06.06.2013 Views

STOCHASTIC

STOCHASTIC

STOCHASTIC

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

as T-* oo. If {A",} is a stationary process, ergodic arguments may often be applied to obtain<br />

the limit. If {X,} is nonstationary but is asympotically stationary, ergodic arguments will<br />

sometimes be applicable, but may require tedious and technical limiting arguments. In this<br />

problem, an important special case is developed using "elementary" arguments based on<br />

Exercises 14 and 15.<br />

Suppose that {X„, n = 0,1,...} is a stationary, irreducible Markov chain with finitely<br />

many states {\,...,K}, all positive recurrent.<br />

Theorem For any finite, real-valued function/on {l,.,.,K}, and for any starting state<br />

X0 = i,<br />

Km i S/W,) = £;/(*,),<br />

W-.00 VVn=0<br />

where E„ is the expected value with respect to the limiting distribution n.<br />

(a) Prove the theorem when f(k) =1 for i = k and f(k) = 0 otherwise. [Hint: Use<br />

Exercises 14(f) and 15(fc).]<br />

(b) Prove the theorem for general /.<br />

(c) Prove that the theorem holds if the lower limit n = 0 in the sum is replaced by any<br />

fixed finite value.<br />

17. Let {X„, H = 0, 1,...} be a stationary Markov chain with finitely many states and<br />

transition matrix P. Let plj be the probability that X„ = j, given X0 = i, and let f"s be the<br />

probability that starting in j the first transition to j occurs at time n S 1, and let ffi = 0.<br />

Define<br />

PiM = f Pii* a^ FtJ(s) = £ /,>"•<br />

n = 0 rt = 0<br />

(a) Show that Pij(s) and Fu(s) exist for all i,j when \s\ < 1.<br />

(b) Show that<br />

F,j(.s) = P,MIPJJ(S) OV;) and F„(s) = 1/(1-P(((i)).<br />

(c) Show that PtJ(s) satisfies the equations<br />

where<br />

PIJ(S) = 6 u + sJ^p,kPk]{s)<br />

h<br />

lo, i * j.<br />

The expected first passage time Uu from i to y is defined to be the expected time of the first<br />

visit toy, starting from i; that is,<br />

M,J= t,nfft.<br />

n = 0<br />

(d) Show that the utJ satisfy the equations<br />

Ha = 1 + Y. Pikfkjk*l<br />

18. Consider a Bernoulli random walk {A^} which drifts between barriers at 0 and h, such<br />

that the process returns instantly to z (0 < z < h) upon hitting either barrier. Let the probabilities<br />

of positive and negative unit steps be p and q, respectively, where p, q > 0 and<br />

p + q = \.<br />

692<br />

PART V DYNAMIC MODELS

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!