ST 302 Stochastic Processes Thorsten Rheinlander London School of Economics and Political Science August 17, 2006 1 Information and Conditioning 1.1 Sigma-Algebras and Filtrations Let us recall that a -algebra is a collection F of subsets of the sample space such that (i) ;, 2 F (ii) if A 2 F then also Ac 2 F (here Ac denotes the complementary set to A) (iii) if we have a sequence A1, A2, ... of sets in F , then [1i=1Ai 2 F . We say that a collection of sets generates a -algebra F if F is the smallest -algebra which contains all the sets. The trivial -algebra is the one containing only ; and . Interpretation: We perform a random experiment. The realized outcome is an element ! of the sample space . Assume that we are given some partial infor- mation in form of a -algebra F . If F does not contain all subsets of , we might not know the precise !, but we may narrow down the possibilities. We know with certainty whether a set A 2 F either contains ! or does not contain it these sets are resolved by our given information. Example 1.1 We toss a coin three times, with the result of each toss being H (head) or T (tail). Our time set consists of t = 0; 1; 2; 3. The sample space is the set of all eight possible outcomes. At time 0 (just before the
rst coin toss), 1 we only know that the true ! does not belong to ; and does belong to , hence we set F0 = f;; g : At time 1, after the
rst coin toss, in addition the following two sets are resolved: AH = fHHH;HHT;HTH;HTTg AT = fTHH; THT; TTH; TTTg : We can say with certainty whether ! belongs to AH or AT . In contrast, the information about the
rst coin toss only is not enough to determine with certainty whether ! is contained e.g. in fHHH;HHTg for this we would have to wait until the second coin toss. As the complement of AH is AT , and the union of AH and AT equals , we set F1 = f;; ; AH ; ATg : At time 2, after the second coin toss, in addition to the sets already contained in F1, the sets AHH = fHHH;HHTg ; AHT = fHTH;HTTg ; ATH = fTHH; THTg ; ATT = fTTH; TTTg get resolved, together with their complements and unions. Altogether, we get F2 = ;; ; AH ; AT ; AHH ; AHT ; ATH ; ATT ; AcHH ; AcHT ; AcTH ; AcTT ; AHH [ ATH ; AHH [ ATT ; AHT [ ATH ; AHT [ ATT : At time 3, after the third coin toss, we know the true realization of !, and can therefore tell for each subset of whether ! is a member or not. Hence F3 = The set of all subsets of : As we have F0 F1 F2 F3 these four -algebras are said to form a
ltration. The random variable X = number of tails among the rst two coin tosses is said to be F2-measurable, but not F1-measurable (as we cannot determine the value of X after only one coin toss). 2 Consider now a discrete time stochastic process X = (X0; X1; X2; :::), i.e. a collection of random variables indexed by time. Hence X is a function of a chance parameter ! and a time parameter n. We would write Xn(!) for a function value, but typically suppress the chance parameter, otherwise the notation gets too heavy. Moreover, the two variables play quite di¤erent roles, since the chance parameter ! comes from the sample space (which can be a very large set, with no natural ordering) whereas the time parameter n is an element of the ordered set N+. We denote with Fn a family of sets containing all information about X up to time n. In more detail, Fn is the -algebra generated by sets of the form (we assume that the starting value X0 is just a deterministic number) fX1 = i1; X2 = i2; :::; Xn = ing if the state space is discrete. If the state space is R then Fn is the -algebra generated by sets of the form fX1 2 (a1; b1) ; X2 2 (a2; b2) ; :::; Xn 2 (an;bn)g with intervals (a1; b1), (a2; b2), ... We have F0 = f;; g (at time zero, we know nothing) F0 F1 F2 ::: Fn (the more time evolves, the more we know) (Fn) = (F0;F1;F2; :::) is called a
ltration. Sometimes we write