Symmetrization associated with hyperbolic reflection principle
 138 Downloads
Abstract
In this paper, in view of application to pricing of Barrier options under a stochastic volatility model, we study a reflection principle for the hyperbolic Brownian motion, and introduce a hyperbolic version of ImamuraIshigakiOkumura’s symmetrization. Some results of numerical experiments, which imply the efficiency of the numerical scheme based on the symmetrization, are given.
Keywords
Hyperbolic Brownian motion Reflection principle Symmetrization Barrier option EulerMaruyama schemeIntroduction
Reflection principle and the static hedge of barrier options

T is the expiry date of the option,

K is the exercise price,

S is the price process of a risky asset, with S_{0}>K,

τ:= inf{s>0:S_{ s }<K^{′}}, the first hitting time of S to K^{′}, the knockout boundary, with K^{′}<K.

at τ equals the call, and

is zero at T on τ>T.

zero at T on τ≤T since at τ it is liquidated, and

(S_{ T }−K) at T on τ>T.
by the reflection principle. In general, the property is referred to as (arithmetic) putcall symmetry at K [4], which is weaker than the reflection principle that ensures putcall symmetry for any K.
The interpretation is first proposed in [3], and there are vast literatures since then (see e.g. [1] and references therein). Among these, we just mention a multidimensional extension proposed in [11], where the reflection principle with respect to reflection groups is applied to the pricing of multiasset barrier options, barrier being the boundary of a Weyl chamber. To the best of our knowledge, it is the first attempt to deal with the cases where the barrier(=knockout/in boundary) is not a one point set.
Symmetrization and its application to numerical calculation of the price of barrier options
A new point of view in the literature, the symmetrization, was first introduced in [10], and further generalized in [2]. The symmetrization is a procedure to convert a given diffusion into the one with a weaker version of reflection principle, aiming at obtaining a precise numerical value of the price of barrier options in a reasonable computational time, rather than statichedge in the market described in the previous section.
The following is proven in [10].
Theorem 1
(ImamuraIshigakiOkumura [10]) The lawunique solution \( \tilde {S} \) of (4) satisfies the putcall symmetry at K, and \( (\tilde {S}_{t})_{0 \leq t \leq \tau } \) has the same law as (S_{ t })_{0≤t≤τ}.

An expectation with stopping time is converted to the one without it.
A numerical calculation of an expectation with stopping time often is a tough challenge due to its pathdependent nature. On the other hand, an expectation with respect to one dimensional marginal of a diffusion process is in most cases numerically tractable. Thus the Eq. (6) gives a new insight to the numerical analysis of barrier options/stopping times.
EulerMaruyama approximation of the price of barrier options
The most common technique to numerically approximate an expectation with respect to a diffusion process would be socalled “EulerMaruyama” scheme. Here we briefly recall the scheme.
where \( \tilde {S}^{n} \) is obtained by the same procedure as (7).
The discretization error, by which we mean the difference between the true value of the expectation and its EulerMaruyama approximation like (8) or (9), is known to be of O(n^{−1/2})in general when t_{ k }−t_{k−1}=T/n for all k. It is reported in [5] that the one with stopping time like (8) cannot be improved, while the one with onedimensional marginal like (9) is, provided some continuity of the coefficients, known to be of O(n^{−1}).
The symmetrized drift coefficient (5) may not be continuous in general even if the original one is very smooth, and as far as we know, no existing result ensures the order is of O(n^{−1}) though recently there have been several papers ([12, 13], and [14]) to deal with discontinuous coefficients in line with the problem posed here. In [10], however, they conjecture that it is the case by performing numerical experiments.
References for more detailed and precise results of the order can be found in [10].
SABR model and hyperbolic Brownian motion
In the present paper, we study a hyperbolic version of the symmetrization, with a view to the application of the pricing of barrier options under SABR model, which is known to be transformed to hyperbolic Brownian motion with drift.

A driftless local volatility model is obtained by setting ν=0, and

\( Z_{t} := \psi (S_{t/\nu ^{2}}, V_{t/\nu ^{2}}) + \sqrt {1} V_{t/\nu ^{2}}\) with \( \psi (x,y) = (x\rho y)/\sqrt {1\rho ^{2}} \) is a hyperbolic Brownian motion with drift, a solution to (13) in “Hyperbolic symmetrization” section (for details see [7]).
The following is a “motto” widely accepted among researchers and practitioners in finance (see e.g. [8]): as tractability of one dimensional diffusion processes is attributed to the reduction to the standard Brownian motion with drift by the Lamperti transform, so the analysis of SABR model will be converted to that of hyperbolic Brownian motion with drift, where we can still work on symmetries from linear fractional transformations. We shall observe a realization of this idea in the present paper.
The contents of the present paper
We start with introducing a hyperbolic version of the reflection principle that parallels the one with the standard Brownian motion in “Hyperbolic reflection principle” section. We introduce in “Hyperbolic symmetrization” section a weak version of the reflection principle, which also parallels with the classical putcall symmetry. Associated symmetrization is then introduced. “Numerical experiments” section is devoted to numerical studies. As in the case of the ImamuraIshigakiOkumura’s scheme using classical symmetrization, the error is not proven to be O(n^{−1}) mathematically but the numerical results support the conjecture of the hyperbolic case as well.
Hyperbolic reflection principle
Invariant property of hyperbolic Brownian motion
Proposition 1
Let \(f:\mathbb {H}\to \mathbb {H}\) be such that \(f(z):=\frac {az+b}{cz+d}\), where \(\left (\begin {array}{ll} a & b \\ c & d \\ \end {array}\right)\in \text {SL}(2,\mathbb {R})\). Then (f(Z_{ t }))_{t≥0} and (Z_{ t })_{t≥0} are equivalent in law provided that f(Z_{0})=Z_{0}.
Proof
where \(d\widetilde {W}^{\mathbb {C}}_{t}=\frac {{cZ}_{t}+d^{2}}{({cZ}_{t}+d)^{2}}dW^{\mathbb {C}}_{t}\), which is another complex Brownian motion. Hence Z_{ t } and f(Z_{ t }) are equivalent in law if they start from the same point, as they are defined by the same SDE. □
Hyperbolic reflections
Let Open image in new window be the totality of such isometries π on the upperhalf plane \(\mathbb {H}\) that π^{2}=Id and that the invariant set \(\text {Inv}_{\pi }:=\left \{z \in \mathbb {H}: \pi (z)=z \right \}\) is a geodesic on \(\mathbb {H}\).
Proposition 2
Proof
It is wellknown that an isometry on \( \mathbb {H} \) is either Φ_{ A } or Φ_{ A }∘Φ_{0} for some \( A \in \text {SL}(2,\mathbb {R})\). By the fundamental theorem of algebra, we know that the equation Φ_{ A }(z)=z has at most two solutions of complex for \(A \in \text {SL}(2,\mathbb {R})\). So Open image in new window .

If a=d, for any b and c, (10) is satisfied.

If a≠d, the Eq. (10) is
 If$$A= \left(\begin{array}{ll} a & b \\ \frac{a^{2}1}{b} & a \\ \end{array}\right), $$
The last equation means that it is a half circle, with center \(\left (\frac {ab}{a^{2}1},0\right)\) and radius of \(\left \frac {b}{a^{2}1}\right \).
 If$$A= \left(\begin{array}{cc} \pm 1 & 0 \\ c & \pm 1 \\ \end{array}\right), $$
The invariant set is a circle with center \(\left (\frac {1}{c},0\right)\) and the radius \(\frac {1}{c}\).
If c=0, the invariant set is the lines perpendicular to the real line. □
Hyperbolic reflection principle
Let Open image in new window . Then, \( \mathbb {H} = D_{+} \cup \text {Inv}_{\pi } \cup D_{} \), where D_{±} are the connected components of \( \mathbb {H} \setminus \text {Inv}_{\pi } \).
Proposition 3
(Hyperbolic Reflection Principle) Let Z_{0}∈D_{+} and τ=inf{t≥0:Z_{ t }∉D_{+}}=inf{t≥0:Z_{ t }∈Inv_{ π }}. If we put \(\widetilde {Z_{t}}=Z_{t} 1_{\left \{t<\tau \right \}}+\pi \left (Z_{t}\right)1_{\left \{t\geq \tau \right \}}\), then we have \( \left (Z_{t}\right)=(\widetilde {Z_{t}}) \) in law.
Proof
It suffices to show that if π is a reflection of \(\mathbb {H}\), then (π(Z_{ t }))_{t≥0}=(Z_{ t })_{t≥0} in law if Z_{0}∈Inv_{ π } since Z is a strong Markov process and Z_{ τ }∈Inv_{ π }. As we have seen that π=Φ_{ A }∘Φ_{0} for some specific \( A \in \text {SL} (2, \mathbb {R}) \), and by Proposition 1, we only need to check that \( (\overline {Z}_{t}) \) is identically distributed as (Z_{ t }) as a stochastic process, but this is obvious since (X_{ t }) is identically distributed as (−X_{ t }). □
Hyperbolic symmetrization
Hyperbolic putcall symmetry
where \( W^{\mathbb {C}} := W^{1} + i W^{2} \) and μ(Z)=μ_{1}(Re(Z),Im(Z))+iμ_{2}(Re(Z),Im(Z)).
Theorem 2
and (13) has a unique weak solution. Then (π(Z_{ t })) and (Z_{ t })have the same law as a stochastic process, provided that Z_{0}∈Inv_{ π }
Proof
Symmetrization
Here we present a hyperbolic version of the symmetrization introduced in [2] and [10].
Theorem 3
 (i) the law unique solution of the SDE, if it exists,satisfies (π(Z_{ t }))=(Z_{ t }) in law, provided that Z_{0}∈Inv_{ π }.$${dZ}_{t} = \text{Im} (Z_{t}) dW^{\mathbb{C}} + \tilde{\mu} (Z_{t}) \,dt\qquad $$

(ii) Let Z_{0}∈D_{+} and τ=inf{t≥0:Z_{ t }∉D_{+}}=inf{t≥0:Z_{ t }∈Inv_{ π }}. If we put \(\widetilde {Z_{t}}=Z_{t} 1_{\left \{t<\tau \right \}}+\pi \left (Z_{t}\right)1_{\left \{t\geq \tau \right \}}\), then we have \( \left (Z_{t}\right)=(\widetilde {Z_{t}}) \) in law.
 (iii) [Conversion Formula] Suppose that F is a bounded measurable function on \( \mathbb {H} \) with support in D_{+}. Then,$$\begin{aligned} &E [F(Z_{t}) 1_{\{\tau > t\}}]\\ &= E [F(Z_{t}) ]  E [F(\pi(Z_{t})) ]. \end{aligned} $$
Proof
(i) and (ii) are direct consequences of Theorem 2 and Proposition 3. (iii) can be proven in the same manner as in [10]. □
Example 1
Numerical experiments
In the hyperbolic symmetrization proposed in the present paper the symmetrized drift may not be continuous in general, as in the case of the symmetrization in [10]. This means that no rigorous mathematical result guarantees the efficiency— (high) order of convergence— in EulerMaruyama approximation. In [10], it is claimed, however, that numerical experiments show the efficiency. In this section we present some simulation results of the Example 1 with c=1, t=1, and F(z)=(z−1)_{+}∧N with N=10^{4}, which suggest that in the hyperbolic case the conjecture is still likely to be true.
 1.
Let n be the number of discretization; we put t_{ k }=k/n, k=0,1,⋯,n.
 2.Let Z be the original process and \( \widetilde {Z} \) be the symmetrized one. We approximate Z and \( \widetilde {Z} \) by Z^{ n }=(X^{ n },Y^{ n }) and \( \widetilde {Z}^{n} = (\widetilde {X}^{n}, \widetilde {Y}^{n}) \), defined asand$$\begin{aligned} & X^{n}_{t_{k}}X^{n}_{t_{k1}} \\ &= Y^{n}_{t_{k1}}\Delta W^{n}_{t_{k}} + \mu \left(Y^{n}_{t_{k1}}\right) n^{1}, \\ Y^{n}_{t_{k}} &= Y^{n}_{t_{k1}} \text{exp} \left(\Delta W^{n}_{t_{k}}  (2n)^{1}\right), \\ \,k&=1,2,\cdots,n \end{aligned} $$where \(\tilde {\mu _{1}}\) and \(\tilde {\mu }_{2}\) are such that \(\tilde {\mu } = \tilde {\mu }_{1} + i \tilde {\mu _{2}}\). Here \( \left \{ \Delta W^{n}_{t_{k}}: k=1,2, \cdots, n\right \} \) simulates, by pseudo random numbers, independent copies of centered Gaussian random variables with variance n^{−1}.$$\begin{aligned} \widetilde{X}^{n}_{t_{k}}\widetilde{X}^{n}_{t_{k1}} =& \widetilde{Y}^{n}_{t_{k1}}\Delta W^{n}_{t_{k}} + \tilde{\mu}_{1}\\ & \left(\widetilde{X}^{n}_{t_{k1}}, \widetilde{Y}^{n}_{t_{k1}}\right) n^{1}, \\ \widetilde{Y}^{n}_{t_{k}}\widetilde{Y}^{n}_{t_{k1}} =& \widetilde{Y}^{n}_{t_{k1}}\Delta W_{t_{k}}^{n} +\tilde{\mu_{2}}\\ &\left(\widetilde{X}^{n}_{t_{k1}}, \widetilde{Y}^{n}_{t_{k1}}\right) n^{1}, \\ k=&1,2,\cdots,n, \end{aligned} $$
 3.The MonteCarlo simulation of PathWise EulerMaruyama approximation of E[F(Z_{1})1_{{τ>1}}] is obtained bywhere Z^{ n,m } stands for the mth simulation of Z^{ n }, and$$\begin{aligned} &\text{PWEM} (n) \\ &:= \frac{1}{M} \sum_{m=1}^{M} F\left(Z^{n,m}_{1}\right) 1_{\{\tau^{n,m} > 1 \} }, \end{aligned} $$$$\tau^{n,m} = \text{min} \left\{ t_{k} : Z^{n,m}_{t_{k}} \leq 1 \right\} $$
 4.The MonteCarlo simulation of \( E[F(\widetilde {Z}_{1}) ]  E [F(\pi (\widetilde {Z}_{1}))] \) is given by$$\begin{aligned} & \text{Symmetrization} (n) \\ &:= \frac{1}{M} \sum_{m=1}^{M} \left(F\left(\widetilde{Z}^{n,m}_{1}\right)  F \left(\pi \left(\widetilde{Z}^{n,m}_{1}\right)\right)\right), \end{aligned} $$
 5.
The “true” value Tr(n) is set to be Symmetrization(n) for some large n.
 6.The errors are calculated accordingly asand$$\begin{aligned} &\text{PW EM Error} (n)\\ &:= \text{log} \text{Tr}(n)  \text{PW EM} (n) \end{aligned} $$$$\begin{aligned} &\text{Sym Error} (n)\\ &:= \text{log} \text{Tr}(n)  \text{Symmetrization} (n). \end{aligned} $$
(X_{0},Y_{0})=(0.75,0.7), Tr1000=0.116674
M:No. of simmulation trials  n:No. of time steps  Symmetrization  PW EM  Sym Error  PW EM Error 

11000  10  0.494071  0.212777  −2.342335  −0.974458 
18000  20  0.424634  0.165305  −3.023494  −1.177785 
27000  30  0.374365  0.15732  −3.202855  −1.355994 
64000  40  0.350142  0.138187  −3.839098  −1.454710 
125000  50  0.33173  0.135724  −3.960688  −1.536857 
216000  60  0.316112  0.128837  −4.409357  −1.612252 
343000  70  0.301227  0.13156  −4.207334  −1.689819 
512000  80  0.290315  0.130349  −4.292186  −1.750765 
729000  90  0.281466  0.128077  −4.473879  −1.803071 
1000000  100  0.273353  0.126315  −4.641730  −1.853556 
(X_{0},Y_{0})=(1.0,1.0), Tr1000=1.253903
M:No. of simmulation trials  n:No. of time steps  Symmetrization  PW EM  Sym Error  PW EM Error 

18000  20  1.305908  1.22153  −3.430431  −2.956415 
27000  30  1.302775  1.235505  −3.995513  −3.018551 
64000  40  1.294995  1.234476  −3.941091  −3.191942 
125000  50  1.294358  1.2407  −4.327311  −3.207565 
216000  60  1.290095  1.23953  −4.242404  −3.318917 
343000  70  1.293495  1.247477  −5.047403  −3.229128 
512000  80  1.295325  1.249126  −5.343942  −3.183943 
729000  90  1.293081  1.248791  −5.276165  −3.239640 
1000000  100  1.290695  1.250402  −5.654707  −3.302475 
Footnotes
 1.
A barrier option is a financial derivative with an additional condition that is made active when the underlying price process goes beyond/below a certain level. For details, see e.g. [9].
 2.
An option is called of “knockout” type if the payoff becomes zero if the underlying price process hits a certain value.
Notes
Funding
There’s no funding.
Availability of data and materials
The experiments are reproducible except the pseudo random numbers used in the MonteCarlo simulation.
Authors’ contributions
Introduced a hyperbolic version of ImamuraIshigakiOkumura’s symmetrization, and by numerical experiments showed efficiency of the scheme. All authors read and approved the final manuscript.
Competing interests
The authors declare that they have no competing interests.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
 1.Akahori, J, Barsotti, F, Imamura, Y: The Value of Timing Risk, working paper (2017). arXiv:1701.05695 [qfin.PR].Google Scholar
 2.Akahori, J, Imamura, Y: On symmetrization of diffusion processes. Quant. Finance. 14(7), 1211–1216 (2014).MathSciNetCrossRefMATHGoogle Scholar
 3.Bowie, J, Carr, P: Static Simplicity. Risk. 7(8), 44–50 (1994).Google Scholar
 4.Carr, P, Lee, R: PutCall Symmetry: Extensions and Applications. Math. Finance. 19(4), 523–560 (2009).MathSciNetCrossRefMATHGoogle Scholar
 5.Gobet, E: Weak approximation of killed diffusion using Euler schemes. Stoch. Process. Appl. 87(2), 167–197 (2000).MathSciNetCrossRefMATHGoogle Scholar
 6.Hagan, P, Lesniewski, A, Woodward, D: Managing smile risk. Wilmott Mag. 1, 84–108 (2002).Google Scholar
 7.Hagan, P, Lesniewski, A, Woodward, D: Probability Distribution in the SABR Model of Stochastic Volatility. In: Large Deviations and Asymptotic Methods in Finance Volume 110 of the series Springer Proceedings in Mathematics & Statistics, pp. 1–35 (2015). http://www.springer.com/us/book/9783319116044.
 8.HenryLabordère, P: A General Asymptotic Implied Volatility for Stochastic Volatility Models (2005). arXiv:condmat/0504317.Google Scholar
 9.Hull, J: Options, Futures, and Other Derivatives. 9th edn. Pearson/Prentice Hall, Upper Saddle River (2014).Google Scholar
 10.Imamura, Y, Ishigaki, Y, Okumura, T: A numerical scheme based on semistatic hedging strategy. Monte Carlo Methods Applic. 20(4), 223–235 (2014).MathSciNetCrossRefMATHGoogle Scholar
 11.Imamura, Y, Takagi, K: SemiStatic Hedging Based on a Generalized Reflection Principle on a Multi Dimensional Brownian Motion. AsiaPacific Finan. Markets. 20(1), 71–81 (2013).CrossRefMATHGoogle Scholar
 12.KohatsuHiga, A, Lejay, A, Yasuda, K: Weak approximation errors for stochastic differential equations with nonregular drift. J. Comput. Applied Math. 326, 138–158 (2017).MathSciNetCrossRefMATHGoogle Scholar
 13.Ngo, HL, Taguchi, D: Strong convergence for the Euler–Maruyama approximation of stochastic differential equations with discontinuous coefficients. Stat. Probab. Lett. 125, 55–63 (2017).MathSciNetCrossRefMATHGoogle Scholar
 14.Taguchi, D: Stability Problem for OneDimensional Stochastic Differential Equations with Discontinuous Drift. In: DonatiMartin, C, Lejay, A, Rouault, A (eds.) Séminaire de Probabilités XLVIII. Lecture Notes in Mathematics, vol 2168. Springer, Cham (2016).Google Scholar
Copyright information
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.