Rosenthal's inequality for point process martingales (Q1593634)

The Rosenthal's inequality for discrete-time martingales may be stated as follows: if \(p\geq 2\), then \( c_p A_{p,n} \leq E|S_n|^p = C_p A_{p,n} \) where \(\{S_n,{\mathcal F}_n\}_{n=1,\ldots}\) is a discrete-time martingale with \(S_0=0\), \(c_p\) and \(C_p\) are positive constants depending only on \(p\), and \[ A_{p,n} = E\left \{\left (\sum _{i=1}^n E[X_i^2\mid {\mathcal F}_{i-1}] \right)^{p/2} + \sum _{i=1}^n |X_i|^p\right \} \] with the martingale difference sequence \(X_i = S_i - S_{i-1}\), \(i=1,2,\ldots \) The author proves the continuous-time counterpart, namely he deals with the process \( S_t = S_t(g) = \int _0^t g_s dM_s \) where \(g_s\) is a predictable process, \( M_s = N_s - \Lambda _s \) is a local martingale since \(N_s\) is a counting process and the compensator \(\Lambda _s = \int _0^s \lambda _u du\) with a predictable intensity \(\lambda _s\) is assumed to be almost surely finite. Then the Rosenthal's inequality remains true for \(p>2\) with \[ A_{p,T} = E\left[\left(\int _0^T g_s^2 \lambda _s ds\right)^{p/2} + \int_0^T |g_s|^p\lambda _s ds\right]. \] The classical Burkholder-Davis-Gundy inequality is also considered and the relations are discussed.