Equivalence of Volterra processes. (Q2574600)

Given a nondegenerate kernel \(K\in L^2([0,T]\times [0,T])\) and a Brownian motion \(W\), the corresponding Volterra process is the Gaussian centered process \(X\) on \([0,T]\) defined by \(X_t=\int _0^tK(t)\,dW\). In particular, the fractional Brownian motion with the Hurst parameter \(0<H<1\) is a Volterra process. Two Volterra processes \(X^1\), \(X^2\) are said to be equivalent provided that their laws are absolutely continuous with respect to each other. The first part of the paper deals with necessary conditions for two Volterra processes to be equivalent while in the second part a sufficient condition for equivalence is given. The necessary conditions are originally due to T. Hida and M. Hitsuda [see e.g. \textit{M. Hitsuda}, Osaka J.\ Math.\ 5, 299--312 (1968; Zbl 0174.49302)] but the presented proof is new as well as the description of the Radon-Nikodym density of the respective laws. In the next section, the previous results are applied on a particular case of Volterra processes. Firstly, the fractional Brownian motion \(B^H\) with the Hurst parameter \(0<H<1\) is shown to be equivalent to \(d_H\int _0^t(t-s)^{H-1/2}\,dW_s\) where \(d_H\) is a suitable constant. Secondly, every Volterra process \(X\) with the kernel \(K\) is equivalent to the unique solution \(X^\lambda \) of the linear stochastic Volterra-type equation \(X^\lambda _t=\lambda \int _0^tK(t,s)X^\lambda _s\,ds+X_t\). The process \(X^\lambda \) is called the resolvent process of \(X\) at \(\lambda \), it is a Volterra process and its kernel is characterized. At the end, two particular examples are computed explicitely. The fourth section is devoted to a deterministic time change for a fractional Brownian motion with the Hurst parameter \(1/2<H<1\). It is proven that, given a strictly positive function \(f\) locally in \(W^{1,2}\), the process \(\int _0^tf\,dB^H\) is equivalent to the time-changed process \(B^H_{T(t)}\) where \(T(t)=\int _0^tf^{1/H}(s)\,ds\). The last section deals with an extension of the Cheridito theorem [\textit{P. Cheridito}, Bernoulli 7, 913--934 (2001; Zbl 1005.60053)]: Let \(W\) be a standard Brownian motion and \(X\) a centered Gaussian process, independent of \(W\), with a twice continuously differentiable (except for the diagonal) covariance function \(R(s,t)\) and let the second cross derivative \(R_{st}\) be \(L^2\)-integrable on the square \([0,T]^2\). Then the process \(X+W\) is a semimartingale with respect to its own filtration and it is equivalent to a Brownian motion.