The chaotic-representation property for a class of normal martingales (Q2642930): Difference between revisions

Let \(Z=(Z_t)\) be a normal martingale, i.e., \(Z\) and \((Z^2_t)\) are martingales. Then, for all \(n\geq 1\) and deterministic square-integrable functions \(f\), the iterated stochastic integrals \[ \int_{\{0\leq t_1<\cdots< t_n\}}f(t_1,\dots,t_n) dZ_{t_1}\dots dZ_{t_n} \] are well defined. If these integrals, together with the constant functions, are dense in \(L^2 (\Omega,{\mathcal F},P)\) (here, \((\Omega,{\mathcal F},P)\) is the underlying probability space and \({\mathcal F}\) is generated by \(Z)\), then \(Z\) has the chaotic-representation property (CRP). It is simple to verify that this implies that \(Z\) has the predictable representation property (PRP) so that \[ d[Z]_t=\Phi_tdZ_t+dt\tag{*} \] for some predictable process \(\Phi\). (Here, (*) is called the structure equation for \(Z\).) In the present paper the following kind of converse is proved: Let \(Z\) be a normal martingale satisfying the structure equation \[ d[Z]_t=(\alpha(t)+\beta(t)Z_{t-}) dZ_t+dt \] where \(\alpha\), (being real valued) is locally bounded and \(\beta\) has values in the interval \([-2,0]\). Then \(Z\) is unique in law, possesses the CRP and is strongly Markovian (in an appropriate sense).