Monte Carlo Methods

Trees
IMPA Commodities Course :
Numerical Methods
Sebastian Jaimungal
[email protected]
Department of Statistics and
Mathematical Finance Program,
University of Toronto, Toronto, Canada
http://www.utstat.utoronto.ca/sjaimung
February 22, 2008
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Table of contents
1
Monte Carlo Methods
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
2
Trees
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Basic MC Methods
For option valuation, an expectation is the basic quantity
V
t
= e
r (Tt)
E
Q
t
[(F
T
(U))]
For value-at-risk calculations, quantiles are the object of
interest
VaR
p
max
_
X : P
_

i
a
i
V
i
> X
_
> p
_
E
P
_

i
a
i
V
i
_
For simulation, the price path {F
t
1
(U), F
t
2
(U), . . . } is required
In all cases Monte Carlo simulation can assist
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Basic MC Methods
Idea: simulate from the distribution of relevant quantity and
compute average
Example: Price a 1-year oating strike Asian option on the
1.25-year forward contract with averaging occurring over the
last month of the contract using the Schwartz model
The payo of this option is
=
_
F
1
(1.25)
20

i =1
F
1+(i 20)/251
(1.25)
_
+
Need to simulate the forward price for all days in nal month
Simulate in one step the forward price at time 1 20/251
Simulate daily time-steps (t = 1/251) within month
Compute pay-o and discount
Average over many paths
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Brownian Bridges
Rather than simulating entire path, can simulate the path in a
progressive renement manner
A Brownian bridge is a way of generating a Brownian path
conditional on the end points
Given X(0) = x
0
and X(t) = x
t
generate X(s) for 0 < s < t.
The joint distribution of X(t), X(s) given X(0) is a bivariate
normal
_
X(s)
X(t)
_

X
0
=x
0
N
__
x
0
x
0
_
;
_
s s
s t
__
Can then show that,
X(s)|
X(0)=x
0
,X(t)=x
t
N
_
x
0
+
s
t
(x
t
x
0
);
2
(t s)s
t
_
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Brownian Bridges
A Bridge renement example: 4 renement steps
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
(a)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
(b)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
(c)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
(d)
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Brownian Bridges
Many Brownian Bridge paths: 6 renement steps, 100 paths,
X
0
= 1, X
1
= 1, = 0.2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
1.5
1
0.5
0
0.5
1
1.5
Time
V
a
l
u
e
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Brownian Bridges
To generate Brownian sample paths using a Brownian Bridge:
1
generate random sample of X(t) given X(0):
X(t)|
X(0)=x
0
N(x
0
; t)
2
build bridge from X(0) = x
0
to X(t) = x
t
3
repeat from step 1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
3
2
1
0
1
2
3
Time
V
a
l
u
e
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Mean-Reverting Bridges
For mean-reverting process
dX
t
= ( X
t
) dt + dW
t
The joint of X
t
1
, X
s
given X
t
0
is
_
X
s
X
t
1
_

X
t
0
N
__
+ e
(st
0
)
(X
t
0
)
+ e
(tt
0
)
(X
t
0
)
_
;
_
where
=

2
2
_
1 e
2(st
0
)
1 e
2(st
0
)
1 e
2(st
0
)
1 e
2(tt
0
)
_
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Mean-Reverting Bridges
Can then show that
X
s
|
X
t
0
,X
t
1
N (m; v)
with
m = e
(st
0
)
_
X
0
+(e
(st
0
)
1)
+
e
2(st
0
)
1
e
2(tt
0
)
1
(e
(t
1
t
0
)
X
t
1
(X
t
0
+(e
(t
1
t
0
)
1)))
_
v =

2
2
(e
2(ts)
1)
e
2(st
0
)
1
e
2(tt
0
)
1
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Mean-Reverting Bridges
Many Mean-Reverting Bridge paths: 6 renement steps, 100 paths,
X
0
= 1, X
1
= 1, = 0, = 1, = 0.2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
1.5
1
0.5
0
0.5
1
1.5
Time
V
a
l
u
e
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Mean-Reverting Bridges
To generate Mean-Reverting sample paths using a Mean-Reverting
Bridge:
1
generate random sample of X(t) given X(0)
2
build bridge from X(0) = x
0
to X(t) = x
t
3
repeat from step 1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
1.2
Time
V
a
l
u
e
6 renement steps, 100 paths, X
0
= 1, = 0, = 1, = 0.2
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Least Squares Monte Carlo
Carrier(1994) and Longsta & Schwartz (2000) developed
the least-squares Monte Carlo method for valuing early
exercise clauses.
Basic idea
1
Generate sample paths forward in time
2
Place payo at end nodes
3
Compute discounted value of option
4
Estimate conditional expectation by projection onto basis
functions
5
Determine optimal exercise point using basis functions
6
Repeat from step 3
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Least Squares Monte Carlo
Example: American Put strike= 1, spot= 1, r = 0.05:
Asset prices
Path t=0 t=1 t=2 t=3
1 1 0.95 0.94 0.82
2 1 0.97 1.21 1.15
3 1 0.96 0.91 0.87
4 1 0.84 1.20 0.87
5 1 0.93 0.90 0.91
6 1 1.03 0.99 1.01
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Least Squares Monte Carlo
Example: American Put strike= 1, spot= 1, r = 0.05:
t=2 t=3
asset prices payo
0.94 0.18
1.21 0
0.91 0.13
1.20 0.13
0.90 0.09
0.99 0
Compute payo at t = 3
Focus only on paths which are in the money at t = 2
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Least Square Monte Carlo
Compute discounted value of payos at time t = 2
t = 2 t=2 t=3
discounted payo asset prices payo
0.17 0.94 0.18
0 1.21 0
0.12 0.91 0.13
0.12 1.20 0.13
0.08 0.90 0.09
0 0.99 0
Regress discounted payo onto asset prices at t = 2 using
basis functions (e.g. 1, S, S
2
):
V
2
(S) = 55.04 + 117.7 S 62.75 S
2
Regression gives estimate of E[e
r t
V
t+dt
|S
t
]
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Least Square Monte Carlo
Compare estimate discounted expectation with immediate
exercise value
t = 2 t=2 t=2 t=2
est. disc. exp. asset prices exercise value est. option price
0.1697 0.94 0.06 0.17
- 1.21 0 0
0.1208 0.91 0.09 0.12
- 1.20 0 0.12
0.0795 0.90 0.10 0.10 x
0.0001 0.99 0.01 0.01 x
In this example last two branches are optimal to exercise
Notice that the realized value at node t = 2 are used when
going backwards, not the estimate of the conditional
expectation
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Least Square Monte Carlo
Continue working backwards to obtain estimated prices at
t = 1 and then t = 0
Example: American put strike = 1, term = 1, r = 5%,
= 20%
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.8
0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1


LS method
FST method
250 steps, 300,000 sample paths
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Basic MC Methods
Monte Carlo Bridges
Least Squares Monte Carlo
Least Square Monte Carlo
Example: American put strike = 1, term = 1, r = 5%, = 1,
= 0, = 20%
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.85
0.9
0.95
1
250 steps, 300,000 sample paths
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Tress
Binomial trees are not appropriate for commodities due to
mean-reversion
Trinomial trees are used instead
Branching probabilities choosing to match mean and variance
E
Q
t
[X
t+t
X
t
] = (e
t
1)X
t
M X
t
V
Q
t
[X
t+t
X
t
] =

2
2
(1 e
2t
) V
Branch steps set to X =

3V
Tree is cut at high and low values to avoid negative
probabilities
Topoftree Middleoftree Bottomoftree
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Tress
Zero mean-reversion level
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Tress
Middle of tree branching probabilities:
p
u
=
1
6
+
j
2
M
2
+jM
2
p
m
=
2
3
j
2
M
2
p
d
=
1
6
+
j
2
M
2
jM
2
Top and Bottom of tree branching probabilities:
Top Bottom
p
u
=
7
6
+
j
2
M
2
+3jM
2
p
u
=
1
6
+
j
2
M
2
jM
2
p
m
=
1
3
j
2
M
2
2jM p
m
=
1
3
j
2
M
2
+ 2jM
p
d
=
1
6
+
j
2
M
2
+jM
2
p
d
=
7
6
+
j
2
M
2
3jM
2
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Tress
Shifted mean-reversion level
For simple mean-reversion will shift via + (ln S
0
)e
t
+ X
t
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods
Monte Carlo Methods
Trees
Tress
Comparison of LSM and Trinomial model
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
65
70
75
80
85
90
95
100


LSM
LSM
LSM
= 2
= 1
= 0
Sebastian Jaimungal [email protected] IMPA Commodities Course : Numerical Methods