Abstract
This paper studies the valuation and hedging problems of forward-start rainbow options (FSROs). By combining the characteristics of both multiple assets and forward-start feature, this new type of derivative has many potential applications, for instance, to incorporate the reset provision in rainbow options for investors or hedgers or design more effective executive compensation plans. The main contribution of this paper is a novel martingale pricing technique for options whose payoffs are associated with multiple assets and time points. Equipped with this technique, the analytic pricing formula and the formulae of the delta and gamma of the FSRO are first derived. We conduct numerical experiments to verify these formulae and examine the characteristics of the FSRO’s price and Greek letters. To demonstrate the importance and general applicability of the proposed technique, we also apply it to deriving the pricing formula for the discrete-sampling lookback rainbow options.
Similar content being viewed by others
Notes
Theoretically speaking, both of these two techniques can be classified as martingale pricing methods.
More concretely, to evaluate \(E^Q[S_a(s)\cdot I(S_b(u)\le S_c(v))]\), where \(Q\) denotes the risk-neutral probability measure, \(S_i(z)\) denotes the price of the asset \(i\) at the time point \(z\), and \(I(\cdot )\) is defined as an indicator function.
Although this paper considers only the non-dividend-paying case for simplicity, all results are straightfoward to be extended to underlying assets with constant dividend yields.
Although it is widely accepted that jumps are able to explain the empirical regularities of derivative pricing, this paper focuses on pure diffusion processes. This is because incorporating additional jumps not only complicates the problem substantially but also obscures the contribution of this paper, which proposes a technique to tackle the evaluation of the expected value of an asset price at a specified time point conditional on the comparison results between individual asset prices (or the corresponding Brownian motions) evolving up to different time points. Even so, we highly appreciate the anonymous referee to mention this point.
Nevertheless, we can analyze qualitatively the possible impacts of adding jumps on the values of FSRPOs. Since the reset provision can turn out-of-the-money options to become at-the-money options on reset dates, the reset provision can partially eliminate the effect of unfavorable jump movements. As for put options on the minimum of multiple assets, through the diversification effect, additional jumps could lower the expected value of \(\min [S_1(T),\ldots ,S_n(T)]\). This is because if there is any asset with a net negative jump movement, this harmful effect will be retained by the minimum function. On the other hand, the realized value of \(\min [S_1(T),\ldots ,S_n(T)]\) rises only when all assets are with net positive jump movements. Since the FSRPO considered in this paper is essentially based on a minimum put options plus the reset option for the strike price, due to the above analysis, it can be expected that the additional jump processes will exhibit a generally positive impact on the value of the FSRPO.
For example, Eqs. (A.5) and (A.6) in Liao and Wang (2003) can be rewritten with our notation system as follows.
$$\begin{aligned} \frac{dR_1}{dQ}=\exp \left( \sigma _1 W_1^Q (T)-\sigma _1^2 T/2\right) , \text{ and } dW_1^{R_1}(z)=dW_1^Q (z)-\sigma _1 dt. \end{aligned}$$Since they price a single-asset reset option, the volatility of the only asset is denoted as \(\sigma _1\). With \(\sigma _1\) as the kernel in the Girsanov theorem, an equivalent probability measure \(R_1\) and the Brownian motion under it, \(dW_1^{R_1}(z)\), can be defined. Note that the probability measure \(R_1\) is defined with respect to the single asset but independent of any specified time point.
References
Baxter, M., & Rennie, A. (2000). Financial calculus. Cambridge: Cambridge University Press.
Black, F., & Scholes, M. (1973). The pricing of options and corporate liabilities. The Journal of Political Economy, 81, 637–659.
Cheng, W. Y., & Zhang, S. (2000). The analytics of reset options. The Journal of Derivatives, 8, 59–71.
Curnow, R. N., & Dunnett, C. W. (1962). The numerical evaluation of certain multivariate normal integrals. The Annals of Mathematical Statistics, 33, 571–579.
Genz, A. (1992). Numerical computation of multivariate normal probabilities. Journal of Computational and Graphical Statistics, 1, 141–149.
Gray, S. F., & Whaley, R. E. (1999). Reset put options: Valuation, risk characteristics, and an application. Australian Journal of Management, 24, 1–20.
Johnson, H. (1987). Options on the maximum or the minimum of several assets. Journal of Financial and Quantitative Analysis, 22, 277–283.
Liao, S. L., & Wang, C. W. (2003). The valuation of reset options with multiple strike resets and reset dates. Journal of Futures Markets, 23, 87–107.
Margrabe, W. (1978). The value of an option to exchange one asset for another. The Journal of Finance, 33, 177–186.
Ouwehand, P., & West, G. (2006). Pricing rainbow options. Wilmott Magazine, 5, 74–80.
Rubinstein, M. (1991). Pay now, choose later. Risk, 4, 13.
Stulz, R. (1982). Options on the minimum or the maximum of two risky assets: Analysis and applications. Journal of Financial Economics, 10, 161–185.
Acknowledgments
The authors thank the Ministry of Science and Technology of Taiwan for the financial support.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1: Proof of Theorem 2.1
The martingale pricing method relies heavily on the Cameron–Martin–Girsanov theorem, which is the key to evaluating the conditional expectation of the price processes with the multivariate standard normal cumulative distribution function (CDF). Therefore, the proof starts with the review of the general \(n\)-dimensional Cameron-Martin-Girsanov theorem based on the description in Baxter and Rennie (2000), with a slight change in notations.
Cameron–Martin–Girsanov Theorem
Let \(\vec {W}^Q(z)=(W_1^Q(z),\ldots ,W_{n}^Q(z))\) be a vector of \(n\)-dimensional independent \(Q\)-Brownian motions up to time \(T\) adapted with the natural filtration \(\mathcal {F}\). Suppose that \(\vec {\nu }(z)=(\nu _1(z),\ldots ,\nu _{n}(z))\) is an \(\mathcal {F}\)-previsible \(n\)-vector process that satisfies the condition \(E^Q[\exp (\frac{1}{2}\int _0^{T}\left| \vec {\nu }^2(\tau )\right| d\tau )]<\infty \). Then there is a new measure \(R\) such that
-
1.
\(Q\) is equivalent to \(R\) up to time \(T\),
-
2.
\(\frac{dR}{dQ}=\exp \left( \sum \limits _{j=1}^{n}\int _0^{T}\nu _{j}(\tau )dW_j^Q(\tau )-\frac{1}{2}\int _0^{T}\left| \vec {\nu }(\tau )\right| ^2d\tau \right) \), and
-
3.
\(W_j^{R}(z)=W_j^Q(z)-\int _0^{z}\nu _{j}(\tau )d\tau ,\) and \(\vec {W}^{R}(z)=(W_1^{R}(z),\ldots ,W_{n}^{R}(z))\) is a vector of \(n\)-dimensional \(R\)-Brownian motion up to time \(T\).
Based upon the general \(n\)-dimensional Cameron–Martin–Girsanov theorem, the most important finding in this paper is stated in Theorem 2.1.
Theorem 2.1
Let \(W_j^Q(z)\)and \(W_i^Q(z)\) be two correlated \(Q\)-Brownian motions, and \(\rho _{ji}\) is their correlation coefficient. By defining the Radon–Nikodym derivative as \(dR_{is}/dQ=\exp (\sigma _iW_i^Q(s) -\sigma _i^2 s/2)\), we can obtain that
is an \(R_{is}\)-Brownian motion for any time points \(s\) and \(z\).
Before proving Theorem 2.1, we state a useful identity equation. Suppose \(W_j^{P}(z)\) and \(W_i^{P}(z)\) are correlated Brownian motions with a correlation coefficient \(\rho _{ji}\) under any probability measure \(P\). Once we define \(\widetilde{W}_{j}^{P}(z)\equiv (W_j^{P}(z)-\rho _{ji}W_i^{P}(z))/\sqrt{1-\rho _{ji}^2}\), then \(\widetilde{W}_{j}^{P}(z)\) is an independent \(P\)-Brownian motion with respect to \(W_i^{P}(z)\). Rearranging the above equation, we can obtain
an equation that is helpful in the following proof.
Proof
Suppose \(W_i^Q(z)\) and \(\widetilde{W}_{j}^Q(z)\) are two independent \(Q\)-Brownian motions satisfying Eq. (6.1), that is, \(W_j^Q(z)=\rho _{ji}W_i^Q(z)+\sqrt{1-\rho _{ji}^2}\widetilde{W}_{j}^Q(z)\). By considering \(\vec {\nu }(z)=(\nu _{i}(z),\nu _{j}(z))=(\sigma _i\cdot I(z \le s),0)\) in the Cameron–Martin–Girsanov theorem, the Radon–Nikodym derivative can be derived as
and for any time point \(z\),
Finally, we have the following equations and the proof is complete:
\(\square \)
After proving Theorem 2.1, we would like to illustrate the specialty of Theorem 2.1 regarding the time dimension by evaluating \(E^Q [S_a(s) \cdot I(S_b(u) \ge S_c (v))]\) (given \(u \le s \le v\) for example), which is a problem arising exclusively when pricing FSROs or other path-dependent rainbow options. To evaluate this expectation, we first rewrite it with \(S_a (s)=S_a (0)\exp ((r-\sigma _a^2/2)s+\sigma _a W_a^Q(s))\), i.e.,
By defining the Radon–Nikodym derivative as \(dR_{as}/dQ=\exp (\sigma _a W_a^Q (s) -\sigma _a^2 s/2)\) and applying Theorem 2.1, we can obtain
As a result, Eq. (6.2) can be expressed as
The last equation is derived by replacing \(W_b^Q (u)\) and \(W_c^Q (v)\) according to Eqs. (6.3) and (6.4), respectively. The straightforward algebraic calculation leads to
where \(\text{ var }(\sigma _c W_c^{R_{as}} (v)-\sigma _b W_b^{R_{as}} (u))\) equals \(\sigma _c^2 v - 2\rho _{cb} \sigma _c \sigma _b \min (v,u) + \sigma _b^2 u\) by definition. In addition, since \(Z_{bu,cv}^{R_{as}} \equiv \frac{\sigma _c W_c^{R_{as}} (v) - \sigma _b W_b^{R_{as}} (u)}{\sqrt{\text{ var }(\sigma _c W_c^{R_{as}} (v) - \sigma _b W_b^{R_{as}} (u))}}\) follows the standard normal distribution, the last equation in the above can be derived with defining \(N(\cdot )\) as the cumulative distribution function of the standard normal distribution.
It should be noted that when we transform \(W_b^Q(u)\) to \(W_b^{R_{as}} (u)\) in Eq. (6.3), the adjustment term is \(\rho _{ba} \sigma _a u\), which is proportional to \(u\) as expected. On the other hand, when we transform \(W_c^Q (v)\) to \(W_c^{R_{as}} (v)\) in Eq. (6.4), although both \(W_c^Q (v)\) and \(W_c^{R_{as}} (v)\) are Brownian motions evolving up to the time point \(v\), the adjustment term is \(\rho _{ca} \sigma _a s\) rather than \(\rho _{ca} \sigma _a v\) according to Theorem 2.1. In fact, if we apply, without any modification, the commonly used change-of-probability-measure techniques for option pricing, e.g., those in Ouwehand and West (2006) and Liao and Wang (2003), this special characteristic associated with the time dimension cannot be captured such that we cannot derive the correct formula corresponding to \(E^Q [S_a(s) \cdot I(S_b(u) \ge S_c(v))]\).
Appendix 2: Derivations of the pricing formula
Theorem 2.1 makes it straightforward to derive the pricing formula for the FSRPO. Recall that in Eq. (2.5) the price today is
where \(S_{it}\equiv \left\{ \begin{array}{ll} K &{} i=M \\ S_i(t) &{} i\ne M \end{array} \right. .\)
We evaluate the option value in Eq. (7.1) term by term in the following paragraphs.
1.1 The first summation in Eq. (7.1)
The expectation inside the first summation in Eq. (7.1) can be evaluated directly in the probability measure \(Q\):
where
In the above equations, \(\sigma _{ab}\) is defined as \(\sqrt{\sigma _{a}^2-2\rho _{ab}\sigma _{a}\sigma _{b}+\sigma _{b}^2}\), and the notation \(d_{as,b\tau }^{P}\) is referred to as the parameter of the multivariate standard normal CDF under the event \(\{S_{a}(s)\ge S_{b}(\tau )\}\) with respect to any probability measure \(P\). The notation \(d_{K,b\tau }^{P}\) is defined similarly but corresponding to the event \(\{K \ge S_{b}(\tau )\}\).
For the correlation matrix \(R^{K,m}\), we first concatenate \(\{Z_{K,it}^Q\}_{1\le i\le n}\), \(\{Z_{jT,mT}^Q\}_{1 \le j\ne m\le n}\), and \(Z_{K,mT}^Q\) to form \(\{Z_{p}^{K,m}\}_{1\le p \le 2n}\), i.e., \(\{Z_{p}^{K,m}\}_{1 \le p\le 2n} \equiv \{\{Z_{\!K,it}^Q\}_{1\le i\le n},\!\{Z_{\!jT,mT}^Q\}_{1\le j\ne m\le n},Z_{K,mT}^Q\}\). Similarly, \(\{Z_{q}^{K,m}\}_{1\le q\le 2n} \!\equiv \! \{\{Z_{\!K,kt}^Q\}_{1\le k\le n}\), \(\{Z_{lT,mT}^Q\}_{1\le l\ne m\le n},Z_{K,mT}^Q\}\). In addition, we also define \(\rho _{p,q}^{K,m}\equiv \text {corr}(Z_{p}^{K,m},Z_{q}^{K,m})\), and \(R^{K,m} \equiv (\rho _{p,q}^{K,m})_{1\le p,q\le 2n}\) is a \(2n\times 2n\) correlation coefficient matrix as follows:
where
and other parts listed below can be computed similarly:
1.2 The second summation in Eq. (7.1)
Under the risk-neutral measure \(Q\), the underlying price of asset \(m\) at time \(T\) is
It is convenient to introduce the probability measure \(R_{mT}\) by setting the corresponding Radon–Nikodym derivative to
With Theorem 2.1, \(W_{j}^{R_{mT}}(z)=W_{j}^Q(z )-\rho _{jm}\sigma _m\min (z ,T)=W_{j}^Q(z )-\rho _{jm}\sigma _m z\) is a standard Brownian motion under the probability measure \(R_{mT}\), where \(j\) is the index for the underlying asset. Then we can rewrite the expectation inside the second summation in Eq. (7.1) evaluated with respect to the \(R_{mT}\) measure as follows:
where
It is worth noting that \(Z_{\bullet ,\bullet }^{R_{mT}}\) are essentially the same as \(Z_{\bullet ,\bullet }^Q\) in the first summation of Eq. (7.1) except under a different probability measure. Hence for the expectation inside the second summation, the correlation matrix of the multivariate standard normal CDF is the same as \(R^{K,m}\) in the first summation in Eq. (7.1).
1.3 The third summation in Eq. (7.1)
Following the same technique, we compute the expectation inside the third summation in Eq. (7.1). The underlying price of asset \(M\) at time \(t\) under the risk-neutral measure \(Q\) is
Let \(R_{Mt}\) be the equivalent probability measure defined by
According to Theorem 2.1, \(W_{j}^{R_{Mt}}(z )=W_{j}^Q(z)-\rho _{jM}\sigma _{M}\min (z ,t)\) is a standard Brownian motion under the probability measure \(R_{Mt}\), where \(j\) is the index for the underlying asset. Note that if the examined time point \(z\) is before \(t\), \(W_{j}^{R_{Mt}}(z )=W_{j}^Q(z)-\rho _{jM}\sigma _{M} z\); if the examined time point \(z\) is after \(t\), \(W_{j}^{R_{Mt}}(z )=W_{j}^Q(z)-\rho _{jM}\sigma _{M} t\) even though both \(W_{j}^{R_{Mt}}(z)\) and \(W_{j}^{Q}(z)\) evolve up to \(z\). The expectation inside the third summation in Eq. (7.1) evaluated with respect to the \(R_{Mt}\) measure is as follows:
where
and the notation \(\sigma _{as,b\tau }\) is defined as \(\sqrt{\sigma _{a}^2s-2\rho _{ab}\sigma _{a}\sigma _{b}\min (s,\tau )+\sigma _{b}^2\tau }\).
Next, we define \(\{Z_{p}^{M,m}\}_{1 \le p\le 2n}\equiv \{\{Z_{it}^{R_{Mt}}\}_{1\le i\le n},\{Z_{jT,mT}^{R_{Mt}} \}_{1\le j\ne m\le n},Z_{Mt,mT}^{R_{Mt}}\}\), \(\{Z_{q}^{K,m}\}_{1\le q\le 2n} \equiv \{\{Z_{kt}^{R_{Mt}}\}_{1\le i\le n},\{Z_{lT,mT}^{R_{Mt}}\}_{1\le j\ne m\le n},Z_{Mt,mT}^{R_{Mt}}\}\), \(\rho _{p,q}^{M,m} \equiv \text {corr}(Z_{p}^{M,m},Z_{q}^{M,m})\), and \(R^{M,m} \equiv (\rho _{p,q}^{M,m})_{1\le p,q\le 2n}\) as a \(2n\times 2n\) correlation coefficient matrix. For simplicity, we abuse the notation slightly and set \(\sigma _i=0\) when \(i=M\). Consequently, as \(i=M\), \(\sigma _{iM}=\sqrt{\sigma _{i}^2-2\rho _{iM}\sigma _i\sigma _{M}+\sigma _{M}^2}=\sqrt{\sigma _{M}^2}=\sigma _{M}\). Then the correlation coefficient matrix \(R^{M,m}\) is presented as follows:
where
1.4 The fourth summation in Eq. (7.1)
The probability measure \(R_{mT}\) used for the expectation inside the fourth summation in Eq. (7.1) is the same as that in the second summation of Eq. (7.1). In addition, comparing with the derivation details of the third summation, it can be easily shown that \(Z_{\bullet ,\bullet }^{R_{mT}}\) are in essence the same as \(Z_{\bullet ,\bullet }^{R_{Mt}}\) except under a different probability measure. Hence the correlation matrix for the multivariate standard normal CDF inside the fourth summation will be \(R^{M,m}\). Consequently, the expectation inside the fourth summation in Eq. (7.1) is evaluated as follows:
where
As a result, the pricing formula of the forward-start put option can be summarized as follows.
Appendix 3: Delta and gamma
The delta of an option can be directly derived by differentiating the pricing formula with respect to the initial price of the underlying asset. However, here we employ an alternative method that uses the linear homogeneity of the option pricing formula to derive the formula for delta. For any linearly homogeneous function, that is, \(f(\lambda x,\lambda y)=\lambda f(x,y)\), the Euler’s rule implies
To simplify the expression, we use \(S_i\) as shorthand for \(S_i(0)\) hereafter. Since the general pricing formula (2.6) is linearly homogeneous, that is,
by the Euler’s rule the equation can be written as
To obtain the delta with respect to the \(a\)-th asset, \(\frac{\partial P}{\partial S_{a}}\), we need to collect all the terms that contain \(S_{a}\). Recall that the pricing formula of \(P(0)\) in Eq. (2.6) is
Inside the above formula, \(S_{a}\) appears in the terms of the second and fourth summations when \(m\) is equal to \(a\) and in the third summation when \(M\) is equal to \(a\). The term that contains \(S_{a}\) in the second summation of the formula (8.1) is
Similarly, the terms with \(S_{a}\) in the third summation of the formula (8.1) can be grouped as follows.
where \(a^{\prime }\) is merely an index of the underlying assets. Finally, we consider the terms that contain \(S_{a}\) in the fourth summation of the formula (8.1) and the corresponding sum of these terms is
After rearranging the terms in the formula (8.1) with respect to each \(S_{a}\) and the strike price \(K\), we can obtain the formula for delta with respect to the \(a\)-th asset as follows:
Based on the formula for delta, we can proceed to derive the formulae for gamma with respect to the \(a\)-th asset and cross gamma with respect to the \(a\)-th and \(b\)-th assets, \(\frac{\partial ^2 P}{\partial S_{a}^2}\) and \(\frac{\partial ^2 P}{\partial S_{a}\partial S_{b}}\). Consider the derivation of \(\frac{\partial ^2 P}{\partial S_{a}^2}\) first:
We obtain the derivative of the first term in the above equation by the chain rule.
The key element in deriving the formula for gamma is the calculation of the derivatives of the multivariate standard normal CDFs. The method adopted in this paper is as follows. Let \(\{d_{u}\}_{1\le u\le 2n}\equiv \{\{d_{K,it}^{R_{aT}}\}_{1\le i\le n},\{d_{jT,aT}^{R_{aT}}\}_{1\le j\ne a\le n},d_{K,aT}^{R_{aT}}\}\), and \(\rho _{u,v}^{K,a}\) be the correlation coefficient element of \(R^{K,a}\). Taking \(\partial N_{2n}/\partial d_{K,aT}^{R_{aT}}\) as an example, we rewrite the above multivariate standard normal CDF according to the results in Curnow and Dunnett (1962).
where
and \(\phi (x)\) is the standard normal probability density function. Hence, \(\partial N_{2n}/\partial d_{2n}=\partial N_{2n}/\partial d_{K,aT}^{R_{aT}}\) can be calculated as
After similarly calculating other terms, we obtain the formula for the FSRPO’s gamma. This paper only briefly describes the method to derive the formulae of gamma and cross gamma, and the detailed formulae are omitted for brevity but available upon request.
Appendix 4: Formulae of discrete-sampling lookback options
This appendix demonstrates the generality of our main theorem by applying it to discrete-sampling lookback rainbow options. In order to describe the formulae more clearly, the notation is slightly different from that in Sect. 2. Let \(S_i(t_j)\) denote the price of the \(i\)-th stock at time \(t_{j}\) for \(i=1,2,\ldots ,m\), \(j=1,2,\ldots ,n\), and \(0=t_0<t_1<\cdots <t_n=T\). The return dynamics of \(S_i\) for \(i=1,2,\ldots ,m\) follow the same stochastic differential equations as Eq. (2.2). In addition, \(M\) denotes the historical highest price up to now. Except for above differences, the rest of the notation, such as \(r\), \(\sigma _i\), \(K\), \(T\), and \(\rho _{ij}\), is the same.
Consider the discrete-sampling lookback rainbow option that is a call option on the maximum among all the prices of every risky asset at every monitoring time during the option’s life. Let \(C_{\max }(T)\) be the payoff of this option at maturity \(T\) with the historical highest price \(M\) until \(t_0\) and constant strike price \(K\):
where
By the risk-neutral valuation, the option value today can be expressed as follows:
As \(M\ge K\), \(I_{A}=1\), the equation (9.1) can be expressed as
where
and
and
In the other case of \(M<K\), \(I_{A_{M}}=0\), the formula is similar to the case of \(M\ge K,\) and Eq. (9.1) can be expressed as
where
This appendix shows that the pricing technique of Theorem 2.1 can be employed to price different types of discrete-sampling path-dependent rainbow options. Therefore, the technique proposed in this paper can broach many possible directions in the design of new contracts with both the path dependent and multiple asset features.
Rights and permissions
About this article
Cite this article
Chen, CY., Wang, HC. & Wang, JY. The valuation of forward-start rainbow options. Rev Deriv Res 18, 145–188 (2015). https://doi.org/10.1007/s11147-014-9105-0
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11147-014-9105-0
Keywords
- Discrete-sampling path-dependent option
- Rainbow option
- Forward-start option
- Reset option
- Lookback option