Iterated tempered stable process (Q6880302)

A subordinator is a one-dimensional increasing Lévy process \(S_t\) (i.e.\ a process with stationary and independent increments and right-continuous sample paths with left limits). The Laplace transform of \(S_t\) is of the form \(\mathbb{E} e^{-\lambda S_t} = \exp(-tf(\lambda))\) where \(f\) is a Bernstein function with characteristic tuple \((b,\mu)\): \N\[\Nf(\lambda) = b\lambda + \int_{(0,\infty)}\left(1-e^{-s\lambda}\right)\nu(ds)\N\]\Nwith \(b\geq 0\) and the jump or Lévy measure \(\mu\). If \(\nu(ds)\) is replaced by \(e^{-\mu s}\,\nu(ds)\), one obtains the tempered Bernstein function \(f_\mu(\lambda) = f(\lambda+\mu)-f(\mu)\) and the corresponding subordinator is a tempered subordinator. Tempering has the advantage that in small time the features of the original subordinator are preserved, while in large time one has e.g.\ all moments.\N\NThe authors study two independent stable subordinators (i.e.\ \(b=0\), \(\nu_i(ds) = c_{\alpha_i} s^{-1-\alpha_i}\,ds\), \(f_i(\lambda) = \lambda^{\alpha_i}\)), temper each of them to get the tempered processes \(S_t, T_t\) and then subordinate them, i.e.\ make the random time change \(U_t = S_{T_t}\). The authors call the process \(U_t\) (which is again a subordinator with Bernstein function \((f_2\circ f_1)(\lambda)\) an iterated tempered stable process.\N\NThe paper establishes asymptotic formulae (as \(t\to\infty\)) for the (fractional) moments of \(U_t\) and the tails \(\mathbb{P}(U_t > x)\), and they establish a ``master equation'' for the transition density \(h(t,x)\) of \(U_t\) of the form \N\[\Nf_1\left(\frac d{dx}\right)h(t,x) = f_2^{-1}\left(-\frac d{dt}\right) h(t,x).\N\]\NSimilar formulae are obtained for the process \(V_t = S_{E_t}\) where \(E_t\) is the pathwise (generalized) inverse of the process \(T_t\).\N\NIf one observes that \(U_t\) is itself a subordinator (with Bernstein function \(g=f_2\circ f_1\)), most results can be immediately derived from more general well-known results from the literature, e.g.\ Theorem 3.2 follows from\N\begin{gather*}\N\mathbb{E} U_t = tg'(0),\quad \mathrm{Var}(U_t)= -tg''(0),\quad \mathrm{Cov}(U_t,U_s) = \mathrm{Var}(U_{s\wedge t}) = -(s\wedge t)g''(0).\N\end{gather*}\NTheorem 3.4 is an application of dominated convergence and the SLLN \(U_t / t \to \mathbb{E}U_1\) almost surely, so\N\begin{gather*}\N\lim_{t\to\infty} \mathbb{E}\left[\left(\frac{U_t}{t}\right)^r\right] = \left(\mathbb{E} U_1\right)^r,\quad r>0.\N\end{gather*}\NThe tail estimates are typical results form the theory of regular variation, see the monograph [\textit{N. H. Bingham} et al., Regular variation. Cambridge: Cambridge University Press (1987; Zbl 0617.26001)] or, for a different proof, [\textit{A. Mimica}, Proc. Lond. Math. Soc. (3) 113, No. 5, 627--648 (2016; Zbl 1362.60068)].