MA 8101 Stokastiske metoder i systemteori

Similar documents
MA8109 Stochastic Processes in Systems Theory Autumn 2013

Theoretical Tutorial Session 2

I forgot to mention last time: in the Ito formula for two standard processes, putting

Stochastic integral. Introduction. Ito integral. References. Appendices Stochastic Calculus I. Geneviève Gauthier.

Stochastic Processes

1 A complete Fourier series solution

An introduction to quantum stochastic calculus

An Overview of the Martingale Representation Theorem

Stochastic Processes

(x + y) ds. 2 (1) dt = p Find the work done by the force eld. yzk

1. Stochastic Processes and filtrations

The concentration of a drug in blood. Exponential decay. Different realizations. Exponential decay with noise. dc(t) dt.

Testing for Regime Switching: A Comment

(B(t i+1 ) B(t i )) 2

ECONOMETRICS FIELD EXAM Michigan State University May 9, 2008

Corrections to Theory of Asset Pricing (2008), Pearson, Boston, MA

Stochastic Differential Equations.

ISOMETRIC APPROACH TO STOCHASTIC INTEGRAL OF CERTAIN RANDOM MEASURES

Exercises. T 2T. e ita φ(t)dt.

Stochastic Calculus (Lecture #3)

THE FOURTH-ORDER BESSEL-TYPE DIFFERENTIAL EQUATION

Ordinary and Stochastic Di erential Equations Lecture notes in Mathematics PhD in Economics and Business. B. Venturi G. Casula, A.

ECONOMETRICS II (ECO 2401S) University of Toronto. Department of Economics. Winter 2014 Instructor: Victor Aguirregabiria

The multidimensional Ito Integral and the multidimensional Ito Formula. Eric Mu ller June 1, 2015 Seminar on Stochastic Geometry and its applications

p 1 ( Y p dp) 1/p ( X p dp) 1 1 p

Stochastic Calculus Made Easy

MATH 2203 Final Exam Solutions December 14, 2005 S. F. Ellermeyer Name

MATH 1190 Exam 4 (Version 2) Solutions December 1, 2006 S. F. Ellermeyer Name

The main purpose of this chapter is to prove the rst and second fundamental theorem of asset pricing in a so called nite market model.

Bernardo D Auria Stochastic Processes /12. Notes. March 29 th, 2012

Stochastic Calculus. Kevin Sinclair. August 2, 2016

Stochastic solutions of nonlinear pde s: McKean versus superprocesses

Stochastic Differential Equations

4.3 - Linear Combinations and Independence of Vectors

Supplemental Material 1 for On Optimal Inference in the Linear IV Model

Tools of stochastic calculus

Brownian Motion. 1 Definition Brownian Motion Wiener measure... 3

Stochastic Processes (Master degree in Engineering) Franco Flandoli

Stochastic Calculus February 11, / 33

From Random Variables to Random Processes. From Random Variables to Random Processes

3 Random Samples from Normal Distributions

IEOR 4701: Stochastic Models in Financial Engineering. Summer 2007, Professor Whitt. SOLUTIONS to Homework Assignment 9: Brownian motion

1 Which sets have volume 0?

Solution for Problem 7.1. We argue by contradiction. If the limit were not infinite, then since τ M (ω) is nondecreasing we would have

ECE 673-Random signal analysis I Final

Chapter 6 Higher Dimensional Linear Systems

PROBABILITY: LIMIT THEOREMS II, SPRING HOMEWORK PROBLEMS

ECONOMET RICS P RELIM EXAM August 24, 2010 Department of Economics, Michigan State University

1 Simulating normal (Gaussian) rvs with applications to simulating Brownian motion and geometric Brownian motion in one and two dimensions

Bernardo D Auria Stochastic Processes /10. Notes. Abril 13 th, 2010

Multiple Random Variables

ECON2285: Mathematical Economics

a11 a A = : a 21 a 22

CAE Working Paper # Fixed-b Asymptotic Approximation of the Sampling Behavior of Nonparametric Spectral Density Estimators

Verona Course April Lecture 1. Review of probability

Man Kyu Im*, Un Cig Ji **, and Jae Hee Kim ***

University of Toronto

The Wiener Itô Chaos Expansion

Rough paths methods 4: Application to fbm

Introduction: structural econometrics. Jean-Marc Robin

Stochastic Integration and Stochastic Differential Equations: a gentle introduction

Lecture 8: Basic convex analysis

Economics 620, Lecture 9: Asymptotics III: Maximum Likelihood Estimation

Introduction Optimality and Asset Pricing

Partial Di erential Equations

NEW SIGNS OF ISOSCELES TRIANGLES

Warped Products. by Peter Petersen. We shall de ne as few concepts as possible. A tangent vector always has the local coordinate expansion

Real Analysis: Homework # 12 Fall Professor: Sinan Gunturk Fall Term 2008

The Kuhn-Tucker Problem

Expectation. DS GA 1002 Statistical and Mathematical Models. Carlos Fernandez-Granda

Malliavin calculus and central limit theorems

Stochastic Differential Equations

SOME THEORY AND PRACTICE OF STATISTICS by Howard G. Tucker CHAPTER 2. UNIVARIATE DISTRIBUTIONS

Introduction to Random Diffusions

Dynamic Semiparametric Models for Expected Shortfall (and Value-at-Risk)

Week 2 Spring Lecture 3. The Canonical normal means estimation problem (cont.).! (X) = X+ 1 X X, + (X) = X+ 1

Stochastic Hamiltonian systems and reduction

Taylor series - Solutions

Introduction to numerical simulations for Stochastic ODEs

Quasi-invariant Measures on Path Space. Denis Bell University of North Florida

Final Exam. Monday March 19, 3:30-5:30pm MAT 21D, Temple, Winter 2018

Stein s method and weak convergence on Wiener space

5 Transcendental Functions

Transition Density Function and Partial Di erential Equations

Malliavin Calculus: Analysis on Gaussian spaces

Widely applicable periodicity results for higher order di erence equations

q-series Michael Gri for the partition function, he developed the basic idea of the q-exponential. From

Convoluted Brownian motions: a class of remarkable Gaussian processes

Divergence Theorems in Path Space. Denis Bell University of North Florida

On pathwise stochastic integration

Expectation. DS GA 1002 Probability and Statistics for Data Science. Carlos Fernandez-Granda

Skew Brownian Motion and Applications in Fluid Dispersion

1 Review of di erential calculus

Mathematical Methods for Neurosciences. ENS - Master MVA Paris 6 - Master Maths-Bio ( )

Maximum Likelihood Drift Estimation for Gaussian Process with Stationary Increments

Lecture 12: Diffusion Processes and Stochastic Differential Equations

ANALYTICAL AND NUMERICAL RESULTS ON ESCAPE OF BROWNIAN PARTICLES. Carey Caginalp. Submitted to the Graduate Faculty of

PROBABILITY: LIMIT THEOREMS II, SPRING HOMEWORK PROBLEMS

Math 212-Lecture 8. The chain rule with one independent variable

1.1 Definition of BM and its finite-dimensional distributions

Transcription:

MA 811 Stokastiske metoder i systemteori AUTUMN TRM 3 Suggested solution with some extra comments The exam had a list of useful formulae attached. This list has been added here as well. 1 Problem In this problem we are considering a standard Brownian motion B t in R 1 starting at : (a) State the basic properties of the Brownian motion. De ne, for a xed a >, the process Verify that also X t is a standard Brownian motion. X t ab ta : (1) A standard Brownian motion starting at is a Gaussian stochastic process de ned for t [; 1) ful lling 1. B t ;. Cov (B t B s ) min (t; s) : From () it follows that a B.M. has orthogonal increments. There exists a version of B.M. with continuous paths. It is obvious that X t is a Gaussian process (This actually requires that all nite collections (X t1 ; ; X tn ) are multivariate Gaussian, but this follows since B t has such a property). Moreover, 1. clearly true. Finally, Cov (X t ; X s ) a Cov B sa B ta s a min a ; t a () min (s; t) : (b) Let t < t 1 < < t N+1 T be a partition of the interval [; T ] and ' the elementary function NX ' (t;!) e j (!) [tj+1 t j ) (t) : (3) j What does it mean that ' is in the class V[; T ], and what is then the value of the Itô integral Show that for all f V[; T ]. ' (t;!) db t (!)? (4) f (t;!) db t (!) : (5) 1

The class V[; T ] consists of B F-measurable functions f (t;!) L ( [; T ]) such that f (t;!) is F t -measurable for all t [; T ]. Here this will be the case if e j < 1 and e j is F tj -measurable for all j ; ; N. Also, ' (t;!) db t (!) Since e j and B tj+1 B tj are independent, Thus, NX e j (!) B tj+1 (!) B tj (!) : (6) j e j Btj+1 B tj (ej ) B tj+1 B tj : (7) ' (t;!) db t (!) : (8) In general, the Itô integral is a limit of integrals of simple functions. This is about all we requite for the exam, but the full argument is as follows: We nd a sequence f' n g such that R 'n db R fdb! n!1 : Then Z ' n db Z Z fdb Z fdb Z (' n (' n Z (' n f) db f) db (9) f) db!! 1! : n!1 (c) Compute the variance of the integral tb t (!) db t (!) : (1) This follows immediately from Itô s Isometry since we know that the expectation is : Problem Var tb t (!) db t (!) tb t (!) db t (!) We consider two Itô processes X t and Y t on R 1. (tb t ) dt (a) Let X t and Y t be two Itô processes X t and Y t on R 1. Prove that t tdt 1 4 : (11) d (X t Y t ) X t dy t + Y t dx t + dx t dy t : (1)

For this formula we apply the D Itô formula for the function g (x; y) xy:then (b) Let d (X t Y t ) @g @x dx t + @g @y dy t + @ g @x@y dx t dy t Show that X t can be written as an Itô integral. We compute dx t using Ito s formula: Y t dx t + X t dy t + 1 dx t dy t : (13) X t e t sin (B t ) : (14) dx t 1 X tdt + e t cos (B t ) db t + 1 et ( sin B t ) dt Hence, e t cos (B t ) db t : (15) X t Z t since it is clear that e t cos (B t ) V [; T ]. e s cos (B s ) db s ; (16) (c) The conclusion in (.b) implies that X t is a Martingale with respect to the ltration of the Brownian motion, F t. Prove this directly by applying the de nition of a Martingale to the expression for X t in qn. 14. The rst is to observe that (X t ) (jx t j) e t jsin (B t )j e t < 1: (17) Since X t is a determininistic, continuous function of B t, X t is clearly F t -measureable. We nally need to show that (X t+t jf t ) X t for t >. Let us write B t+t B t + B. Then (X t+t jf t ) e (t+t) sin (B t + B) jf t e (t+t) (sin (B t ) cos (B) + cos (B t ) sin (B) jf t ) (i) e (t+t) [sin (B t ) (cos (B) jf t ) + cos (B t ) (sin (B) jf t )] (ii) e (t+t) [sin (B t ) (cos (B)) + cos (B t ) (sin (B))] (18) (iii) h i e (t+t) sin (B t ) e t + e t sin (B t ) X t : Here (i) and (ii) are formulae for the conditional expectation. Moreover, for (iii), (cos (B)) is listed and (sin (B)) is obviously. 3

3 Problem (a) Show that a linear stochastic di erential equation dx t p (t) X t dt + q (t) db t (19) may be solved by an integrating factor h (t) such that d [h (t) X t ] h (t) q (t) db t : () It follows from Itô s formula that d (h (t) X t ) h (t) X t dt + h (t) dx t : (1) We then multiply qn. 19 by h (t): h (t) dx t d (h (t) X t ) h (t) X t dt h (t) p (t) X t dt + h (t) q (t) db t : () The stated form follows if we nd h such that h (t) h (t) p (t) : (3) Then the solution to qn. 19 is: or h (t) X t h (t ) X t Z t t h (t) q (t) db t ; (4) X t h (t ) X t + R t t h (t) q (t) db t : (5) h (t) (b) Apply the method in (a) to solve the equation dx t 1 t X tdt + tdb t ; t 1; X 1 1: (6) The function h has to satisfy the equation that is, h (t) t 1. In this case the start is at t 1 such that 4 Problem X t X 1 + R t 1 1 s sdb s 1t h + 1 h ; (7) t t (1 + B t B 1 ) t ~ B 1;1 t ; t 1: (8) Consider the following geometric Brownian motion in R 1 : X t x exp ( t + B t ) ; t. (a) Show that X t is an Itô di usion with generator A x d dx + x d dx : (9) 4

The di usion representation of X t is dx t X t dt + X t db t + 1 X tdt The formula for the generator follows from when dx t dt + db t. (b) We start X t at x 1 and de ne, for x < 1, X t dt + X tdb t : (3) A d dx + 1 d dx ; (31) (!) minft ; X t (!) x g: (3) Prove that log x : (33) We are going to apply Dynkin s Formula and need to know that < 1. Here the part t will e ectively "kill" the Brownian motion part Bt. Of course, X t will also cross the level x a.s. This argument is su cient for the exam, but the full proof could be based on the law of the iterated logarithm, or the following simple estimate: P ( u) P (X u x ) P (B u log x + u) Bu P u 1 log x + u u 1 O u 1 e u ; (34) when u! 1; using the inequality in the list. This is su cient for to be nite. The next step is to nd an f such that Af is equal to a constant. Here f (x) log x will do since x d log x + x d log x dx dx 1: (35) By Dynkin s Formula we then have Z 1 (f (X )) log (x ) log 1 + 1 ( 1) ds ; (36) or log (x ) : (37) (c) Let a; b be two positive numbers, a < b. We start X t at x [a; b] and let Determine p for all x [a; b]. p Pr (X t hits level b before it hits level a) 5

We need to nd a function f (x) such that Af, and try x : x dx dx + x d x dx x + 1 ( 1) ; (38) for. We then apply Dynkin s formula with the function f (x) x,. The expected escape time from the interval [a; b] is clearly nite since the situation will be similar to the case in (b) for the lower level a. Thus (X ) p b + (1 p) a x + ; (39) and hence, a p x b a : (4) 6

A list of useful formulae Note: The list does not state requirements for the formulae to be valid. The probability density of a 1D Gaussian variable with mean and variance :! (x) p 1 (x ) exp (41) A de nite integral: 1 x cos (x) exp dx p e : (4) An inequality: Let X be N (; 1) and x >. Then P (X x) Two formulae for Conditional xpectations: The Itô isometry: r 1 1 x e x : (43) (i) If Y is H-measurable, then (XY jh) Y (XjH) : (ii) If X is independent of H, then (XjH) (X) : Itô s D formula: f (t;!) db t (!) dg (t; X t ; Y t ) @g @g dt + @t @x dx t + @g @y dy t + 1 jf (t;!)j dt kfk L ([;T ]) (44) @ g @x (dx t) + @ g @x@y dx tdy t + 1 @ g @y (dy t) : (45) and "the rules". The generator: A (f) (x) Dynkin s Formula: nx i1 i (x) @f @x i (x) + 1 nx i;j1 x (f (X )) f (x) + x Z (x) (x) @ f (x) : (46) i;j @x i @x j Af (X s ) ds : (47) 7