Some aspects of robust estimation in time series analysis.

In this thesis, a number of robust methods have been developed for estimating the parameters in a time series setting. To estimate the power spectrum of an ARMA process, an M estimation method has been introduced which maximizes the robust likelihood function of the discrete Fourier transforms of the process. This robust method is useful in estimating the parameters of the continuous spectrum ARMA process by downweighting the influence of possible discrete spectrum harmonic components on the data. The proposed M estimation method has been applied to some actual time series data sets of sea level records, where a strong presence of tidal (harmonic) components is observed along with the continuous spectrum surge process. Here robust estimation of the power spectrum of the surge process has been considered assuming that the surge follows an ARMA process.