, where the last two arguments have
c do K t! S H
energy and nonenergy businesses, and weather derivatives
offer a means of hedging the associated risks. Securities been added to capture the market structure (interest rate
with payoffs linked to the occurrence or nonoccurrence of and dividend rate).
earthquake, hurricane, or other events are called catastrophe In-options can be valued by complementarity. Simple
derivatives. References in these areas include Litzenberger inspection reveals that a standard option is the sum of an
et al. (1996), Canter et al. (1997) and Geman (1999). in-option plus an out-option with identical characteristics.
For instance, for calls we have
The term real options refers to the opportunity to ini-
tiate, expand, or contract a capital investment in a real c do S t! K H T + c di S t! K H T = cS t! K T
asset, such as land, physical plant, or equipment. Diffi-
culties arise because market prices might not be contin- c uo S t! K H T + c ui S t! K H T = cS t! K T
uously observable. Investment decisions can be discrete
rather than continuous; decisions might be reversible or where the expressions on the right-hand side are the prices
irreversible; and current decisions could lead to different of standard call options. Similar relations hold across puts.
potential future decisions. A seminal paper in this area is Valuation formulas for European barrier options, in the
Brennan and Schwartz (1985), and representative research BSM setting, are found in Rubinstein and Reiner (1991).
includes Cortazar et al. (1998), Kulatilaka and Perotti (1998), American barrier options are priced by Rubinstein and
Bollen (1999), and Schwartz and Zozaya-Gorostiza (2003). Reiner (1991) in the binomial model and Gao et al. (2000) in
Books on the subject include Dixit and Pindyck (1994), Tri- the BSM setting. The latter paper establishes an EEP decom-
georgis (1996), Amram and Kulatilaka (1999), and Copeland position of the price and derives an integral equation for
and Antikarov (2001). the exercise boundary.
A.2.2. Capped Options. Capped options were intro-
A.2. Path-Dependent Options duced on the CBOE in 1991 (see the capped-style index
A path-dependent option is an option whose payoff depends options (CAPS) contract) and have also appeared as com-
on prices not only at, but also prior to the exercise ponents of hybrid securities issued by firms for financ-
time. Some of the most popular path-dependent contracts, ing purposes (such as the Mexican index-linked euro secu-
reviewed in §§A.2.1–A.2.3, are barrier options, capped rity (MILES) contract). Capped options can be European or
options, and Asian options. Occupation time derivatives are American style, and can include various types of automatic
examples of recent products that are experiencing a surge exercise provisions in the event that the underlying price
of interest in the academic community. Definitions and ref- breaches the cap. The caps involved can be constant, time-
erences for those contracts will be provided in §A.2.4. dependent functions, or stochastic processes.
A typical capped option involves a limit on the amount A.2.4. Occupation Time Derivatives. Occupation time
realized upon exercise. A standard capped call option, for derivatives are fairly new products attracting some atten-
instance, pays minS L − K+ upon exercise, where L > K tion from investors and researchers. A defining character-
is a constant cap on the underlying price (L − K is the cap istic of these contracts is an exercise payoff that depends
on the payoff). European-style capped options, with payoff on the time spent by the underlying asset in some pre-
at the maturity date, are easily valued. A European capped determined region(regions). Typical specifications of the
call is identical to a bull spread and is therefore simply the occupation region involve barriers, thus an occupation time
difference between two call option prices: cS t! K L T = derivative can be thought of as a type of barrier option.
cS t! K T − cS t! L T . Various claims with features of this type have been
American-style contracts are usually more difficult to studied. One example is the quantile option. A European
price. A relatively simple case is when the cap is a constant. -quantile call pays off M T − K+ at exercise where
In that case it is clear that immediate exercise is optimal at M T is the smallest (constant) barrier such that the frac-
the cap, because the maximal value of the payoff, L − K, is tion of time spent by the underlying price at or below
attained. This is true under the relatively weak assumption M T , during 0 T , exceeds . Formally,
of a positive interest rate. Of course exercise can also be T
optimal prior to hitting the cap, but the optimality of exer- M T ≡ inf x6 1Sv ≤x dv > T
cise at the cap enables us to conclude that the capped-call 0
option is in fact an up-and-out barrier option with barrier T

where 0 1Sv ≤x dv is the occupation time of the set S ≤ x
L and rebate L − K. Valuation methods for barrier options during the period 0 T .
can then be applied to this contract. Quantile options were suggested by Miura (1992) as an
In the context of the BSM model one can also use sim- alternative to standard barrier options, which have the
ple dominance arguments to show that the exercise bound- drawback of losing all value at the first touch of the bar-
ary of the capped option, B·! L T , is the minimum of the rier. Quantile options lose value more gradually. The bar-
cap L and of the exercise boundary of the corresponding rier implied by the quantile decreases as the underlying
uncapped option B·! T . That is Bt! L T = minL Bt! T . asset spends more time at lower levels. The option loses all
The price of the claim is the present value of the payoff value only if M T ≤ K. Pricing formulas are provided
stopped at this first hitting time of this boundary. by Akahori (1995) and Dassios (1995). Formulas for the dis-
Some of these insights and extensions to growing caps tribution of the quantiles of a Brownian motion with drift
can be found in Broadie and Detemple (1995) for the can be found in Embrecht et al. (1995) and Yor (1995).
BSM framework. Results for some diffusion models are in A second example is the Parisian option. A Parisian out
Detemple and Tian (2002). option with window D, barrier L, and maturity date T will
A.2.3. Asian Options. The payoff of an Asian option lose all value if the underlying price has an excursion of
depends on an average of prices over time. An Asian call duration D above or below the barrier L during the option’s
option has the payoff A − K+ , where A is the underlying life. If the loss of value is prompted by an excursion above
average price and K the strike price. An average strike call (below) the barrier, the option is said to be an up-and-
has payoff S − A+ based on the difference between the out (down-and-out) Parisian option. A Parisian in option
current and average prices. Asian puts are defined in a sym- with window D, barrier L, and maturity date T comes
metric fashion. Arithmetic averages are common, although alive if the underlying price has an excursion of duration D
geometric averages are sometimes used. Continuously aver- before maturity. If specification involves an excursion above
aged prices are of great theoretical interest, whereas aver- (below) the barrier the option is an up-and-in (down-and-
ages at discrete time intervals are typically used in practice. in) Parisian option.
Asian options were first proposed in Boyle and Emanuel Parisian contractual forms were introduced and stud-
(1980) and have become very popular in practice. For firms ied by Chesney et al. (1997). The motivation parallels the
with periodic revenues or expenses tied to an underlying motivation for quantile options. Parisian options are more
price or index level an Asian option is a better hedging sturdy, because they lose value more gradually than do
instrument than a European option whose payoff depends standard barrier options. Contracts of this type are more
on a price at a single point in time. The pricing of Asian robust to eventual price manipulations. The pricing formu-
options in the BSM model is quite challenging. A numerical las in Chesney et al. (1997) involve inverse Laplace trans-
approach based on Monte Carlo simulation for the pricing forms. Implementation issues are examined by Chesney
of Asian options was developed in Kemna and Vorst (1990), et al. (1997). A variation of the Parisian option is the
and additional numerical approaches (including the case of cumulative Parisian option, which pays off based on the
American Asian options) were presented in Ritchken et al. cumulative amount of time above or below a barrier. Pric-
(1993), Curran (1994), Hansen and Jorgensen (2000), and ing formulas are provided by Chesney et al. (1997) and
Ben-Ameur et al. (2002). An analytical formula for continu- Hugonnier (1999).
ously averaged Asian options was first published by Geman Our last example is the step option. This derivative’s pay-
and Yor (1993).22 Series solutions for continuously averaged off is discounted at some rate that depends on the amount
Asians have been developed in Dufresne (2000), Schröder of time spent above or below a barrier. Various discounting
(2002), and Linetsky (2004). schemes can be used. For instance, a proportional step call,
with strike K and barrier L, pays off
22
A similar formula was presented in Reiner (1990). exp−Eot S LSt − K+
at exercise, for some constant E > 0, where ot S L is the Figure A1 Exercise Region of a Max-Call Option
amount of time spent by time t above (or below) the
barrier L. The discount rate E is the knock-out rate, and S2 E X2(t)
exp−Eot S L is the knock-out factor. Proportional step E 1X(t)
options are also referred to as geometric or exponential step
options. Simple step options use a piecewise linear amorti-
zation scheme instead of an exponential one. A down-and-
out simple step call, with strike K and barrier L, pays off
max1 − Eot S L 0St − K+
This option is knocked out at the first time at which Bt2
ov S L ≥ 1/E. This stopping time is the knock-out-time.
Step options were studied by Linetsky (1999). Laplace B 2X
transforms and their inverses are central to the valuation B 1X
formulas provided.
A.3. Multiasset Options 0 Bt1 S1

Derivatives written on multiple underlying assets have long
been of interest to investors. Contracts such as options on
the maximum (max-options) or the minimum (min-options) for some boundaries B 1 S 2 t and B 2 S 1 t, such that
of several underlying prices have long been quoted over the B 1 S 2 t > S 2 for all S 2 t ∈ + × 0 T and B 2 S 1 t > S 1
counter. Options on the spread between two prices (spread
for all S 2 t ∈ + × 0 T . Indeed, for max-calls immediate
options) can now be traded on organized exchanges such as
exercise along the diagonal, where S 1 = S 2 , is never optimal
the New York Mercantile Exchange (NYMEX—an example
prior to maturity; this is true independently of the values
is the Gasoline Crack Spread Option). Options to exchange
taken by underlying prices, i.e., even if they become very
one asset for another (exchange options) appear in vari-
ous financial contexts including convertible securities and large, S 1 = S 2 → ! Intuition for this property can perhaps
takeover attempts. be gained by considering the case of independent and sym-
Pricing formulas for European contracts were provided metric price processes. Consider a point along the diagonal
early on. Margrabe (1978), for instance, prices exchange S 1 = S 2 and suppose that the holder of the security must
options, whereas Johnson (1981) and Stulz (1982) deal with decide whether to exercise or wait. By independence the
max- and min-options. Results for American options are probability of an increase in the maximum of the two prices,
more recent. Broadie and Detemple (1997) study max-call over the next time increment, is roughly equal to 3/4. With
options, spread options, and related contracts (dual max- identical price volatilities value increases if the decision to
calls, min-put options, capped exchange options, etc.) exercise is postponed by an infinitesimal amount of time.
written on dividend-paying assets. Min-call options are This argument also prevails for prices off the diagonal but
examined by Villeneuve (1999) for non-dividend-paying sufficiently close to it. At those points immediate exercise
assets and Detemple et al. (2003) for dividend-paying assets. can be improved on by capitalizing on likely increases in
The main difference between options on a single underly- the max of the two prices over the next short time period.
ing price and those written on multiple prices rests with the The pricing methods presented for single-asset options
structure of the exercise region. In the single-asset case opti- also apply to multiasset options. The price of the max-
mal exercise can be described in terms of a single exercise call can be characterized as the solution of a free-boundary
boundary. This simple structure fails in the multiasset case: problem, or a variational inequality problem. The EEP rep-
Multiple exercise boundaries are typically needed in order resentation also applies. This representation gives rise to a
to describe the optimal exercise decision. This section illus- pair of coupled integral equations for the boundary compo-
trates the complexity of the exercise policy for the special
nents B 1 S 2 t and B 2 S 1 t.
case of a call on the maximum of two prices. The properties
Several contracts display exercise properties similar to
described were established in the BSM framework.
those of the max-call. For a dual max-call, with payoff
A max-call option written on two underlying prices pays
off S 1 ∨ S 2 − K+ at exercise. The function S 1 ∨ S 2 is not S 1 − K1 ∨ S 2 − K2 + , immediate exercise is suboptimal
differentiable along the diagonal S 1 = S 2 , thus this payoff along the translated diagonal S 1 = S 2 + K1 − K2 . The immedi-
does not have the usual smoothness properties. This feature ate exercise region can also be described by two subregions,
is a source of surprising exercise properties, described next. each defined by its own boundary. A spread call option,
Figure A1 illustrates the structure of the exercise region with payoff S 1 −S 2 −K+ , can be viewed as a one-sided ver-
for the contract. The most notable aspect is that this region sion of a dual max-call. Immediate exercise is suboptimal
is the union of two subregions, corresponding to sets where on or below the translated diagonal S 1 = S 2 + K; the exercise
Asset 1 is the maximum or Asset 2 is the maximum, with region is characterized by a single boundary, parametrized
disjoint time sections for t < T . That is, max = 1max ∪ 2max by S 1 t, that lies above this line. A min-put option, paying
where off K − S 1 ∧ S 2 + , has two exercise boundaries issued from
the origin and lying either above or below the diagonal.
1max = S 1 S 2 t ∈ 2+ ×0T 6 S 1 = S 1 ∨S 2 and S 1 ≥ B 1 S 2 t
Immediate exercise along the diagonal is also suboptimal
2max = S 1 S 2 t ∈ 2+ ×0T 6 S 2 = S 1 ∨S 2 and S 2 ≥ B 2 S 1 t for this contract.
The min-call option, with payoff S 1 ∧ S 2 − K+ , presents Barraquand, J. 1995. Numerical valuation of high dimensional
additional challenges for valuation. For this contract imme- multivariate European securities. Management Sci. 41(12)
diate exercise could be optimal along the diagonal. The 1882–1891.
exercise region has two exercise boundaries that meet and Bates, D. 1988. Pricing options under jump-diffusion processes.
Working paper, Rodney L. White Center for Financial Research,
merge along the diagonal (see Detemple et al. 2003). In the
University of Pennsylvania, Philadelphia, PA.
absence of dividend payments it collapses to a subset of the
Bates, D. 1996. Jumps and stochastic volatility: Exchange rate pro-
diagonal (see Villeneuve 1999). Valuation can be performed cesses implicit in Deutsche Mark options. Rev. Financial Stud. 9
using any of the methods previously discussed. One novel 69–107.
aspect is the appearance of a local time component in the Bates, D. 2000. Post-’87 crash fears in S&P 500 futures options.
EEP representation of the price. This term is related to the J. Econometrics 94 181–238.
discontinuities in the derivatives of the payoff function in Bates, D. 2003. Empirical option pricing: A retrospection. J. Econo-
the exercise region. metrics 116(1/2) 387–404.
Ben-Ameur, H., M. Breton, P. L’Ecuyer. 2002. A dynamic pro-
gramming procedure for pricing American-style Asian options.
References Management Sci. 48(5) 625–643.
Akahori, J. 1995. Some formulae for a new type of path-dependent Bensaid, B., J. P. Lesne, H. Pagès, J. Scheinkman. 1992. Derivative
option. Ann. Appl. Probab. 5 383–388. asset pricing with transaction costs. Math. Finance 2(2) 63–86.
Akesson, F., J. Lehoczky. 2000. Path generation for quasi–Monte Bensoussan, A. 1984. On the theory of option pricing. Acta Appl.
Carlo simulation of mortgage-backed securities. Management Math. 2 139–158.
Sci. 46(9) 1171–1187. Bensoussan, A., J. L. Lions. 1978. Applications des Inequations Varia-
Alford, J., N. Webber. 2001. Very high order lattice methods for one tionelles en Contrôle Stochastique. Dunod, Paris, France.
factor models. Working paper, University of Warwick, Coven- Bielecki, T., M. Rutkowski. 2002. Credit Risk: Modeling, Valuation and
try, U.K. Hedging. Springer-Finance, Berlin, Germany.
Amin, K. 1991. On the computation of continuous time option Birge, J. R. 1994. Quasi–Monte Carlo approaches to option pricing.
prices using discrete approximations. J. Financial Quant. Anal. Technical report 94-19, Department of Industrial and Opera-
26(4) 477–496. tions Engineering, University of Michigan, Ann Arbor, MI.
Amin, K. 1993. Jump-diffusion option valuation in discrete time. Bismut, J. M., B. Skalli. 1977. Temps d’arrêt, théorie générale de
J. Finance 48(5) 1833–1863. processus et processus de Markov. Z. Wahrscheinlichkeitstheorie
Amram, M., N. Kulatilaka. 1999. Real Options: Managing Strate- verwandte Gebiete 39 301–313.
gic Investment in an Uncertain World. Harvard Business School Bjerksund, P., G. Stensland. 1993. American exchange options and
Press, Boston, MA. a put-call transformation: A note. J. Bus. Finance, Accounting 20
Andersen, A., J. Andreasen. 2000. Jump-diffusion processes: Volatil- 761–764.
ity smile fitting and numerical methods for option pricing. Rev. Black, F. 1976. The pricing of commodity contracts. J. Financial
Derivatives Res. 4 231–262. Econom. 3 167–179.
Andersen, L. 2000. A simple approach to the pricing of Bermudan Black, F., J. Cox. 1976. Valuing corporate securities: Some effects of
swaptions in the multi-factor Libor market model. J. Comput. bond indenture provisions. J. Finance 31 351–367.
Finance 3(1) 5–32. Black, F., E. Derman, W. Toy. 1990. A one-factor model of interest
Andersen, L., M. Broadie. 2004. A primal-dual simulation algorithm rates and its applications to treasury bond options. Financial
for pricing multi-dimensional American options. Management Analysts J. 46(1) 33–39.
Sci. 50(9). Black, F., M. Scholes. 1973. The pricing of options and corporate
Andersen, L., D. Buffum. 2004. Calibration and implementation of liabilities. J. Political Econom. 81 637–654.
convertible bond models. J. Computational Finance 7(2) 1–34. Bollen, N. 1999. Real options and product life cycles. Management
Andersen, L., R. Brotherton-Ratcliffe. 2001. Extended Libor market Sci. 45 670–684.
models with stochastic volatility. Working paper, General Re Boyle, P. P. 1986. Option valuation using a three-jump process. Inter-
Financial Products, New York. nat. Options J. 3 7–12.
Andersen, L., J. Andreasen, R. Brotherton-Ratcliffe. 1998. The pass- Boyle, P. P., D. Emanuel. 1980. The pricing of options on the gen-
port option. J. Comput. Finance 1(3) 15–36. eralized mean. Working paper, University of British Columbia,
Anderson, R., S. Sundaresan. 1996. Design and valuation of debt Vancouver, BC, Canada.
contracts. Rev. Financial Stud. 9 37–68. Boyle, P. P., J. Evnine, S. Gibbs. 1989. Numerical evaluation of multi-
Andreasen, J. 1998. The pricing of discretely sampled Asian and variate contingent claims. Rev. Financial Stud. 2 241–250.
lookback options: A change of numeraire approach. J. Comput. Boyle, P. P., S. Lau. 1994. Bumping up against the barrier with the
Finance 2(1) 5–30. binomial method. J. Derivatives 1(4) 6–14.
Apel, T., G. Winkler, U. Wystup. 2002. Valuation of options in Boyle, P. P., T. Vorst. 1992. Option replication in discrete time with
Heston’s stochastic volatility model using finite element meth- transaction costs. J. Finance 47(1) 271–294.
ods. J. Hakala, U. Wystup, eds. Foreign Exchange Risk: Models, Boyle, P. P., M. Broadie, P. Glasserman. 1997. Monte Carlo meth-
Instruments and Strategies. Risk Books, London, U.K. ods for security pricing. J. Econom. Dynam. Control 21(8–9)
Artzner, P., F. Delbaen, J.-M. Eber, D. Heath. 1999. Coherent mea- 1276–1321.
sures of risk. Math. Finance 9(3) 203–228. Brace, A., D. Gatarek, M. Musiela. 1997. The market model of inter-
Babbs, S. 1992. Binomial valuation of lookback options. Working est rate dynamics. Math. Finance 7 127–154.
paper, Midland Global Markets, London, U.K. Brennan, M., E. Schwartz. 1977. The valuation of American put
Bakshi, G., C. Cao, Z. Chen. 1997. Empirical performance of alter- options. J. Finance 32 449–462.
native option pricing models. J. Finance 52 2003–2049. Brennan, M., E. Schwartz. 1978. Finite difference methods and jump
Barlow, M. T. 2002. A diffusion model for electricity prices. Math. processes arising in the pricing of contingent claims: A synthe-
Finance 12(4) 287–298. sis. J. Financial Quant. Anal. 13 461–474.
Brennan, M., E. Schwartz. 1985. Evaluating natural resource invest- Cont, R., P. Tankov. 2004. Financial Modelling with Jump Processes.
ments. J. Bus. 58(2) 135–157. Chapman & Hall/CRC, Boca Raton, FL.
Brezinski, C., M. Zaglia. 1991. Extrapolation Methods: Theory and Copeland, T., V. Antikarov. 2001. Real Options: A Practitioner’s Guide.
Practice. North-Holland, New York. Texere, New York.
Broadie, M., J. B. Detemple. 1995. American capped call options on Cortazar, G., E. Schwartz, M. Salinas. 1998. Evaluating environmen-
dividend-paying assets. Rev. Financial Stud. 8 161–191. tal investments: A real options approach. Management Sci. 44(8)
Broadie, M., J. B. Detemple. 1996. American option valuation: New 1059–1070.
bounds, approximations, and a comparison of existing meth- Cox, J. C. 1975. Notes on options I: Constant elasticity of variance
ods. Rev. Financial Stud. 9 1211–1250. diffusion. Working paper, Stanford University, Stanford, CA.
Broadie, M., J. B. Detemple. 1997. The valuation of American Cox, J. C. 1996. The constant elasticity of variance option pricing
options on multiple assets. Math. Finance 7 241–285. model. J. Portfolio Management 22 15–17.
Broadie, M., P. Glasserman. 1996. Estimating security price deriva- Cox, J. C., S. Ross. 1976. The valuation of options for alternative
tives using simulation. Management Sci. 42(2) 269–285. stochastic processes. J. Financial Econom. 3 145–166.
Broadie, M., P. Glasserman. 1997. Pricing American-style securities Cox, J. C., J. Ingersoll, S. Ross. 1985. A theory of the term structure
using simulation. J. Econom. Dynam. Control 21(8–9) 1323–1352. of interest rates. Econometrica 53 385–407.
Broadie, M., P. Glasserman. 2004. A stochastic mesh method for Cox, J. C., S. A. Ross, M. Rubinstein. 1979. Option pricing: A sim-
pricing high-dimensional American options. J. Comput. Finance plified approach. J. Financial Econom. 7 229–263.
7(4) 35–72. Cuoco, D. 1997. Optimal consumption and equilibrium prices with
Broadie, M., Y. Yamamoto. 2003. Application of the fast Gauss trans- portfolio constraints and stochastic income. J. Econom. Theory
form to option pricing. Management Sci. 49(8) 1071–1088. 72 33–73.
Broadie, M., J. Cvitanić, M. Soner. 1998. Optimal replication of con- Curran, M. 1994. Valuing Asian and portfolio options by condition-
tingent claims under portfolio constraints. Rev. Financial Stud. ing on the geometric mean price. Management Sci. 40 1705–1711.
11(1) 59–79.
Cvitanić, J., W. Schachermayer, H. Wang. 2001. Utility maximiza-
Broadie, M., P. Glasserman, G. Jain. 1997. Enhanced Monte Carlo tion in incomplete markets with random endowment. Finance
estimates for American option prices. J. Derivatives 5(1) 25–44. Stochastics 5 259–272.
Canter, M., J. Cole, R. Sandor. 1997. Insurance derivatives: A new Cvsa, V., P. Ritchken. 2001. Pricing claims under GARCH-level
asset class for the capital markets and a new hedging tool for dependent interest rate processes. Management Sci. 47(12)
the insurance industry. J. Appl. Corporate Finance 10(3) 69–83. 1693–1711.
Carr, P., R. Jarrow. 1990. The stop-loss start-gain paradox and option Dai, Q., K. Singleton. 2000. Specification analysis of affine term
valuation: A new decomposition into intrinsic and time value. structure models. J. Finance 55(5) 1943–1978.
Rev. Financial Stud. 3 469–492.
Das, S., R. Sundaram. 2000. A discrete-time approach to arbitrage-
Carr, P., D. Madan. 1998. Towards a theory of volatility trading. free pricing of credit derivatives. Management Sci. 46(1) 46–62.
R. A. Jarrow, ed. Volatility: New Estimation Techniques for Pricing
Derivatives. Risk Books, London, U.K., 417–427. Dassios, A. 1995. The distribution of the quantile of a Brownian
motion with drift and the pricing of related path-dependent
Carr, P., L. Wu. 2003. The finite moment logstable process and options. Ann. Appl. Probab. 5 389–398.
option pricing. J. Finance 58(2) 753–777.
Davydov, D., V. Linetsky. 2001. The valuation and hedging of bar-
Carr, P., R. Jarrow, R. Myneni. 1992. Alternative characterizations
rier and lookback options under the CEV process. Management
of American put options. Math. Finance 2 87–106.
Sci. 47 949–965.
Carr, P., H. Geman, D. Madan, M. Yor. 2002. The fine structure of
Debreu, G. 1959. Theory of Value. Wiley, New York.
asset returns: An empirical investigation. J. Bus. 75 305–332.
Delbaen, F., W. Schachermayer. 1994. A general version of the fun-
Carr, P., H. Geman, D. Madan, M. Yor. 2003. Stochastic volatility for
damental theorem of asset pricing. Math. Ann. 300 463–520.
Lévy processes. Math. Finance 13(3) 345–382.
Delbaen, F., W. Schachermayer. 1996. The variance-optimal martin-
Carriere, J. 1996. Valuation of the early-exercise price for options
gale measure for continuous processes. Bernoulli 2 81–105.
using simulations and nonparametric regression. Insurance,
Math., Econom. 19 19–30. Demeterfi, K., E. Derman, M. Kamal, J. Zou. 1999. A guide to
volatility and variance swaps. J. Derivatives 6 9–32.
Chen, R.-R., L. Scott. 1992. Pricing interest rate options in a two-
factor Cox-Ingersoll-Ross model of the term structure. Rev. Derman, E., I. Kani. 1994. Riding on a smile. Risk 7 32–39.
Financial Stud. 5 613–636. Detemple, J. B. 2001. American options: Symmetry properties.
Chernov, M., E. Ghysels. 2000. A study towards a unified approach J. Cvitanić, E. Jouini, M. Musiela, eds. Handbooks in Mathe-
to the joint estimation of objective and risk neutral measures matical Finance: Topics in Option Pricing, Interest Rates and Risk
for the purpose of options valuation. J. Financial Econom. 56 Management. Cambridge University Press, Cambridge, U.K.,
407–458. 67–104.
Chernov, M., E. Ghysels, A. Gallant, G. Tauchen. 2003. Alternative Detemple, J., C. Osakwe. 2000. The valuation of volatility options.
models for stock price dynamics. J. Econometrics 116 225–257. Eur. Finance Rev. 4(1) 21–50.
Chesney, M., R. Gibson. 1995. State space symmetry and two-factor Detemple, J. B., S. Sundaresan. 1999. Non-traded asset valuation
option pricing models. Adv. Futures Options Res. 8 85–112. with portfolio constraints: A binomial approach. Rev. Financial
Chesney, M., M. Jeanblanc-Picqué, M. Yor. 1997. Brownian excur- Stud. 12 835–872.
sions and Parisian barrier options. Adv. Appl. Probab. 29 Detemple, J., W. Tian. 2002. The valuation of American options for
165–184. a class of diffusion processes. Management Sci. 48 917–937.
Chesney, M., J. Cornwall, M. Jeanblanc-Picqué, G. Kentwell, M. Yor. Detemple, J. B., S. Feng, W. Tian. 2003. The valuation of American
1997. Parisian pricing. Risk 1 77–79. call options on the minimum of two dividend-paying assets.
Clewlow, L., C. Strickland. 2000. Energy Derivatives: Pricing and Risk Ann. Appl. Probab. 13 953–983.
Management. Lacima Publications, London, U.K. Diener, F., M. Diener. 2004. Asymptotics of the price oscillations of
Collin-Dufresne, P., R. Goldstein. 2002. Do bonds span fixed income a vanilla option in a tree model. Math. Finance 13(2) 271–294.
markets? Theory and evidence for “unspanned” stochastic Dixit, A., R. Pindyck. 1994. Investment under Uncertainty. Princeton
volatility. J. Finance 57(4) 1685–1730. University Press, Princeton, NJ.
Druskin, V., L. Knizhnerman, T. Tamarchenko, S. Kostek. 1997. Fouque, J.-P., G. Papanicolaou, R. Sircar. 2000. Derivatives in Finan-
Krylov subspace reduction and its extensions for option pric- cial Markets with Stochastic Volatility. Cambridge University
ing. J. Comput. Finance 1(1) 63–79. Press, Cambridge, U.K.
Duan, J. C. 1995. The GARCH option pricing model. Math. Finance Fritelli, M. 2000. The minimal entropy martingale measure and the
5 13–32. valuation problem in incomplete markets. Math. Finance 10
Duan, J.-C., J.-G. Simonato. 1998. Empirical martingale simulation 39–52.
for asset prices. Management Sci. 44(9) 1218–1233. Gao, B., J.-Z. Huang, M. Subrahmanyam. 2000. The valuation of
Duan, J.-C., G. Gauthier, J.-G. Simonato. 2001. Asymptotic distribu- American barrier options using the decomposition technique.
tion of the EMS option price estimator. Management Sci. 47(8) J. Econom. Dynam. Control 24 1783–1827.
1122–1132. Garcia, R., E. Ghysels, E. Renault. 2004. The econometrics of option
Duffie, D. 1986. Stochastic equilibria: Existence, spanning number, pricing. Y. Ait-Sahalia, L. Hansen, eds. Handbook of Finan-
and the “no expected financial gain from trade” hypothesis. cial Econometrics. Elsevier-North Holland, Amsterdam, The
Econometrica 54 1161–1183. Netherlands, Forthcoming.
Geman, H., ed. 1999. Insurance and Weather Derivatives From Exotic
Duffie, D., R. Kan. 1996. A yield-factor model of interest rates. Math.
Options to Exotic Underlyings. Risk Books, London, U.K.
Finance 6 379–406.
Geman, H., M. Yor. 1993. Bessel processes, Asian options and per-
Duffie, D., K. Singleton. 1999. Modeling term structures of default-
petuities. Math. Finance 3 349–375.
able bonds. Rev. Financial Stud. 12 687–720.
Geman, H., N. El Karoui, J.-C. Rochet. 1995. Changes of numeraire,
Duffie, D., K. Singleton. 2003. Credit Risk. Princeton University
changes of probability measure and option pricing. J. Appl.
Press, Princeton, NJ.
Probab. 32 443–458.
Duffie, D., J. Pan, K. Singleton. 2000. Transform analysis and asset
Geske, R., H. Johnson. 1984. The American put option valued ana-
pricing for affine jump-diffusions. Econometrica 68 1343–1376. lytically. J. Finance 39 1511–1524.
Duffie, D., W. Fleming, M. Soner, T. Zariphopoulou. 1997. Hedg- Ghysels, E., A. Harvey, E. Renault. 1996. Stochastic volatility. G. S.
ing in incomplete markets with Hara utility. J. Econom. Dynam. Madal, ed. Handbook of Statistics, Vol. 14. Statistical Methods in
Control 21 753–782. Finance. North Holland, Amsterdam, The Netherlands.
Dufresne, D. 2000. Laguerre series for Asian and other options. Glasserman, P. 2004. Monte Carlo Methods in Financial Engineering.
Math. Finance 10 407–428. Springer-Verlag, New York.
Dunn, K., C. Spatt. 1985. An analysis of mortgage contracting: Pre- Glasserman, P., S. G. Kou. 2003. The term structure of simple for-
payment penalties and the due-on-sale clause. J. Finance 40 ward rates with jump risk. Math. Finance 13 383–410.
293–308.
Glasserman, P., P. Heidelberger, P. Shahabuddin. 2000. Variance
Dupire, B. 1994. Pricing with a smile. Risk 7 18–20. reduction techniques for estimating value-at-risk. Management
Dybvig, P. H., S. A. Ross. 1987. Arbitrage. J. Eatwell, M. Milgate, Sci. 46 1349–1364.
P. Newman, eds. The New Palgrave: Finance. Norton, New York. Grabbe, O. 1983. The pricing of call and put options on foreign
El Karoui, N. 1981. Les aspects probabilistes du contrôle stochas- exchange. J. Internat. Money Finance 2 239–253.
tique. Lecture notes in Mathematics 876, Springer-Verlag, Gukhal, C. R. 2001. Analytical valuation of American options on
Berlin, Germany, 73–238. jump-diffusion processes. Math. Finance 11(1) 97–115.
El Karoui, N., I. Karatzas. 1991. A new approach to the Skorohod Hagan, P., D. Woodward. 1999. Equivalent Black volatilities. Appl.
problem and its applications. Stochastics Stochastics Rep. 34 Math. Finance 6(3) 147–157.
57–82. Hansen, A., P. Jorgensen. 2000. Analytical valuation of American-
Embrecht, P., L. C. G. Rogers, M. Yor. 1995. A proof of Dassios’ style Asian options. Management Sci. 46(8) 1116–1136.
representation of the -quantile of Brownian motion with drift. Harrison, M., D. Kreps. 1979. Martingales and arbitrage in multi-
Ann. Appl. Probab. 5 757–767. period security markets. J. Econom. Theory 20 381–408.
Engle, R. F. 1982. Autoregressive conditional heteroscedasticity Harrison, M., S. Pliska. 1981. Martingales and stochastic integrals
with estimates of the variance of United Kingdom inflation. in the theory of continuous trading. Stochastic Processes Appl.
Econometrica 50(4) 987–1008. 11 215–260.
Eraker, B., M. Johannes, N. Polson. 2003. The impact of jumps in Haugh, M., L. Kogan. 2004. Approximating pricing and exercising
equity index volatility and returns. J. Finance 58 1269–1300. of high-dimensional American options: A duality approach.
Fakeev, A. G. 1971. Optimal stopping of a Markov process. Theory Oper. Res. 52(2) 258–270.
Probab. Appl. 16 694–696. Heath, D., R. Jarrow, A. Morton. 1990. Bond pricing and the
Föllmer, H., M. Schweizer. 1991. Hedging of contingent claims term structure of interest rates: A discrete time approximation.
under incomplete information. M. H. A. Davis, R. J. Elliott, J. Financial Quant. Anal. 25(4) 419–440.
eds. Applied Stochastic Analysis Stochastics Monographs, Vol. 5. Heath, D., R. Jarrow, A. Morton. 1992. Bond pricing and the term
Gordon and Breach, London, U.K., 389–414. structure of interest rates: A new methodology for contingent
Föllmer, H., D. Sondermann. 1986. Hedging of non-redundant con- claims valuation. Econometrica 60 77–105.
tingent claims. W. Hildenbrand, A. MasColell, eds. Contribu- Heston, S. L. 1993. A closed-form solution for options with stochas-
tions to Mathematical Economics: Essays in Honor of G. Debreu. tic volatility with applications to bond and currency options.
North-Holland, Amsterdam, The Netherlands, 205–223. Rev. Financial Stud. 6 327–343.
Forsyth, P., K. Vetzal. 2002. Quadratic convergence for valuing Heston, S., G. Zhou. 2000. On the rate of convergence of discrete-
American options using a penalty method. SIAM J. Sci. Com- time contingent claims. Math. Finance 10(1) 53–75.
put. 23(6) 2095–2122. Hilliard, J., J. Kau, D. Keenan, W. Muller III. 1995. Pricing a class
Forsyth, P., K. Vetzal, R. Zvan. 1999. A finite element approach to of American and European path-dependent securities. Manage-
the pricing of discrete lookbacks with stochastic volatility. Appl. ment Sci. 41(12) 1892–1889.
Math. Finance 6 87–106. Hirsa, A., D. Madan. 2004. Pricing American options under vari-
Forsyth, P., K. Vetzal, R. Zvan. 2002. Convergence of lattice and ance gamma. J. Comput. Finance 7(2) 63–80.
PDE methods for valuing path-dependent options using inter- Ho, T., S.-B. Lee. 1986. Term structure movements and the pricing
polation. Rev. Derivatives Res. 5 273–314. of interest rate contingent claims. J. Finance 41 1011–1029.
Hodges, S., A. Neuberger. 1989. Optimal replication of contingent Kemna, A. G. Z., A. C. F. Vorst. 1990. A pricing method for options
claims under transaction costs. Rev. Futures Markets 8 222–239. based on average asset values. J. Banking Finance 14 113–129.
Hugonnier, J. 1999. The Feynman-Kac formula and pricing occu- Kholodnyi, V. A., J. F. Price. 1998. Foreign Exchange Option Symmetry.
pation time derivatives. Internat. J. Theoret. Appl. Finance 2 World Scientific Publishing Co., London, U.K.
153–178. Kim, I. J. 1990. The analytic valuation of American options. Rev.
Hugonnier, J., D. Kramkov. 2004. Optimal investment with random Financial Stud. 3 547–572.
endowments in incomplete markets. Ann. Appl. Probab. 14(2) Kinderlehrer, D., G. Stampacchia. 1980. An Introduction to Variational
845–864. Inequalities and their Applications. Academic Press, New York.
Hull, J. 2003. Options, Futures and Other Derivatives, 5th ed. Prentice Kloeden, P. E., E. Platen. 1999. Numerical Solution of Stochastic Dif-
Hall, Englewood Cliffs, NJ.
ferential Equations, 3rd ed. Springer-Verlag, Berlin, Germany.
Hull, J., A. White. 1987. The pricing of options on assets with
Kou, S. G. 2002. A jump-diffusion model for option pricing. Man-
stochastic volatilities. J. Finance 42 281–300.
agement Sci. 48 1086–1101.
Hull, J., A. White. 1994. Numerical procedures for implementing
Kreps, D. M. 1981. Arbitrage and equilibrium in economies with
term structure models I: Single factor models. J. Derivatives 2(1)
infinitely many commodities. J. Math. Econom. 8 15–35.
7–16.
Hull, J., A. White. 1995. The impact of default risk on the prices Kulatilaka, N., E. Perotti. 1998. Strategic growth options. Manage-
of options and other derivative securities. J. Banking Finance 19 ment Sci. 44(8) 1021–1031.
299–322. Kurpiel, A., T. Roncalli. 2000. Hopscotch methods for two-state
Ingersoll, J. 1987. Theory of Financial Decision Making. Rowman and financial models. J. Computational Finance 3(2) 53–89.
Littlefield Publishers, NJ. Lee, R. 2004. Option pricing by transform methods: Extensions, uni-
Jacka, S. D. 1991. Optimal stopping and the American put. Math. fication, and error control. J. Computational Finance 7(3) 51–86.
Finance 1 1–14. Leland, H. 1985. Option pricing and replication with transaction
Jacod, J., A. N. Shiryaev. 1998. Local martingales and the funda- costs. J. Finance 40(5) 1283–1301.
mental asset pricing theorem in the discrete-time case. Finance Leland, H. 1994. Corporate debt value, bond covenants, and opti-
Stochastics 2 259–273. mal capital structure. J. Finance 49 1213–1252.
Jaillet, P., D. Lamberton, B. Lapeyre. 1990. Variational inequali- Lewis, A. 2000. Option Valuation under Stochastic Volatility. Finance
ties and the pricing of American options. Acta Appl. Math. 21 Press, Newport Beach, CA.
263–289. Li, D. X. 2000. On default correlation: A copula function approach.
James, J., N. Webber. 2000. Interest Rate Modelling. John Wiley and J. Fixed Income 9(4) 43–54.
Sons, Chichester, U.K. Linetsky, V. 1999. Step options. Math. Finance 9 55–96.
Jamshidian, F. 1989. An exact bond option pricing formula. Linetsky, V. 2004. Spectral expansions for Asian (average price)
J. Finance 44 205–209. options. Oper. Res. Forthcoming.
Jamshidian, F. 1991. Forward induction and construction of yield
Little, T., V. Pant, C. Hou. 2000. A new integral representation of the
curve diffusion models. J. Fixed Income 1 62–74.
early exercise boundary for American put options. J. Comput.
Jarrow, R., A. Rudd. 1983. Option Pricing. Dow Jones-Irwin, Finance 3(3) 73–96.
Homewood, IL.
Litzenberger, R., D. Beaglehole, C. Reynolds. 1996. Assessing catas-
Jarrow, R., S. Turnbull. 1995. Pricing options on financial securities trophe reinsurance-linked securities as a new asset class.
subject to default risk. J. Finance 50 53–86. J. Portfolio Management (Winter Special Issue) 76–86.
Johannes, M. 2004. The statistical and economic role of jumps in Longstaff, F., E. Schwartz. 2001. Valuing American options by sim-
continous-time interest rate models. J. Finance 59(1) 227–260. ulation: A simple least squares approach. Rev. Financial Stud.
Johnson, H. 1981. The pricing of complex options. Working paper, 14 113–148.
Louisiana State University, LA.
Lucas, R. E. 1978. Asset prices in an exchange economy. Economet-
Jorion, P. 2001. Value at Risk: The Benchmark for Controlling Market rica 46 1429–1445.
Risk, 2nd ed. McGraw-Hill, New York.
Lucia, J., E. Schwartz. 2002. Electricity prices and power derivatives:
Joy, C., P. P. Boyle, K. S. Tan. 1996. Quasi–Monte Carlo methods in Evidence from the Nordic power exchange. Rev. Derivatives Res.
numerical finance. Management Sci. 42 926–938. 5(1) 5–50.
Kabanov, Y., C. Stricker. 2001. The Harrison-Pliska arbitrage pricing Madan, D., P. Carr, E. Chang. 1998. The variance gamma process
theorem under transaction costs. J. Math. Econom. 35 185–196. and option pricing. Eur. Financial Rev. 2 79–105.
Kamrad, B., P. Ritchken. 1991. Multinomial approximating mod- Margrabe, W. 1978. The value of an option to exchange one asset
els for options with k state variables. Management Sci. 37 for another. J. Finance 33 177–186.
1640–1652.
McConnell, J., M. Singh. 1993. Valuation and analysis of collateral-
Karatzas, I. 1988. On the pricing of American options. Appl. Math.
ized mortgage obligations. Management Sci. 39(6) 692–709.
Optim. 17 37–60.
McDonald, R., M. Schroder. 1998. A parity result for American
Karatzas, I., S. G. Kou. 1996. On the pricing of contingent claims
options. J. Comput. Finance 1 5–13.
under constraints. Ann. Appl. Probab. 6(2) 321–369.
McKean, H. P. 1965. A free boundary problem for the heat equation
Karatzas, I., S. Shreve. 1988. Brownian Motion and Stochastic Calculus.
arising from a problem in mathematical economics. Indust.
Springer Verlag, New York.
Management Rev. 6 32–39.
Karatzas, I., S. Shreve. 1998. Methods of Mathematical Finance.
Springer Verlag, New York. Merton, R. C. 1973. Theory of rational option pricing. Bell J. Econom.
Management Sci. 4 141–183.
Karatzas, I., G. Zitkovic. 2003. Optimal consumption from invest-
ment and random endowment in incomplete semimartingale Merton, R. C. 1974. On the pricing of corporate debt: The risk struc-
markets. Ann. Probab. 31(4) 1821–1858. ture of interest rates. J. Finance 29 449–470.
Kau, J., D. Keenan, W. Muller, III, J. Epperson. 1990. The valua- Merton, R. C. 1976. Option pricing when underlying stock returns
tion and analysis of adjustable rate mortgages. Management Sci. are discontinuous. J. Financial Econom. 3 125–144.
36(12) 1417–1431. Merton, R. C. 1992. Continuous Time Finance. Blackwell Publishing.
Miltersen, K., K. Sandmann, D. Sondermann. 1997. Closed form Schröder, M. 2002. Mathematical ramifications of option valua-
solutions for term structure derivatives with lognormal interest tion: The case of Asian options. Habilitationsschrift Universitat,
rates. J. Finance 52 409–430. Mannheim, Germany.
Miura, R. 1992. A note on look-back option based on order statis- Schwartz, E. S., C. Zozaya-Gorostiza. 2003. Investment under uncer-
tics. Hitosubashi J. Commerce Management 27 15–28. tainty in information technology: Acquisition and develop-
Morton, K. W., D. F. Mayers. 1994. Numerical Solutions of Partial ment. Management Sci. 49(1) 57–70.
Differential Equations. Cambridge University Press, Cambridge, Schwartz, E. S., J. Smith. 2000. Short-term variations and long-term
U.K. dynamics in commodity prices. Management Sci. 46(7) 893–911.
Naik, V., M. Lee. 1990. General equilibrium pricing of options on Schwartz, E. S., W. Torous. 1989. Prepayment and the valuation of
the market portfolio with discontinuous returns. Rev. Financial mortgage-backed securities. J. Finance 44 375–392.
Stud. 3(4) 493–521. Schweizer, M. 1992. Mean-variance hedging for general claims.
Neuberger, A. 1994. The log contract. J. Portfolio Management Ann. Appl. Probab. 2 171–179.
20(Winter) 74–80. Schweizer, M. 1995a. On the minimal martingale measure and
Niederreiter, H. 1992. Random Number Generation and Quasi–Monte the Föllmer-Schweizer decomposition. Stochastic Anal. Appl. 13
Carlo Methods. Society for Industrial and Applied Mathematics, 573–599.
Philadelphia, PA. Schweizer, M. 1995b. Variance-optimal hedging in discrete time.
Owen, A. B. 1997. Scrambled net variance for integrals of smooth Math. Oper. Res. 20 1–32.
functions. Ann. Statist. 25 1541–1562. Scott, L. 1997. Pricing stock options in a jump-diffusion model with
Pan, J. 2002. The jump-risk premia implicit in options: Evidence stochastic volatility and stochastic interest rates: Application of
from an integrated time-series study. J. Financial Econom. 63 Fourier inversion methods. Math. Finance 7 413–426.
3–50. Shiller, R. 2003. The New Financial Order: Risk in the 21st Century.
Parkinson, M. 1977. Option pricing: The American put. J. Bus. 50 Princeton University Press, Princeton, NJ.
21–36. Shiryaev, A. N. 1978. Optimal Stopping Rules. Springer-Verlag,
Paskov, S., J. Traub. 1995. Faster valuation of financial derivatives. New York.
J. Portfolio Management 22 113–120. Shreve, S., J. Vecer. 2000. Options on a traded account: Vacation
Pelsser, A. 2000. Efficient Methods for Valuing Interest Rate Derivatives. calls, vacation puts and passport options. Finance Stochastics 4
Springer Finance, London, U.K. 255–274.
Pham, H., N. Touzi. 1999. The fundamental theorem of asset pricing Stein, E. M., J. C. Stein. 1991. Stock price distributions with stochas-
under cone constraints. J. Math. Econom. 31 265–280. tic volatility: An analytical approach. Rev. Financial Stud. 4(4)
727–752.
Pilipović, D. 1998. Energy Risk. McGraw-Hill, New York.
Stulz, R. M. 1982. Options on the minimum or the maximum of
Pratt, J. W. 1964. Risk aversion in the small and the large. Econo- two risky assets. J. Financial Econom. 10 161–185.
metrica 32 122–136.
Sundaresan, S. 2002. Fixed Income Markets and Their Derivatives, 2nd
Randall, C., E. Kant, A. Chhabra. 1997. Using program synthesis to ed. South-Western, Cincinnati, OH.
price derivatives. J. Comput. Finance 1(2) 97–129.
Tavella, D., C. Randall. 2000. Pricing Financial Instruments: The Finite
Rannacher, R. 1984. Finite element solution of diffusion problems Difference Method. John Wiley and Sons, New York.
with irregular data. Num. Math. 43 309–327.
Taylor, H. M. 1967. Evaluating a call option and optimal timing
Reiner, E. 1990. Valuation of path-dependent options. Presenta- strategy in the stock market. Management Sci. 14 111–120.
tion at the Waterloo Options Conference, Waterloo, Ontario,
Trigeorgis, L. 1996. Real Options: Managerial Flexibility and Strategy
Canada.
in Resource Allocation. MIT Press, Cambridge, MA.
Ritchken, P., L. Sankarasubramanian, A. Vijh. 1993. The valuation Tsitsiklis, J., B. Van Roy. 1999. Optimal stopping of Markov pro-
of path-dependent contracts on the average. Management Sci. cesses: Hilbert space theory, approximation algorithms, and an
39(10) 1202–1213. application to pricing high-dimensional financial derivatives.
Rogers, C. 2002. Monte Carlo valuation of American options. Math. IEEE Trans. Automatic Control 44 1840–1851.
Finance 12 271–286. Tsitsiklis, J., B. Van Roy. 2001. Regression methods for pricing com-
Rogers, C., Z. Shi. 1995. The value of an Asian option. J. Appl. plex American-style options. IEEE Trans. Neural Networks 12(4)
Probab. 32 1077–1088. (Special issue on computational finance) 694–703.
Rubinstein, M. 1994. Implied binomial trees. J. Finance 49 Vasicek, O. 1977. An equilibrium characterization of the term struc-
771–818. ture. J. Financial Econom. 5 177–188.
Rubinstein, M., E. Reiner. 1991. Breaking down the barriers. Risk 4 Vecer, J. 2001. A new PDE approach for pricing arithmetic average
28–35. Asian options. J. Comput. Finance 4 105–113.
Rudd, A., M. Schroeder. 1982. The calculation of minimum margin. Villaneuve, S. 1999. Exercise regions of American options on several
Management Sci. 28 1369–1379. assets. Finance Stochastics 3(3) 295–322.
Rutkowski, M. 1994. The early exercise premium representa- Wilmott, P., J. Dewynne, S. Howison. 1993. Option Pricing: Mathe-
tion of foreign market American options. Math. Finance 4 matical Models and Computation. Oxford Financial Press, Oxford,
313–325. U.K.
Samuelson, P. A. 1965. Rational theory of warrant pricing. Indust. Yor, M. 1995. The distribution of Brownian quantiles. J. Appl. Probab.
Management Rev. 6 13–31. 32 405–416.
Schönbucher, P. 2003. Credit Derivatives Pricing Models: Model, Pric- Zipkin, P. 1993. Mortgages and Markov chains: A simplified eval-
ing and Implementation. John Wiley and Sons, New York. uation model. Management Sci. 39(6) 683–691.
Schröder, M. 1999. Changes of numeraire for pricing futures, for- Zvan, R., P. Forsyth, K. Vetzal. 1998. Robust numerical methods for
wards and options. Rev. Financial Stud. 12 1143–1163. PDE models of Asian options. J. Comput. Finance 2 39–78.

2004 Option Pricing Valuation Models and Applications

Caricato da

Informazioni sul documento

Descrizione originale:

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

2004 Option Pricing Valuation Models and Applications

Caricato da

Copyright:

Formati disponibili

MANAGEMENT SCIENCE informs ®

Vol. 50, No. 9, September 2004, pp. 1145–1177 doi 10.1287/mnsc.1040.0275

50th Anniversary Article

1. Derivatives Markets: maturity date T , for a prespeciﬁed price K, called the

1 S 1 2 %c St t units of the underlying asset. Because delta

same as the portfolio value. That is, (17)

will retain the martingale property relative to a suit- j j j j j

where cSt t! K is the Black-Scholes Formula (6) and where d W

almost any underlying asset (equities, commodi- 6%

30% 40 Asian Currency

tion of K is often called the implied volatility skew

characteristic function under the new measure can be

BSM Implied Volatility

T > lnK − e−rT KQxT > lnK

sum of BSM formulas. Although the expressions (48) 25%

Lattice Outcome Probability

option is in fact an up-and-out barrier option with barrier T

A.3. Multiasset Options 0 Bt1 S1

Potrebbero piacerti anche

2004 Option Pricing Valuation Models and Applications

Caricato da

Informazioni sul documento

Descrizione originale:

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

2004 Option Pricing Valuation Models and Applications

Caricato da

Copyright:

Formati disponibili

MANAGEMENT SCIENCE informs ®

Vol. 50, No. 9, September 2004, pp. 1145–1177 doi 10.1287/mnsc.1040.0275

50th Anniversary Article

1. Derivatives Markets: maturity date T , for a prespeciﬁed price K, called the

1 S 1 2 %c St  t units of the underlying asset. Because delta

same as the portfolio value. That is, (17)

will retain the martingale property relative to a suit- j j j j j

where cSt  t! K is the Black-Scholes Formula (6) and where d W 

almost any underlying asset (equities, commodi- 6%

30% 40 Asian Currency

tion of K is often called the implied volatility skew

characteristic function under the new measure can be

BSM Implied Volatility

 T > lnK − e−rT KQxT > lnK

sum of BSM formulas. Although the expressions (48) 25%

Lattice Outcome Probability

option is in fact an up-and-out barrier option with barrier T

A.3. Multiasset Options 0 Bt1 S1

Potrebbero piacerti anche

1 S 1 2 %c St t units of the underlying asset. Because delta

where cSt t! K is the Black-Scholes Formula (6) and where d W

T > lnK − e−rT KQxT > lnK

option is in fact an up-and-out barrier option with barrier T