parametric option pricing with chebyshev ... - acmac.uoc… · parametric option pricing = pop we...
TRANSCRIPT
Parametric Option Pricingwith Chebyshev Interpolation
Kathrin GlauTechnische Universitat Munchen
joint work with
Maximilian Gaß, Mirco Mahlstedt and Maximilian Mair
Stochastic Methods in Physics and Finance:July 24, 2015, Crete
Parametric Option Pricing = POP
Day-to-day tasks in the financial industry:
calibration, risk assessment, quotation
Respecting credit and liquidity risk results in nonlinear pricingequations, BSDEs and PIDEs. However, today...
...”it’s impossible to implement on large portfolios in an efficient way”...
D. Brigo, Conference Munich 2015
We need reliable complexity reduction techniques.
2
Parametric Option Pricing = POP
We exploit the recurrent nature of the option pricing problem inan efficient, reliable and general way.
Fast Fourier Transformfor plain vanilla prices — for varying strikes K .
Carr, Madan (1999), Raible (2000),Lee (2004), Lord, Fang, Bervoets and Oosterlee (2008),Feng and Linetsky (2008), Kudryavtsev and Levendorskiy (2009),Boyarchenko and Levendorskiy (2014),. . .
3
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 1
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
-0.02
0
0.02
Figure : Call prices in the BS model4
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 2
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
-0.01
0
0.01
Figure : Call prices in the BS model5
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 3
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-3
-2
0
2
Figure : Call prices in the BS model6
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 4
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-3
-2
0
2
Figure : Call prices in the BS model7
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 5
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-3
-2
0
2
Figure : Call prices in the BS model8
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 5
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-3
-2
0
2
Figure : Call prices in the BS model9
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 6
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-3
-1
0
1
Figure : Call prices in the BS model10
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 7
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-4
-5
0
5
Figure : Call prices in the BS model11
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 8
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-4
-2
0
2
Figure : Call prices in the BS model12
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 9
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-4
-2
0
2
Figure : Call prices in the BS model13
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 10
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-4
-2
0
2
Figure : Call prices in the BS model14
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 11
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-4
-1
0
1
Figure : Call prices in the BS model15
Parametric Option Pricing = POP
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
pri
ce
0
0.2
0.4
0.6N = 12
BS price
Cheby approx
BS σ0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
∆ p
rice
×10-5
-5
0
5
Figure : Call prices in the BS model16
Parametric Option Pricing = POP
Crucial observations:
• The same pricing problem needs to be solved for a large set ofdifferent parameter values: POP Parametric Option Pricing.
• Model prices typically are continuous or even smooth in theparameter values.
We exploit the functional dependence
x 7→ Pricex , x = parameters
and approximate this function by interpolation.
Related literature: Interpolation of Black Scholes’s call prices inthe volatility, Pistorius, Stolte (2012)
17
Parametric Option Pricing = POP
0.5
maturity T
1
1.5
2108
6
strike K
42
0
2
3
4
5
0
1
Price
T,K
Figure : Call prices in the CGMY model18
Chebyshev Interpolation
19
Chebyshev Interpolation Points
-1 -0.5 0 0.5 1
0
0.2
0.4
0.6
0.8
1
Chebyshev nodes for D=1
20
Chebyshev Polynomial Interpolation for POP
We approximate
K 7→ PriceK = E[f K (X )
]by Chebyshev polynomial interpolation for K = [−1, 1]
PriceK ≈ 2∑
0≤j≤N
′cj(Price)Tj(K ),
with
nodes Kk := cos(π 2k+1
2N+2
)Chebyshev polynomials Tj(K ) := cos
(j arccos(K )
)coeffients cj(Price) :=
∑′0≤k≤N PriceKk cos
(jπ 2k+1
2N+2
).
∑′ indicates that the first summand is multiplied with 1/2.
21
Accuracy of Chebyshev Interpolation
By implementing Chebyshev interpolation Platte & Trefethen (2008) aim
. . . “to combine the feel of symbolics with the speed of numerics.”
<
i<
−1 1
ab
a + b = ρ
Theorem (Bernstein (1912), Trefethen (2013))
If g : [−1, 1]→ R is analytic and bounded in the ellipse with foci ±1 andsemiaxis lengths summing to %, then for IN(g)(x) :=
∑′0≤j≤N cj(g)Tj(x),
‖g − IN(g)‖∞ ≤ C%−N
%− 1.
22
Tensorized Chebyshev Interpolation
pk1-1 -0.5 0 0.5 1
pk2
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Chebyshev nodes for D=2
Figure : A set of d-variate Chebyshev points x ∈ [−1, 1]d for d = 2 andN = 20 as defined in (??). Comparing the arrangement of points xi ford = 2 to the arrangement for d = 1 as shown in Figure ??, thetensorizing structure is clear to see.
The basis functions Tj , j ∈ Nd0 , possess a tensorizing structure, and
can thus be expressed by their univariate siblings defined in (??),
T(j1,...,jd )(x1, . . . , xd) =d∏
i=0
Tji (xi ). (1)
23
Tensorized Chebyshev Interpolation
We approximate
x = (K ,T , p) 7→ PriceK ,T ,p := E[f K (X p
T )]
by tensorized Chebyshev polynomial interpolation for x ∈ [−1, 1]D
IN1...ND(Price)(x) :=
N1∑j1=0
. . .
ND∑jD=0
c(j1,...,jD)
D∏i=1
Tji (xi ),
with coefficients
c(j1,...,jD ) =( D∏
i=1
21{ji>0}
(Ni + 1)
) N1∑k1=0
. . .
ND∑kD=0
Pricex(k1,...,kD )
D∏i=1
cos
(jiπ
2ki + 1
2N + 2
).
24
Tensorized Chebyshev Interpolation
For % = (%1, . . . , %D) ∈ (1,∞)D let the tensorized Bernstein ellipse
B([−1, 1]D , %) = B([−1, 1]D , %1)× . . .× B([−1, 1]D , %D).
Theorem (Gaß, G., Mahlstedt, Mair (2015))
If g has an analytic and bounded extension to B([−1, 1]D , %) forsome parameter vector % ∈ (1,∞)D , then∥∥g − IN(g)
∥∥∞ ≤ C%−N ,
where % = min1≤i≤D
%i and N = min1≤i≤D
Ni .
Proof by a generalization of the univariate case by induction over D.
25
Tensorized Chebyshev Interpolation
The error decay is of order O(%M
1/D)in the number M of degrees
of freedom.
The result naturally extends to functions
f : P → R
with hyperrectangular domain P. We call B(P, %) theappropriately transformed tensorized Bernstein ellipse.
26
Application to POP
27
Chebyshev Interpolation Applied to POP
Precomputation phase:
Determine the approximate option prices
Pricex=(K ,T ,p)
at the Chebyshev nodes xk1,...,kD = (x1k1, . . . , xDkD ) e.g. with
• Fourier techniques• Partial (integral) differential equation methods• Monte Carlo
Evaluation phase:
Evaluate the approximate polynomial
IN1...ND(Price)(x) :=
N1∑j1=0
. . .
ND∑jD=0
c(j1,...,jD)
D∏i=1
Tji (xi ),
with explicit coefficients.28
Chebyshev Interpolation Applied to POP
0.5
maturity T
1
1.5
2108
6
strike K
42
0
2
3
4
5
0
1
Price
T,K
Figure : Call prices in a CGMY model, N1 = N2 = 15 Chebyshev nodes29
Chebyshev Interpolation Applied to POP
0.5
maturity T
1
1.5
2108
6
strike K
42
0
0
2
-4
-2
4
×10-5
∆ P
rice
T,K
Figure : Error plot for N1 = N2 = 15, call prices in a CGMY model30
Chebyshev Interpolation Applied to POP
strike K
1 2 3 4 5 6 7 8 9 10
Price
K
0
1
2
3
4
5
Option price
Cheby price (N=20)
strike K
1 2 3 4 5 6 7 8 9 10
∆ P
rice
K
×10-6
-5
0
5
Abs. error
Figure : Price and error plot, N=20, call prices in a CGMY model.31
Conditions on Options and Models for
(Sub)Exponential Convergence
32
Convergence of Chebyshev Interpolation for POP
Task: Determine accessible sufficient conditions for
p = (K ,T , π) 7→ PriceK ,T ,π := E[f K (Xπ
T )]
for p = (K ,T , π) ∈ P to have an analytic bounded extension tosome generalized Bernstein ellipse.
Observe in PriceK ,T ,p = E[f K (X p
T )]
typically
K 7→ f K (X pT ) e.g. K 7→ (X p
T − K )+
is not even differentiable.
33
Convergence of Chebyshev Interpolation for POP
Our approach: Fourier representation of option prices,
PriceK ,T ,p =1
(2π)d
∫Rd+iη
f K (−z)ϕT ,p(z) dz
with the characteristic functions f K of f K and ϕT ,p of X pT .
34
Conditions
Let the parameter set P = P1 × P2 ⊂ RD be a hyperrectangle.Let % ∈ (1,∞)D with %1 := (%1, . . . , %m) and %2 := (%m+1, . . . , %D)and let weight η ∈ Rd .
(A1) Integrability of x 7→ e〈η,x〉 f K (x) for all K ∈ P1.
(A2) Analyticity of K 7→ f K (z − iη) in B(P1, %1) for all z ∈ Rd .
(A3) Exponential moment E(
e−〈η,XpT 〉)<∞ for all (T , p) ∈ P2.
(A4) Analyticity of (T , p) 7→ ϕT ,p(z + iη) in B(P2, %2) for all z ∈ Rd .
(A5) Uniform bound: There exists h ∈ L1(Rd) such that
max(K ,T ,p)∈B(P,%)
∣∣f K (−z − iη)ϕT ,p(z + iη)∣∣ ≤ h(z) for all z ∈ Rd .
35
Convergence of Chebyshev Interpolation for POP
PriceK ,T ,p = E[f K (X p
T )]
Proposition (Eberlein, G., Papapantoleon (2010))
Assume (A1), (A3) and (A5). Then each PriceK ,T ,p = E[f K (X p
T )]
has the Fourier representation
Price(K ,T ,p) =1
(2π)d
∫Rd+iη
f K (−z)ϕT ,p(z) dz .
36
Convergence of Chebyshev Interpolation for POP
Theorem (Gaß, G., Mahlstedt, Mair (2015))
Let % ∈ (1,∞)D and weight η ∈ Rd . Conditions (A1)–(A5) imply
max(K ,T ,p)∈P
|Price(K ,T ,p) − IN(Price(·))(K ,T , p)| ≤ C%−N ,
where % = min1≤i≤D
%i and N = min1≤i≤D
Ni .
37
Examples investigated in more detail
Models
• Black Scholes, Merton, NIG, CGMY,. . .
• time-inhomogeneous Levy
• Heston, affine (jump) models
Payoff profiles
• European: call, put, digital down&out, power,. . .
• Baskets: call on basket, call on minimum,. . .
38
Detailed analysis in Levy and affine models
39
European options in Levy models
Spt = S0 eL
pt with Lp a Levy process and special semimartingale for
varying parameters p = (b, σ)
E(
eizLpt)
= etψp(z),
ψp(z) =σ2z
2+ ibz +
∫R
(eizx −1− izx
)F (dx)︸ ︷︷ ︸
=:ψ(z)
.
Martingale condition
b = b(r , σ) = r − σ2
2−∫R
(ex −1− x
)F (dx).
Fair value of a European option
Price(r ,K ,S0,T ,p) = e−rT E(f K (S0 eL
pT )).
40
European options in Levy models
Let P = [K ,K ]× [T ,T ]× [b, b]× [σ, σ] with T > 0.
Corollary (Gaß, G., Mahlstedt, Mair (2015))
Let η ∈ R, ρ1 > 1. Assume (A2), that∫|x|>1
(e−ηx ∨ ex)F (dx) <∞ and
(A1’) supK∈B([K ,K ],%1)
∥∥ e〈·,η〉 f K∥∥L1(Rd )
<∞ and∣∣∣f p1 (−z − iη)
∣∣∣ ≤ c1ec2|z|
for some c1, c2 > 0.
If additionally
(i) σ > 0, or
(ii) there exist α ∈ (1, 2] and C1,C2 > 0 such that
<(ψ)(z + iη) ≤ C1 − C2|z |α for all z ∈ R,
then there exist constants C > 0 and % > 1 such that for N = min1≤i≤4
Ni ,
max(K ,T ,b,σ)∈P
∣∣Price(K ,T ,b,σ) − IN(Price(·))(K ,T , b, σ)∣∣ ≤ C%−N .
41
Basket Options in Affine Models
Parametrized affine process X π′with state space D ⊂ Rd , π′ ∈ Π′
and with C-valued ϕπ′
and Cd -valued φπ′
solutions of generalizedRicatti equations satisfying
ϕ(t,x ,π′)(z) = E(
ei〈z,Xπ′t 〉∣∣Xπ′
0 = x)
= eϕπ′ (t,iz)+〈φπ′ (t,iz),x〉, (2)
for every t ≥ 0, z ∈ Rd and x ∈ D.
Fair value of a European option
Price(K ,T ,x ,π′) = E(f K (X π′
T )|Xπ′0 = x
)(3)
where f K is a parametrized family of measurable payoff functionsf K : Rd → R+ for K ∈ P1.
42
Basket Options in Affine Models
Corollary (Gaß, G., Mahlstedt, Mair (2015))
Assume (A1’), (A2), (A3’) for η ∈ Rd and hyperrectangular P ⊂ RD. Let(i) for every p2 = (t, x , π′) ∈ P2 ⊂ RD−m and every z ∈ R+ iη,
ϕp2=(t,x,π′)(z) = E(
ei〈z,Xπ′t 〉∣∣Xπ′
0 = x)
= eφπ′ (t,iz)+〈ψπ
′(t,iz),x〉,
(ii) (t, π′) 7→(φπ′(t, iz − η), ψπ
′(t, iz − η)
)has an analytic extension to
a Bernstein ellipse B(Π′, %′) with %′ ∈ (1,∞)D−m for every z ∈ Rd ,
(iii) there exist α ∈ (0, 2] and C1,C2 > 0 such that for everyp2 = (t, x , π′) ∈ P2 and every z ∈ R,
<(φπ′(t, iz) + 〈ψπ
′(t, iz), x〉
)(z + iη) ≤ C1 − C2|z |α.
Then there exist constants C , % > 0 such that
maxp∈P1×P2
∣∣Pricep − IN(Price(·))(p)∣∣ ≤ C%−min1≤i≤D Ni .
The existence of exponential moments of affine processes is investigatedin [?] where moreover criteria are provided under which formula (2) andthe related generalized Riccati system can be extended to complexexponential moments z ∈ Cd . The question has already been treated forimportant special cases where more explicit conditions are available [?]treat affine diffusions and [?] investigate affine processes with killingwhen the jump measures possess exponential moments of all orders.
43
Numerical Results
44
Numerical Experiments: European Call
Fixed strike, K = 1
p2 K ∈ T ∈
BS σ = 0.2 [1, 10] [0.5, 2]
Merton σ = 0.15, [1, 10] [0.5, 2]α = −0.04,β = 0.02,λ = 3
CGMY C = 0.6, [1, 10] [0.5, 2]G = 10,M = 28,Y = 1.1
Heston T = 2, [1, 10] v0 ∈ [0.12, 0.42]κ = 1.5,θ = 0.22,σ = 0.25,ρ = 0.1
45
Numerical Experiments: European Call
2
T
1.5
Chebyshev price error, BS
10.50
5K
-1
0
1
10
×10-4
∆Price
2
T
1.5
Chebyshev price error, Merton
10.50
5K
-5
0
5
10
×10-4
∆Price
2
T
1.5
Chebyshev price error, CGMY
10.50
5K
1
-1
0
10
×10-6
∆Price
10
K
Chebyshev price error, Heston
50
0.050.1
v0
0.15-5
0
5
×10-5
∆Price
Figure : Absolute pricing error. We compare the Chebyshev interpolationwith N1 = N2 = 20 to classic Fourier pricing by numerical integration.
46
Numerical Experiments: European Call
Chebyshev N
0 5 10 15 20 25 30 35 40 45 50
error
10-14
10-12
10-10
10-8
10-6
10-4
10-2
100
Chebyshev error decay
εL∞ (BS)
εL2 (BS)
εL∞ (Merton)
εL2 (Merton)
εL∞ (CGMY)
εL2 (CGMY)
εL∞ (Heston)
εL2 (Heston)
Figure : Convergence study for the BS, Merton, CGMY, Heston model
47
Numerical Experiments: Digital Down&Out
Chebyshev N
0 5 10 15 20 25 30 35 40 45 50
error
10-14
10-12
10-10
10-8
10-6
10-4
10-2
100
Chebyshev error decay
εL∞ (BS)
εL2 (BS)
εL∞ (Merton)
εL2 (Merton)
εL∞ (CGMY)
εL2 (CGMY)
εL∞ (Heston)
εL2 (Heston)
Figure : Convergence study for prices for the BS, Merton, CGMY, Heston
48
Numerical Experiments:Basket, Lookback, Barrier
Basket option for d underlyings
f (S1(T ), . . . ,Sd(T )) =
(( 1
d
d∑j=1
Sj(T ))− K
)+
.
Lookback option for d underlyings, S j(T ) := maxt≤T Sj(t),
f(S(t)0≤t≤T
)=
(( 1
d
d∑j=1
S j(T ))− K
)+
.
Barrier option for d underlyings, down&out, S j(T ) := mint≤T
Sj(t),
f(S(t)0≤t≤T
)=
(( 1
d
d∑j=1
S j(T ))− K
)+
· 1{S j (T )≥80, j=1,...,d}.
49
Numerical Experiments:Basket, Lookback, Barrier, American Put
r = 0, Free parameters K ∈ [83.33, 125] and T ∈ [0.5, 2]
Model fixed parameters:p1 p2
BS Sj,0 = 100, σj = 0.2
Heston Sj,0 = 100, κj = 2,θj = 0.22,σj = 0.3,ρj = −0.5,vj,0 = 0.22
Merton Sj,0 = 100, σj = 0.2,αj = −0.1,βj = 0.45,λj = 0.1
Monte Carlo simulation of 1 Mio sample paths, antithetic variates, 400time steps per year. 95% confidence bound for a price level of 10.
50
Numerical Experiments:Basket, Lookback, Barrier, American Put
N = 30 Chebyshev nodes in each varying parameter
Model Option d εL∞ MC price CB price
BS Basket 5 2.0 · 10−4 10.2988 10.2986Heston 1.4 · 10−3 2.8189 2.8203Merton 2.4 · 10−3 4.734 4.7364
BS Lookback 5 5.3 · 10−3 18.8028 18.7975Heston 3.1 · 10−3 24.1618 24.1587Merton 5 · 10−3 21.275 21.28
BS Barrier 5 7.4 · 10−3 10.0076 10.0149Heston 4.5 · 10−3 5.4898 5.4942Merton 7.0 · 10−3 9.2406 9.2476
BS American 1 8.6 · 10−3 9.5515 9.5602
51
Numerical Experiments: Digital Down&Out
2
T
1.5
Chebyshev price error, BS
10.50
5K
-1
1
0
10
×10-3
∆Price
2
T
1.5
Chebyshev price error, Merton
10.50
5K
2
-2
0
10
×10-3
∆Price
2
T
1.5
Chebyshev price error, CGMY
10.50
5K
0
2
-210
×10-6
∆Price
10
K
Chebyshev price error, Heston
50
0.050.1
v0
0.15-5
0
5
×10-4
∆Price
Figure : Absolute pricing error for a European digital down&out option
52
Applications and further developments
53
Benefits
Accurate Acceleration in models with low number of parameter
• PricingPrecomputation of prices once with a fine resolution, thenalways a fast realization available.
• CalibrationInterpolation of the objective function for fast optimization.
• Hedging, risk assessment, parameter uncertainty,...
Gain of freedom in modeling with a low number of parameters:
From ”For liquid options a fast pricer (Fourier-representation) isnecessary.”To ”An accurate pricer is available and the parameterdependence is analytic.”Thus one may include more realistic features, e.g. dividends.
54
Contribution
• Propose Chebyshev interoplation for parametric option pricing
• Show that the method is applicable for a variety of modelsand option types.(a) Theoretically:
- Exponential convergence for analytic parameter dependence.- Explicit conditions on the Fourier transform of payoff and
distribution.- Examples
(b) Numerically:
- Experimental order of convergence for Call and digital in BS.Heston, Merton, CGMY model. (CI combined with Fouriermethod)
- (Path-dependent) multivariate options: CI combined withMonte-Carlo and Finite Differences.
55
Outlook
56
Limitations of Chebyshev Interpolation
Methodological constraints
• Curse of dimensionality: degree of freedom of order ND forD-dimensional parameter space
• Requirement: Tensor structure of parameter space!
Realistic requirements
• Dimension of the parameter space is typically large.
• Parameter constraints often do not result inhyperrectangulars, e.g. entries covariance matrices.
How to find a generic interpolation that is flexible in theparameter space?
57
Literature
S. Bernstein. Sur l’ordre de la meilleure approximation des fonctionscontinues par des polynomes de degre donne. Acad. Royale de Belgique.Classe d. sc. Memoires. Coll. in-4”. Hayez, imprimeur des academiesroyales, 1912.
S. Boyarchenko and S. Levendorskiy. Efficient variations of the Fouriertrans- form in applications to option pricing. Journal of ComputationalFinance, 18(2), 2014.
P. Carr and D. B. Madan. Option valuation and the fast Fouriertransform. Journal of Computional Finance, 2(4): 61-73, 1999.
E. Eberlein, K. Glau, and A. Papapantoleon. Analysis of Fouriertransform valuation formulas and applications. Applied MathematicalFinance, 17(3): 211-240, 2010.
L. Feng and V. Linetsky. Pricing Discretely Monitored Barrier Optionsand Defaultable Bonds in Levy Process Models: A Fast Hilbert TransformApproach. Mathematical Finance, 18(3): 337-384, 2008.
M. Gaß, K. Glau, M. Mahlstedt, M. Mair. Chebyshev Interpolation forParametric Option Pricing. Preprint on Arxiv, 2015
58
Literature
M. Pistorius and J. Stolte, Fast computation of vanilla prices intime-changed models and implied volatilities, International Journal ofTheoretical and Applied Finance, 15, 1250031-1250031, 2012
O. Kudryavtsev and S. Z. Levendorskiy. Pricing of first touch digitalsunder normal inverse Gaussian processes. Finance and Stochastics, 13(4):531-562, 2009.
R. W. Lee. Option Pricing by Transform Methods: Extensions,Unification, and Error Control. Journal of Computational Finance, 7(3):51-86, 2004.
R. Lord, F. Fang, F. Bervoets, and C. W. Oosterlee. A Fast and AccurateFFT-Based Method for Pricing Early-Exercise Options under LevyProcesses. SIAM Journal on Scientific Computing, 30(4): 1678-1705,2008.
R. B. Platte and N. L. Trefethen. Chebfun: A New Kind of NumericalComputing, pages 69-86. Springer, 2008.
S. Raible. Levy Processes in Finance: Theory, Numerics, and EmpiricalFacts. PhD thesis, Universitat Freiburg, 2000.
L. N. Trefethen. Approximation Theory and Approximation Practice.SIAM books, 2013.
59