ESTIMATION OF A REGRESSION WITH THE PULSE TYPE NOISE FROM DISCRETE DATA

This paper considers the problem of estimating parameters in a periodic regression in continuous time with a semimartingale noise by discrete time observations. Improved estimates for the regression parameters are proposed. It is established that under some general conditions these estimates have an advantage in the mean square accuracy over the least squares estimates. The asymptotic minimaxity of the improved estimates has been proved in the robust risk sense. The properties of the proposed procedure for the models with non-Gaussian noises of pulse type have been studied. The pulse disturbances have random intensity and occur at random times which form a Poisson process.