A Novel Technique for Estimation the Coefficients of ARMA Model Using Higher Order Moments

Time series data is observed in many different areas such as communication systems, signal processing, climate data and earthquake data. Statistical modeling and analysis of time series data includes transformation of the data into stationary times series and fit the time series model to the transformed data. Autoregressive Moving Average (ARMA) model is one of the most often used to fit the time series data. The proper estimation of the coefficients in the time series model is one of the important steps of modeling. In this study, a novel technique for estimating the coefficients of non-Gaussian ARMA model using higher order moments of the observed data. The proposed ARMA coefficients estimator is based on building a special matrix with entries of higher order moments of the observed output only. The observed output data may be corrupted with additive white Gaussian noise. Simulation results promise that the proposed method achieves performance comparable to existing well-known methods even when the available output signal is heavily corrupted with additive white Gaussian noise.