Simulation methods finance_1

Introduction to Monte Carlo in Finance
Giovanni Della Lunga
WORKSHOP IN QUANTITATIVE FINANCE
Bologna - May 12-13, 2016
Giovanni Della Lunga (WORKSHOP IN QUANTITATIVE FINANCE)Introduction to Monte Carlo in Finance Bologna - May 12-13, 2016 1 / 93

Introduction
Outline
1 Introduction
Some Basic Ideas
Theoretical Foundations of Monte
Carlo Simulations
2 Single Asset Path Generation
Deﬁnitions
Exact Solution Advancement
Numerical Integration of SDE
The Brownian Bridge
3 Variance Reduction Methods
Antithetic Variables
Moment Matching
4 Multi Asset Path Generation
Choleski Decomposition
Copula Functions
5 Valuation of European Option with
Stochastic Volatility
Square-Root Diﬀusion: the CIR
Model
The Heston Model

Introduction Some Basic Ideas
What is Monte Carlo?
From a quite general point of view (not a really precise one, actually)
with the term Monte Carlo usually one means a numerical technique
which makes use of random numbers for solving a problem.
For the moment we assume that you can understand, at least
intuitively, what a random number is.
Later we will return to the deﬁnition of a random number, and, as we
shall see, this will lead to absolutely not trivial issues.
Let’s start immediately with some practical examples (we’ll try to give
a more formal deﬁnition later).

Let’s consider two problems apparently very different in nature;
The first one is of probabilistic nature: the assessment of the
premium for an option of European type option on a stock that does
not pay dividends;
The second one is an issue of purely deterministic nature: the
determination of the area enclosed by a plane figure, such as a circle.

Let’s start with the ﬁrst problem. The pricing of an option is usually
dealt with in the context of so-called risk-neutral valuation.
Indicating with f [S(T)] where S is the value of the underlying asset,
the value of the option at maturity T, the value today, f [S(t)], is
given by
f (S(t)) = EQ
[P(t, T)f [S(T)]]
EQ being the risk-neutral expectation value and P(t, T) the discount
function between t and T.
Let’s assume, for simplicity, to know with certainty the value of the
discount function so the problem can be put in the form
f (S(t)) = e−r(T−t)
EQ
[f [S(T)]]

The formulation of the problem makes clear its inherently probabilistic
nature.
The application of the Monte Carlo method in this case is reduced
essentially to the generation of a suﬃciently high number of estimates
of f [S(T)] from which to extract the average value.
To this end it is necessary ﬁrst to introduce a hypothesis on how the
underlying stock price evolves over time;

Let’s suppose for example that the asset price follows a geometric
Brownian motion, according to this hypothesis the rate of change of
the price in a range of inﬁnitesimal time is described by
dS = rSdt + Sσdw
where r is the risk free rate, σ is the volatility of S returns and dw is
a brownian motion;
First of all we choose a discrete version of this SDE:
∆S = rS∆t + Sσ
√
∆T
where is a random number drawn from a normal distribution (we
assume for the moment to have some procedure that allows us to
generate random numbers with probability distribution assigned);

Once we have the simulated value of the underlying at time T, we are
able to derive the value of the option at the same date;
Assuming for example that the option is a CALL we simply write
f [S(T)] = max(S(T) − K, 0)
where K is the strike price.
By repeating the above procedure a very large number of times we
are able to obtain a distribution of values for f [S(T)] from which it is
possible to extract the expectation value ...

Interlude - Let’s start coding...

Out toolbox: Jupyter, Python, R
Python, R
The Python world developed the IPython notebook system.
Notebooks allow you to write text, but you insert code blocks as
”cells” into the notebook.
A notebook is interactive, so you can execute the code in the cell
directly!
Recently the Notebook idea took a much enhanced vision and scope,
to explicitly allow languages other than Python to run inside the cells.
Thus the Jupyter Notebook was born, a project initially aimed at
Julia, Python and R (Ju-Pyt-e-R). But in reality many other
languages are supported in Jupyter.

The IRKernel
To enable support of a new language means that somebody has to
write a ”kernel”.
The kernel for R is called IRKernel (available at github).
How do you use Jupyter?
Once Jupyter is up and running, you interact with it on a web page.

Benefits of using Jupyter
Jupyter was designed to enable sharing of notebooks with other
people. The idea is that you can write some code, mix some text with
the code, and publish this as a notebook. In the notebook they can
see the code as well as the actual results of running the code.
This is a nice way of sharing little experimental snippets, but also to
publish more detailed reports with explanations and full code sets. Of
course, a variety of web services allows you to post just code snippets
(e.g. gist). What makes Jupyter different is that the service will
actually render the code output.
One interesting benefit of using Jupyter is that Github magically
renders notebooks. See for example, the github Notebook gallery.

Notebook
GitHub : polyhedron-gdl;
Notebooks : mcs 1;

Let’s now consider the second
problem;
Take a circle inscribed in a unit
square. Given that the circle
and the square have a ratio of
areas that is π/4, the value of π
can be approximated using a
Monte Carlo method:

1 Draw a square on the ground,
then inscribe a circle within it.
2 Uniformly scatter some objects
of uniform size (grains of rice or
sand) over the square.
3 Count the number of objects
inside the circle and the total
number of objects.
4 The ratio of the two counts is an
estimate of the ratio of the two
areas, which is π/4. Multiply
the result by 4 to estimate π.

In this procedure the domain of
inputs is the square that
circumscribes our circle.
We generate random inputs by
scattering grains over the square
then perform a computation on
each input (test whether it falls
within the circle).
Finally, we aggregate the results
to obtain our ﬁnal result, the
approximation of π.

Notebook
Notebooks : mcs 2;

Monte Carlo is Integration!
There is a formal connection between the use of the Monte Carlo
method and the concept of integration of a function.
First of all we observe how the problems discussed in the previous
paragraph can be attributed both to the calculation of integrals.
The case related to the area of the circle is evident
The price of an option as we have seen is nothing more than the
discounted value of the expectation value of the price at maturity, the
underlying risk factor (the stock price) is distributed according to a
log-normal distribution, therefore, we have (for the CALL case):
C(t, S) =
e−r(T−t)
√
2π
+∞
−∞
h S exp (r − q − σ2
/2)(T − t) − σx
√
T − t) e− 1
2
x2
dx
where h(S) = |S − K|+.

More in general we can state that each extraction of a sample of
random numbers can be used as an estimator of an integral.
As an example consider the case relating to the integration of a
function of a real variable;
by a suitable change of variable, we can always bring us back to the
simplest case in which the integration interval is between 0 and 1:
I =
1
0
f (x)dx

The key point of our argument is to recognize that the expression
written above is also the expectation value of the function f at values
of a random variable uniformly distributed in the range [0, 1].
It becomes possible to estimate the value of our integral using an
arithmetic mean of n values of f (Ui ) where each Ui is a sample from
a uniform distribution in [0, 1].
In other words we can say that the quantity
˜In =
1
n
n
i=1
f (Ui )
is an unbiased estimator of I.

The variance of the estimator is
var ˜In =
var(f (Ui )
n
the mean square error of the estimator, which can be interpreted as
the mean square error of the Monte Carlo simulation, decreases with
increasing n.
This result is completely independent of the dimensionality of the
problem.
It’s this last characteristic that makes attractive the Monte Carlo
method for solving problems with a large number of dimensions.
In this case typically the Monte Carlo method converge to the ﬁnal
value faster than the traditional numerical methods.

Pricing a Call Option
It’s worth to recast the pricing problem into a simple integral
formulation in order to gain some insight into the general problem;
So let’s consider again the payoﬀ of a simple plain vanilla option
e−rT
EQ
[h(ST )] = e−rT
EQ
h S0elog(ST /S0)
By a simple application of Ito’s lemma is easy to demonstrate that
the variable X = log(ST /S0) has a normal distribution with mean
m = (r − 1
2σ2)T and variance s = σ2T.
So we can write
C(S, t) = e−rT
S0
+∞
0
max[eX
− K, 0]e−
(X−m)2
2s2 dX

Let’s now recall the fundamental probability integral transform
which relates to the result that data values that are modelled as being
random variables from any given continuous distribution can be
converted to random variables having a uniform distribution
This result states that if a random variable X has a continuous
distribution for which the cumulative distribution function (CDF) is
FX . Then the random variable Y deﬁned as
Y = FX (X)
has a uniform distribution;

This means that, in our case, we can say that the variable
u = Φ[X; m, u], u → 1 when X → +∞
has a uniform distribution;
From the previous relation we ﬁnd (within a normalization factor)
du =
dΦ[X; m, u]
dX
dX ⇒ dX =
1
e−
(X−m)2
2s2
du
and ﬁnally we can write our integral in the form
C(S, t) =
1
0
f (u)du
where f (u) = e−rT max[S0 exp(Φ−1(u; m, s)) − K, 0]

Pricing a Call Option - The Python Code
def f (u , S0 , K, r , sigma , T) :
m = ( r − .5∗ sigma ∗ sigma )∗T
s = sigma ∗ s q r t (T)
f u = exp(−r ∗T) ∗
np . maximum( S0∗exp ( scnorm . ppf (u , m, s ))−K, 0 )
return f u
u = rand (1000000)
f u = f (u , S0 ,K, r , sigma ,T)
print mean( f u )

Pricing a Call Option - The Integrand Function

Notebook
Notebooks : mcs 3;

Introduction Theoretical Foundations of Monte Carlo Simulations
Feynman–Kac formula
The Feynman–Kac formula named after Richard Feynman and
Mark Kac, establishes a link between parabolic partial diﬀerential
equations (PDEs) and stochastic processes.
It oﬀers a method of solving certain PDEs by simulating random paths
of a stochastic process. Conversely, an important class of expectations
of random processes can be computed by deterministic methods.
Consider the PDE
∂u
∂t
(x, t)+µ(x, t)
∂u
∂x
(x, t)+ 1
2 σ2
(x, t)
∂2
u
∂x2
(x, t)−V (x, t)u(x, t)+f (x, t) = 0
subject to the terminal condition
u(x, T) = ψ(x)

Feynman–Kac formula
Then the Feynman–Kac formula tells us that the solution can be
written as a conditional expectation
u(x, t) = EQ
T
t
e− r
t V (Xτ ,τ) dτ
f (Xr , r)dr + e− T
t V (Xτ ,τ) dτ
ψ(XT ) Xt = x
under the probability measure Q such that X’ is an Ito process driven
by the equation
dX = µ(X, t) dt + σ(X, t) dW Q

Valuing a derivative contract
A derivative can be perfectly replicated by means of a self-financing
dynamic portfolio whose value exactly matches all of the derivative
flows in every state of the world. This approach shows that the values
of the derivative (and of the portofolio) solves the following PDE
∂f
∂t
+
∂f
∂S
rS +
1
2
∂f
∂S
σ2
S2
= fr (1)
with the terminal condition at T that is the derivative’s payoff
f (T, S(T)) = payoff

According to the Feynmann-Kac formula, if f (t0, S(t0)) solves the
B-S PDE, then it is also solution of
f (t0, S(t0)) = E e−r(T−t0)
f (T, S(T)|Ft0
i.e. it’s the expected value of the discounted payoﬀ in a probability
measure where the evolution of the asset is
dS = rSdt + σSdw
This probability measure is the Risk Neutral Measure

Since there exist such an equivalence, we can compute option prices
by means of two numerical methods
PDE: ﬁnite diﬀerence (explicit, implicit, crank-nicholson)suitable for
optimal exercise derivatives;
Integration
Quadrature Methods;
Monte Carlo Methods

Single Asset Path Generation
Outline
1 Introduction
Some Basic Ideas
Carlo Simulations
Deﬁnitions
The Brownian Bridge
Moment Matching
Copula Functions
Model
The Heston Model

Single Asset Path Generation Definitions
Scenario Nomenclature
As we have have repeated ad nauseam, the value of a derivative
security with payoffs at a known time T is given by the expectation of
its payoff, normalized with the numeraire asset;
The value of an European derivative whose payoff depends on a single
underlying process, S(t), is given by
V [S(0), 0] = B(0)EB V (S(T), T)
B(T)
(2)
where all stochastic processes in this expectation are consistent with
the measure induced by the numeraire asset B(t).

We consider an underlying process S(t) described by the sde
dS(t) = a(S, t)dt + b(S, t)dW (3)
A scenario is a set of values ˆSj (ti ), i = 1, . . . , I that are an
approximation to the j − th realization,Sj (ti ), of the solution of the
sde evaluated at times 0 ≤ ti ≤ T, i = 1, . . . , I;
A scenario is also called a trajectory
A trajectory can be visualized as a line in the state-vs-time plane
describing the path followed by a realization of the stochastic process
(actually by an approzimation to the stochastic process).

For example, the Black and Scholes model assumes a market in which
the tradable assets are:
A risky asset, whose evolution is driven by a geometric brownian motion
dS = µSdt + σSdw ⇒ S(T) = S(t0)e(µ− 1
2 σ2
)(T−t0)+σ[w(T)−w(t0)]
(4)
the money market account, whose evolution is deterministic
dB = Brdt ⇒ B(T) = B(t0)er(T−t0)
(5)

Scenario Contruction
There are several ways to construct scenario for pricing
Constructing a path of the solution to the SDE at times ti by exact
advancement of the solution;
This method is only possible if we have an analytical expression for the
solution of the stochastic diﬀerential equation
Approximate numerical solution of the stochastic diﬀerential
equation;
This is the method of choice if we cannot use the previous one; Just as
in the case of ODE there are numerical techniques for discretizing and
solving SDE.

Single Asset Path Generation Exact Solution Advancement
Example: Log-normal process with constant drift and volatility
SDE ⇒
dS
S
= µdt + σdw
SOLUTION ⇒ S(T) = S(t0)e(µ−1
2
σ2)(T−t0)+σ[w(T)−w(t0)]
How to obtain a sequence of Wiener process?
w(ti ) = w(ti−1) + ti − ti−1Z Z ∼ N(0, 1)

Deﬁning the outcomes of successive drawings of the random variable
Z corresponding to the j − th trajectory by Zj
i , we get the following
recursive expression for the j − th trajectory of S(t):
Sj
(ti ) = Sj
(ti−1)exp µ −
1
2
σ2
(ti − ti−1) + σ ti − ti−1Zj
i

Some observations are in order...
The set w(ti ) must be viewed as the components of a vector of
random variables with a multidimensional distribution. This means
that for a ﬁxed j Zi
j are realizations of a multidimensional standard
normal random variable which happen to be independent;
Wheter we view the Zi
j as coming from a multidimensional
distribution of independent normals or as drawings from a single
one-dimensional distribution does not aﬀect the outcome as long as
the Zi
j are generated from pseudo-random numbers;
This distinction, however, is conceptually important and it becomes
essential if we generate the Zi
j not from pseudo-random
numbers but from quasi-random sequences.

Single Asset Path Generation Numerical Integration of SDE
The numerical integration of the SDE by finite difference is another
way of generating scenarios for pricing;
In the case of the numerical integration of ordinary differential
equations by finite differences the numerical scheme introduces a
discretization error that translates into the numerical solution
differing from the exact solution by an amount proportional to a
power of the time step.
This amount is the truncation error of the numerical scheme.

In the case of the numerical integration of SDE by ﬁnite diﬀerences
the interpretation of the numerical error introduced by the
discretization scheme is more complicated;
Unlike the case of ODE where the only thing we are interested in
computing is the solution itself, when dealing with SDE there are two
aspects that interest us:
One aspect is the accuracy with which we compute the trajectories or
paths of a realization of the solution
The other aspect is the accuracy with which we compute functions of
the process such as expectations and moments.

The order of accuracy with which a given scheme can approximate
trajectories of the solution is not the same as the accuracy with which
the same scheme can approximate expectations and moments of
functions of the trajectories;
The convergence of the numerically computed trajectories to the
exact trajectories is called strong convergence and the order of the
corresponding numerical scheme is called order of strong convergence;
The convergence of numerically computed functions of the stochastic
process to the exact values is called weak convergence and the related
order is called order of weak convergence.

We assume that the stock price St is driven by the stochastic
diﬀerential equation (SDE)
dS(t) = µ(S, t)dt + σ(S, t)dWt (6)
where Wt is, as usual, Brownian motion.
We simulate St over the time interval [0; T], which we assume to be
is discretized as 0 = t1 < t2 < · · · < tm = T, where the time
increments are equally spaced with width dt.
Equally-spaced time increments is primarily used for notational
convenience, because it allows us to write ti − ti−1 as simply dt. All
the results derived with equally-spaced increments are easily
generalized to unequal spacing.

Euler Scheme
The simplest way to discretize the process in Equation (6) is to use
Euler discretization
EULER ⇒ ˆS(ti+1) = ˆS(ti )+µ[ˆS(ti ), ti ]∆t+σ[ˆS(ti ), ti ] (w(ti+1) − w(ti ))
Milshstein Scheme
MILSHSTEIN ⇒ EULER+
1
2
σ[ˆS(ti )]
∂σ[ˆS(ti )]
∂S
(w(ti+1) − w(ti ))2
− ∆t
This scheme is described in Glasserman and in Kloeden and Platen for general
processes, and in Kahl and Jackel for stochastic volatility models. The scheme
works for SDEs for which the coeﬃcients µ(St ) and σ(St ) depend only on S, and
do not depend on t directly

Notebook
Notebooks :
mcs sde solution;
Code : mcs sde solution.py;

Single Asset Path Generation The Brownian Bridge
The Brownian Bridge
Assume you have a Wiener process deﬁned by a set of time-indexed
random variables W (t1), W (t2), ..., W (tn).
How do you insert a random variable W (tk) where ti ≤ tk ≤ ti+1 into
the set in such a manner that the resulting set still consitutes a
Wiener process?
The answer is: with a Brownian Bridge!
The Brownian Bridge is a sort of interpolation tat allows you to
introduce intermediate points in the trajectory of a Wiener process.

The Brownian Bridge
Brownian Bridge Construction
Given W (t) and W (t + δt1 + δt2) we want to ﬁnd W (t + δt1);
We assume that we can get the middle point by a weighted average
of the two end points plus an independent normal random variable:
W (t + δt1) = αW (t) + βW (t + δt1 + δt2) + λZ
where α, β and λ are constants to be determined and Z is a standard
normal random variable.

The Brownian Bridge
We have to satisfy the following conditions:



cov[W (t + ∆t1), W (t)] = min(t + ∆t1, t) = t
cov[W (t + ∆t1), W (t + ∆t1 + ∆t2)] = t + ∆t1
var[W (t + ∆t1)] = t + ∆t1



α + β = 1
αt + β(t + ∆t1 + ∆t2) = t + ∆t1
α2
t + 2αβt + β2
(t + ∆t1 + ∆t2) + λ2
= t + ∆t1
which are equivalent to:



α = ∆t2
∆t1+∆t2
β = 1 − α
γ =
√
∆t1α

The Brownian Bridge
We can use the brownian bridge to generate a Wiener path and then
use the Wiener path to produce a trajectory of the process we are
interested in;
The simplest strategy for generating a Wiener path using the
brownian bridge is to divide the time span of the trajectory into two
equal parts and apply the brownian bridge construction to the middle
point. We then repeat the procedure for the left and right sides of the
time interval.

The Brownian Bridge
Notice that as we fill in the Wiener path, the additional variance of
the normal components we add has decreasing value;
Of course the total variance of all the Wiener increments does not
depend on how we construct the path, however the fact that in the
brownian bridge approach we use random variables that are multiplied
by a factor of decreasing magnitude means that the importance of
those variables also decreases as we fill in the path;
The dimension of the random variables with larger variance need to
be sampled more efficiently than the dimension with smaller variance;

Notebook
Code :
mcs brownian bridge.py;

Variance Reduction Methods
Outline
1 Introduction
Some Basic Ideas
Carlo Simulations
Deﬁnitions
The Brownian Bridge
Moment Matching
Copula Functions
Model
The Heston Model

In numerical analysis there is an informal concept known as the curse
of dimensionality. This refers to the fact that the computational load
(CPU time, memory requirements, etc...) may increase exponentially
with the numer of dimensions of the problem;
The computational work needed to estimate the expectation through
MC does not depend explicitly on the dimensionality of the problem,
this means that there is no curse of dimensionality in MC
computation when we are only interested in a simple expectation (this
is the case with european derivatives, thinghs are more complicated
with early exercise features).

The efficiency of a simulation refers to the computational cost of
achieving a given level of confidence in the quantity we are trying to
estimate;
Both the uncentainty in the estimation of the expectation as well as
the uncertainty in the error of our estimation depend on the variance
of the population from which we sample;
However whatever we do to reduce the variance of the population will
most likely tend to increase the computational time per MC cycle;
As a result in order to make a fair comparison between different
estimators we must take into account not only their variance but also
the computational work for each MC cycle.

Suppose we want to compute a parameter P, for example the price of
a derivative security, and that we have a choice between two types of
Monte Carlo estimates which we denote by
ˆP1i i = 1, . . . , n ˆP2i i = 1, . . . , n
Suppose that both are unbiased, so that
E[ˆP1i ] = P1 E[ˆP2i ] = P2
but
σ1 < σ2

A sample mean of n replications of P1 gives a more precise estimate
of P than does a sample mean of n replications of P2;
This oversimplifies the comparison because it fails to capture possible
differences in the computational effort required by the two estimators;
Generating n replications of P1 may be more time-consuming than
generating n replications of P2;
Smaller variance is not sufficient grounds for preferring one estimator
over another!

To compare estimators with diﬀerent computational requirements, we
argue as follows;
Suppose the work required to generate one replication of Pj is a
constant, bj (j = 1, 2);
With computing time t , the number of replications of Pj that can be
generated is t/bj ;
The two estimators available with computing time t are therefore:
b1
t
t/b1
i=1
ˆP1
i
b2
t
t/b2
i=1
ˆP2
i

For large t these are approximately normally distributed with mean P
and with standard deviations
σ1
b1
t
σ2
b2
t
Thus for large t the ﬁrst estimator should be preferred over the
second if
σ2
1b1 < σ2
2b2
The important quantity is the product of variance and work per run;

If we do nothing about eﬃciency, the number of MC replications we
need to achieve acceptable pricing acccuracy may be surprisingly large;
As a result in many cases variance reduction techiques are a practical
requirement;
From a general point of view these methods are based on two
principal strategies for reducing variance:
Taking advantage of tractable features of a model to adjust or correct
simulation output
Reducing the variability in simulation input

The most commonly used strategies for variance reduction are the
following:
Antithetic variates
Moment Matching
Control variates
Importance sampling
Stratiﬁcation
Low-discrepancy sequences

Variance Reduction Methods Antithetic Variables
Variance Reduction Methods - Antithetic Variates
In this case we construc the estimator by using two brownian
trajectories that are mirror images of each other;
This causes cancellation of dispersion;
This method tends to reduce the variance modestly but it is
extremely easy to implement and as a result very commonly used;
For the antithetic method to work we need V + and V − to be
negatively correlated;
this will happen if the payoﬀ function is a monotonic function of Z;

Variance Reduction Methods Antithetic Variables
Variance Reduction Methods - Antithetic Variates
To apply the antithetic variate technique, we generate standard
normal random numbers Z and define two set of samples of the
undelying price
S+
T = S0e(r−σ2/2)T+σ
√
TZ
S−
T = S0e(r−σ2/2)T+σ
√
T(−Z)
Similarly we define two sets of discounted payoff samples ...
V +
T = max[S+
(T) − K, 0] V −
T = max[S−
(T) − K, 0]
... and at last we construct our mean estimator by averaging these
samples
¯V0 =
1
n
n
j=1
1
2
V +
j + V −
j

Variance Reduction Methods Moment Matching
Variance Reduction Methods - Moment Matching
Let zi , i = 1, ..., n, denote an independent standard normal random
vector used to drive a simulation.
The sample moments will not exactly match those of the standard
normal. The idea of moment matching is to transform the zi to
match a finite number of the moments of the underlying population.
For example, the first and second moment of the normal random
number can be matched by defining
˜zi = (zi − ˜z)
σz
sz
+ µz, i = 1, .....n (7)
where ˜z is the sample mean of the zi and σz is the population
standard deviation, sz is the sample standard deviation of zi , and
µz”¿ is the population mean.

Multi Asset Path Generation
Outline
1 Introduction
Some Basic Ideas
Carlo Simulations
Deﬁnitions
The Brownian Bridge
Moment Matching
Copula Functions
Model
The Heston Model

Multi Asset Path Generation Choleski Decomposition
The Choleski Decomposition makes an appearance in Monte Carlo
Methods where it is used to simulating systems with correlated
variables.
Cholesky decomposition is applied to the correlation matrix, providing
a lower triangular matrix A, which when applied to a vector of
uncorrelated samples, u, produces the covariance vector of the
system. Thus it is highly relevant for quantitative trading.
The standard procedure for generating a set of correlated normal
random variables is through a linear combination of uncorrelated
normal random variables;
Assume we have a set of n independent standard normal random
variables Z and we want to build a set of n correlated standard
normals Z with correlation matrix Σ
Z = AZ, AAt
= Σ

We can ﬁnd a solution for A in the form of a triangular matrix





A11 0 . . . 0
A21 A22 . . . 0
...
...
... . . .
An1 An2 . . . Ann





diagonal elements
aii = Σii −
i−1
k=1
a2
ik
oﬀ-diagonal elements
aij =
1
aii
Σij −
i−1
k=1
aikajk

For example, for a two-dimension random vector we have simply
A =
σ1 0
σ2ρ σ2 1 − ρ2
say one needs to generate two correlated normal variables x1 and x2
All one needs to do is to generate two uncorrelated Gaussian random
variables z1 andz2 and set
x1 = z1
x2 = ρz1 + 1 − ρ2z2

Multi Asset Path Generation Copula Functions
Copula Functions
Why Copula?
A copula is a function that links univariate marginals to their
multivariate distribution;
Non-linear dependence
Be able to measure dependence for heavy tail distributions
Very ﬂexible: parametric, semi-parametric or non-parametric
Be able to study asymptotic properties of dependence structures
Many others... (ask to Cherubini for everything you would like to
know about copula)

Copula Functions
In most applications, the distribution is assumed to be a multivariate
gaussian or a log-normal distribution for tractable calculus, even if the
gaussian assumption may not be appropriate.
Copulas are a powerful tool for finance, because the modelling
problem can be splitted into two steps:
• the first step deals with the identification of the marginal
distributions;
• and the second step consists in defining the appropriate copula in
order to represent the dependence structure in a good manner.

Copula Functions

Notebook
Notebook :
mcs multi asset path;

Valuation of European Option with Stochastic Volatility
Outline
1 Introduction
Some Basic Ideas
Carlo Simulations
Deﬁnitions
The Brownian Bridge
Moment Matching
Copula Functions
Model
The Heston Model

Valuation of European Option with Stochastic Volatility Square-Root Diffusion: the CIR Model
CIR Model
In this section, we consider the stochastic short rate model MCIR85
of Cox- Ingersoll-Ross which is given by the SDE:
drt = κr (θr − rt)dt + σr
√
rtdZt (8)
To simulate the short rate model, it has to be discretized. To this
end, we divide the given time interval [0, T] in equidistant
sub-intervals of length t such that now t ∈ {0, ∆t, 2∆t, . . . , T}, i.e.
there are M + 1 points in time with M = T/t.
The exact transition law of the square-root diffusion is known.
Consider the general square- root diffusion process
dxt = κ(θ − xt)dt + σ
√
xtdZt (9)

CIR Model
It can be show that xt, given xs with s = t − ∆t, is distributed
according to
xt =
σ2(1 − e−κ∆t)
4κ
χ 2
d
4−κ∆t
σ2(1 − e−κ∆t)
xs
where χ 2
d denotes a non-central chi-squared random variable with
d =
4θκ
σ2
degrees of freedom and non-centrality parameter
l =
4−κ∆t
σ2(1 − e−κ∆t)
xs

CIR Model
For implementation purposes, it may be convenient to sample a
chi-squared random variable χ2
d instead of a non-central chi-squared
one, χ 2
d .
If d > 1, the following relationship holds true
χ 2
d (l) = (z +
√
l)2
+ χ2
d−1
where z is an independent standard normally distributed random
variable.
Similarly, if d ≤ 1, one has
χ 2
d (l) = χ2
d+2N
where N is now a Poisson-distributed random variable with intensity
l/2. For an algorithmic representation of this simulation scheme refer
to Glasserman, p. 124.

CIR Model: Pricing ZCB
A MC estimator for the value of the ZCB at t is derived as follows.
Consider a certain path i of the I simulated paths for the short rate
process with time grid t ∈ {0, ∆t, 2∆t, . . . , T}.
We discount the terminal value of the ZCB, i.e. 1, step-by-step
backward. For t < T and s = t − ∆t we have
Bs,i = Bt,i e−rt +rs
2
∆t
The MC estimator of the ZCB value at t is
BMC
t =
1
I
I
i=1
Bt,i

CIR Model: Pricing ZCB
The present value of the ZCB in the CIR model takes the form:
B0(T) = b1(T)e−b2(T)r0
where
b1(T) =
2γ exp((κr + γ)T/2)
2γ + (κr + γ)(eγT − 1)
2κr θr
σ2
r
b2(T) =
2(eγT − 1)
2γ + (κr + γ)(eγT − 1)
γ = κ2
r + 2σ2
r

Notebook
Notebook : n06 mcs cir;

Valuation of European Option with Stochastic Volatility The Heston Model
The Heston Model
Stochastic volatility models are those in which the variance of a
stochastic process is itself randomly distributed.
The models assumes that the underlying security’s volatility is a
random process, governed by state variables such as the price level of
the underlying security, the tendency of volatility to revert to some
long-run mean value, and the variance of the volatility process itself,
among others.
Stochastic volatility models are one approach to resolve a
shortcoming of the Black–Scholes model.
In particular this model cannot explain long-observed features of the
implied volatility surface such as volatility smile and skew, which
indicate that implied volatility does tend to vary with respect to strike
price and expiry.
By assuming that the volatility of the underlying price is a stochastic
process rather than a constant, it becomes possible to model
derivatives more accurately.

The Heston Model
By assuming that the volatility of the underlying price is a stochastic
process rather than a constant, it becomes possible to model derivatives
more accurately.
Heston model
CEV model
SABR volatility model
GARCH model

The Heston Model
In this section we are going to consider the stochastic volatility model
MH93 with constant short rate.
This section values European call and put options in this model by
MCS.
As for the ZCB values, we also have available a semi-analytical pricing
formula which generates natural benchmark values against which to
compare the MCS estimates.

The Heston Model
The basic Heston model assumes that St, the price of the asset, is
determined by a stochastic process:
dSt = µSt dt +
√
νtSt dW S
t
where νt, the instantaneous variance, is a CIR process:
dνt = κ(θ − νt) dt + ξ
√
νt dW ν
t
and dW S
t , dW ν
t are Wiener process with correlation ρ, or equivalently,
with covariance ρdt.

The Heston Model
The parameters in the above equations represent the following:
µ is the rate of return of the asset.
θ is the long variance, or long run average price variance; as t tends
to inﬁnity, the expected value of νt tends to θ.
κ is the rate at which νt reverts to θ.
ξ is the volatility of the volatility, or vol of vol, and determines the
variance of νt.
If the parameters obey the following condition (known as the Feller
condition) then the process νt is strictly positive
2κθ > ξ2

The Heston Model
The correlation introduces a new problem dimension into the
discretization for simulation purposes.
To avoid problems arising from correlating normally distributed
increments (of S) with chi-squared distributed increments (of v), we
will in the following only consider Euler schemes for both the S and v
process.
This has the advantage that the increments of v become normally
distributed as well and can therefore be easily correlated with the
increments of S.

The Heston Model
we consider two discretization schemes for S and seven discretization
schemes for v.
For S we have the simple Euler discretization scheme (with
s = t − ∆t)
St = Ss er∆t
+
√
vt
√
∆tz1
t
As an alternative we consider the exact log Euler scheme
St = Sse(r−vt /2)∆t+
√
vt
√
∆tz1
t
This one is obtained by considering the dynamics of log St and
applying Ito’s lemma to it.

The Heston Model
These schemes can be combined with any of the following Euler
schemes for the square-root diﬀusion (x+ = max[0, x]):
Full Truncation
˜xt = ˜xs + κ(θ − ˜x+
s )∆t + σ ˜x+
s
√
∆tzt, xt = ˜x+
t
Partial Truncation
˜xt = ˜xs + κ(θ − ˜xs)∆t + σ ˜x+
s
√
∆tzt, xt = ˜x+
t
Truncation
xt = max 0, ˜xs + κ(θ − ˜xs)∆t + σ ˜xs
√
∆tzt
Reﬂection
˜xt = |˜xs| + κ(θ − |˜xs|)∆t + σ |˜xs|
√
∆tzt, xt = |˜xt|

The Heston Model
Hingham-Mao
˜xt = ˜xs + κ(θ − ˜xs)∆t + σ |˜xs|
√
∆tzt, xt = |˜xt|
Simple Reflection
˜xt = ˜xs + κ(θ − ˜xs)∆t + σ ˜xs
√
∆tzt
Absorption
˜xt = ˜x+
s + κ(θ − ˜x+
s )∆t + σ ˜x+
s
√
∆tzt, xt = ˜x+
t
In the literature there are a lot of tests and numerical studies
available that compare efficiency and precision of different
discretization schemes.

Notebook
Notebook : n07 mcs heston;

Simulation methods finance_1

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Simulation methods finance_1 (20)

More from Giovanni Della Lunga (20)

Recently uploaded (20)

Simulation methods finance_1