Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees...

28
Constructing Brownian Motions and Radon-Nikodym Derivatives Steven R. Dunbar Standard Brownian Motion Binomial Trees Using the Radon-Nikodym derivative Constructing Brownian Motions and Radon-Nikodym Derivatives Steven R. Dunbar February 5, 2016 1 / 28

Transcript of Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees...

Page 1: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Constructing Brownian Motions andRadon-Nikodym Derivatives

Steven R. Dunbar

February 5, 2016

1 / 28

Page 2: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Outline

1 Standard Brownian Motion

2 Binomial Trees

3 Using the Radon-Nikodym derivative

2 / 28

Page 3: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Comment

I tend to use Brownian Motion and Wiener Processinterchangeably, but

I like to use Brownian Motion for a physicalmanifestation (stock prices, motion of a particle in afluid)I like to use Wiener Process for the mathematicalmodel.

3 / 28

Page 4: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Definition

Wiener ProcessThe Standard Wiener Process is a stochastic processW (t), for t ≥ 0, with the following properties:

1 Increments W (t)−W (s) are normally distributedW (t)−W (s) ∼ N(0, t− s).

2 For t1 < t2 ≤ t3 < t4, increments W (t4)−W (t3)and W (t2)−W (t1) are independent randomvariables.

3 W (0) = 0.4 W (t) is continuous for all t (with probability 1).

4 / 28

Page 5: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Random Walk

Let

Yi =

+1 with probability 1/2−1 with probability 1/2

be a sequence of

independent, identically distributed Bernoulli randomvariables. Let Y0 = 0 for convenience and let

Tn =n∑

i=0

Yi

Note that Var [Yi] = 1, which we will need to use in amoment.

5 / 28

Page 6: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Random Walk Graph

n 0 1 2 3 4 5 6 7 8 9 10Yn 0 1 1 1 -1 -1 -1 1 -1 -1 -1Tn 0 1 2 3 2 1 0 1 -1 -2 -3

0 2 4 6 8 10-2

-1

0

1

2

3

6 / 28

Page 7: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Random Walk as continuous function

Sketch the random fortune Tn versus time using linearinterpolation between the points (n− 1, Tn−1) and(n, Tn).

The interpolation defines a function W (t) defined on[0,∞) with W (n) = Tn.

The notation W (t) reminds us of the piecewise linearnature of the fu

7 / 28

Page 8: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Random Walk Continuous FunctionExample

0 2 4 6 8 10-2

-1

0

1

2

3WcaretN

8 / 28

Page 9: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Scaling Random Walk

Compress time, rescale the space in a connected way. LetN be a large integer, and consider the rescaled function

WN(t) =

(1√N

)W (Nt).

This has the effect of taking a step of size ±1/√N in

1/N time unit. For example,

WN(1/N) =

(1√N

)W (N · 1/N) =

T1√N

=Y1√N.

9 / 28

Page 10: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Distribution of the Scaled Walk

Now consider

WN(1) =W (N · 1)√

N=W (N)√

N=

TN√N.

According to the Central Limit Theorem, this quantity isapproximately normally distributed, with mean zero, andvariance 1. More generally,

WN(t) =W (Nt)√

N=√tW (Nt)√

Nt

If Nt is an integer, WN(t) is normally distributed withmean 0 and variance t. Furthermore, WN(0) = 0 andWN(t) is a continuous function.10 / 28

Page 11: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Limit Theorem

Theorem (essentially Donsker, 1951)The limit of the rescaled random walk definingapproximate Brownian Motion is Brownian Motion in thefollowing sense:

P[WN(t1) < x1, WN(t2) < x2, . . . WN(tn) < xn

]→

P [W (t1) < x1,W (t2) < x2, . . .W (tn) < xn]

as N →∞ where t1 < t2 < · · · < tn.

11 / 28

Page 12: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Example

Octave Source (compressed for space)p = 0.5;global N = 400; global T = 1; global SS = zeros(N+1, 1);S(2:N+1) = cumsum( 2 * (rand(N,1)<=p) - 1);function retval = WcaretN(x)

global N; global T; global S;Delta = T/N;prior = floor(x/Delta) + 1; # add 1 since arrays are 1basedsubsequent = ceil(x/Delta) + 1;retval = sqrt(Delta)*(S(prior) + ((x/Delta+1) - prior).*(S(subsequent)-S(prior)));

endfunctionfplot(@WcaretN, [0,T])

12 / 28

Page 13: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Example

13 / 28

Page 14: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

The goal

Build more intuition about Brownian Motions by lookingat probability on several discrete binomial trees.

The goal is to be very strongly intuitive and motivationalin contrast to rigorous, but still to do honest examples.

14 / 28

Page 15: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

A standard binomial tree

Note that the expected value at time n is 0.

H

1/2

T

1/2

HH

1/2

HT

1/2

TH1/2

TT

1/2

HHH

1/2

HHT

1/2

HTH1/2

HTT

1/2

THH1/2

THT

1/2

TTH1/2

TTT

1/2

15 / 28

Page 16: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

A skewed binomial tree

The expected value at n is 13n, this process is “drifting”

upward.

H

2/3

T

1/3

HH

2/3

HT

1/3

TH2/3

TT

1/3

HHH

2/3

HHT

1/3

HTH2/3

HTT

1/3

THH2/3

THT

1/3

TTH2/3

TTT

1/316 / 28

Page 17: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

A general binomial tree

Instead of branch probabilities, give the path probabilitymeasure instead. Each can be recovered from the other.

πH

p11

πT

p10

πHH

p21

πHT

p20

πTHp21

πTT

p20

πHHH

p31

πHHT

p30

πHTHp31

πHTT

p30

πTHHp31

πTHT

p30

πTTHp31

πTTT

p3017 / 28

Page 18: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Another general binomial tree

Another probability measure. What is the relation to theprevious measure?

0

φH

φT

φHH

φHT

φTH

φTT

φHHH

φHHT

φHTH

φHTT

φTHH

φTHT

φTTH

φTTT

18 / 28

Page 19: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

The Radon-Nikodym derivative on the tree

The Radon-Nikodym derivative measures the likelihoodratio.

φH⁄πH

φT ⁄πT

φHH⁄πHH

φHT⁄πHT

φTH⁄πTH

φTT ⁄πTT

φHHH⁄πHHH

φHHT ⁄πHHT

φHTH ⁄πHTH

φHTT ⁄πHTT

φTHH ⁄πTHH

φTHT ⁄πTHT

φTTH⁄πTTH

φTTT ⁄πTTT

19 / 28

Page 20: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Recover φ from π

Given P on paths, and the R-N derivative dQdP ,

Q =dQdP

P

Note also: the R-N derivative dQdP is defined on paths, so

it too is a random variable on the space Ω.

20 / 28

Page 21: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Equivalent Measures

Equivalent measuresTwo probability measures on the space Ω with σ-algebraF are equivalent if for any set B ∈ F , P [B] > 0 if andonly if Q [B] > 0. This is, the probability measures areequivalent if P Q and Q P.

21 / 28

Page 22: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Expectation

The Radon-Nikodym derivative is defined on paths, andis F -measurable, so it is also a random variable.

Even more, the Radon-Nikodym derivative is Ft-adapted.

Let ζt = dQdP folowing paths to time t.

ζt = EP

[dQdP|Ft

]for every t.

The expectation, knowing the information up to time tmeasured by P, represents the amount of change ofmeasure so far up to time t along the current path as ζt.

22 / 28

Page 23: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Using the R-N Derivative

If we want to know EQ [F (Xt)], it would be EP [ζtF (Xt)].

If we want to know EQ [F (Xt)|Fs], then we would needthe amount of change from time s to t which is justζt/ζs, which is change up to time t with the change upto time s removed. In other words

EQ [F (Xt)|Fs] = ζ−1s EP [ζtF (Xt)|Fs] .

23 / 28

Page 24: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Passing to the Limit

Suppose P and Q are equivalent measures on a tree ofpaths. Given scaled path points (t1, x1), . . . (tn, xn) withtn = 1 then dQ

dP up to time T = 1 is the limit of thelikelihood ratios:

dQdP

= limn→∞

fnQ(x1, . . . , xn)

fnP (x1, . . . , xn)

Then for Brownian Motions let ζt = dQdP . Then

ζt = EP

[dQdP|Ft

]and

EQ [F (Xt)|Fs] = ζ−1s EP [ζtF (Xt)|Fs] .24 / 28

Page 25: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Simple Example

The point of the example is to apply the defintions andnotations to the pair of binomial trees with P defined byP [H] = 1

2and P [T ] = 1

2and Q [H] = 2

3and Q [T ] = 1

3.

This has minimal mathematical content, but is a goodillustration of defintions and notation.

ζt =

(4

3

) t+Xt2

·(

2

3

) t−Xt2

25 / 28

Page 26: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Simple Example Continued

EQ[1[X3≥1]|F2

]Node Expectation X2 ProbHH 2

3· 1 + 1

3· 1 2 1

HT 23· 1 + 1

3· 0 1 2

3

TH 23· 1 + 1

3· 0 1

TT 23· 0 + 1

3· 0 0 0

Conditional probabilities the “old-fashioned way”

EQ[1[X3≥1]|F2

]X=0

= Q [X3 = 1 |HT,HT ]

= Q [HHH |HT, TH] =Q [HHH]

Q [HT, TH]

=8/27

4/9=

2

326 / 28

Page 27: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Simple Example Continued

EP[1[X3≥1]ζ3|F2

]Node Expectation X2 ProbHH 1

2· 1 ·

(43

)3+ 1

2· 1 ·

(43

)2 (23

)2(43

)2HT 1

2· 1 ·

(43

)2 (23

)+ 1

2· 0 ·

(43

) (23

)2 1 12

(43

)2 23

TH 12· 1 ·

(43

)2 (23

)+ 1

2· 0 ·

(43

) (23

)2 1 12

(43

)2 23

TT 12· 0 ·

(43

) (23

)2+ 1

2· 0 ·

(23

)3 0 0

27 / 28

Page 28: Constructing Brownian Motions and Radon-Nikodym Derivativessdunbar1/Mathematical...Binomial Trees Using the Radon-Nikodym derivative Ageneralbinomialtree Insteadofbranchprobabilities,givethepathprobability

ConstructingBrownian

Motions andRadon-Nikodym

Derivatives

Steven R.Dunbar

StandardBrownianMotion

Binomial Trees

Using theRadon-Nikodymderivative

Simple Radon-Nikodym DerivativeExample Summarized

EQ [F (Xt)|Fs] = ζ−1s EP [ζtF (Xt)|Fs] .

X2 EQ[1[X3≥1]|F2

]EP[1[X3≥1]ζ3|F2

]ζ2 Prob

2 1(43

)2 (43

)2 11 2

3

(12

) (43

)2 23

(43

) (23

)23

0 0 0(23

)2 0

28 / 28