Arxiv:1605.00307V1 [Q-Fin.CP] 1 May 2016 Arteei H Etnadshoe-H 2]Models

arXiv:1605.00307v1 [q-fin.CP] 1 May 2016 ARteei h etnadShoe-h 2]models. Schöbel-Zhu [24] and Heston the is there SABR lg oma eetn oes hl o the for while models, reverting method- mean same to the applying ology discuss we stochastic Second correlated three variables. on functional are prices option or eedn aaeest rc ail pin ( options time- Vanilla with price equations to parameters SABR dependent of solutions path-integral case. mean-reverting ue rdcs htaeas isn r generaliza- are the missing of Struc- also solutions and are and options What tions Asian Products. as path tured for such needed contracts are which expressions dependent trajectories, analytic stochastic explicit the providing for not are solutions the [14]. for discussed al mastered et was finally Korn integrals and by multiple SABR [11] literature of so- the form analytical a in of the problems concept in selected approxi- The lutions in needed. are yet is using accuracy solutions applications, higher options These most in Vanilla useful [10]. solutions of mated/asymptotic pricing established [26] accurate model Stein and equations Stein Heston the the earlier. of of and structure the [12] follows This cess. ARdnmc stelnaiyi the in the linearity of the characteristic is important dynamics SABR very The model. Normal sacnnclSohsi oaiiymodel: Volatility Stochastic canonical a is hfe o-omleuto.A icse n[5 17], [15, the in discussed of As setting and equation. Log-Normal models Shifted Black-Scholes Bachelier/Normal, eiaayi ahitga ouino ARadHso equ Heston and SABR of solution integral path Semi-analytic h agto hswr stood is euse we First twofold. is work this of target The niefrteSitdLgNra oe,tee‘base’ these model, Shifted-Log-Normal the for Unlike very provided model SABR the of Authors The h ARmdlitoue yHgne l 1,20] [10, al. et Hagan by introduced model SABR The hsfaeokpoie aua xeso fthe of extension natural a provides framework This )tgte with together 1) β d d S σ ˜ ˜ and 0 = t t = = iedpnigprmtr.Frhr tcnas epractica be also can the it in Further, pricing the Vanilla parameters. solving depending when correlate lite time efficient three the and on in practical functional analyti is known expressions integrated compact method, is to e This variables leads SABR-type of Monte-Carlo. solving set via one for numerically approach, method this semi-analytical In a discuss We ewrs ARHso eiaayia ouin,Aino Asian solutions, semi-analytical SABR/Heston Keywords: S ν σ 0 ˜ .INTRODUCTION I. β t d ARmdladt other to and model SABR 0 = S S ˜ V ˜ 0 t β t ρ sa pin.W hwthat show We options. Asian 0 = autedsHC nvri´ eLuan,C-05Lausann CH-1015 Lausanne, Université de HEC, Faculté des β = σ ˜ t { ± β ρd ed oteSitdLog- Shifted the to leads 1 V AReutost a to equations SABR 0 = ˜ t + p 1 W − ˜ t ρ a ulnk n ei Tyloo Kevin and Kuklinski Jan tcatcpro- stochastic 2 d W ˜ Dtd etme 4 2018) 14, September (Dated: n sa options Asian and t } β yemodels. type 0 = β β β 1 = 0 = 0 = (1) (2) M1 n M2 o ail option. Vanilla for (MC2) Monte-Carlo 2 between and convergence (MC1) the 1 of Comparison 1. FIG. yais[3[] e h raeteothsbe put been has effort greatest the Yet Heston the [23][1]. numerically havedynamics solving arithmetic including authors the of options, Many pricing Asian the approximate or literature. solve to the tried in discussed widely very simulate to having paths. without of number priced large be out-of-the-money strongly can that It options note FIG.1). to (see interesting paths error, than also few the is a smaller for And is even approach square-root, classical used. twice the in is is converging method still task MC2 although computational the The when task smaller numerical means. ex- the and analytical diminishing useful by indicates Monte- MC2) it semi-analytical as as term referred planatory the (also found integration We Carlo Monte- Conditional [31]. inte- as semi- literature MC2 Carlo the also Such in it known is call analytically. gration), (we integrated technique the integration be as analytic can integrated Monte-Carlo one be second to dimension one only MC1. as fast as twice h su fpiigaihei sa pin svery is options Asian arithmetic pricing of issue The requires using are we methodology the that stress We

Std. error smile and skew with ptions 10 10 10 10 10 10 10 10 tcatcvrals h methodology The variables. stochastic d -3 -2 -1 -7 -6 -5 -4 10 0 al hl h eodsti integrated is set second the while cally AR etnadBtsmdl with models Bates and Heston SABR, l ple opiigAinotosin options Asian pricing to applied lly 2 2.0 3.0 2.5 3.5 auea odtoa Monte-Carlo, Conditional as rature 10 utosbsdo ahintegrals. path on based quations 2 10 10 3 3 10 speed speed ratio MC2/MC1 4 ,Switzerland e, 10 10 5 10 4 tos rcn Vanilla pricing ations: 6 10 7 10 N 10 5 8 10 Inset: 6 C sabout is MC2 10 7 N MC2 MC1 − 0 . 5 10 8 2 on finding or approximating the distribution of a sum Where: of log-normally distributed random variables. This sum arises naturally for arithmetic Asian options under the t ˜ 2 Black-Scholes model due to the average in the payoff. ˜x 2νVs−ν s ft = σ0 e ds (8) Milevsky and Posner [19] approximates the finite sum s 0 Z of Lognormals by an inverse Gamma distribution which σ0ρ ˜ − 2 2 f˜y = eνVt ν t/ 1 (9) is the true density if the sum was infinite [6, 33, 34]. t ν { − } A similar idea was applied by Levy [16] and Turnbull t ˜ − 2 2 ˜z 2 νVs ν s/ ˜ and Wakeman [29] but using a log-normal distribution ft = σ0 1 ρ e dWs (10) − 0 [2]. However the efforts done for the Black-Scholes model p Z cannot account for market skew and smile. ˜z We can further rewrite the variables ft using time In contrast to the rather complex calculations made changed integrals τ(t)= ν2t e.g.: for Asian options in the Black-Scholes model, the semi- analytic the β = 0 SABR pricing for Asian options pre- ϑ=ν2t sented herein is very simple and efficient and to the best σ0 ˜ ˜z 2 2Vτ −τ ˜ ft = 1 ρ e dτQW (11) of our knowledge was not discussed prior in the literature. ν − s 0 It should be stressed that the methods applied to other p Z β = 0 models and that such framework can be considered and Q˜W being an unit Gaussian variable (E[Q˜W ]=0 as generalizations of the Heston, Bates [4] and Schöbel- ˜2 and E[QW ] = 1), uncorrelated with the Wiener process Zhu models. In contrast to Vanilla pricing, the Asian V˜τ . options pricing simplicity cannot be reproduced outside Finally the β = 1 SABR solution has the form: the SABR β = 0 model. For Vanilla pricing, the semi-analytic MC2 method can ˜ 1 ˜x 2 ˜y ˜z be applied for both β = 0 and β = 1. This type of cal- St = S0 exp[ (ft ) + ft + ft ] (12) culation can be used to check the accuracy of the Hagan −2 solutions for the SABR model. It is worth stressing that The same pattern applied to the β = 0 SABR: for β = 1, solutions discussed by Korn et al. are to the best of our knowledge not available. Most impor- ˜ ˜ 2 ˜ tantly, the MC2 calculation provides a reliable solution dSt = S0σ˜t ρdVt + 1 ρ dWt (13) { 2 − } to stochastic equations with time dependent parameters. pν σ˜ = σ0 exp[νV˜ t] (14) Last but not least it can also be used as a comparative t t − 2 tool to calibrate general Monte-Carlo pricing scheme. leads to the following solution: The presented calculation procedure is that it can be extended to mean-reverting models including jumps. We ˜ ˜y ˜z briefly discuss its application to the Heston, Bates and St = S0 1+ ft + ft (15) Schöbel-Zhu models. { } The formal path-integral solutions of the SABR equations as displayed in Eqs. 12 and 15 have a simple form 2 II. PATH INTEGRAL SOLUTION OF THE β = 1 depending on three parameters ϑ = ν t, σ0√T and ρ. AND β = 0 SABR EQUATIONS Further for ρ = 0 the SABR solutions are depending only on two parameters ϑ andσ ¯ = σ0√T . For β = 0 and ρ = 1 we recover the Shifted Log-Normal model We start with β = 1 SABR model: ± ˜y depending only on ft . This generic form of SABR solutions was partially discussed in the literature [18, 25, 31]. 2 dS˜t = S˜tσ˜t ρdV˜t + 1 ρ dW˜ t (3) { 2 − } pν σ˜t = σ0 exp[νV˜t t] (4) III. RE-FORMATIING THE SOLUTIONS AND − 2 SEMI-ANALYTICAL INTEGRATION

We use the log-variable Φ˜ t: When considering stochastic problems depending on a ﬁxed time horizon t = T , such as Vanilla pricing, we Φ˜ t(S˜t,t)= ln[S˜t/S0] (5) can further simplify the structure of the solutions using 2 1 another set of stochastic variables (ϑ = ν T ): dΦ˜ = (˜σ )2dt +˜σ ρdV˜ + 1 ρ2dW˜ (6) t −2 t t{ t − t} p We integrate the three parts: y˜ϑ = exp[V˜ϑ ϑ/2] 1 (16) − − ϑ ˜ 1 x 2 y z z˜ϑ = dτ exp[2Vτ τ] (17) Φ˜ = (f˜ ) + f˜ + f˜ (7) s 0 − t {−2 t t t } Z 3

The solutions have then the form: that cannot be adjusted to the market. To accommodate for a wider range of skew and smile a framework such as σ0 the SABR model needs to be considered. We focus in f˜x = z˜ (18) t ν ϑ this section on the β = 0 SABR model. σ0ρ For the arithmetic Asian options , under the Bache- f˜y = y˜ (19) t ν ϑ lier model, we can use the sum of Gaussian variables to σ0 characterize its average of the underlying spot price. The f˜z = 1 ρ2z˜ Q˜ (20) t ν − ϑ W formula for an Asian option based on monthly prices and p yearly averaging, is equivalent to Vanilla form but the The solution provided by Eq. 20 follows with the logic volatility is now adapted (β = 0, ν = 0, ρ = 0). of the formalism introduced first by Bougerol [5] and follows by other [30, 32, 33]. Bachelier σ0 z z z S˜ = S0 1+ (f˜ + f˜ + + f˜ ) (22) Following Eq. 20 we see that for t = T a double numer- Average 12 t1 t2 ··· t12 ˜ ˜ ical Monte-Carlo integration over Vt and Wt trajectories j=12 can be replaced with a Monte-Carlo integration involving σ0 = S0 1+ (n j + 1)[W˜ W˜ − ] ˜ ˜ 12 − tj − tj 1 trajectories Vt and a single Gaussian variable QW . j=1 n X o As next step is made, following Eq. 10, for pricing (23) problems in which the integration over the Q˜W variable ˜ = S0 1+ σ Q˜ (24) can be performed analytically. The we generate Vt tra- { A B} =12 jectories via Monte-Carlo and for each volatility trajec- j 2 1 n j +1 σ0 tory solve analytically the linear diffusion/pricing prob- σ = σ0 − = (25) A v12 12 ∼ √ lem eventually averaging exact Vanilla prices over the set u j=1 3 u X of trajectories. The calculation details together with nu- t merical computations are discussed in the Appendix. We Eventually: call this semi-analytic Monte-Carlo calculations, while Bachelier Bachelier the literature use the name of Conditional Monte-Carlo CAsian (F,K,σ0, ∆T =1)= CV anilla (F,K,σA, ∆T = 1) [18, 31] or Rao-Blackwellization in the context of the Rao- (26) Blackwell-Kolmogorov theorem [21]. For the SABR β = 0 two adjustments need to be done: ˜y the volatilityσ ˜t is variable and we have ft terms shifting each compounding Gaussian distribution. Yet we can IV. COMPARING TO HAGAN ET AL easily generate a set of volatility trajectories and solve the ASYMPTOTIC SOLUTIONS Bachelier problem with variable volatility. Eventually we are able to implement the semi-analytic Monte-Carlo The MC2 solution allows evaluating the accuracy of calculation. the Hagan et al approximated solution. The control parameter of this approximation is: j=12 SABR 1 y z S˜ = S0 1+ (f˜ + f˜ ) (27) Asian 12 tj tj T j=1 2 n X o =12 =12 ϑ = νt dt (21) j j 0 1 ρ z Z = S0 1+ [˜σ σ0]+ (n j + 1)df˜ 12 ν tj − − tj 2 j=1 j=1 For constant ν we get ϑ = ν T . The solutions of n X X o SABR equations obtained via the two Monte-Carlo meth- (28) ods MC1 and MC2 are compared on FIG.2. As we see, tj ˜z 2 ˜ for ϑ =0.2 the Hagan solution has a good match for re- dftj = 1 ρ σ˜sdWs (29) − tj−1 alistic option prices. For ϑ = 0.7 the mismatch for very p Z out-of-the-money options is significant. From a practical The terms related toσ ˜tj are following via Monte-Carlo point of view, the discrepancies above 10−2 matter as the ˜ ˜z on Vt. Incorporating the terms dftj requires a further price of an option is quoted up to the cents. time discretization of the time increments and decompo- The same conclusion can be drawn when the correla- sition into independent Gaussian variables (see Appendix tion (ρ) is different from 0. for details).

V. PRICING ASIAN OPTION USING SABR VI. EXTENSION TO MEAN REVERTING β = 0 EQUATIONS MODELS

Most of the literature reviewed on Asian options treats The MC2 computation can be eﬃciently applied for the case of the Black-Scholes model. It is a well-known Vanilla pricing in both the Heston/Bates and Sch¨obel- fact that by using this dynamic, we assume a rigid skew Zhu models (hence we can solve both models with time 4

Hagan Hagan 0 100 β =0 MC1 β =1 MC1 10 =0 MC2 =0 MC2 -1 10-1 =02 =02 10 10-2 10-2 10-3 10-3 -4 price 10 price 10-4 10-5 -5 10 -6 10 -6 10 10-7

60 70 80 90 100 110 120 130 140 60 70 80 90 100 110 120 130 140 strike strike

Hagan Hagan 100 0 MC1 MC1 10 -1 10 MC2 MC2 10-1 10-2 10-2 10-3 10-3 -4 10 -4 -5 10 price 10 price 10-5 10-6 β =0 β =1 -6 10-7 =−05 =−05 10 -7 10-8 =02 =02 10 -8 10-9 10

60 70 80 90 100 110 120 130 140 60 70 80 90 100 110 120 130 140 strike strike

Hagan Hagan β =0 MC1 β =1 MC1 =0 MC2 =0 MC2 100 100 =07 =07

10-1 10-1 price price 10-2

10-2 10-3

60 70 80 90 100 110 120 130 140 60 70 80 90 100 110 120 130 140 strike strike

Hagan Hagan MC1 MC1 100 MC2 MC2 100

10-1 10-1 10-2 price price 10-2 10-3 β =0 β =1 =−05 =−05 10-3 10-4 =07 =07

60 70 80 90 100 110 120 130 140 60 70 80 90 100 110 120 130 140 strike strike

FIG. 2. Out-of-the-money Vanilla options obtained by Hagan et al. asymptotic solution (Hagan), Monte-Carlo 1 (MC1) and 2 (MC2). 5 dependent coeﬃcients). As discussed in the Appendix, solutions. Separately, we can price arithmetic Asian op- the option price has the form of a straightforward av- tions on markets with skew and smile using the β = 0 erage over Black-Scholes prices. The various method of SABR. The latter seems to be an advantage over stan- simulating the Heston dynamics is presented in [3]. dard calculations performed within the scope of the To resolve pricing of Asian options, we need to consider Black-Scholes model. a β = 0 generalization of these models. In such a case The MC2 scheme is limited to Vanilla and weakly ex- the stochastic equations are (w ˜t=0 = w0,β = 0): otic options. Its applicability is limited to ρ = 0 for the pricing of option having a barrier condition.

˜ β St 2 dS˜t = S0 w˜t ρdV˜t + 1 ρ dW˜ t + dJ˜t + µdt S0 { − } Appendix A: Numerical MC2 calculations p p (30) dw˜ = κ(w w˜ )dt + ν w˜ dV˜ (31) Vanilla pricing for SABR β = 0 t L − t t t p where J˜t is a compounded Poisson jump process and For the SABR model we can solve explicitly the tra- µ is the correcting drift factor. jectory of the volatility variable functional on the path Another similat model is reltaed to the Sch¨obel-Zhu of the Brownian motion. For the SABR β = 0 we get: model (ζt=0 = σ0,β = 0) 2 dS˜ = S0σ¯ ρdV¯ + 1 ρ dW˜ (A1) t t{ t − t} ˜ β 2 ˜ St ˜ ˜ 2 ˜ ˜ pν dSt = S0 ζt ρdVt + 1 ρ dWt + dJt + µdt σ¯t = σ0 exp[νV¯t t] (A2) S0 { − } − 2 p (32) The solution of Eq. A1 can be partially integrated and dζ˜ = κ(ζ ζ˜ )dt + νdV˜ (33) written as: t L − t t

The models discussed above allow MC2 integration for t ρ 2 Asian option pricing. While technically stochastic inte- S˜t = S0 1+ [¯σt σ0]+ σ0 1 ρ σ¯sdW˜ s (A3) gration is straightforward, the choice and particular ap- ν − − 0 n p Z o plicability of these models to ﬁtting market option prices To calculate the second part we proceed with a time is not clear for the Authors at the moment of writing this discretization tn = n∆t deﬁning the two Brownian mo- document. tions functional on two sets of independent incremental v w uni Gaussian variables (E[¯gj g˜j ]= δjiδvw):

VII. CONCLUSION j=n ¯ ¯ √ v Vtn = Vn = ∆t g¯j (A4) In this paper we have discussed a methodology for pric- j=1 X ing options in frameworks similar to the SABR and He- j=n ston models. These dynamics are characterized by a lin- W˜ = W˜ = √∆t g˜w (A5) ear secondary stochastic driver. This linearity coupled tn n j j=1 with the β parameter allows to switch between a Bache- X v lier or Black-Scholes model. Alternatively, these models We generate a vector of Gaussian random numbersg ¯j can base on a Shifted Log-Normal basis (“Shifted Log- and this gives us a trajectory for the volatility given by Normal Backbone”). Eq. A2. For such a time dependent volatility we can The calculation methodology is based on integrating solve the Bachelier diﬀusion problem as: analytically on a set of stochastic variables while the second set is integrated numerically. This methodology is j=n σ0ρ 2 w known in the literature as Conditional Monte-Carlo while S˜ = S0 1+ [¯σ σ0]+ √∆T 1 ρ g˜ σ¯ n ν n − − j j we call it Semi-Analytic Monte-Carlo or MC2. j=1 n X o The MC2 scheme can be applied to calculate option p (A6) prices for all SABR/Heston models including those with The above can be re-written in the form of a sum of time-dependent coeﬃcients. These solutions are some- independent Gaussian variables: how better organized in the sense that only one variable is simulated and it relies on closed-form formulae = ˜ j n (either Bachelier or Black-Scholes solution depending on Sn √ 2 w 2 =1+ A + ∆T 1 ρ g˜j σ˜j (A7) S0 − β). The semi-analytic procedure has also a better con- j=1 vergence than basic Monte-Carlo schemes. p X ˜ The semi-analytic MC2 calculating Vanilla option Sn =1+ A + √∆TBQ˜W (A8) prices beyond the scope of the Hagan et al. asymptotic S0 6

where ρ S˜ = S0 exp [¯σ σ0] t ν t − ρ n t A = [¯σn σ0] (A9) 1 2 ν − σ¯s ds (A17) − 2 0 j=n Z t 2 2 2 B = 1 ρ σ¯j (A10) + 1 ρ σ¯sdW˜ s − v − 0 uj=1 Z p uX p t o t ρ 1 2 2 = S0 exp [¯σt σ0] ρ σ¯s ds ν − − 2 0 We recovered the solution of the Bachelier process but n Z 1 t with a variable volatility (B term) and a corrected S0 (A 2 2 (1 ρ ) σ¯s ds (A18) term). The above allows calculating the Vanilla price for − 2 − 0 th Zt a selected m trajectory: 2 + 1 ρ σ¯sdW˜ s − 0 p Z o

Num Bachelier We solve the integral parts of the above equation us- Cm (F,K,σ0, ∆T )= CV (F + Am,K,Bm, ∆T ) ing discretization for the two Wiener trajectories. The (A11) discretization must be done avoiding lattice errors which Eventually we average over the M Monte-Carlo sam- checked by getting a Black-Scholes type formula: pled volatility trajectories:

˜ Sn BS 1 2 ln = A ∆TB + √∆TBQ˜W (A19) SABR β=0 S0 − 2 CV (F,K,σ0, ∆T )= h i m=M 1 where CBachelier(F + A ,K,B , ∆T ) (A12) M V m m m=1 X j=n BS ρ 1 2 2 A = [¯σ σ0] ρ σ¯ √∆t (A20) ν n − − 2 j The price of a Vanilla call under the Bachelier model j=1 is as follows: X j=n B = 1 ρ2 σ¯2 (A21) − v j uj=1 p uX 2 B F σ0√∆T − 1 d t C (F,K,σ0, ∆T ) = (F K) [d0]+ e 2 0 V − N √2π The Black-Scholes price for a single volatility trajec- (A13) tory is: K F d0 = − (A14) F σ0√∆T Num BS BS Cm (F,K,σ0, ∆T )= CV (F + Am ,K,Bm, ∆T ) (A22) And the SABR price avergaed over the volatility trajectories is: Vanilla pricing for SABR β = 1

SABR β=1 C (F,K,σ0, ∆T )= For the SABR β = 1 we get: V m=M 1 (A23) CBS(F + ABS,K,B , ∆T ) M V m m m=1 X 2 dS˜t = S˜tσ¯t ρdV¯t + 1 ρ dW˜ t (A15) { 2 − } pν σ¯t = σ0 exp[νV¯t t] (A16) − 2 Asian options for SABR β = 0

We proceed along similar lines and solve partially Eq. We select a trajectory for the volatility variable and A15: follow on the result given by Eq. 23: 7

100 MC1 MC1 MC2 MC2 10-1 N−0.5 10-2

10-3 2.5

speed ratio MC2/MC1 10-4 Std. error Std. 2.0 10-5 -MC )/MC 1,2 2,33m 2,33m -6 10 1.5 102 103 104 105 106 107 108 1 -7 10 1 1 1 1 1 1 1 102 103 104 105 106 107 108 N

FIG. 3. Relative error of the Monte-Carlo methods compared FIG. 4. Comparison of the convergence between Monte-Carlo 25 to MC2 with 33 (2 ) millions trajectories for an out-of-the- 1 (MC1) and 2 (MC2) for Asian option based on SABR β = 0 money Vanilla option (K=80%). The MC2 error is more sta- dynamic. Inset: MC2 is about twice as fast as MC1. ble compared to MC1, even for small number of simulations.

SABR β=0 CAsian (F,K,σ0, ∆T )= j=12 ˜SABR = SA 1 ¯y m M (A29) =1+ ft 1 Bachelier ′ ′ S0 12 j C (F + A ,K,B , ∆T ) j=1 M V m m X (A24) m=1 j=12 X 1 + 1 ρ2√∆T (n j + 1)df˜z 12 − − tj j=1 p X We calculate the shift term:

j=12 ′ 1 ρ A = [¯σ σ0] (A25) 12 ν tj − j=1 X ˜z And decompose fti into elementary variables:

i=Nj i=Nj−1 df˜z = g˜wσ¯ g˜wσ¯ (A26) tj i i − i i i=1 i=1 X X Please note that this decomposition accounts for the dynamic aspect of volatility. Eventually we calculate:

j=12 Nj Nj−1 ′ B = 1 ρ2 σ¯2 σ¯2 (n j + 1)2 − v i − i − u j=1 i=1 i=1 p u X X X t (A27)

˜SABR SA ′ ′ =1+ A + √∆TB Q˜W (A28) S0

This means that the price for an Asian option has the form: 8

100 MC1 MC1 β =0 MC2 β =0 0 -1 MC2 10 10 =0 =0 10-2 =02 =07 -1 10-3 10 10-4 10-2 10-5 -6 price 10 price 10-3 10-7 -8 10 10-4 10-9 -10 10 10-5 10-11 60 70 80 90 100 110 120 130 140 60 70 80 90 100 110 120 130 140 strike strike

MC1 MC1 -1 0 10 MC2 MC2 10 -3 10 10-1 10-5 10-2 10-7 10-3 10-9 -4 price 10 price 10-11 β =0 β =0 10-5 10-13 =−05 =−05 -6 10-15 =02 =07 10 -7 10-17 10

60 70 80 90 100 110 120 130 140 60 70 80 90 100 110 120 130 140 strike strike

FIG. 5. Out-of-the money arithmetic Asian options obtained by Monte-Carlo 1 (MC1) and 2 (MC2) based on SABR β = 0 dynamic. The two methods have a good match. 9

[1] An extensive review of the pricing od Asian options is [19] M. Milevsky and S. Posne. Asian options, the sum of log- given in [18]. normals and the reciprocal gamma distribution. Journal [2] The sum of correlated lognormals is also of important of Financial and Quantitative Analysis, 33/3, 1998. topics in other fields such as telecommunication. [20] J. Obloj. Fine-tune your smile, correction to hagan et al. [3] L. Andersen. Efficient simulation of the heston stochas- Wilmott Magazine 35, 2008. tic volatility model. http://ssrn.com/abstract=946405, [21] A. Owen. Monte Carlo theory, methods and examples. 2007. 2013. [4] D. Bates. Jumps and stochastic volatility: Exchange rate [22] P. Praetz. The distribution of share price changes. Jour- processes implicit in deutsche mark options. Review of nal of Business 49, 1972. Financial Studies, 1996. [23] J. Printems. Pricing asian options in the heston model. [5] Ph Bougerol. Exemples de thormes locaux sur les groupes http:www.quantize.maths-fi.comheston asian, 2014. rsolubles. Ann. Inst. H. Poincaré, 1983. [24] R. Schöbel and Zhu J. Stochastic volatility with an [6] D. Dufresne. Algebraic properties of beta and gamma dis- ornstein-uhlenbeck process: An extension. European Fi- tributions, and applications. Advances in Applied Math- nance Review, 1999. ematics, 1997. [25] L. Scott. Option pricing when the variance changes ran- [7] D. Dufresne. Laguerre series for asian and other options. domly: Theory, estimation, and an application. The Mathematical Finance 10, 2000. Journal of Financial and Quantitative Analysis 22, 1987. [8] F. Dufresne. The integral of geometric brownian motion. [26] E. M. Stein, , and J. C. Stein. Stock price distributions Advances in Applied Probability, 33, 2001. with stochastic volatility: An analytic approach. Review [9] Jondeau E., S.H. Poon, and M. Rockinger. Financial of Financial Studies 4, 727-752, 1991. Modeling Under Non-Gaussian Distributions. 2007. [27] C. Tsallis. Possible generalization of boltzmann-gibbs [10] P. Hagan, D. Kumar, and A. Lesniewski. Managing smile statistics. Journal of Statistical Physics 52, 1988. risk. Wilmott Magazine, 2002. [28] C. Tsallis, C. Anteneodo, L. Borland, and R. Osorio. Non [11] P. Henry-Labordère. Encyclopedia of quantitative fi- extensive statistical mechanics and economics. Physica nance - SABR model. John Wiley & Sons Ltd, 2010. A, 2003. [12] S. Heston. A closed-form solution for options with [29] S. Turnbull and Wakeman L. A quick algorithm for pric- stochastic volatility with applications to bond and cur- ing european average options. Journal of Financial and rency options. The Review of Financial Studies 6, 1993. Quantitative Analysis 26, 1991. [13] H. Kleinert. Path Integrals - Path Integrals and Financial [30] S. Vakeroudis. Bougerol’s identity in law and extensions. Markets. 2009. arXiv:1201.5823, 2012. [14] R. Korn and S. Tan. Exact analytical solution for the [31] G. A. Willard. Calculating prices and sensitivities for normal SABR model. Wilmott Magazine, 2014. pth-independent derivative securities in multifactor mod- [15] J. Kuklinski, D. Negru, and Pliszka P. Modelling the els. The Journal of Derivatives, Vol 5, 1997. skew and smile of spx and dax index options using [32] M. Yor. On some exponential functionals of brownian the shifted log-normal and SABR stochastic models. motion. Advances in Applied Probability 24, 1992. arXiv:1404.4659, 2014. [33] M. Yor. Sur les lois des fonctionnelles exponentielles [16] E. Levy. Pricing european average rate currency options. du mouvement brownien, considérées en certains instants Journal of International Money and Finance, 1992. aléatoires. Comptes Rendus de l’Académie des Sciences [17] R. Lord. Fifty shades of SABR simulation. In Bachelier - Series I -Mathematics, 1992. th Finance Society, 8 World Congress. [34] M. Yor. Bessel processes, asian options, and perpetuities. [18] R. Lord. Partially exact and boundedapproximations for Exponential Functionals of Brownian Motion and Related arithmetic asian options. Efficient pricing algorithms for Processes Springer Finance, 2001. exotic derivatives, Tinbergen Institute Research Series, 2008.