The tail estimation of the quadratic variation of a quasi left continuous local martingale (Q2472331)

Let \(M= (M_t)\) be a càdlàg locally square integrable martingale such that \(M_0= 0\) which is defined on some filtered probability space \((\Omega,{\mathcal F},({\mathcal F}_t),P)\) satisfying the usual conditions. Let \(\mu\) denote the random measure on \(\mathbb{R}_+\times(\mathbb{R}\setminus\{0\})\) such that for all \(t> 0\) and Borel sets \(U\subset\mathbb{R}\setminus\{0\}\), \[ \mu(\cdot,(0,t]\times U)= \sum_{0< s\leq t} 1_U(\Delta M_s)\qquad (\Delta M_s= M_s- M_{s-\mu},s> 0). \] Let \(\widehat\mu\) denote the predictable compensator of \(\mu\). In a paper by \textit{R. Liptser} and \textit{A. Novikov} [Tail distributions of supremum and quadratic variation of local martingales. From stochastic calculus to mathematical finance. Berlin: Springer. 421--432 (2006; Zbl 1110.60044)] the following result was obtained (\({\mathcal T}\) denoting set of all stopping times). Theorem. Let \(\langle M\rangle_\infty= \lim_{t\to\infty} \langle M\rangle_t< \infty\) a.s., and let \(\{M^+_\tau:\tau\in{\mathcal T}\}\) be uniformly integrable. Then (i) \(0\leq E[M_\infty]\leq E[M^+_\infty]<\infty\). (ii) If \(\{\Delta M_\tau:\tau\in{\mathcal T}\}\) is uniformly integrable, then \[ \lambda P\Biggl(\sup_{t\geq 0} (M^-_t)>\lambda\Biggr)\to E[M_\infty] \] as \(\lambda\to\infty\). (iii) If \(|\Delta M|\leq K\) and \(E[\exp(\varepsilon M_\infty)]< \infty\) for some \(K> 0\) and \(\varepsilon> 0\), then \[ \lim_{\lambda\to\infty}\lambda P(\langle M\rangle_\infty>\lambda^2)= \lim_{\lambda\to\infty}\lambda P([M]_\infty> \lambda^2)= (2/\pi)^{1/2}E[M_\infty]. \] The main results of the present paper include the following result (here, \(\varphi_\lambda(x)= \exp(-\lambda x)- 1+ \lambda x-(\lambda^2/2) x^2)\): Theorem. Let \(M\) be additionally quasi-left continuous, \(\langle M\rangle_\infty<\infty\) a.s. and let \(\{M^-_\tau:\tau\in{\mathcal T}\}\) be uniformy integrable. Assume that, for some \(\lambda_0> 0\) and \(K> 0\), \[ E\Biggl[\int_{\mathbb{R}_+\times\{|x|> K\}}|\varphi_{\lambda_0}(x)| \widehat\mu(\cdot,ds\,dx)\Biggr]< \infty. \] Then, as \(\lambda\to\infty\), \[ \lambda P(\langle M\rangle_\infty> \lambda^2)\to -(2/\pi)^{1/2}E[M_\infty]. \] A similar result is obtained for \([M]_\infty\).