On the small-time behavior of subordinators (Q442077)

The authors examine the behavior of a random process \((Y_t)_{t>0}\) with positive values near \(t=0\). And, in order to more accurately describe the objectives and results of their work, they mention the following consideration of a previous work of the first author. Proposition. Let \(F_t(y)=\operatorname{P}(Y_t\leq y)\), \(\psi_t(u)=\operatorname{E}(e^{-uY_t})\) and \(\overset{d}{\longrightarrow}\) denote convergence in distribution. Let also be assumed that \(F_t(0)=0\) for all \(t\) and \(Y^*\) be a random vector with distribution function \(F^*\) which is not concentrated at one point. Then \(Y_t^{-t}\overset{d}{\longrightarrow}Y^*\) as \(t\to 0\) if and only if \(\psi_t(u^{1/t})\to 1-F^*(u)\) as \(t\to 0\) at all continuity points \(u\) of \(F^{*}\). In the light of this result, it is possible to say that, in this paper, the authors study the case when \(\psi_t(u)=\psi(u)^t\) for some Laplace-Stieltjes transform \(\psi\). This means that \(F\) is infinitely divisible, and the \((0,\infty)-\)valued process \((Y_t)_{t>0}\) (with \(Y_0\equiv 0\)) can then be interpreted as a subordinator (real-valued Levy process with nondecreasing sample paths) with Laplace exponent \(\Phi(u)\) in \(\psi_t(u)=e^{-t\Phi(u)}\). As for the results, we can note three things. First, in another study by the same authors, it was proved that, for a subclass of exponential dispersion models generated by an infinitely divisible probability measure \(\mu\) on \([0,\infty)\) and associated with an unbounded Levy measure \(\nu\) satisfying \(\nu((x,\infty))\sim -\gamma \log x\) as \(x\to 0\), the limit \(F^*\) is a Pareto-type law supported on \([1,\infty)\). The main result here shows that this is indeed the only limit law that can occur for any subordinator. Second, they give several necessary and sufficient conditions for this convergence to occur. Third, the results presented in this paper enable an approximation of the distribution of \(Y_t\) for relatively small values of \(t\). The useful effect of this is that, while the distribution of \(Y_t\) can be quite complex, the specific limiting Pareto law is rather simple to handle.