0% found this document useful (0 votes)
52 views35 pages

Lectures08-09 Stoch Calchandout

The document discusses probability theory, stochastic processes, Itô's formula, and Monte Carlo simulation. It provides definitions of key concepts like probability space, random variables, discrete and continuous random variables, and information filtrations. It also outlines how stochastic processes can be used to model variables like stock prices that change over time. Monte Carlo simulation is introduced as a method to analyze stochastic processes.

Uploaded by

e.mahler1997
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
52 views35 pages

Lectures08-09 Stoch Calchandout

The document discusses probability theory, stochastic processes, Itô's formula, and Monte Carlo simulation. It provides definitions of key concepts like probability space, random variables, discrete and continuous random variables, and information filtrations. It also outlines how stochastic processes can be used to model variables like stock prices that change over time. Monte Carlo simulation is introduced as a method to analyze stochastic processes.

Uploaded by

e.mahler1997
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 35

Probability theory

Stochastic processes
Itô’s formula
Monte Carlo simulation

Derivatives and Risk Management


Lectures 8-9: Stochastic calculus and Monte Carlo simulation

Søren Hesel

Department of Business and Management

Fall 2022

Søren Hesel DRM


Probability theory
Stochastic processes
Itô’s formula
Monte Carlo simulation

What type of process describes reality best?

Source: http://www.google.com/finance

Søren Hesel DRM


Probability theory
Stochastic processes
Itô’s formula
Monte Carlo simulation

What is a stochastic process?

Random variable Value assigned some uncertain outcome:


• Coin flip: 1 for heads, zero for tails
Stochastic process:
• collection of random variables, one for each “relevant” point in
time
• tracks how the value of an object changes over time
• can be classified as discrete time or continuous time
▶ discrete time: can change only at fixed points in time
▶ continuous time: can change at any point in time
• can be classified as discrete variable or continuous variable
▶ discrete variable: can only take discrete values
▶ continuous variable: can take any value within a certain range

Søren Hesel DRM


Probability theory
Stochastic processes
Itô’s formula
Monte Carlo simulation

Outline

1 Probability theory

2 Stochastic processes

3 Itô’s formula

4 Monte Carlo simulation

Søren Hesel DRM


Probability theory
Probability space, (Ω, F, P)
Stochastic processes
Random variables
Itô’s formula
Filtrations
Monte Carlo simulation

Probability space

The basic object for studies of uncertain events is a probability space,


(Ω, F, P), with
• the state space, Ω, consisting of all possible states/outcomes of
all relevant uncertain objects, e.g.
▶ coin toss: Ω = {heads, tails}

▶ throw of a die: Ω = , , , , ,
• the σ-algebra, F, consisting of the “probabilizable” events, e.g.
▶ coin toss: F = {∅, heads, tails, Ω}
 
▶ throw of a die: e.g. , , , ∈ F, but there are many
events A ∈ F
• the probability measure, P : F → [0, 1], assigning a probability,
P(A), to each probabilizable event A ∈ F, e.g.
▶ coin toss: P({heads}) = P({tails}) = 1
  1 2
▶ throw of a die: P
 1
= 6 and P , , = 2

Søren Hesel DRM


Probability theory
Probability space, (Ω, F, P)
Stochastic processes
Random variables
Itô’s formula
Filtrations
Monte Carlo simulation

A few words about probability measures

Many different probability measures can be defined on the same


σ-algebra, F, of events

In the example of the dice, two probability measures, P and Q, can be


defined by

ω
1 1 1 1 1 1
P(ω) 6 6 6 6 6 6
1 1 1 1 1 3
Q(ω) 12 6 6 6 6 12

Do you know a financial example with multiple probability measures?

Søren Hesel DRM


Probability theory
Probability space, (Ω, F, P)
Stochastic processes
Random variables
Itô’s formula
Filtrations
Monte Carlo simulation

Random variables
A random variable is a function X : Ω → R mapping ω 7→ X(ω) such
that {ω ∈ Ω | X(ω) ∈ I} ∈ F (measurability) for any interval I ⊆ R, e.g.

1 if ω = heads
• coin toss: X(ω) =
0 if ω = tails
• throw of a die: X(ω) = number of eyes in the realization ω
• finance: the stock price in one year, S1 (ω)

Random variables are the natural way of representing the (possibly)


abstract outcomes ω ∈ Ω by a real number

The (cumulative) distribution function, FX , of a random variable, X, is


defined as

FX (x) = P(X ⩽ x) := P({ω ∈ Ω | X(ω) ⩽ x})

Søren Hesel DRM


Probability theory
Probability space, (Ω, F, P)
Stochastic processes
Random variables
Itô’s formula
Filtrations
Monte Carlo simulation

Discrete random variables

• Only take on finitely many different values x1 , x2 , . . . , xm ∈ R


• Can be represented by the probabilities

pi = P(X = xi ) := P({ω ∈ Ω | X(ω) = xi }), i = 1, . . . , m

which naturally sums to one. NB: Ω need not be finite.


The expected value is then (naturally) defined as
Z m Z
X X
m
E[g(X)] := g(X)dP = g(X)dP = pi g(xi )
Ω i=1 (X=xi ) i=1

for any “nice” function g, e.g., g(x) = x.

Søren Hesel DRM


Probability theory
Probability space, (Ω, F, P)
Stochastic processes
Random variables
Itô’s formula
Filtrations
Monte Carlo simulation

Continuous random variables

• Take values in a continuum


• If FX is absolutely continuous, it can be represented by a
probability density function, fX : R → R+ ,
Zx
fX (x) = FX′ (x) ⇒ FX (x) = fX (y)dy
−∞
R∞ R∞
and −∞ fX (x)dx = −∞ FX′ (x)dx = FX (∞) − FX (−∞) = 1.
The expected value is then defined as
Z Z∞ Z∞
E[g(X)] := g(X)dP = g(x)dFX (x) = g(x)fX (x)dx
Ω −∞ −∞

for any “nice” function g, e.g., g(x) = x.

Søren Hesel DRM


Probability theory
Probability space, (Ω, F, P)
Stochastic processes
Random variables
Itô’s formula
Filtrations
Monte Carlo simulation

Information filtrations
A filtration F = (Ft )t∈T is an increasing family of σ-algebras, i.e.,
Ft ⊆ Ft ′ whenever t ⩽ t ′ .

Ft contains all “decidable” events at time t, i.e., the events which we


at time t can verify has occurred or not. At time 0 we know nothing
and hence F0 = {∅, Ω}. In an economy with a time horizon of T we will
assume that all uncertainty is resolved over time, i.e., FT = F.

Our formal model of uncertainty and information is therefore a filtered


probability space (Ω, F, F, P), where (Ω, F, P) is a probability space
and F = (Ft )t∈T is a filtration.

How do we specify a filtered probability space in which the behavior


of stock prices, interest rates, etc., can be studied? We don’t! Instead
we describe the dynamics of the key variables, which then implicitly
defines the probability space and its natural filtration.
Søren Hesel DRM
Probability theory General concepts
Stochastic processes Standard Brownian motion
Itô’s formula Stochastic integrals
Monte Carlo simulation Itô processes

General concepts

• Interested in a dynamic economy, ⇝ the evolution of uncertain


objects must be modeled. A stochastic process is a family of
random variables, x = (xt )t∈T .
• Discrete-time stochastic processes take T = {t0 , t1 , . . . , tN } but we
will work with continuous-time stochastic processes taking
T = [0, T] or T = [0, ∞).
• A process is adapted to F if xt is Ft -measurable for all t ∈ T, i.e. if
the value of the process at time t is known at time t for all t ∈ T
• A process is a (P, F)-martingale, if for all t, t ′ ∈ T with t ⩽ t ′

Et [xt ′ ] := E[xt ′ | Ft ] = xt ,

i.e., no change in the value is expected, Et [xt ′ − xt ] = 0.

Søren Hesel DRM


Probability theory General concepts
Stochastic processes Standard Brownian motion
Itô’s formula Stochastic integrals
Monte Carlo simulation Itô processes

General concepts, cont’d

• For a fixed realization ω ∈ Ω, the function t 7→ xt (ω) is a sample


path of the process x
• The value space of a process x is defined as the smallest set S
with the property that P(xt ∈ S) = 1 for all t ∈ T.
• The history of a process x at a time t is the process (xs )0⩽s⩽t . The
future values are still stochastic.
• The natural filtration of a process x is the smallest σ-algebra
containing the history of x, i.e., Ftx := σ({xs | 0 ⩽ s ⩽ t})
• A process x is Markov if the history contains no information about
the future value that cannot be extracted from the current value,
i.e.
P(xt ′ ∈ A | (xs )0⩽s⩽t ) = P(xt ′ ∈ A | xt )

Søren Hesel DRM


Probability theory General concepts
Stochastic processes Standard Brownian motion
Itô’s formula Stochastic integrals
Monte Carlo simulation Itô processes

Standard Brownian motion (Wiener process)


A stochastic process z = (zt )t⩾0 is called a standard Brownian motion
if it satisfies the following conditions:
• z0 = 0
• for all t ′ > t ⩾ 0: zt ′ − zt ∼ N(0, t ′ − t)
• independent increments: for all 0 ⩽ t0 < t1 < · · · < tn :
zt1 − zt0 , . . . , ztn − ztn−1 are mutually independent
Properties:
• z can take any values in R
• z is a Markov process
• z is a martingale
• sample paths are nowhere differentiable
Is this a good process for stock/bond/currency prices or interest
rates?

Søren Hesel DRM


Probability theory General concepts
Stochastic processes Standard Brownian motion
Itô’s formula Stochastic integrals
Monte Carlo simulation Itô processes

The stochastic integral (Itô integral)


Rt ′
Is defined, t σu dzu , of a “nice” process σu
• if σu = σ is a constant
Z t′ Z t′
σdzu = σ dzu = σ(zt ′ − zt )
t t
• if σu is a piece-wise constant process, i.e.,


 σt0 u ∈ [t0 , t1 )


σt1 u ∈ [t1 , t2 )
σu = .
..




σtn−1 u ∈ [tn−1 , tn ]
for t = t0 < t1 < · · · < tn = t ′ the integral can be computed as
Z t′ n−1 Z ti+1
X X
n−1
σu dzu = σu dzu = σti (zti+1 − zti )
t i=0 ti i=0

Søren Hesel DRM


Probability theory General concepts
Stochastic processes Standard Brownian motion
Itô’s formula Stochastic integrals
Monte Carlo simulation Itô processes

The stochastic integral, cont’d


• a general process σu can be approximated by a sequence,
(1) (2)
σu , σu , . . . , of piece-wise constant processes which converge
to σu ⇝ the stochastic integral of σu defined as the limit of the
stochastic integrals of the piece-wise constant processes.

Properties:
• Theorem 3.2: If σ = (σu ) is a “nice” process, then
"Z ′ # "Z ′ # Z′
t t t
Et σ2u du
 
Et σu dzu = 0 and Vart σu dzu =
t t t

⇝ martingale
• Theorem 3.3: If σu = σ(u) is a deterministic function of time, the
stochastic integral,
Z t′ Z t′ !
σ(u)dzu ∼ N 0, σ(u) du
2
t t

Søren Hesel DRM


Probability theory General concepts
Stochastic processes Standard Brownian motion
Itô’s formula Stochastic integrals
Monte Carlo simulation Itô processes

Itô processes
• A process on the form
Zt Zt
xt = x0 + µu du + σu dzu ,
0 0

where x0 is a constant and µ = (µu )u∈T and σ = (σu )u∈T are


“nice” adapted processes
• In the short-hand form,stochastic differential equation (SDE)
dxt = µt dt + σt dzt

• Informally,
Et [dxt ] = µt dt and Vart [dxt ] = σ2t dt
so µt and σ2t are called the instantaneous drift and variance.
• In finance applications, the instantaneous standard deviation σt
is also known as the volatility. When is x a martingale?
Søren Hesel DRM
Probability theory General concepts
Stochastic processes Standard Brownian motion
Itô’s formula Stochastic integrals
Monte Carlo simulation Itô processes

Itô processes, cont’d


Properties:

• x has continuous paths and the paths are nowhere differentiable


• the value space and the distribution of x will depend on the
processes µ and σ
Examples:
• standard Brownian motion: x0 = 0, µ = 0 and σ = 1
• generalized Brownian motion: µ and σ constants
▶ Markov process
▶ xt = x0 + µt + σzt ∼ N(x0 + µt, σ2 t)
▶ the value space is R
• diffusion processes: µu = µ(xu , u) and σu = σ(xu , u)
▶ Markov process
• geometric Brownian motion: µ(xu , u) = µ · xu , σ(xu , u) = σ · xu
▶ Markov process
▶ the value space is R+
Are any of these good models for prices or interest rates? (see
Figure 3.4)?
Søren Hesel DRM
Probability theory General concepts
Stochastic processes Standard Brownian motion
Itô’s formula Stochastic integrals
Monte Carlo simulation Itô processes

Example
(Tutorial: Hull’s Question 14.5)

Consider a variable S that follows the process



(2, 3) for 0 ⩽ t ⩽ 3
dSt = µ(t)dt + σ(t)dzt , S0 = 5, (µ(t), σ(t)) =
(3, 4) for 3 < t ⩽ 6
What is the probability distribution of S6 ?

Hints:
• Integrate both sides of the SDE from [0, 6] to obtain:
Z6 Z6 Z6 Z6
S6 − S0 = dSt = (µ(t)dt + σ(t)dzt ) = µ(t)dt + σ(t)dzt
0 0 0 0
• Split up the integral from [0, 3] and (3, 6] and exploit that µ and σ
are constants on these intervals
• Use that increments to Brownian motion are independent and
normally distributed.
Søren Hesel DRM
Probability theory
Stochastic processes The one-dimensional Itô’s formula
Itô’s formula The two-dimensional Itô’s formula
Monte Carlo simulation

The chain rule from ordinary calculus


Let x 7→ g(x) and t 7→ h(t) be two functions and consider the
composite function y(t) = g(h(t)). Then

dy(t) dh(t)
= g ′ (h(t))
dt dt
or in short-hand notation: dy(t) = g ′ (h(t))dh(t)

Think of this as coming from a Taylor-approximation

g(h(t + dt)) − g(h(t))


dh(t) 1
≈ g ′ (h(t)) dt + [g ′′ (h(t))h ′ (t) + g ′ (h(t))h ′′ (t)] (dt)2
dt 2
where terms involving dt of order greater than one vanish, i.e.
(dt)2 = · · · = 0 since these terms are much smaller than dt for small
dt.
Søren Hesel DRM
Probability theory
Stochastic processes The one-dimensional Itô’s formula
Itô’s formula The two-dimensional Itô’s formula
Monte Carlo simulation

A similar argument for stochastic processes


Now, take x 7→ g(x) and a stochastic process xt with representation
dxt = µt dt + σt dzt . Consider the stochastic process yt = g(xt ). Is it true
that
dyt = g ′ (xt )dxt ?

Unfortunately not, but y will still be an Itô process with a suitable drift
and volatility.

Taylor approximation
g ′′ (xt )
g(xt+dt ) − g(xt ) ≈ g ′ (xt )dxt + (dxt )2 + . . .
2
Here
(dxt )2 = (µt dt + σt dzt )2 = µ2t (dt)2 + σ2t (dzt )2 + 2µt σt (dt)(dzt )
Taking the mean we see that Et [(dzt )2 ] = dt and Et [(dt)(dzt )] = 0, while
the variance will be of dt-order higher than one.
Søren Hesel DRM
Probability theory
Stochastic processes The one-dimensional Itô’s formula
Itô’s formula The two-dimensional Itô’s formula
Monte Carlo simulation

A similar argument, cont’d


Ignoring terms involving dt of order higher than one gives
g ′′ (xt )σ2t
dyt = g ′ (xt )dxt + dt
2
g ′′ (xt )σ2t
 
= g ′ (xt )µt + dt + g ′ (xt )σt dzt
2
Generalizing further to functions yt = g(xt , t) we get the
one-dimensional Itô’s formula as in Theorem 3.6:
1 ∂2 g
 
∂g ∂g 2 ∂g
dyt = (xt , t) + (xt , t)µt + (x t , t)σt dt + (xt , t)σt dzt
∂t ∂x 2 ∂x2 ∂x
∂g ∂g 1 ∂2 g
= (xt , t)dt + (xt , t)dxt + (xt , t)(dxt )2 (3.10)
∂t ∂x 2 ∂x2
where (dxt )2 is computed according to the rules (dt)2 = dt · dzt = 0 and
(dzt )2 = dt.

Buzz assignment: The geometric Brownian motion.


Søren Hesel DRM
Probability theory
Stochastic processes The one-dimensional Itô’s formula
Itô’s formula The two-dimensional Itô’s formula
Monte Carlo simulation

Example
(Tutorial: Hull’s Question 14.9)

Suppose that a stock price S follow geometric Brownian motion with


expected return µ and volatility σ:

dSt = µSt dt + σSt dzt

What is the process followed by Stn ? Show that Stn also follows
geometric Brownian motion.
• Apply Itô’s formula with the function g(x, t) = xn applied to the
process St .
• Compute the derivatives ∂g ∂g ∂2 g
∂t , ∂x , and ∂x2
• Insert in Itô’s formula:
1 ∂2 g
 
∂g ∂g 2 2 ∂g
dg(St , t) = (St , t) + (St , t)µSt + (St , t)σ St dt + (St , t)σSt dzt
∂t ∂x 2 ∂x2 ∂x

Søren Hesel DRM


Probability theory
Stochastic processes The one-dimensional Itô’s formula
Itô’s formula The two-dimensional Itô’s formula
Monte Carlo simulation

Example
(Tutorial: Exercise 4 in Tutorial plan)

Consider a process y with the following dynamics

dyt = κ(ȳ − yt )dt + σdzt

where κ, ȳ, σ are positive constants and (zt ) is a Brownian motion.


1 What are the dynamics of the process Bt = e−yt (T−t) ?
2 What is the value of BT for any yT ?

If yt is an interest rate what is the interpretation of Bt = e−yt (T−t) ?

Søren Hesel DRM


Probability theory
Stochastic processes The one-dimensional Itô’s formula
Itô’s formula The two-dimensional Itô’s formula
Monte Carlo simulation

The two-dimensional Itô’s formula


Now, take stochastic processes xt and yt with representations
dxt = µxt dt + σx1t dz1t + σx2t dz2t , dyt = µyt dt + σy1t dz1t + σy2t dz2t (3.17)

where z1 and z2 are independent standard Brownian motions.

With similar arguments as before, the process Wt = g(xt , yt , t) can be


shown to be an Itô process with dynamics
∂g ∂g ∂g 1 ∂2 g 2 1 ∂2 g 2
dWt = + µxt + µyt + (σ + σ2x2t ) + (σ + σ2y2t )
∂t ∂x ∂y 2 ∂x2 x1t 2 ∂y2 y1t
!
∂2 g
+ (σx1t σy1t + σx2t σy2t ) dt
∂x∂y
   
∂g ∂g ∂g ∂g
+ σx1t + σy1t dz1t + σx2t + σy2t dz2t
∂x ∂y ∂x ∂y
∂g ∂g ∂g 1 ∂2 g 1 ∂2 g ∂2 g
= dt + dxt + dyt + 2
(dxt )2 + 2
(dyt )2 + (dxt )(dyt )
∂t ∂x ∂y 2 ∂x 2 ∂y ∂x∂y

where (dt)2 = (dt)(dz1t ) = (dt)(dz2t ) = (dz1t )(dz2t ) = 0.


Søren Hesel DRM
Probability theory
Stochastic processes The one-dimensional Itô’s formula
Itô’s formula The two-dimensional Itô’s formula
Monte Carlo simulation

Example
(Tutorial: Munk’s Exercise 3.5)

Consider the two general stochastic processes x1 = (x1t ) and


x2 = (x2t ) defined by the dynamics
dx1t = µ1t dt + σ1t dz1t ,
q
dx2t = µ2t dt + ρt σ2t dz1t + 1 − ρ2t σ2t dz2t ,
where z1 and z2 are independent one-dimensional standard Brownian
motions.
• Interpret µit , σit , and ρt .
• Define the processes y = (yt ) and w = (wt ) by yt = x1t x2t and
wt = x1t /x2t . What is the dynamics of y and z?
• Concretize your answer for the special case where x1 and x2 are
geometric Brownian motions with constant correlation.

Hint: Apply the two-dimensional Itô formula with g(x1 , x2 , t) = x1 x2 and


g(x1 , x2 , t) = x1 /x2 .
Søren Hesel DRM
Probability theory
Stochastic processes The basic method
Itô’s formula Precision and improvements
Monte Carlo simulation

Simulating a standard Brownian motion

Partition the relevant time interval [t, T] into N subintervals of length


∆t = T−t
N , i.e.,
t = t0 < t1 < · · · < tN = T,
with tn = t + n∆t.

Then {ztn − ztn−1 }Nn=1 forms a sequence of i.i.d. N(0, ∆t)-variables.



Distributional equivalent to a sequence, {εn ∆t}Nn=1 , where {εn }Nn=1 are
i.i.d. N(0, 1)-variables (why?)

Søren Hesel DRM


Probability theory
Stochastic processes The basic method
Itô’s formula Precision and improvements
Monte Carlo simulation

Simulating a standard Brownian motion, cont’d


How do we generate N(0, 1)-variates ?
1 Built-in functionality in some software package
2 Transform a sequence of i.i.d. uniformly distributed variates:
▶ Let U ∼ U(0, 1) and let Φ denote the distribution function of a
N(0, 1) random variable
▶ The distribution of X = Φ−1 (U) is then

P(X ⩽ x) = P(Φ−1 (U) ⩽ x) = P(U ⩽ Φ(x)) = Φ(x)

3 Box-Muller: Given U1 , U2 ∼ U(0, 1) it can be shown that


p p
ε1 = −2 ln U1 cos(2πU2 ), ε2 = −2 ln U1 sin(2πU2 )

are two i.i.d. N(0, 1) variables.


4 From scratch. . .
Søren Hesel DRM
Probability theory
Stochastic processes The basic method
Itô’s formula Precision and improvements
Monte Carlo simulation

Simulating a standard Brownian motion, cont’d

Excel (US) Excel (DK) R Matlab


N(0, 1) Data Analysis Data Analysis rnorm randn
U(0, 1) Data Analysis Data Analysis runif rand
or RAND or SLUMP
Φ−1 NORMSINV STANDARDNORM.INV

Note: Using set.seed(number) in R/rng in Matlab will fix the starting


point
⇝ random numbers will be the same

Time for you to work:


• Simulate 10 paths of a standard Brownian motion and depict
them in a figure
• In your simulation use a time span of [0, 1] and N = 52 time points

Søren Hesel DRM


Probability theory
Stochastic processes The basic method
Itô’s formula Precision and improvements
Monte Carlo simulation

The idea
Assume that the Q-dynamics of the underlying state variable is

dxt = µ(xt , t) dt + σ(xt , t) dzQ


t .

If we assume a constant risk-free rate, r, risk-neutral valuation gives:

1 X −r(T−t)
h i M
−r(T−t)
f (x, t) = EQ
t e F(xT ) ≈ e F(xTm )
M
m=1

where F is the payoff function and m denotes a simulated path of x.

Dividing the time period [t, T] into N sub-intervals of length ∆t, the
SDE can now be approximated by the Euler scheme

xtn+1 = xtn + µ(xtn , tn )∆t + σ(xtn , tn )εtn+1 ∆t, εtn ∼ N(0, 1)

for the time points tn = t + n∆t, n = 0, 1, . . . , N.


Søren Hesel DRM
Probability theory
Stochastic processes The basic method
Itô’s formula Precision and improvements
Monte Carlo simulation

Procedure

1 Repeat for m = 1, . . . , M:
▶ Draw/simulate a sequence {εmt | n = 1, . . . , N} of i.i.d. N(0, 1) random
n
variables
▶ Calculate the discounted payoff of the derivative, e−r(T−t) F(xTm )
2 Calculate the average discounted payoff to get an estimate of the
value of the derivative

Example: Compute the expected value and variance of a geometric


Brownian motion in a year, x1 , when

dxt = µxt dt + σxt dzt ,

with µ = 0.05 and σ = 0.2.

Søren Hesel DRM


Probability theory
Stochastic processes The basic method
Itô’s formula Precision and improvements
Monte Carlo simulation

Accuracy

• Given the M simulated discounted payoffs f 1 , f 2 , . . . , f M , where


f m = e−r(T−t) F(xTm )
P
• Estimate of the option price: f̄ = M1 Mm=1 f
m
q PM
• Sample standard deviation: σ̂f = M−1 1 m
m=1 (f − f̄ )
2

• 95% confidence interval for the option price is:


 
1.96σ̂f 1.96σ̂f
f̄ − √ , f̄ + √
M M
(if normally distributed option values or asymptotically)
• To double the precision, use 4 times as many simulations
• Often M = 10, 000 or more is necessary to obtain good precision

Søren Hesel DRM


Probability theory
Stochastic processes The basic method
Itô’s formula Precision and improvements
Monte Carlo simulation

Variance reduction techniques

• To estimate f with a reasonable precision a very large number of


trials is necessary if the simulation is carried out as described so
far
• To reduce the computation time we can use some variance
reduction procedures, e.g.,
▶ Antithetic variable technique: simulate one path with
{εmtn | n = 1, . . . , N} and get one for free, namely, the one with
{−εmtn | n = 1, . . . , N}
▶ Control variate technique: use method to price a “similar” security
which has an analytic pricing formula and assume that the pricing
error is the same

Søren Hesel DRM


Probability theory
Stochastic processes The basic method
Itô’s formula Precision and improvements
Monte Carlo simulation

Antithetic variable technique

• Note that if ε ∼ N(0, 1) ⇒ −ε ∼ N(0, 1)


• Idea: Use both series (εi )M M
i=1 and (−εi )i=1 to price the derivative
• For m = 1, . . . , M
m
F+ : use εm m
1 , . . . , εN
m
F− : use − εm m
1 , . . . , −εN

• Finally calculate the average of F+


m m
and F−
m m
F+ + F−
Fm =
2
• This often reduces the standard error significantly and is often
much more efficient than doubling the number of trials

Søren Hesel DRM


Probability theory
Stochastic processes The basic method
Itô’s formula Precision and improvements
Monte Carlo simulation

Control variate technique


Assume we have two similar derivatives, A and B
• Derivative A is the security being valued, which can only be done
numerically
• Derivative B is similar to Derivative A, but has an analytic
solution available (this could be the BS-formula)

Idea:
• Determine fB analytically and fBMC and fAMC by MC simulation
(using same ε’s)
• Assume we make the same error for both derivatives in the
numerical procedure
• The error of derivative B is
fB − fBMC
so if this error equals the error of derivative A we must have that
fA − fAMC = fB − fBMC ⇔ fA = fAMC + (fB − fBMC )
So we adjust our estimate by theDRM
Søren Hesel known error.
Probability theory
Stochastic processes The basic method
Itô’s formula Precision and improvements
Monte Carlo simulation

Pros and cons


Pros:
• Can be used to price path-dependent derivatives (Asian options,
Barrier options, look-back options, . . . )
• Can be used to price options depending on more than one
underlying variable
▶ The computation time increases approximately linearly with the
number of variables, whereas the computation time for procedures
like binomial trees and finite difference methods increases
exponentially with the number of variables

Cons:
• Cannot be used directly to, e.g., American/Bermudan options:
▶ Problem: We don’t know when we have hit the critical barrier
• Computation time

Søren Hesel DRM

You might also like