Scribd
Upload
70 views

EC744 Lecture Note 6 Stochastic Models: Mathematical Preliminaries

The document discusses mathematical preliminaries for stochastic models, including: 1. The definition of a σ-algebra and measurable space. 2. The definition of a measure and properties such as σ-finiteness. 3. Measurable functions and integration with respect to a measure.

Uploaded by

binicle
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
70 views

EC744 Lecture Note 6 Stochastic Models: Mathematical Preliminaries

The document discusses mathematical preliminaries for stochastic models, including: 1. The definition of a σ-algebra and measurable space. 2. The definition of a measure and properties such as σ-finiteness. 3. Measurable functions and integration with respect to a measure.

Uploaded by

binicle
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 18

EC744 Lecture Note 6 Stochastic Models: Mathematical Preliminaries

Prof. Jianjun Miao

Motivation Optimal growth model v (x, z) = for IID shock z How to dene E []? Finite many values of z or continuous values? Measurability?
0yf (x)z

max

U (f (x) z y) + E v y, z 0

Let S be a set and let S be a family of subsets of S. S is called a -algebra if a. , S S; b. A S implies Ac = S\A S; c. An S, n = 1, 2, ..., implies An S. n=1 The pair (S, S) is called a measurable space. Any set A S is called a measurable set. Examples: Finite or countable set S, uncountable set Given any other metric space (S, d), the Borel algebra B(S) is the smallest -algebra containing the open balls, A = {s S : d(s, s0) < } where s0 S and > 0. An element of B(S) is called a Borel set.

Let (S, S) be a measurable space. A measure is a real valued function : S R such that a. () = 0; b. (A) 0, all A S; c. if {An} is a countable, disjoint sequence of subsets in S, then ( An) = n=1 n=1 P n=1 (An). (S, S, ) is called a measure space. If (S) = 1, it is called a probability measure. (S, S, ) is called a probability space.

Let S be a set, and let A be a family of its subsets. Then A is called an algebra if a. , S A; b. A A implies Ac = S\A A; and c. A1, ..., An A implies n Ai A. i=1 Dene measure on algebra A.

Caratheodory Extension Theorem. Let S be a set, A an algebra of its subsets, and a measure on A. Let B be the smallest -algebra containing A. Then there exits a measure on B such that (A) = (A) , for all A A. Denition: If there is a sequence of sets {Ai}n in A with (Ai) < , all i=1 i, and S = Ai, then is -nite. i=1 Hahn Extension Theorem. If is -nite, then the extension to B is unique. The completion of a -algebra is a -algebra. Lebesgue measurable sets, Lebesgue measure, Borel measure.

Given a measurable space (S, S), a real-valued function f : S R is measurable with respect to S or (S-measurable) if {s S : f (s) a} S, all a R. Can be dened in terms of , , <, and > . Examples: For S = {0, 1} , S = {, S, {0} , {1}} and S = {, S} are both -algebras. All functions on S are S-measurable, but only the constant functions are S-measurable. There are functions that are not Borel measurable.

Any monotone or continuous function is measurable. Pointwise convergence preserves measuability. Approximation of a measuable function by a simple function (s) =
n X

i=1

aiAi (s)

If f and g are S-measuable, then f + g, f g, |f | , and cf are S-measurable. The composition of Borel measurable functions are Borel measurable, but not true for Lebesgue measurable functions.

Let (S, S) and (T, T ) be measurable spaces. Then the function f : S T is measurable if the inverse image of every measurable set is measurable. Let be a correspondence of S into T. Then the function h : S T is a measurable selection from if h is measurable and h (s) (s) for all s S. (Measurable Selection Theorem) Let : S T be a nonempty compactvalued and u.h.c. correspondence. Then there exists a measurable selection from .

Integration: Lebesgue integral and Riemann integral The Lebesgue integral of f, f (x) d (x) on [a, b] is dened in terms of Pn i=1 yi (Ai) where y1 y2 ... yn, Ai = {x : yi f (x) yi+1} when yis are closer together.
R Dene integration S f (s) (ds) on any measurable space (S, S, ) R

Monotone Convergence Theorem Dominated Convergence Theorem Fatous Lemma

is absolutely continuous w.r.t. ( << ) if (A) > 0 implies (A) > 0 for all A.

Radon-Nikodym Theorem Suppose << . Then there is an integrable function h such that (A) =
Z
A

h (s) (ds) all A

Let (, F, ) be a probability space; let A F be a algebra; and let f : R be integrable. Then the conditional expectation of f relative to A is an A-measurable function E [f |A] : R such that
Z
C

E [f |A] () (d) =

f () (d) , all C A

Product Spaces Let (X, X ) and (Y, Y) be measurable spaces, and let Z = X Y. A set C = AB Z is a measurable rectangle if A X and B Y. Let C be the set of all measurable rectangles, and A the set of all nite unions of measurable rectangles. Then A is an algebra. Let Z = X Y be the algebra generated by the algebra A. (Z, Z) is called the product space.

Theorem. Let : C R+ have the following properties: a. () = 0; and b. if {Ci} ={Ai Bi} is a sequence of disjoint sets in C and Ci is i=1 i=1 C = P (C ) . in C, then i=1 i i i=1 Then there is a measure on A that coincides with on C. Apply the Caratheodory and Hahan Extension Theorems to dene a measure on Z, which is a -algebra generated by A Can be extended to any space that is the product of a nite number of measurable spaces.

Markov Processes Let (Z, Z) be a measurable space. A transition function is a function Q : Z Z [0, 1] such that a. for each z Z, Q (z, ) is a probability measure on (Z, Z) ; b. for each A Z, Q (, A) is a Zmeasurable function. Dene an operator T : B(Z) B(Z) as T f (z) =
Z

f (z 0)Q(z, dz 0).

Dene the adjoint operator T : (Z) (Z) as T (A) =


Z

Q(z, A)(dz).

Q satises the Feller property if T f () is bounded and continuous for any continuous and bounded function f .

Q is monotone if T f () is increasing for any increasing function f . Examples: A nite state Markov chain with transition matrix = ij An AR(1) process zt+1 = zt + t+1

Probability Measures on Spaces of Sequences Let


t, Z t = (Z Z, Z Z) (t times) Z Z Z Z

Dene t (z0, ) : Z t [0, 1] as follows: t (z0, B) =


A1

At1 At

Q (zt1, dzt) Q (zt2, dzt1) Q (z0, dz1)

for any rectangle B = A1 At Z t. Can dene probability measure (z0, ) over Z . Use nite measurable rectangle B in Z : B = A1 A2 AT Z Z , At Z (z0, B) =
Z
A1

AT 1 AT

Q (zt1, dzt) Q (zt2, dzt1)Q (z0, dz1)

A stochastic process on (, F, P ) is an increasing sequence of -algebras F1 F2 F; a measurable space (Z, Z) ; and a sequence of functions t : Z such that each t is Ft-measurable. A nite dimensional probability measure is Pt+1,...,t+n (C) = P ({ : ( t+1 () , ..., t+n ()) C}) . A stochastic process is stationary if Pt+1,...,t+n (C) is independent of t for all n and C. A (rst-order) Markov process is a stochastic process with the property that Pt+1,...,t+n (C|ats, ..., at) = Pt+1,...,t+n (C|at) for all t, n and s.

Iterated Integral (Fubinis Theorem)

and

t t z , dz t F z 0 t Z Z Z t1, z Q (z t1 z , dz t1 = F z t t1, dzt) 0 Z t1 Z t t z , dz t F z 0 t Z Z Z t t = F z1, z2 t1 z1, dz2 Q (z0, dz1) Z Z t1 Z

You might also like