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

# (Revised version, December 2, 1995) Exotic options are a generic name given to derivative securities which have more

complex cash- ow structures than standard puts and calls. The principal motivation for trading exotic options is that they permit a much more precise articulation of views on future market behavior than those o ered by \vanilla" options. Like options, exotics can be used as part of a risk-management strategy or for speculative purposes. From the investor's perspective, some exotics provide high leverage because they can focus the payo structure very precisely (this is the case of barrier options discussed below). Exotics are usually traded over the counter and are marketed to sophisticated corporate investors or hedge funds. Exotic option dealers are generally banks or investment houses. They manage their risk-exposure by making two-way markets and attempting to be market-neutral as much as possible, and hedging with their \vanilla" option book and cash instruments. The risk-management of exotics is more delicate than that of standard options because they are less liquid. This means that the seller of an exotic may not be able to buy it back if his theoretical hedging strategy failed without having to pay a large premium. Therefore, making a market in exotic options requires acute timing skills in hedging and the use of options to manage volatility risk. Roughly speaking, we can say that \ exotic options are to standard options what options are to the cash market". By this we mean that exotic options are very sensitive to higher-order derivatives of option prices such as Gamma and Vega. Some exotics can be seen essentially as bets on the future behavior of higher order \Greeks" Gamma and Vega. Another important issue is the notion of pin risk: since some exotics have discontinuous payo s, they can have huge Deltas and Gammas near expiration that make them very di cult, if not impossible, to Delta-hedge.

EXOTIC OPTIONS I

**1. List of the most common exotics
**

Digital options Barrier options Look-back options Average-rate options (Asian options)

Typeset by AMS-TEX

2

Options on baskets Forward-start options Compound options (options on options) This lecture gives an introduction to these derivatives. We will discuss various aspects of exotics, namely (i) binomial tree pricing, (ii) closed-form solutions assuming the spot price follows a Geometric Brownian Motion, (iii) price sensitivity and hedging. The second point will require that we develop several mathematical results on the distribution of rst-passage times and of the supremum of Brownian motion with drift over a given time-interval. Aside from providing an introduction to these instruments, this study is interesting because it gives us a better perspective on the risks associated with hedging derivative products in general, including the risk-management of portfolios of standard options.1

2. Digital options

A digital, or binary option is a contingent claim on some underlying asset or commodity that has an \all-or-nothing" payo . A digital call has a the payo

F (ST ) =

81 > < >0 :

if ST if ST

K K

(1)

(A digital put has payo 1 ; F (ST )). Like standard options, digitals can be classi ed as European or American style. The European digital provides a payo of $1 if the asset end above the strike price at the option's maturity date and zero otherwise. The American digital has a payo of $1 if the underlying asset reaches the value K before or at the expiration date T .

2.1 European Digitals

The fair value of the digital call with payo (1) can be derived easily under the assumptions of lognormal prices (the \Black-Scholes world'). In fact, the fair value of the digital is given by

V (S T ) = e;r T E f H (ST

1

;

K )g = e;r T P fST

Kg

(2)

The material for this lecture was taken from various research publications and my own notes. Recommended reading: (a) J.Hull: An introduction to ..., chapter on exotics (b) Mark Rubinstein: Exotic options, preprint Berkeley University, 1992, (a compilation of his articles in RISK Magazine.) and (c) \From Black-Scholes to black holes", anther compilation of articles by several authors from RISK Magazine.

1 2 ) T 2 = K ZK = Therefore.q. The calculation of the last probability in (2) is straightforward. p T (3) we conclude that the fair value of the European digital call is given by V (S T ) = e.. 1 ZK T p +(r.q) T K T . namely .ZK ) : Here N ( ) is the cumulative distribution function of the standard normal and ZK is de ned by the equation Se i. ZK = p 1 ln S e(r. q . Since the terminal price of the underlying asset satis es H (X ) = 8 1 if X > < > 0 if X < : 0 ST = S e Z pT 1 +(r. Here.q.3 where r is the interest rate (assumed constant) and the expectation is taken with respect to a risk-neutral probability. we have P fST Kg = + Z1 p ZK 1 e. 1 2 ) T 2 1 2 : d2 . 2 2 ) T where Z is normal with mean zero and variance 1.r T N (d2) : (4) This formula resembles strongly the expression derived previously for for the cash-amount to be held in the equivalent hedging portfolio for a vanilla call. z22 dz 2 = N (. 0 is the Heaviside step function. de ning p T ln K S .e. (r .

the risk is just one-sided because the position is short Gamma at all levels of spot. Mathematically.r T . from the point of view of risk-management. the standard call can be viewed as a \portfolio" of two digital options (one digital with payo consisting of one share and . if the hedger is short a standard option. On the other hand. from the Black-Scholes formula. In particular. to time-decay. On the other hand.4 . The holder of a European call is { by equivalence of the nal cash. There are two fundamental di erences: Digital options have mixed convexity Digitals have discontinuous payo s. when making a market in digital options. 22 = .K digitals with payo of $ 1).q T N (d1 ) where . K e. In other words. Di erentiation with respect to S in (5) gives digital e. the value of the call is S e. Recall that if the hedger is short Gamma he is vulnerable to large moves in the underlying asset whereas if he is long Gamma he is vulnerable to small moves. Let us examine these issues by looking at the Greeks of the European digital.r T N (d2 ) p (5) d1 = p 1 ln S e(r. European digital options are even simpler to price than standard options. agents may not gain a market advantage by quoting a price with an implied volatility that is higher than the one of standard options.q) T + 1 K 2 T T: We can interpret the two terms in this formula as the values of the two digital payo s that \make up" the standard option.ows { long a contingent claim that delivers one share if ST K or nothing if ST < K and short K European digital options. K e.r T N (d2) : The resemblance is not accidental. e2 2 e p d1 : S T 2 (7) . In contrast. The issue of mixed convexity is important for the hedger because it means that the risk-exposure is complex as volatility changes. i. the di erence between vanillas and digitals is substantial.e.digital .r T e. 22 = p 2 S 2 T d2 d2 (6) and .

At some point. If S > S (T ). For S < S (T ). particular.(r. This has two consequences: rst. The discrete nature of price movements can make continuous-time hedging techniques very risky when Delta changes rapidly as the spot moves. e. As T converges to zero.q) T K .(r. This observation applies also to highly leveraged option portfolios. When S = S (T ) the hedger is subject to risk from both large and small moves! The sensitivity of the price of the digital with respect to the volatility parameter is ( V ega digital = @V @S T ) = . the amount of shares bought or sold can be so large that the risk due to a small change of the stock price may exceed the maximum liability of the digital.r T e. The second consequence is pin risk: if the price of the underlying asset oscillates around the strike price near expiration. Gamma is large and negative. the hedger is subject to time-decay risk: he must rebalance his position frequently in order to o set time-decay. we should recall that prices do not change continuously | the lognormal approximation is just a convenient device for generating simple pricing formulas. the hedger will have to buy and sell large numbers of shares very quickly to replicate the option. far away from expiration the value of the digital option is small compared to $1 and the Deltas and Gammas are small. Let denote a small number. At this point. One way to understand the European digital option in terms of standard options is in term of call spreads. As the expiration date approaches.Then. the position To understand better pin risk.5 These sensitivities become large as T ! 0 for S e.q) T e 1 2 T : (8) (This is the value of spot for which the Delta of the standard call is exactly 1/2). Delta-hedging becomes extremely risky. Recall that a call spread is a position which consists of being long one call with a given strike and short another call with a di erent strike. This means that the hedger is exposed to signi cant risk near expiration if there is a big move in the spot price. In. 2 . 22 p 2 d2 d1 : (9) Vega also changes sign at S = S (T ). the Delta of the Digital option approaches the Dirac delta-function.2 The Gamma of the option vanishes for 2 S = S (T ) = K e. the seller of the option is vulnerable to an increase or a decrease in market volatility according to whether S is smaller or greater than S (T ). hedging the digital becomes much more complicated due to the unbounded Deltas and Gammas.

V should satisfy e. to be paid if and only if ST > K . the idea of using call spreads is a useful one. Example: An important example where digitals appear in nance is in the pricing of contingent premium options. and short 1= calls with strike K where the calls have the same expiration date as the digital. the di erence between the payo s of the digital and the option spread. can be hedged in the cash market at a lesser risk. If is large. then. this will require relatively few options. options with strikes very close to K may not exist in the market.e. in particular. i. This observation suggests that a good hedging strategy for digital options would be to use call-spreads instead of Delta-hedging.6 long 1= calls with strike K . A contingent premium option can be viewed as a portfolio consisting of Long one standard option with strike K and maturity T . K ) call-spreads. Nevertheless. according to equations (4) and (5). V for ST > K .K 0)g . that the investor makes money only if ST > K + V and will actually lose money if K < ST < K + V . hedging with an option spread which approximates the binary payo (but not necessarily dominates it or replicates it exactly) can help o set pin risk by diminishing the magnitude of the jump.q) T N (d1 ) 2 . i. Moreover. The intrinsic value of a contingent-premium call option is therefore zero for ST K and ST . e. the \fair" deferred premium should be N (d ) V = S e(r. Notice. These are derivative securities which are structured as standard European options. This implies that the value of the digital is less than that of 1= (K . Short V binary calls with strike K and maturity T .K ) for all ST .e. In other words. We say that the call spreads dominate the digital. a portion of the risk can be diversi ed by hedging with options and the residual.) If the option is structured so that no downpayment is required. diminishing will make the di erence in prices arbitrarily small but the hedge requires many options. has a nal payo Min f Max ST . except for the fact that the holder pays the premium at maturity and only if the option is in the money. K (12) . ) 0 ] 1 g : (10) This function is greater than H (ST . In fact.rT E f Max ( ST . (The situation is analogous to that of a person which has \free" medical insurance but with with a large \deductible". This may be di cult and costly to execute. but the fair value of the spread may be much greater than that of the digital. to adopt a static hedging strategy. On the other hand. where V represents the premium. (K .r T V P f ST > K g = 0: (11) Therefore. K . from (4) and (5).

The \fair value" of the American digital is.r Tg (13) where expectation is taken with respect to a risk-neutral probability measure and represents the rst hitting time of the strike price S = K .25 % if LIBOR < 5% on the coupon date. we have j Vnj = 1 if Sn K (14a) . aside from being useful to \benchmark" the binomial tree calculation. It is worthwhile to consider the two valuation methods that we are familar with: the binomial pricing model and the lognormal approximation. This introduces additional time-optionality to the problem. the value of the American binary is $1 if the option is in the money. the investor does not pay for the oor when he buys the structured note. To implement the binomial approach. the structure may be desirable to investors seeking protection if interest rates drop but who are unwilling to nance the interest-rate insurance upfront. with the usual notation. This derivative security could be desirable to investors that believe that if interest rates go below 5% then they will be signi cantly below 5% for some period of time. 2. ir requires introducing new tools from Probability Theory. according to the general principles. A simple example of a structured note would consist of a note with coupons indexed to LIBOR with the following characteristics: Coupon payment = Max (LIBOR 5%) Contingent premium = 0. he can (and also must) take advantage of the interest-rate oor by paying 0. All such options have \embedded" European binaries.1 American digitals The payo of the American digital option is similar. but now the holder receives $1 if and when the the underlying asset trades above $K for the rst time. V (S T ) = E f e. However. The lognormal approximation gives insight into the sensitivities of the price with respect to the model parameters. the former is relatively easy to program. The structure resembles that of a oating-rate note with an interest-rate oor (series of puts on interest rates. Instead. As we shall see.) However.7 More generally.25% ( 25 basis points) if LIBOR goes below 5% on any given coupon date. Therefore. we proceed as follows: by de nition. Also. The idea was implemented in recent years for designing debt securities known as structured notes. This note guarantees the holder a \ oor" of 5% on the interest rate income. such options can be structured so that a portion of the premium is paid upfront and another is contingent on the option being in-the-money at maturity .

using an \e ective" mean-square volatility T such that 2 = T . the main di erence between European and American binaries is due to the fact that the relevant volatility parameter for European binaries is the annualized standard deviation of the change of price between now and the maturity date. is known as a Dirichlet problem or boundary-value problem in the theory of Partial Di erential Equations. that involves a lateral boundary condition.rn dt n +1 PU Vnj+1 + (n) PD Vnj+1 (n) o : (16) The only di erence. This point is signi cant because the hitting time of the barrier is unknown. Closed-from expressions for American binary options can be obtained under the assumption that the volatility and interest rates remain constant. the e ect on pricing due to a time-varying volatility is nontrivial. setting Vnj+1 = 1 ).e.3 From the point of view of interest-rate and volatility term structures. In contrast.e. American binaries are sensitive to the entire volatility path. T 3 0 .1 t dt. the value of the American digital in equation (13) can be written as In contrast. European binary options can be priced in a time-dependent volatility environment T R 2 like standard options.8 and j j VN = 0 if SN < K : (14b) The portion of the binomial tree which needs to be determined by roll-back corresponds to the nodes (n j ) such that j Sn < K with n < N : (15) The value of the binary option at these nodes is calculated using the familiar recursive relation Vnj = e. Let f ( ) represent the probability density function of the random variable . Therefore. This type of problem. This requires introducing new mathematical tools. Pf < T g = ZT 0 f( ) d : (17) Then. i. Notice that the formulation (16) allows for term-structures of volatility and interest rates. from the numerical point of view. between American and European binaries resides in the fact that the value at the nodes which are one step away from the boundary fS = K g are computed using equation (14a) for the value of +1 +1 Vnj+1 (i.

where A > 0.q) K ( 2 . Zt = Zt + t where Zt is a Brownian motion and is a constant.r f ( ) d : (18) An explicit expression for the probability distribution of the rst-exit time can be derived from Lemma 1.9 V (S T ) = ZT 0 e. Then. 1 2 A 1 ln K S : The reader will verify easily that if represents the rst hitting time of S = K .q and . Pf A < Tg = N .A .q. Let A represent the rst time the path Z hits A. To apply this result to the case of a lognormal random walk of the form St = S e we set Zt + 1 (r. we conclude that 2(r. 2 2) t (20) r.A + p T T + e2 A N . then P f < Tg = P f A < Tg : Using equation (19). p T T : (19) We defer the proof of this Lemma fro the moment. 1) N (d ) 3 S (21) P f < T g = N (d2) + where (22) . Let Zt represent a Brownian motion with drift . i.e.

More precisely.q) K ( 2 . The same is true if the option is modi ed so that the holder collects $1 at time T if the price ever touches K (and not at time ).r P f < : (24) The nal expression for V (S T ) is.r dd P f < = ZT 0 gd = e.r Pf < g =T =0 +r ZT 0 e.r T 2(r. Notice that in the special case r = 0.10 d3 p 1 T S ln K . the formula simpli es further.r T N (d2 ) + e. 1 2 2 T : (23) Equation (22) gives a closed-from expression for the probability distribution of the rst-exit probability of the set fS < K g. the value of the American digital is V (S T ) = e. V (S T ) = e. The probability density of can then be computed by di erentiating (22) with respect to T . 1 ) N (d ) + r Z e.q) K ( 2 .r T N (d2 ) + e. r.r f ( ) d e.r T T 2(r. 1 ) N (d ) : 3 S (26) Next. from (22). we discuss the option's sensitivities to changes in spot price or market volatility.r T P f < Tg + r ZT 0 e.q.r P f < 3 S 0 gd (25) where the probability inside the integral is given by (22) (with T replaced by ). . In both cases.r P f < gd gd = e. This last integral can be computed numerically by quadrature. we have V (S T ) = ZT 0 e.

Near the \barrier" S = K . Therefore. say. or barrier. however. In this event. These are contingent claims that expire automatically Recall that the European digital has two-sided volatility risk. V (K . Delta-hedging builds up a large spot position in the rally (long the market). since increasing the volatility increases the premium. In fact.4 Just like with standard options. . T ) (K T ) decreases as increases.) Thus. since the option expires after K is hit for the rst time. He must therefore charge above the market volatility or else \set aside" some of his other funds to nance the strategy. The most common types or barrier options are when the spot price touches one or more predetermined barriers. Of course. over the lifetime of the option. increasing the volatility parameter (with respect to. the option has signi cant pin risk. The \worst-case scenario" for the hedger would be a market rallying slowly towards the strike level which collapses immediately before the option's maturity.11 Both the Delta and Gamma of the American digital are monotone increasing for 0 < S < K and become unbounded as T ! 0 in a neighborhood of S = K . we can make a straightforward analysis of the sensitivity of the Delta of the option to volatility (what some traders call colloquially \D-Delta-D-Vol"). the hedger may incur a signi cant loss. defeating the purpose of Delta-hedging. The Vega of the American digital option is positive at all values of spot.volatility strategy. There is. (This follows from the convexity with respect to the spot price. the implied volatility of vanilla options traded in the market) will provide protection against slippage when the spot is away from K improve the exposure to pin risk by decreasing the Delta at the barrier. we know that V (S T ) is non-decreasing as a function of the volatility and that V (K T ) = 1. The exposure to Gamma and the pin risk are simpler than for the European counterpart. 4 5 Knock-out options. Therefore. Barrier options Barrier options are a generic name given to derivative securities with payo s which are contingent on the spot price reaching a given level. the risk-exposure due to an incorrect estimate of the volatility is only one sided. the seller fears an increase in volatility and the buyer a decrease in volatility. Hence.5 3. If the market falls suddenly. This implies that the di erence quotient V (K T ) . the seller will have to charge more if he wishes to follow this augmented. varies inversely to in a neighborhood of K . an important di erence with respect to European binaries: the hedger does not have to worry about market \whipping" around the strike price.

provided that the spot price goes below $H between now and the maturity date. Its treasury department could have anticipated the problem by purchasing standard options but did not do this. the company fears a decrease in revenues in DEM terms. These contingent claims are activated when the spot price touches one or more predetermined barriers. G. 1995 reported that exotic options now constitute around 10% of the currency option business.27 DEM/USD.00% a U. the cost of this option is therefore approximately $ 3. The most common barrier options are structured as standard European puts and calls with one knock-in or knock-out barrier.000.80% and a German deposit rate of 4.S. provided that the spot price never goes below $H between now and the maturity date. The value of this option is instead $ 0.42 expiring in 180 days is $ 0. The London Financial Times of November 16. or $ 1. This is illustrated in the following example described to me by a trader. the value of a dollar put with strike 1. An up-and-in call with strike K . barrier H and maturity T is an option to buy the underlying asset for $K at time T . barrier H and maturity T is an option to buy the underlying asset for $K at time T . equivalently.) Thus.181. business revenue into DEM periodically. barrier H and maturity T is an option to buy the underlying asset for $K at time T . A rst observation regarding barrier options is that are much cheaper than standard options. For instance. Barrier options are especially used in foreignexchange derivatives markets. (We will derive below a pricing formula for knockout options. Nov. If the spot exchange rate is 1. The result was rounded to the nearest $ 10.000 to the nearest $ 1. the company would like to have an at-themoney DEM call/Dollar put with six months to expiration. 16. An up-and-out call with strike K . provided that the spot price never goes above $H between now and the maturity date.S. 6 7 Example: A large multinational corporation based in Europe must convert its . Given the weakness of the Dollar with respect to the Deutschemark in the past years and drop of the Dollar at the beginning of 1995.00%. Bowley:\ New Breed of exotics thrives". We used a volatility of 13. deposit rate of 5.12 Knock-in options. Supplement on derivatives. A down-and-in call with strike K .4225 DEM/USD. provided that the spot price goes above $H between now and the maturity date. suppose that the company purchases now a down-an-out dollar put (or. Ideally. a down-and-out call with strike K . On a $ 100 million notional.01181 per dollar notional. which are relatively simple variations on the European put and call enjoy a great popularity. LFT. Similar de nitions apply to puts. an up-and-out Mark call) with a knockout barrier at 1.000.000.910.7 On th other hand.0391 per dollar notional. The optionality feature can be targeted more precisely by introducing a barrier. U.6 Barrier options. barrier H and maturity T is an option to buy the underlying asset for $K at time T .

the value of this derivative security is determined recursively by solving the problem: h ( +1 i ( j Vnj = e.13 the knockout option with a 1. The option described in the above example. We will therefore discuss primarily barrier options which knock in or out when the option is in-the-money. if the treasurer believes that the dollar will not drop below $ 1. In the case of an up-and-out call. In contrast.27 barrier is nearly 4 times cheaper than the vanilla. We note two cases that were mentioned to us by professional traders: Double knock-in or double knock-out options. Thus. We can therefore reduce the question pricing barrier options to the pricing of knockouts. We note that in some cases.rn dt PUn) Vnj+1 + PDn) Vnj+1 if Sn < H j Vnj = 0 if Sn (28a) ( ( where PUn) and PDn) are risk-neutral probabilities. Knock-in and knock-out options are related by the simple formula KI + KO = Vanilla : (27) This formula is self-evident: the holder of a portfolio consisting of one knock-in call and one knock-out call with same strike. . barrier and maturity will e ectively hold a call at maturity regardless of whether the barrier was crossed or not. options with out-of-the-money barriers do not seem to be very interesting from a hedging perspective. which have two barriers knock-out options with rebate.27 over the next six months. similar to the one encountered in digitals.8 3. the structure of barrier options is more complicated. The holder receives a \consolation prize" in the form of a cash rebate on the premium paid if the option knocks out. Delta-hedging reverse knock-in and knock-out options may lead to signi cant pin risk.1 Pricing barrier options Barrier options are priced by solving a boundary-value problem similar to the one for American digitals. H (28b) and 8 This option consists of a regular knockout option with an attached American digital option. The di erence between in-themoney and out-of-the-money barriers is signi cant because the former have discontinuous payo s at expiration. which knocks out when the option is in-the-money is often called a reverse knock-out. Therefore. the knockout option provides a cheaper alternative with the \same" terminal payo .

Then. To apply this result.B)2 2T 2 eB . we consider the pricing assuming that St is a lognormal random walk with constant q and r constant (geometric Brownian Motion). As in the case of the American binary option. and (ii) the absence of arbitrage. 2 T T dB dB 1: (30) We shall prove this Lemma later. we will need some auxiliary results on the properties of Brownian motion with drift.14 j j VN = Max SN h . 1 2 2 (B B + dB) 1 e. the value of a down-and-out put with strike price K . Let Zt t P = 0 0 represent a Brownian motion with drift .r T E Max K . Lemma 2. . let denote the rst time that the lognormal walk (20) hit the level S = H . knockout at H (H < K ) and maturity T satis es PKO (S T K H ) = e.r T E f Max K = e. Then. which determine their value at the barrier and at maturity. j K 0 if SN i K: (28c) In the case of an up-and-in call. The values of barrier-puts are determined by making obvious modi cations. if A and B are positive numbers with B A. ST 0 ] ST 0 ] 0 > T g MinT St > H t (31) . Next. which implies (28a). Max T Zt t p A and ZT (2A. the boundary conditions (28b) and (28c) are replaced by j ~ Vnj = Vnj if Sn H K: (29a) ~ where Vnj represents the value of a vanilla call at the node (n j ). and j j VN = 0 if SN (29b) The validity of these equations follows from (i) the terms of the barrier options.

r T E ST Notice that the rst term corresponds to the value of a standard European put with strike K . using Lemma 2.r T E f K . ST ST < H g P H < ST < K 0 MinT St t H < ST < K 0 MinT St t H H : (32) . K e. ST H < ST < K g .r T + e. To calculate the two remaining terms.r T E f K .q . we will use the result of Lemma 2. . ST e. The second term can be calculated easily using the same reasoning as in the derivation of the Black-Scholes formula. 1 2 we have. e. ST ST < K g E f K . Introducing the parameters AH AK and 1 ln H S 1 ln K S r.r T E K H < ST < K 0 MinT St t H = e. .15 This last expression can be rewritten as e.r T . ST H < ST < K 0 MinT St > H t .r T E K = e.

1) S .dH ) + S e. t 2 T . < ZT < .B .AK 0 MinT St t . .AK 0 MinT St t H = P AH < ZT < AK 0 MinT Zt t AH .r T N (. 2 e. 1 dB T p 2 T : (34) Calculating explicitly the two integrals in (33) and (34) and using the Black-Scholes formula to calculate the rst two terms in (32).dK ) 1 K e.dH ) 2 1 2(r.r T f N (d4 ) .Zt distribution.AH = S .B)2 2T dB T p .16 P H < ST < K = P = . : (33) Here.B .q T N (. ZT . . S e. 2 2 0 MaxT Zt.r T N (. (.AH 1 e .q T N (.AH Z .2AH .q) H ( 2 . Similarly. N (d5 ) g . B e.AH e. we use the fact that . t and Zt t have the same probability H = E ST H < ST < K S E e. K e.AH Z . < ZT < .AK .AK e.AH (2AH +B)2 2T Max T Zt. . 0 t 2 .dK ) 2 . we arrive at the nal result PKO (S T K H ) = K e.

q) H ( 2 ) + 1 fN (d ) 6 S . 2 T T T T T p . 1 q+2 d7 = p 1 H2 ln S K T + r . 2 p 1 .2 2 p 1 H2 ln S K T ln H S T . q+1 2 q.q T where 2(r. with the underlying asset viewed as the unit of account.1 2 q+1 2 q . 1 q. q. N (d7 ) g (35) dK = 1 dK = 2 dH = 1 dH = 2 d4 = d5 = d6 = and p 1 1 1 S ln K T S ln K T S ln H T S ln H T + r + r + r + r + r + r + r .1 2 2 2 T T p 1 . q+1 2 2 T : The formula for an up-and-out call is obtained immediately by a change of numeraire: an up-and-out call on the risky asset with strike K is nothing but a down-and-out put on cash. The pricing formulas for up-and-out puts/ down-and-out calls are obtained using very similar techniques. For instance. 2 p . Finally. We leave them as an exercise for the interested reader. the fair values of knock-in options can be obtained using the parity relation (27). 1 2 2 p 1 ln H S T .17 + S e. using (27) and (35) we nd that the value of a downand-in put is .

0193 .0133 . 1.30 Val.q) H ( 2 + 1) fN (d ) 6 S N (d7 ) g To end this section we present some numerical values for a particular barrier option (this example was mentioned earlier).28 1.38 1.dH ) 1 .0315 -.0116 .28 1.32 1.0105 .1623 . 1.27 DEM/USD 180 days to maturity Spot 1.r T 2(r.r T N (.4225 1.30 Val. USD interest rate: 5.0015 . 2(r.2012 .r T .40 1.44 .34 1.00 % volatility: 13.0194 .0035 .1823 .0148 -. 1) S .42 DEM/USD Knockout at: 1.0068 . f N (d4 ) .0090 .1326 .4084 . S e.00 % Strike: 1.0210 .0043 .3185 .85 % DEM interest rate: 4.0118 .4511 .0117 . Example: A reverse-knockout dollar put / DEM call.0111 90 days to maturity Spot 1.36 1.0169 .40 1.0196 .18 PKI (S T K H ) = + K e.0968 .32 1.0782 -.q) H ( 2 .q T N (.0174 -.36 1.38 1.dH ) 2 + K e. N (d5 ) g S e.0102 .1160 -.1893 .0588 .0219 -.0211 .0146 .44 .0115 .34 1.4225 1.

at the barrier the Gamma risk is complex: if the spot price is just below the barrier.2.0411 .30 1.4676 -.34 1.4863 -.4225 1.36 1. whereas near S = H the \digital" dominates and the holder is short Gamma/Vega.0135 1. The Delta increases without bounds near the barrier. the hedger must adjust his Delta in order to \earn time decay" (his liability is that of a standard put if the options fails to knock-out).28 1.0445 .0737 . Val.19 30 days to maturity Spot Val.0298 .0107 .4645 -.0411 .3226 -.0169 .44 . The risk-exposure of a reverse knockout put option can be understood intuitively as follows: from equation (27) the holder of this option is Long a standard put with strike K short an American digital option with barrier at H which pays one put with strike K upon hitting the barrier (in other words.3610 1.6459 1.4225 1. We then see immediately that the option has mixed Gamma exposure: far away from the barrier. 1.0250 -. a knock-in put) Ignoring the di erence between a knock-in put and an American digital option with payo H .44 .0145 .0232 .0320 .0398 .9666 -.40 1.5579 -. The seller that wishes to hedge faces the mirror-image position: short Vega and Gamma near the strike or in-the-money and long Gamma/Vega closer to the barrier. .5668 -.0327 . if the option does not knock-out. 1.38 1.0390 -.4989 -.30 1. The risk-management of barrier options should take into consideration the mixedGamma exposure of these instruments (for reverse-knockouts and knock-ins) as well as the pin risk at the barrier.32 1.0067 . the large Delta position may be detrimental in case of a large market because this would lead to a loss in th spot market.0438 .36 1.2282 3.3544 7 days to maturity Spot.28 1.9900 -.9417 -.7798 -.0600 .32 1.0025 3.34 1. Hedging barrier options.0713 . the standard put dominates and the holder of the knockout is long Gamma/Vega.38 1.40 1. K at the barrier is not a bad approximation near the expiration date. On the other hand. However.0308 .

the liquidity issue in the trading of barrier options is discussed in great depth. his spot position in USD would be long $ 3. Lemma 3.000. Consider the option described in the previous section.000 dollars. where.000 ! To have a better idea of the likelihood that this happens. on the other hand is (a whopping) $ 364. this would be a three-standard deviation move in one day. 4. A selling order of nearly 400 million dollars as the exchange-rate goes through the barrier may cause a further drop in the dollar as there will be few buyers and many sellers.1 A consequence of the invariance of Brownian Motion under re ections. when the spot price was 1:4225. see N. To make this more concrete. Let Zt denote standard Brownian motion on the interval 0 T ]. Proofs of Lemmas 1 and 2 This section contains sketches of the proofs of the two Lemmas used to derive closed-form solutions for barrier options and American digitals. Taleb: Dynamic Hedging (1995. The event has low probability but is not impossible. for all A > 0 and B < A.20 an agent sold the option with 180 days to expiration.6459 per dollar notional. let us mention the important point of liquidity. we have P 0 maxT ZT > A ZT t 9 2 (B B + dB ) = p 1 e.TB)2 dB : (36) 2 T One should also take into account that the annualized volatility may very well underestimate the daily move of the exchange rate. The spot position. This will have dire consequences for the \hedger". 10 To nd out more about the risk-management of exotic options. note that this represents a 2.000. assume that the notional amount is $ 100. Then.1094 per dollar notional.10 Example. The loss in the dollar position if the market moves suddenly down by 0. at 0:0118 per dollar notional. .03 in the exchange rate in one day will result in a loss of $ 0.28 DEM/USD and the agent Delta-hedges according to the above tables.940. At an annual volatility of 13%. (2A2. If seven days before expiration the spot trades at 1.4% move of spot in one day.180. Would you risk a loss of $ 9 million given the odds?9 Moreover. A drop of 0. assuming that 4. in particular.000.559. manuscript in preparation). The premium collected for the option was $ 1.03 through the barrier would be $ 10.

we conclude from (37) that equation (36) holds. B g : (This last equality holds because 2A0 . B > A0 . XN by ZT and j t A0 by A. The probabilities for + dt and . Therefore. we conclude that for B < A0 P max Xj 1 j N A0 XN = B = P max Xj 1 j N A0 XN = 2A0 .2 The case 6= 0: To prove Lemma 2.) Let T = N dt. = A0 for m > n occurs with the same probability as 1 the original one (namely 2 N ). see for instance Billingsley: \Convergence of Probability Measures" .1 p dt n = 1 2 ::: N p p where dt represents a small positive number. By the Central Limit Theorem. the joint distribution of the random variables X nt] approaches that or a Brownian Motion as dt ! 0 N ! n +1. A0 represents the largest p integer multiple of dt which is A. Set A0 p A dt p dt where X ] represents the integer part of X .21 Proof: Consider a simple random walk de ned by Xn = Xn. 1968. of the random walk is such that Xn = 0 for some n A N and that Xm < A0 for m < n. or realization. Therefore. We observe that the path which coincides with this realization of the random walk for m n and which is re ected about the line X. Wiley. we will need the following result about Brownian Motion with drift: For a rigorous proof of the formal passage to the limit (37) =) (36). 4. if we replace formally 1 maxN Xj by 0 maxT Zt . 11 . Therefore. dt are assumed to be 1=2. The Lemma is proved. B (37) = P f XN = 2A0 .11 Notice that we have established the analogue of Lemma 2 in the case = 0. Assume that a given path.

1 ) (t1 .1 Ztj. namely e ZT . tn o : (38) This result states that the expectation of a function of Brownian motion with drift is equal to the expectation of the same function of regular Brownian multiplied by an exponential factor.1 ))2 = dy1 dy2 :::dyn G (y1 y2 ::: yn ) exp : . (tj .1 + (tj . . set . tn. Ztj.1 ) : G (y1 y2 ::: yn ) F (y1 y1 + y2 ::: y1 + y2 + ::: yn) Using the explicit form of the Gaussian distribution and the fact that the increments Yj are independent random variables with mean (tj . tj. Theorem (Cameron . . tj.1 .Martin). tj. tj . Zt1 Zt2 ::: Ztn E f G ( Y1 Y2 ::: Yn ) g = = Z Rn 9 8 n < X (yj . tj. Then.22 tion. t0 ) ::: (tn . 1 2 2 T : Proof of the Theorem: De ne the increments of the Brownian path Yj = Ztj = Ztj Also.1 ) and variance tj . Zt1 Zt2 ::: Ztn E F ( Zt1 Zt2 ::: Ztn ) e = 1 2 2 n Ztn . Let (F (z1 z2 ::: zn ) be a continuous funcE F .1) j =1 . 2 (tj . we obtain E F .

then P E = = 12 0 Max T Zt t A ZT < C ZT . 1 2 2 tn o which is what we wanted to show. Proof of Lemma 2: Using an approximation argument which we omit. ibid. = 1 2 2 0 Max T Zt t Max T Zt t e A ZT < C e A ZT = B 1 2 2 T ZC 0 E e. 0 e B. This concludes the proof of the Theorem. B)2 2T B. we nd that the last integral in (39) is equal to E F ( Zt1 Zt2 ::: Ztn ) e We are now ready for the n Ztn . it can be shown that the above theorem can be also applied to the functional of the path F (Z ) 0 Max T Zt : t (This is a continuous function of the path that can be approximated in a suitable sense by continuous functions of n variables. G : j=1 2 (tj .23 = Z Rn 8 n 9 < X 2 = yj dy1 dy2 :::dyn ~ (y1 y2 ::: yn ) exp . T pdB 2 T See Billingsley. . tj. 2 1 where ~ G (y1 y2 ::: yn) = G (y1 y2 ::: yn) e j=1 2 (39) tn : Making a change of variables.1 ) (t1 . tn. we conclude that if C < A. t0 ) ::: (tn . as in the previous theorem12) Applying the Theorem to this functional. 1 2 2 T dB (40) ZC 0 (2A .1) n P yj .

ZA 0 e. we prove Lemma 1 on the distribution of the rst-passage time for Brownian motion with drift. ZA 0 P Max T Zt t ZA 0 P fZT = B g dB 2 2 e. BT pdB ZA 0 P 0 Max T Zt < A ZT = B dB t dB : 2 T (41) ZA 0 2 T . Using the notation of Lemmas 1 and 2.24 where we used Lemma 3 to derive the last equality. Equation (30) follows immediately by di erentiating both sides of (40).B)2 2T p The conclusion of Lemma 1 follows by evaluating this last expression in terms of the cumulative normal distribution. (2A. Finally. . we nd that Pf A < Tg = P = = = 0 Max T Zt t 0 A A ZT = B dB . Proof of Lemma 1.