Efficient estimation of parameters for non-Gaussian autoregressive processes

The problem of estimating the parameters of a non-Gaussian autoregressive process is addressed. Departure of the driving noise from Gaussianity is shown to have the potential for improving their accuracy of the estimation of the parameters. While the standard linear prediction techniques are computationally efficient, they show a substantial loss of efficiency when applied to non-Gaussian processes. A maximum-likelihood estimator is proposed for more precise estimation of the parameters of these processes coupled with a realistic non-Gaussian model for the driving noise. The performance is compared to that of the linear prediction estimator and, as expected, the maximum-likelihood estimator displays a marked improvement. >