Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles...
Transcript of Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles...
![Page 1: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/1.jpg)
Multilevel Monte Carlopath simulation
Mike Giles
Oxford University Mathematical Institute
Oxford-Man Institute of Quantitative Finance
Acknowledgments: research funding from Microsoft and EPSRC, and
collaboration with Paul Glasserman (Columbia) and Ian Sloan, Frances Kuo (UNSW)
Multilevel Monte Carlo – p. 1/41
![Page 2: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/2.jpg)
Outline
Long-term objective is faster Monte Carlo simulation of pathdependent options to estimate values and Greeks.
Several ingredients, not yet all combined:
multilevel method
quasi-Monte Carlo
adjoint pathwise Greeks
parallel computing on NVIDIA graphics cards
Emphasis in this presentation is on multilevel method
Multilevel Monte Carlo – p. 2/41
![Page 3: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/3.jpg)
Generic Problem
Stochastic differential equation with general drift andvolatility terms:
dS(t) = a(S, t) dt + b(S, t) dW (t)
We want to compute the expected value of an optiondependent on S(t). In the simplest case of Europeanoptions, it is a function of the terminal state
P = f(S(T ))
with a uniform Lipschitz bound,
|f(U) − f(V )| ≤ c ‖U − V ‖ , ∀ U, V.
Multilevel Monte Carlo – p. 3/41
![Page 4: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/4.jpg)
Simplest MC Approach
Euler discretisation with timestep h:
Sn+1 = Sn + a(Sn, tn)h + b(Sn, tn) ∆Wn
Estimator for expected payoff is an average of Nindependent path simulations:
Y = N−1N∑
i=1
f(S(i)T/h
)
weak convergence – O(h) error in expected payoff
strong convergence – O(h1/2) error in individual path
Multilevel Monte Carlo – p. 4/41
![Page 5: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/5.jpg)
Simplest MC Approach
Mean Square Error is O(N−1 + h2
)
first term comes from variance of estimator
second term comes from bias due to weak convergence
To make this O(ε2) requires
N = O(ε−2), h = O(ε) =⇒ cost = O(N h−1) = O(ε−3)
Aim is to improve this cost to O(ε−p), with p as small as
possible, ideally close to 1.
Note: for a relative error of ε = 0.001, the difference betweenε−3 and ε−1 is huge.
Multilevel Monte Carlo – p. 5/41
![Page 6: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/6.jpg)
Standard MC Improvements
variance reduction techniques (e.g. control variates,stratified sampling) improve the constant factor in frontof ε−3, sometimes spectacularly
improved second order weak convergence (e.g. throughRichardson extrapolation) leads to h = O(
√ε), giving
p=2.5
quasi-Monte Carlo reduces the number of samplesrequired, at best leading to N ≈O(ε−1), giving p≈2 withfirst order weak methods
Multilevel method gives p=2 without QMC, and at bestp ≈ 1 with QMC.
Multilevel Monte Carlo – p. 6/41
![Page 7: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/7.jpg)
Other Related Research
In Dec. 2005, Ahmed Kebaier published an article inAnnals of Applied Probability describing a two-level methodwhich reduces the cost to O
(ε−2.5
).
Also in Dec. 2005, Adam Speight wrote a workingpaper describing a very similar multilevel use of controlvariates.
There are also close similarities to a multileveltechnique developed by Stefan Heinrich for parametricintegration (Journal of Complexity, 1998)
Multilevel Monte Carlo – p. 7/41
![Page 8: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/8.jpg)
Multilevel MC Approach
Consider multiple sets of simulations with differenttimesteps hl = 2−l T, l = 0, 1, . . . , L, and payoff Pl
E[PL] = E[P0] +L∑
l=1
E[Pl−Pl−1]
Expected value is same – aim is to reduce variance ofestimator for a fixed computational cost.
Key point: approximate E[Pl−Pl−1] using Nl simulationswith Pl and Pl−1 obtained using same Brownian path.
Yl = N−1l
Nl∑
i=1
(P
(i)l −P
(i)l−1
)
Multilevel Monte Carlo – p. 8/41
![Page 9: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/9.jpg)
Multilevel MC Approach
Discrete Brownian path at different levels
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−1
−0.5
0
0.5
1
1.5
2
2.5
3
3.5
P7
P6
P5
P4
P3
P2
P1
P0
Multilevel Monte Carlo – p. 9/41
![Page 10: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/10.jpg)
Multilevel MC Approach
Using independent paths for each level, the variance of thecombined estimator is
V
[L∑
l=0
Yl
]=
L∑
l=0
N−1l Vl, Vl ≡ V[Pl−Pl−1],
and the computational cost is proportional toL∑
l=0
Nl h−1l .
Hence, the variance is minimised for a fixed computationalcost by choosing Nl to be proportional to
√Vl hl.
The constant of proportionality can be chosen so that thecombined variance is O(ε2).
Multilevel Monte Carlo – p. 10/41
![Page 11: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/11.jpg)
Multilevel MC Approach
For the Euler discretisation and a Lipschitz payoff function
V[Pl−P ] = O(hl) =⇒ V[Pl−Pl−1] = O(hl)
and the optimal Nl is asymptotically proportional to hl.
To make the combined variance O(ε2) requires
Nl = O(ε−2Lhl).
To make the bias O(ε) requires
L = log2 ε−1 + O(1) =⇒ hL = O(ε).
Hence, we obtain an O(ε2) MSE for a computational costwhich is O(ε−2L2) = O(ε−2(log ε)2).
Multilevel Monte Carlo – p. 11/41
![Page 12: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/12.jpg)
Multilevel MC Approach
Theorem: Let P be a functional of the solution of a stochastic o.d.e.,
and Pl the discrete approximation using a timestep hl = M−l T .
If there exist independent estimators Yl based on Nl Monte Carlo
samples, and positive constants α≥ 12 , β, c1, c2, c3 such that
i) E[Pl − P ] ≤ c1 hαl
ii) E[Yl] =
E[P0], l = 0
E[Pl − Pl−1], l > 0
iii) V[Yl] ≤ c2 N−1l hβ
l
iv) Cl, the computational complexity of Yl, is bounded by
Cl ≤ c3 Nl h−1l
Multilevel Monte Carlo – p. 12/41
![Page 13: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/13.jpg)
Multilevel MC Approach
then there exists a positive constant c4 such that for any ε<e−1 thereare values L and Nl for which the multi-level estimator
Y =L∑
l=0
Yl,
has Mean Square Error MSE ≡ E
[(Y − E[P ]
)2]
< ε2
with a computational complexity C with bound
C ≤
c4 ε−2, β > 1,
c4 ε−2(log ε)2, β = 1,
c4 ε−2−(1−β)/α, 0 < β < 1.Multilevel Monte Carlo – p. 13/41
![Page 14: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/14.jpg)
Milstein Scheme
The theorem suggests use of Milstein scheme —better strong convergence, same weak convergence
Generic scalar SDE:
dS(t) = a(S, t) dt + b(S, t) dW (t), 0<t<T.
Milstein scheme:
Sn+1 = Sn + a h + b∆Wn + 12 b′ b
((∆Wn)2 − h
).
Multilevel Monte Carlo – p. 14/41
![Page 15: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/15.jpg)
Milstein Scheme
In scalar case:
O(h) strong convergence
O(ε−2) complexity for Lipschitz payoffs – trivial
O(ε−2) complexity for Asian, lookback, barrier anddigital options using carefully constructed estimatorsbased on Brownian interpolation or extrapolation
Multilevel Monte Carlo – p. 15/41
![Page 16: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/16.jpg)
Milstein Scheme
Key idea: within each timestep, model the behaviour assimple Brownian motion conditional on the two end-points
S(t) = Sn + λ(t)(Sn+1 − Sn)
+ bn
(W (t) − Wn − λ(t)(Wn+1−Wn)
),
where
λ(t) =t − tn
tn+1 − tn
There then exist analytic results for the distribution of themin/max/average over each timestep.
Multilevel Monte Carlo – p. 16/41
![Page 17: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/17.jpg)
Results
Geometric Brownian motion:
dS = r S dt + σ S dW, 0 < t < T,
with parameters T =1, S(0)=1, r=0.05, σ=0.2
European call option: exp(−rT ) max(S(T ) − 1, 0)
Asian option: exp(−rT ) max
(T−1
∫ T
0S(t) dt − 1, 0
)
Lookback option: exp(−rT )
(S(T ) − min
0<t<TS(t)
)
Down-and-out barrier option: same as call providedS(t) stays above B=0.9
Multilevel Monte Carlo – p. 17/41
![Page 18: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/18.jpg)
MLMC Results
GBM: European call
0 2 4 6 8−30
−25
−20
−15
−10
−5
0
l
log 2 v
aria
nce
Pl
Pl− P
l−1
0 2 4 6 8−20
−15
−10
−5
0
l
log 2 |m
ean|
Pl
Pl− P
l−1
Multilevel Monte Carlo – p. 18/41
![Page 19: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/19.jpg)
MLMC Results
GBM: European call
0 2 4 6 810
2
104
106
108
l
Nl
ε=0.00005ε=0.0001ε=0.0002ε=0.0005ε=0.001
10−4
10−3
10−2
10−1
100
101
ε
ε2 Cos
t
Std MCMLMC
Multilevel Monte Carlo – p. 19/41
![Page 20: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/20.jpg)
MLMC Results
GBM: Asian option
0 2 4 6 8−30
−25
−20
−15
−10
−5
0
l
log 2 v
aria
nce
Pl
Pl− P
l−1
0 2 4 6 8−20
−15
−10
−5
0
l
log 2 |m
ean|
Pl
Pl− P
l−1
Multilevel Monte Carlo – p. 20/41
![Page 21: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/21.jpg)
MLMC Results
GBM: Asian option
0 2 4 6 810
2
104
106
108
l
Nl
ε=0.00005ε=0.0001ε=0.0002ε=0.0005ε=0.001
10−4
10−3
10−2
10−1
100
101
ε
ε2 Cos
t
Std MCMLMC
Multilevel Monte Carlo – p. 21/41
![Page 22: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/22.jpg)
MLMC Results
GBM: lookback option
0 5 10−30
−25
−20
−15
−10
−5
0
l
log 2 v
aria
nce
Pl
Pl− P
l−1
0 5 10−20
−15
−10
−5
0
l
log 2 |m
ean|
Pl
Pl− P
l−1
Multilevel Monte Carlo – p. 22/41
![Page 23: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/23.jpg)
MLMC Results
GBM: lookback option
0 5 1010
2
104
106
108
l
Nl
ε=0.00005ε=0.0001ε=0.0002ε=0.0005ε=0.001
10−4
10−3
10−2
10−1
100
101
ε
ε2 Cos
t
Std MCMLMC
Multilevel Monte Carlo – p. 23/41
![Page 24: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/24.jpg)
MLMC Results
GBM: barrier option
0 2 4 6 8−30
−25
−20
−15
−10
−5
0
l
log 2 v
aria
nce
Pl
Pl− P
l−1
0 2 4 6 8−20
−15
−10
−5
0
l
log 2 |m
ean|
Pl
Pl− P
l−1
Multilevel Monte Carlo – p. 24/41
![Page 25: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/25.jpg)
MLMC Results
GBM: barrier option
0 2 4 6 810
2
104
106
108
l
Nl
ε=0.00005ε=0.0001ε=0.0002ε=0.0005ε=0.001
10−4
10−3
10−2
10−1
100
101
ε
ε2 Cos
t
Std MCMLMC
Multilevel Monte Carlo – p. 25/41
![Page 26: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/26.jpg)
Milstein Scheme
Generic vector SDE:
dS(t) = a(S, t) dt + b(S, t) dW (t), 0<t<T,
with correlation matrix Ω(S, t) between elements of dW (t).
Milstein scheme:
Si,n+1 = Si,n + ai h + bij ∆Wj,n
+12
∂bij
∂Slblk
(∆Wj,n ∆Wk,n − hΩjk − Ajk,n
)
with implied summation, and Lévy areas defined as
Ajk,n =
∫ tn+1
tn
(Wj(t)−Wj(tn)) dWk − (Wk(t)−Wk(tn)) dWj .
Multilevel Monte Carlo – p. 26/41
![Page 27: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/27.jpg)
Milstein Scheme
In vector case:
O(h) strong convergence if Lévy areas are simulatedcorrectly – expensive
O(h1/2) strong convergence in general if Lévy areas areomitted, except if a certain commutativity condition issatisfied (useful for a number of real cases)
Lipschitz payoffs can be handled well using antitheticvariables
Other cases may require approximate simulation ofLévy areas
Multilevel Monte Carlo – p. 27/41
![Page 28: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/28.jpg)
Results
Heston model:
dS = r S dt +√
V S dW1, 0 < t < T
dV = λ (σ2−V ) dt + ξ√
V dW2,
T =1, S(0)=1, V (0)=0.04, r=0.05,
σ=0.2, λ=5, ξ=0.25, ρ=−0.5
Multilevel Monte Carlo – p. 28/41
![Page 29: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/29.jpg)
MLMC Results
Heston model: European call
0 2 4 6 8−30
−25
−20
−15
−10
−5
0
l
log 2 v
aria
nce
Pl
Pl− P
l−1
0 2 4 6 8−20
−15
−10
−5
0
l
log 2 |m
ean|
Pl
Pl− P
l−1
Multilevel Monte Carlo – p. 29/41
![Page 30: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/30.jpg)
MLMC Results
Heston model: European call
0 2 4 6 810
2
104
106
108
l
Nl
ε=0.00005ε=0.0001ε=0.0002ε=0.0005ε=0.001
10−4
10−3
10−2
10−1
100
101
ε
ε2 Cos
t
Std MCMLMC
Multilevel Monte Carlo – p. 30/41
![Page 31: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/31.jpg)
Quasi-Monte Carlo
well-established technique for approximatinghigh-dimensional integrals
for finance applications see papers by l’Ecuyer andbook by Glasserman
Sobol sequences are perhaps most popular;we use lattice rules (Sloan & Kuo)
two important ingredients for success:randomized QMC for confidence intervalsgood identification of “dominant dimensions”(Brownian Bridge and/or PCA)
Multilevel Monte Carlo – p. 31/41
![Page 32: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/32.jpg)
Quasi-Monte Carlo
Approximate high-dimensional hypercube integral∫
[0,1]df(x) dx
by
1
N
N−1∑
i=0
f(x(i))
where
x(i) =
[i
Nz
]
and z is a d-dimensional “generating vector”.
Multilevel Monte Carlo – p. 32/41
![Page 33: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/33.jpg)
Quasi-Monte Carlo
In the best cases, error is O(N−1) instead of O(N−1/2) butwithout a confidence interval.
To get a confidence interval, let
x(i) =
[i
Nz + x0
].
where x0 is a random offset vector.
Using 32 different random offsets gives a confidenceinterval in the usual way.
Multilevel Monte Carlo – p. 33/41
![Page 34: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/34.jpg)
Quasi-Monte Carlo
For the path discretisation we can use
∆Wn =√
h Φ−1(xn),
where Φ−1 is the inverse cumulative Normal distribution.
Much better to use a Brownian Bridge construction:
x1 −→ W (T )
x2 −→ W (T/2)
x3, x4 −→ W (T/4),W (3T/4)
. . . and so on by recursive bisection
Multilevel Monte Carlo – p. 34/41
![Page 35: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/35.jpg)
Multilevel QMC
rank-1 lattice rule developed by Sloan, Kuo &Waterhouse at UNSW
32 randomly-shifted sets of QMC points
number of points in each set increased as needed toachieved desired accuracy, based on confidenceinterval estimate
results show QMC to be particularly effective on lowestlevels with low dimensionality
Multilevel Monte Carlo – p. 35/41
![Page 36: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/36.jpg)
MLQMC Results
GBM: European call
0 2 4 6 8−40
−35
−30
−25
−20
−15
−10
−5
0
l
log 2 v
aria
nce
1162564096
0 2 4 6 8−20
−15
−10
−5
0
l
log 2 |m
ean|
Pl
Pl− P
l−1
Multilevel Monte Carlo – p. 36/41
![Page 37: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/37.jpg)
MLQMC Results
GBM: European call
0 2 4 6 810
1
102
103
104
105
106
l
Nl
ε=0.00005ε=0.0001ε=0.0002ε=0.0005ε=0.001
10−4
10−3
10−4
10−3
10−2
10−1
ε
ε2 Cos
t
Std QMCMLQMC
Multilevel Monte Carlo – p. 37/41
![Page 38: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/38.jpg)
MLQMC Results
GBM: barrier option
0 2 4 6 8−40
−35
−30
−25
−20
−15
−10
−5
0
l
log 2 v
aria
nce
1162564096
0 2 4 6 8−20
−15
−10
−5
0
l
log 2 |m
ean|
Pl
Pl− P
l−1
Multilevel Monte Carlo – p. 38/41
![Page 39: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/39.jpg)
MLQMC Results
GBM: barrier option
0 2 4 6 810
1
102
103
104
105
106
107
l
Nl
ε=0.00005ε=0.0001ε=0.0002ε=0.0005ε=0.001
10−4
10−3
10−3
10−2
10−1
100
ε
ε2 Cos
t
Std QMCMLQMC
Multilevel Monte Carlo – p. 39/41
![Page 40: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/40.jpg)
Conclusions
Results so far:
much improved order of complexity
fairly easy to implement
significant benefits for model problems
However:
lots of scope for further developmentmulti-dimensional SDEs needing Lévy areasadjoint Greeks and “vibrato” Monte Carlonumerical analysis of algorithmsexecution on NVIDIA graphics cards (128 cores)
need to test ideas on real finance applications
Multilevel Monte Carlo – p. 40/41
![Page 41: Multilevel Monte Carlo path simulation - NAG · Multilevel Monte Carlo path simulation Mike Giles giles@comlab.ox.ac.uk Oxford University Mathematical Institute Oxford-Man Institute](https://reader036.fdocuments.in/reader036/viewer/2022062602/5e899d92aae127615c7dea3a/html5/thumbnails/41.jpg)
Papers
M.B. Giles, “Multilevel Monte Carlo path simulation”,to appear in Operations Research, 2007.
M.B. Giles, “Improved multilevel convergence using theMilstein scheme”, to appear in MCQMC06 proceedings,Springer-Verlag, 2007.
M.B. Giles, “Multilevel quasi-Monte Carlo path simulation”,submitted to Journal of Computational Finance, 2007.
www.comlab.ox.ac.uk/mike.giles/finance.html
Email: [email protected]
Multilevel Monte Carlo – p. 41/41