You are on page 1of 31

Exotic Options Variance reduction Antithetic variates

wi3425TU—Monte Carlo methods

L.E. Meester

Week 4

1/ 31
Exotic Options Variance reduction Antithetic variates

Week 4—Program for this week

1 Pricing exotic options by simulation


Exotic options: quick overview (from Chapter 19)
Pricing, an example
Some advantages and disadvantages
2 Variance reduction—the big picture (§21.2)
Efficiency from several angles
3 Antithetic variates
Principle: applied to the π-example
Analysing the uniform case
Analysing the normal case
How to tell when/whether AV reduce the variance?
Interpretation for path dependent options
Don’t forget: Do the weekly practice material and upload your
solutions and we’ll provide you with feedback!
2/ 31
Exotic Options Variance reduction Antithetic variates

Points of attention for Monte Carlo simulations (cont’d)


5 Consistency. In some situations a quantity may be estimated in
more than one way or using several methods. The resulting
estimates should then be consistent: differences between them,
expressed in the number of (appropriately computed) standard
errors, should not be too big. If they are, then something is wrong
and should be checked. . .
6 Mistakes, checks. Everybody makes mistakes: errors of reasoning,
programming errors, etcetera. So, use every opportunity you have to
catch those mistakes. Never blindly believe the final answer your
program spits out, but scrutinize intermediate results for errors.
Usually, there are enough things around that you might check. If
necessary, create a testing opportunity, for example a special case
for which you know the answer.
7 Random seeds. Initialise the randomgenerator(s) so that your
results are reproducible. Set a seed chosen by yourself: if we all
mimick Higham and start our program with rand(’state’,100)
and randn(’state’,100), how random is that?
3/ 31
Exotic Options Variance reduction Antithetic variates Overview Pricing Pros and cons

1 Pricing exotic options by simulation


Exotic options: quick overview (from Chapter 19)
Pricing, an example
Some advantages and disadvantages

2 Variance reduction—the big picture (§21.2)


Efficiency from several angles

3 Antithetic variates
Principle: applied to the π-example
Analysing the uniform case
Analysing the normal case
How to tell when/whether AV reduce the variance?
Interpretation for path dependent options

4/ 31
Exotic Options Variance reduction Antithetic variates Overview Pricing Pros and cons

Chapter 19: Exotic options

1 American options: early exercise possible. Good idea?


Call: no In the money? (St > E ): sell one short.
Put: depends, it can be better in certain situations.
Exercise strategy: involves (debating) the decision whether to
exercise at time t; this involves

the (possibly whole) price path {Su , 0 ≤ u ≤ t}.

2 Barrier options. For example, the up-and-out-call: a call


option that becomes void if a certain price level B is crossed.
There are many variations:
(up/down)-(out/in)-(call/put) options;
double or partial barriers;
Parisian options.

5/ 31
Exotic Options Variance reduction Antithetic variates Overview Pricing Pros and cons

3 Lookback options:
Define price maximum S max := max0≤t≤T St
and price minimum S min := min0≤t≤T St .
Floating strike lookback call: payoff at time T of S(T ) − S min .
Fixed strike lookback call: payoff at T of max(S max − E , 0).
4 Asian options: based on average price over whole price path.
5 Bermudan options, Shout options, Indonesian options, et
cetera.

What makes exotic options special/complex:


the nature of the path dependency—often the whole path
plays a role;
early exercise may be possible as well.

6/ 31
Exotic Options Variance reduction Antithetic variates Overview Pricing Pros and cons

1 Pricing exotic options by simulation


Exotic options: quick overview (from Chapter 19)
Pricing, an example
Some advantages and disadvantages

2 Variance reduction—the big picture (§21.2)


Efficiency from several angles

3 Antithetic variates
Principle: applied to the π-example
Analysing the uniform case
Analysing the normal case
How to tell when/whether AV reduce the variance?
Interpretation for path dependent options

7/ 31
Exotic Options Variance reduction Antithetic variates Overview Pricing Pros and cons

Asset path model: geometric Brownian motion

Model for (asset) prices S(t0 ), . . . , S(tn ) at time instances


t0 = 0 < t1 < · · · < tn = T based on empirical “fact” that the
returns S(ti+1 )−S(ti )
S(ti ) (appear to) have a normal distribution.
This motivates the geometric Brownian motion model, whose
discrete grid values can be represented as

S(ti+1 ) = S(ti ) exp (µ − σ 2 /2)(ti+1 − ti ) + σ ti+1 − ti · Zi ,


p 

with µ: the drift,


σ: the volatility,
Z0 , . . . , Zn : independent N (0, 1) random variables.

8/ 31
Exotic Options Variance reduction Antithetic variates Overview Pricing Pros and cons

Pricing exotic options by simulation

For many this is the only way to determine their prices.


Simulate paths on a grid i∆t, i = 0, 1, 2, . . . and according to
the risk-neutral distribution:
 √ 
S((i + 1)∆t) = S(i∆t) exp (r − σ 2 /2)∆t + σ ∆t · Zi .

Estimate prices based on these paths.


N.B.: a discrete time sampling of the continuous time model.

To be chosen: grid size ∆t. Considerations:


Smaller ∆t: more accurate path–representation,
but requires more CPU time: halving ∆t means (roughly)
doubling the work.

9/ 31
Exotic Options Variance reduction Antithetic variates Overview Pricing Pros and cons

Example: pricing the up-and-out-call

Part of code from UpAndOutCall.m:

for i = 1:M
Z= randn(N,1);
Svals=S*cumprod(exp((r-0.5*sigma^2)*Dt+sigma*sqrt(Dt)*Z));
Smax = max(Svals);
notout=Smax<B;
V(i) = notout*exp(-r*T)*max(Svals(end)-E,0);
end

With ∆t = 0.01 and M = 104 we find


an estimated price of 0.1117 with s.e. 0.0031.
Exact price in 4 decimals is 0.0983. Is this OK?

10/ 31
Exotic Options Variance reduction Antithetic variates Overview Pricing Pros and cons

Problem: Some barrier-crossings may be “missed” by the discretely


sampled path; consequence here: price is overestimated.

HigF71sampled.m: coarse (blue) versus fine grid (red)

11/ 31
Exotic Options Variance reduction Antithetic variates Overview Pricing Pros and cons

1 Pricing exotic options by simulation


Exotic options: quick overview (from Chapter 19)
Pricing, an example
Some advantages and disadvantages

2 Variance reduction—the big picture (§21.2)


Efficiency from several angles

3 Antithetic variates
Principle: applied to the π-example
Analysing the uniform case
Analysing the normal case
How to tell when/whether AV reduce the variance?
Interpretation for path dependent options

12/ 31
Exotic Options Variance reduction Antithetic variates Overview Pricing Pros and cons

Pricing exotic options by simulation:


Advantages:
flexibility;
can price options that cannot be priced any other way.
Disadvantages:
often very CPU intensive;
possibly biased (often intractable).

Coming weeks:
More examples.
Analysis of bias in some special cases.

13/ 31
Exotic Options Variance reduction Antithetic variates Efficiency

1 Pricing exotic options by simulation


Exotic options: quick overview (from Chapter 19)
Pricing, an example
Some advantages and disadvantages

2 Variance reduction—the big picture (§21.2)


Efficiency from several angles

3 Antithetic variates
Principle: applied to the π-example
Analysing the uniform case
Analysing the normal case
How to tell when/whether AV reduce the variance?
Interpretation for path dependent options

14/ 31
Exotic Options Variance reduction Antithetic variates Efficiency

Variance reduction—the big picture (§21.2)

Recall the π-example from week 1:


Method 1 to estimate π: based on X1 with
E [X1 ] = π, Var(X1 ) = π(4 − π) ≈ 2.697.

Method 2: X2 = 4 1 − U 2 with standard uniform U,
E [X2 ] = π, Var(X2 ) = 32/3 − π 2 ≈ 0.797.
The variance Var(X2 ) is 0.797
2.697 ≈ 0.296 of Var(X1 ), which means
2.697
that Method 2 is 0.797 ≈ 3.4 times as efficient as Method 1.
Equivalent manifestations:
Method 2 is 3.4 times as efficient as Method 1.

With same M: s.e. of Method 2 is factor 3.4 ≈ 1.8 smaller.
Method 2 at M replications: as accurate as 1 with 3.4 × M.
Method 2 is “3.4 times as fast.” (provided . . . )

15/ 31
Exotic Options Variance reduction Antithetic variates Efficiency

Efficiency is determined by:


the number of replications;
the amount of work (CPU time) per replication.
A more refined comparison:
Method 1:
CPU time (or cost) per replication: 1
Var(X1 ) = σ12
Method 2:
CPU time (or cost) per replication: R2
Var(X2 ) = σ22 = R1 σ22
Supose, the computing budget is N.
σ1
Method 1: can do N runs; s.e. is √ .
N √

Method 2: M = N
R2 runs; s.e. σ2

M
= √R1 σ1 = R1 R 2 · σ1

N
.
N/R2

Conclusion: variance reduction is obtained if R1 R2 < 1.


N.B. Most of the time we only look at R1 . (R2 is “messy”)

16/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

1 Pricing exotic options by simulation


Exotic options: quick overview (from Chapter 19)
Pricing, an example
Some advantages and disadvantages

2 Variance reduction—the big picture (§21.2)


Efficiency from several angles

3 Antithetic variates
Principle: applied to the π-example
Analysing the uniform case
Analysing the normal case
How to tell when/whether AV reduce the variance?
Interpretation for path dependent options

17/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

1st variance reduction method: Antithetic variates (§21.4)

R1 √
Old example: I = 0 e u du. We can see this as
h √ i
I =E e U with U a U (0, 1)-random variable.

Monte Carlo can be done with iid standard uniforms U1 ,. . . ,UM


and
√ M
Ui 1 X
Xi = e and IM = Xi .
M
i=1

It is easy to come up with some alternatives:



Xia = e 1−Ui , 1 − Ui is also U (0, 1)-distributed.
Xi + Xia
E[Xi ]+E[Xia ]
Yi = , E [Yi ] = 2 = I.
2
Try this in Matlab: antithet.m.

18/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

Summary simulation results with 104 replications:

basis for √
MC Iˆ ± s.e. rel.eff. wrt X
X = exp( √ U) 2.00093 ± 0.00442 1
X a = exp( 1 − U) 1.99874 ± 0.00442 1
(X + X a )/2 1.99983 ± 0.00033 ≈ 175

Just as with the common random numbers method for the Greeks,
the success of this approach is explained by the correlation.
We have
ρ (X , X a ) ≈ −0.989.

19/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

1 Pricing exotic options by simulation


Exotic options: quick overview (from Chapter 19)
Pricing, an example
Some advantages and disadvantages

2 Variance reduction—the big picture (§21.2)


Efficiency from several angles

3 Antithetic variates
Principle: applied to the π-example
Analysing the uniform case
Analysing the normal case
How to tell when/whether AV reduce the variance?
Interpretation for path dependent options

20/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

Analysis: the uniform case (§21.5)

Suppose I = E [f (U)] for U ∼ U (0, 1) and some function f .


Key: U and 1 − U have the same distribution.
Each of

X = f (U), X a = f (1 − U), Y = 21 (X + X a )

can serve as basis for simulation to estimate I .


1 PM
If Y1 ,. . . , YM are iid replications of Y , and IˆM = M i=1 Yi ,
then h i   1
E IˆM = I and Var IˆM = Var(Y ) .
M

X and X a are called antithetic variates.

21/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

Analysis:
1
+ X a ) = 14 Var(X + X a )

Var(Y ) = Var 2 (X
= 1
4 [Var(X ) + Var(X a ) + 2 Cov(X , X a )]
= 1
2 [Var(X ) + Cov(X , X a )]
= Var(X ) · 12 [1 + ρ (X , X a )].

So, it would be nice if ρ (X , X a ) were (strongly) negative.


In our example ρ ≈ −0.989, so the reduction factor R1 is

1 1 1
2 [1 − 0.989] ≈ 0.0057 and = ≈ 175
R 1 0.0057
With the same number of replications we get more than 1
extra digit of accuracy!
But note: R2 ≈ 2, so in all we gain about a factor 100.

22/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

1 Pricing exotic options by simulation


Exotic options: quick overview (from Chapter 19)
Pricing, an example
Some advantages and disadvantages

2 Variance reduction—the big picture (§21.2)


Efficiency from several angles

3 Antithetic variates
Principle: applied to the π-example
Analysing the uniform case
Analysing the normal case
How to tell when/whether AV reduce the variance?
Interpretation for path dependent options

23/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

Analysis: the normal case (§21.6)

Suppose I = E [f (Z )] for Z ∼ N (0, 1) and some f .


Key: Z and −Z have the same distribution.
Consider
1
X = f (Z ), X a = f (−Z ),
Y = (X + X a ).
2
The same analysis applies, and we find again

Var(Y ) = Var(X ) · 21 [1 + ρ (X , X a )].

If ρ < 0 we always win something.


If ρ = 0 we break even—we neither win nor lose something.
1
Why? R1 = 2 but R2 = 2. The reduction is R = R1 R2 = 1.

24/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

1 Pricing exotic options by simulation


Exotic options: quick overview (from Chapter 19)
Pricing, an example
Some advantages and disadvantages

2 Variance reduction—the big picture (§21.2)


Efficiency from several angles

3 Antithetic variates
Principle: applied to the π-example
Analysing the uniform case
Analysing the normal case
How to tell when/whether AV reduce the variance?
Interpretation for path dependent options

25/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

When do antithetic variates reduce the variance?

This lemma can help us:


Lemma (covariance inequality for comonotone functions)
If X is a random variabele and the functions f and g are monotone
in the same direction, i.e., both nondecreasing or both
nonincreasing, then:

Cov(f (X ), g (X )) ≥ 0.

If they are monotone in the opposite direction, then:

Cov(f (X ), g (X )) ≤ 0.

Proof: (on board).

26/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

§21.7 Multivariate version of the lemma

Lemma.
If f and g are functions of n variables, monotone in the same
sense, and X1 , . . . , Xn are independent, then:

Cov(f (X1 , . . . , Xn ), g (X1 , . . . , Xn )) ≥ 0.

N.B. monotone: it is sufficient if this is coordinatewise the same,


i.e., for a subset of the coordinates f and g are both increasing,
and both are decreasing in the remaining coordinates.
N.B. Coordinatewise monotone in opposite sense, then: Cov ≤ 0.

Application: Up-and-out call option. UpAndOutCallanti.m


Limited speed gain: about 2.3 times, at twice the work. . .
27/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

1 Pricing exotic options by simulation


Exotic options: quick overview (from Chapter 19)
Pricing, an example
Some advantages and disadvantages

2 Variance reduction—the big picture (§21.2)


Efficiency from several angles

3 Antithetic variates
Principle: applied to the π-example
Analysing the uniform case
Analysing the normal case
How to tell when/whether AV reduce the variance?
Interpretation for path dependent options

28/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

Why AV works for (some) path dependent options


Key: The pay-off is often increasing in the simulated Z ’s.
We have the recursion

S(ti ) = S(ti−1 ) exp (r − 12 σ 2 )(ti − ti−1 ) + σ ti − ti−1 · Zi ,


 p 

which we can solve, and then simplify:


i
Y
exp (r − 12 σ 2 )(tj − tj−1 ) + σ tj − tj−1 · Zj
 p 
S(ti ) = S(t0 )
j=0
i
X 
1 2
 p 
= S(t0 ) exp (r − 2 σ )(tj − tj−1 ) + σ tj − tj−1 · Zj
j=0
 i
X 
1 2
p
= S(t0 ) exp (r − 2 σ )(ti − t0 ) + σ tj − tj−1 · Zj .
j=0

29/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

So, for i = 1, . . . , n:
 i
X 
1 2
p
S(ti ) = S(t0 ) · exp (r − 2 σ )(ti − t0 ) + σ tj − tj−1 · Zj
j=1

This represents the prices S(t0 ), . . . , S(tn ) at time points


t0 = 0 < t1 < · · · < tn = T , as functions of:
the (fixed) parameters r , σ, S(t0 );
(t0 , t1 , . . . , tn = T also fixed);
independent N (0, 1)-random variables Z1 , Z2 , . . . , Zn .

Check this:
S(ti ) is increasing as function of Z1 , Z2 , . . . , Zi , (Zi+1 , . . . Zn ).

30/ 31
Exotic Options Variance reduction Antithetic variates Principle Uniform case Normal When good? Application

The (approximation of the) option price often is of the form


 
e−rT f S(t0 ), S(t1 ), . . . , S(tn ) . (1)

Here the mapping

(s0 , s1 , . . . , sn ) 7→ f (s0 , s1 , . . . , sn )

often is increasing (or at least nondecreasing) in s0 , s1 , . . . , sn .

In those cases:
the payoff (1) is an increasing (or at least nondecreasing)
function of Z1 , Z2 ,. . . , Zn ,
application of antithetic variates will result in some variance
reduction, by the multivariate version of the lemma.
These things also may give insight in other situations, where there
only is some monotonicity.
31/ 31

You might also like