Poisson approximations for time-changed point processes (Q1106547): Difference between revisions

Let N be the point process belonging to a strictly increasing unbounded sequence of finite stopping times \(0<T_ 1<T_ 2<..\). of a filtration \({\mathcal F}=({\mathcal F}(t))_{t\geq 0}\) and denote A its ``compensator'', that is the unique \({\mathcal F}\)-predictable stochastic process with increasing paths such that \(A(0)=0\) and N-A is a local martingale, and \(\hat N\) its ``time transformed point process'', that is the point process belonging to the sequence \(0<A(T_ 1)<A(T_ 2)<\dots\) One important result on time transformed point processes states (Meyer 1971): Let \(N = (N_1, N_2, \dots, N_ k)\) be a vector of point processes. Suppose that A1) for any \(i \neq j\) the probability that \(N_ i\) and \(N_ j\) have common jumps is 0, and A2) the compensators \(A_ i\) of \(N_ i\) are continuous, then the vector \(\hat N = (\hat N_1, \hat N_2, \dots, \hat N_ k)\) of the time transformed point processes is a vector \(\Pi = (\Pi_1, \Pi_2, \dots, \Pi_ k)\) of independent unit rate Poisson processes. This statement is not true, if the two assumptions A1) and A2) do not hold. In that case, this paper provides an explicit bound for the total variation distance \(d(\hat N(t), \Pi (t))\).