Chaotic representation property of certain Azéma martingales (Q2505475)

A normal martingale with values in \(\mathbb{R}^n\) is called ``Azéma martingale'' if it satisfies a structure equation of the form \[ d[X^i,X^j]_t =\delta^{ij}dt+ \sum_{i,j}\varphi_k^{ij}(X_{t-})dX^k_t, \] where \(\Phi^{ij}_k:\mathbb{R}^n\to\mathbb{R}\) are affine functions. The central result of this paper is that, if \(l\) is a linear mapping from \(\mathbb{R}^n\) to some vector space \(E\), if \(Y=l(X)\) has all exponential moments and \(X\) is measurable w.r.t. \(\sigma(Y)\), then \(Y\) is unique in law and \(X\) satisfy the chaotic representation property (CRP). In particular, \(X\) satisfy the CRP if it has all exponential moments. A normal martingale is called a ``good Azéma martingale'' if it satisfies the above structure equation for \(\Phi_k^{ij}(x)=\delta^{ijk}(c_i-a_i\sum_jb_{ij}x^j)\), with \(a_i\geq 0\), the matrix \((b_{ij})\) symmetric and positive and \(a_ib_{ij}\leq 2\). The author shows that the good Azéma martingales are natural candidates for the CRP and deduce from the central result that the CRP holds for a large class of them.