On parameter estimation of the hidden Gaussian process in perturbed SDE (Q2219225): Difference between revisions

The authors propose a new estimator for the parameters of a partially observed linear Gaussian process. Let's look at the problem under consideration. Let $(\Omega,F,P)$ be a probability space, $T>0$, $X$, $Y$, $W$, and $V$ real valued stochastic processes defined over $\Omega\times[0,T]$, $\Theta=\{(\alpha,\beta)/|\alpha|+|\beta|<\infty\}$ such that \[ \begin{aligned} dX_{t} &= f(\theta,t)\cdot Y_{t}\cdot dt+\varepsilon\cdot\sigma(t)\cdot dW_{t},\quad X_0=0,\\ dY_{t} &= a(\theta,t)\cdot Y_{t}\cdot dt+\varepsilon\cdot b(t)\cdot dV_{t},\quad Y_0=y_0\neq 0, \end{aligned} \] where $W$ and $V$ are independent Wiener processes, $f$ and $a$ are (known) real functions defined over $\Theta\times[0,T]$, $\sigma$ and $b$ are (known) real functions defined over $[0,T]$, and $\varepsilon$ is a positive constant. Here the problem is to estimate the parameter $\theta$ from observations of $X$. There are already estimators of the referred parameter, as maximum likelihood estimator (MLE) and Bayesian estimator (BE), but they are hard to calculate. In this paper, the authors propose a new estimator the computation of which is simpler and the consistency and efficiency properties are as reliable as those of the existing estimators. The last section of the paper shows the application of the proposed method in two examples with simulated data. Unfortunately, no results of applications to examples with real data are shown. Most of the main theorems and claims are rigorously and carefully proved. However, understanding this work requires a good knowledge of the mathematics of stochastic differential equations.