Time series AR modeling with missing observations based on the polynomial transformation

Abstract

This paper focuses on parameter estimation problems of auto-regression (AR) time series models with missing observations. The standard estimation algorithms cannot be applied to such AR models with missing observations. The polynomial transformation technique is employed to transform the AR models into models which can be identified from available scarce observations, then the extended stochastic gradient algorithm is proposed to fit the time series with missing observations. The convergence properties of the proposed algorithm are analyzed and an example is given to test and illustrate the conclusions in the paper.