Description

A SIMPLE OPTION FORMULA FOR GENERAL JUMP-DIFFUSION AND OTHER EXPONENTIAL LÉVY PROCESSES ALAN L. LEWIS Envision Financial Systems and OptionCity.net August, 00 Revised: September, 00 Abstract Option values

Information

Category:
## Math & Engineering

Publish on:

Views: 30 | Pages: 20

Extension: PDF | Download: 0

Share

Transcript

A SIMPLE OPTION FORMULA FOR GENERAL JUMP-DIFFUSION AND OTHER EXPONENTIAL LÉVY PROCESSES ALAN L. LEWIS Envision Financial Systems and OptionCity.net August, 00 Revised: September, 00 Abstract Option values are well-known to be the integral of a discounted transition density times a payoff function; this is just martingale pricing. It s usually done in S-space, where S is the terminal security price. But, for Lévy processes the S-space transition densities are often very complicated, involving many special functions and infinite summations. Instead, we show that it s much easier to compute the option value as an integral in Fourier space and interpret this as a Parseval identity. The formula is especially simple because (i) it s a single integration for any payoff and (ii) the integrand is typically a compact expressions with just elementary functions. Our approach clarifies and generalizes previous work using characteristic functions and Fourier inversions. For example, we show how the residue calculus leads to several variation formulas, such as a well-known, but less numerically efficient, Black-Scholes style formula for call options. The result applies to any European-style, simple or exotic option (without pathdependence) under any Lévy process with a known characteristic function. All comments welcome and encouraged. Contact info. : Tel: (949) , Bayside Cove West, Newport Beach, Ca 9660 USA . INTRODUCTION AND SUMMARY The benchmark model for security prices is geometric Brownian motion. A relatively simple yet powerful generalization is the class of all continuous-time process where all non-overlapping increments Xt Xt = log( St / St ) are independent random variables with stationary distributions. This set of processes X t are the Lévy processes or processes with stationary independent increments. [see the monographs by Sato (996) and Bertoin (999)]. They consist of a combination of a linear drift, a Brownian motion, and an independent jump process Here we provide a solution to the associated European-style option valuation problem. These models help explain some, but not all, of the well-documented deviations from the benchmark model. Lévy process models can be good fits to daily stock return distributions which are characterized by wide tails and excess kurtosis. [For example, see Eberlein, Keller, and Prause (998)]. The so-called jump-diffusions (which are members of the class where the jumps are compound Poisson processes) offer a compelling explanation for the relatively steep smiles observed in expiring index options like the S&P500. It should be noted that the independent increment assumption is counter-factual in some respects. Nevertheless, because of their successes, flexibility, and analytic tractability, continued financial applications are likely. For a recent survey of applications, in finance and elsewhere, see Barndorff-Nielsen, Mikosch, and Resnick (00). Jump-diffusion models are distinguished by their jump amplitude distributions. Two examples are Merton (976) who solved for option prices with log-normally distributed jumps, and Kou (000), who did the same for a double exponential distribution. To obtain an option formula, the authors relied upon particular properties of those distributions. Merton s solution relies upon a product of lognormal variates being lognormally distributed. Kou s derivation stresses the importance of the memoryless property of the exponential distribution. Our results make clear that, in fact, no special properties are needed: we obtain an option formula for any jump distribution. In addition, many of the original results for these models are very complicated. Special functions and complicated expressions are required when option formulas are given in S-space or stock price space. 3 However, more recently, it has been recognized that option values for both the Lévy process problem and related proportional returns problems are much simpler in Fourier space 4. For example, Carr and Madan (999) have derived relatively simple formulas for call options on Lévy processes, working in Fourier space. Bakshi and Madan (000), although not working directly on Lévy processes, derive an applicable Black-Scholes-style formula for call options using characteristic functions and some more complicated formulas for general claims. Raible (000) obtained an option formula which is very similar to our general result (see below). However, he presents it as a mixture of Fourier and two-sided Laplace transforms. Because we use the generalized Fourier transform consistently, our strip condition is more transparent. In Lewis (000), we obtained related inversion formulas for options under stochastic volatility, a proportional returns problem. Here we generate the value of the general claim under a Lévy processes as an integral of Fourier transforms. Once you have our main result, the residue calculus provides a standard approach to variations. For example, we show that the Black- Scholes style formula is simply obtained by moving integration contours. The formula can be easily explained with a little background. Assume that under a pricing measure a stock price evolves as ST = S0 exp [ ( r q) T + XT ], where r q is the cost of carry, T is the expiration time for some option, and X T is some Lévy process satisfying E [ exp( X T ) ] =. For Lévy processes, the important role of the characteristic functions φ t( u) = E [ exp( iuxt) ], u R is well-known. Like all characteristic functions, they are Fourier transforms of a density and typically have an analytic extension (a generalized Fourier transform) u z C, regular in some strip S X parallel to the real z-axis. Somewhat less appreciated, yet key to our approach, is the recognition that option payoff functions also have simple generalized Fourier transforms. Using the variable x = log S T, these transforms are ŵz ( ) = exp( izxwxdx ) ( ), where wx ( ) is the payoff function 5. For example, if x K is a strike price, the call option payoff is wx ( ) = ( e K ) + and so, by a simple integration, iz+ wz ˆ ( ) = K /( z iz), Im z . Note that if z were real, this regular Fourier transforms would not exist. As shown in Lewis (000), payoff transforms ŵz ( ) for typical claims exist and are regular in their own strips S w in the complex z-plane, just like characteristic functions. Then, the initial option value V( S 0 ) is given by simply integrating the (conjugate) product of these two transforms times a phase factor. To do the integration legally, one has to keep z within the intersection of the two strips of regularity S w and S X ( S X is the reflection of S X across the real z-axis). With that synopsis, the formula is rt iν+ izy (.) Option values: V( S ) e 0 = e φ T ( z) wˆ ( z) dz π, iν where Y = ln S0 + ( r q) T, z = u+ iν, z S = S S. V w X Although very compact, (.) contains (as special cases), the Black-Scholes model, Merton s jump-diffusion model, Kou s jump-diffusion model, and all of the pure jump models that have been introduced. Indeed, it applies to the entire class of exponential Lévy processes which have E [ exp( X T ) ] . The proof of it and our use of the residue calculus to obtain variations is our primary contribution in this paper. Any path-independent European-style payoff, plain vanilla or exotic, may be valued. The expression (.) is obviously just a single integral. It s real-valued and readily evaluated 6. Typically the integrand is a short expression, often with only elementary functions. See Tables. and 3. for examples of φ ( ) and ŵz ( ) respectively. T z iz+ Consider the call option again. Since wz ˆ ( ) = K /( z iz), the integrand in (.) is a regular function of z in the larger strip S X, except for simple poles at z = 0 and z = i. Using the residue calculus, you can move the integration contour around in S X. Of course, you pick up residue contributions if you move contours across (or along!) Im z = 0., We do this and obtain a number of variations on (.) including the Black-Scholes style formula that we mentioned earlier. Finally, it turns out that exp( izy ) φt ( z) is the (conjugate of the) Fourier transform of the transition density for log S 0 to reach log S T after the elapse of T. This allows us to interpret (.) simply as a Parseval identity. In fact, the proof of that clarifies what space of payoff functions are handled by our theory and which types are excluded. In the next section, we review some fundamental aspects of Lévy processes, their applications to finance, and their analytic characteristic functions. This material is almost entirely standard and experts in those topics could well skim for our notation and then jump to the proof of (.) in Section 3.. BACKGROUND. The Framework We consider a marketplace in which a stock price or security price S t 0 follows an exponential Lévy process X t (defined below) on a continuous-time probability space ( Ω, F, Q ). We stress that Q is a fixed martingale pricing measure. The pricing measure has the same null sets as an objective or statistical measure P, and the two measures are related by an unspecified Girsanov change-of-measure transformation 7. However, P plays no direct role in our discussion all expectations and stochastic processes are defined relative to Q. A stock buyer receives a continuous dividend yield q; she could finance her purchase at the riskless rate of interest r. The net financing cost (the cost of carry) is r q. It is convenient to explicitly remove this constant we then investigate option valuation where St = S0 exp [ ( r q) t+ Xt ] under Q, where X t is a Lévy process. Following Sato (999), we adopt the following definition: 3 Definition (Lévy Process) An adapted real-valued process X t, with X 0 = 0, is called a Lévy process if: (i) it has independent increments; that is, for any choice of n and 0 t0 t tn, the random variables X t 0, Xt X t,, X 0 t X n tn are independent; (ii) it is time-homogeneous; that is, the distribution of { Xt+ s Xs; t 0 } does not depend upon s. (iii) it is stochastically continuous; that is, for any ε 0, Pr { Xs+ t Xs ε} 0 as t 0. (iv) as a function of t, it is right-continuous with left limits. Processes that satisfy (i) and (ii) are called processes with stationary (or time-homogeneous) independent increments (PIIS). Some authors (e.g. Bertoin) simply define a Lévy process to be a PIIS process with X 0 = 0. Such processes can be thought of as analogs of random walks in continuous time. To prevent an arbitrage opportunity, the stock price (net of the cost of carry) must be a local martingale under Q. In fact, we maintain throughout the stronger assumption: S t, net of the carry, is a Q-martingale. That is, E [ St ] = S0 exp[( r q) t] or E [exp X t ] =. For those Lévy processes with E [exp X ] , this normalization can be achieved by a drift adjustment. t Types of Lévy processes. In general, Lévy processes are a combination of a linear drift, a Brownian motion, and a jump process. When jumps occur, X t jumps by Xt = x R \ { 0}, (the notation means that we exclude zero as a possible jump amplitude x). Now consider any closed interval A R that does not contain the origin. Then, the cumulative number of jumps in A the time interval [0,t] with a size that belongs to A, call it N t, is a random variable which is also a measure. This integer-valued Q-random measure is usually written Nt A = ν([ 0, t], A). A With A fixed, then N t has a Poisson distribution with a mean value t Aµ ( x) dx. Here we have introduced the Lévy measure µ ( xdx ), which measures the relative occurrence of different jump amplitudes [see Sato, 00, Theorem.4]. Two types of Lévy processes with a jump component can be distinguished. In type I (the Poisson case), we have R µ ( xdx ) . Then, we can write µ ( x) = λ f( x), where λ = R µ ( xdx ) is the Poisson intensity (the mean jump arrival rate), and f ( x) dx = df( x), where F( x ) is a cumulative probability distribution 8. We will also call this case the jump-diffusion case. An example of this type is Merton s (976) jump-diffusion model, where x is normally distributed: µ ( x) = λf( x) = λexp[ ( x α) / δ]/( πδ) /. 4 In the alternative type II case, R µ ( xdx ) = and no overall Poisson intensity can be defined. A + α simple example is the Lévy α stable process: µ ( x) = c± / x, 0 α , where c ± are two constants for x 0 and x 0. Carr and Wu (000) have proposed a special case of this model for stocks. Notice that in the type II example, the source of the divergence is the failure of µ ( x) to be integrable at the origin: there are too many small jumps. The divergence at the origin is always the source of the integrability failure of µ ( x) in type II models the Lévy-Khintchine representation (see below) guarantees that µ ( x) is always integrable at large x. A general integral representation. One can take a differential of Nt = ν([ 0, t], A), writing dnt = ν( dt, dx) and use these differential random measures in an integration theory [see Jacod and Shiryaev (987)]. With that theory, one can decompose any Lévy process X t into the form 9 : (.) X = ω t + σb + ( xν ( ds, dx) h( x) µ ( x) ds dx ) t h t t 0 R\{ 0} where B t is a Q-Brownian motion, ωh and σ are constants, and hxis ( ) a truncation function, to be explained. The Brownian motion and the jump process are independent. This representation is unique in the sense that, once the truncation function is fixed, then there is only one set of characteristics { ωh, σ, µ ( x)} for a given { Xt : t 0 }. If the truncation function is changed, the drift ω h changes but the pair { σ, µ ( x)} is invariant. The purpose of the truncation function is to make the integral in (.) exist near the origin, where the integrand must be taken as a whole. +ε Such a function is only necessary in some Type II models where µ ( x) diverges as O( / x ), 0 ε ; otherwise it may be set to 0. When a truncation function is needed, hxis ( ) required to behave like x near the origin and it is frequently taken to be a bounded function away from the origin. Some popular choices are: (i) h( x) = x { x } (Sato) ; (ii) hx ( ) = x/( + x ) (Lukacs, Breiman). If R x µ ( x) dx , then the truncation function need not be bounded and the choice hx ( ) = xmay be convenient. In that special case, we see that (.) is the sum of a linear drift, a Brownian motion, and an independent compensated jump-martingale.. Analytic Characteristic Functions. As an application of (.), we show that it leads immediately to the celebrated Lévy-Khintchine representation for the characteristic function φ t ( z). This representation is important to our development because it provides an explicit and simple formula for φ t ( z) for all the Poissontype models and some type II models. First, a definition of φ t ( z) and infinite divisibility, then a remark that we will use later, and then the statement of the theorem. A, 5 Definition (Characteristic Function). For z C (z a complex number) and call φ ( z) = E [ exp( izx ) ] the characteristic function of the process X. t t t a Im z b, we Remark. Let pt ( x ) be the transition probability density for a Lévy process to reach Xt = x after the elapse of time t. For a Im z b, the characteristic function of the process is identical to the characteristic function of this transition density, which is also the generalized Fourier transform of the transition density 0. φ t( z) = F [ pt( x) ] exp( izx) pt( x) dx, a Im z b. R Definition (Infinitely Divisible Characteristic Functions). A characteristic function φ t ( z) is said to be infinitely divisible, if for every positive integer n, it is the nth power of some characteristic function. The characteristic function of Lévy processes are infinitely divisible; this is a simple consequence of the PIIS properties. THEOREM (Lévy-Khintchine Representation). If φ T ( z), a Im z b, is an infinitely divisible characteristic function, then it has the representation (.) φt( z) = exp { izωht z σ T + T e ( ) ( ) } izx izh x µ x dx. R where min(, x ) µ ( x) dx R\{ 0} PROOF: For a proof (when z is real), we refer to Sato (999, Theorem 8.). For an extension to complex z, we refer to Lukacs Theorem below. However, just proceeding formally, it s easy to see how the representation (.) follows from the representation (.). We will only take the simplest case where the deterministic part of the integral in (.) exists on its own. Using that assumption, and the independence of the Brownian motion and the jump process, we have immediately from (.): { ω h µ } E[exp( iz XT)] = exp iz T izt h( x) ( x) ds 0 R \{ } E 0 R\{ 0} T [ exp( izσbt ) ] E exp ( iz xν( dt, dx) ) Now it is well-known that E [ exp( izσbt ) ] = exp( z σ T ). It is also well-known that, 6 T izxt izx ( iz x ν dt dx ) e { Nt Nt } ( T e µ x dx ) E exp (, ) = E = exp ( ) ( ). 0 R\{ 0} \{ 0} 0 t R T Combining these results yields (.). The Lévy-Khintchine representation has the form φ T ( z) = exp [ T Ψ ( z) ], where Ψ ( z) is called the characteristic exponent 3. The normalization φ T ( z = 0) = and the martingale identity φ T ( z = i) = imply that Ψ ( 0) = Ψ ( i) = 0 4. Since, for a sensible stock market model, φ t ( z) must exist at both z = 0 and z = i, it would be helpful if it existed for all z in between. Indeed, all the examples in Table. exist for z within a horizontal strip S X = { z : a Im z b}. Here a and b are real numbers, such that b a. Analyticity in strips is typical, based on this theorem: THEOREM (Lukacs, 970, Theorem 7..): If a characteristic function φ( z) is regular 5 in the neighborhood of z = 0, then it is also regular in a horizontal strip and can be represented in this strip by a Fourier integral. This strip is either the whole z-plane, or it has one or two horizontal boundary lines. The purely imaginary points on the boundary of the strip of regularity (if this strip is not the whole plane) are singular points of φ ( z). Remarks. Because of the representation φ t ( z) = exp [ Ψ t ( z) ], singularities of φ t ( z) are singularities of Ψ ( z). Hence, an immediate corollary of Lukacs theorem is that the strips of regularity for the analytic characteristic functions of Lévy processes are time-independent 6. Clearly, a good Lévy process (good for the purpose of building a stock price model), has an analytic characteristic function regular within a strip: a Im z b, where a, b 0. If a particular Lévy process is a good one, say at t =, then it is a good one for all t. Our option valuation formula only applies to good Lévy processes. Examples. Example of characteristic functions for Lévy processes that have been proposed for the stock market are shown in Table.. [Three entries are adapted from Raible (000)]. For each process, there is a constant drift parameter ω, determined by solving φt ( i) =. We have already mentioned the first two models in the table, both of which have Brownian motion components and are type I or Poisson types. The remaining (pure jump) models are all type II. The third table entry is the Variance Gamma process. The option value was obtained by Madan, Carr, and Chang (998). The process is built up by sampling Brownian motion with drift at random times, time increments which themselves are described by another Lévy process. Clearly, the sampled process is a pure jump process. 7 The next table entry is the Normal Inverse Gaussian (NIG) process, another pure jump process applied to stock returns by Barndorff-Nielsen (998). There are 4 real parameters: ( αβωδ,,, ) where, roughly speaking, α and β are shape parameters (steepness, tail decay), and ω and δ are drift and scale parameters, respectively. [Also see Lillestøl, (998)]. Following is the Generalized Hyperbolic process. This pure jump Lévy process incorporates the VG and NIG process as special cases, as well as another special cas

Related Search

Reading and Writing for General TrainingOntology for General Medical ScienceOrganisation For Economic Co Operation And DeTechnology Diffusion and Adaptation, TechnoloE-learning for Hearing Impairments (Deaf and Medicinal plants for cancer diabetes hypertenDeveloping novel approaches for therapeutic cAMA CATEGORY 1 PRA CME CREDIT FOR USA PODIATRPhotographic records as a tool for research iComputer Aided Design for Helical Compression

Similar documents

We Need Your Support

Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks