Identification d'un processus autor\'{e}gressif gaussien stable par la m\'{e}thode de moyennisation logarithmique

In the present work, we consider a stable one-dimensional gaussian autoregressive model in continous time. Using the limit theorems with logarithmic averaging obtained for continous local martingales, we construct then an estimator of the noise covariance ${\sigma }^2$ and an estimator of $\theta$ different of the one of the least squares estimator. By exploiting the weighting method we ameliorate the convergence rates of these new estimators.