Filtering and parameter estimation in a simple linear system driven by a fractional Brownian motion (Q1265972): Difference between revisions

A normalized fractional Brownian motion \(B_H(t)\) with Hurst parameter \(H\in (1/2, 1)\) is a Gaussian process with continuous paths and stationary increments such that \(B_H(0)=0\), \(EB_H(t)=0\), and \(EB_H^2(t)=| t| ^{2H}\). Let \(\xi\) be a random variable independent of \(\{B_h(t)\}\) and define \[ Y_t=\int_0^t a(s)\xi ds + \int_0^t b(s) dB_H(s),\qquad t\geq 0. \] The author investigates the best estimator \(\hat{\xi}_t\) of \(\xi\) based on \(\{Y_s,\;0\leq s\leq t\}\). It is shown that the problem is connected with the solution of a certain integral equation. The special case \(b(s)=\lambda a(s)\) is analyzed in detail. The results are applied to the deterministic regression model where \(\xi=\theta\) is an unknown parameter. A Girsanov-type formula is proved which shows that the best linear unbiased estimator of \(\theta\) coincides with the maximum likelihood estimator.