Estimating Parameters in Autoregressive Process During Rolling Periods

There are many techniques used to estimate autoregressive (AR) model parameters, but they have disadvantages such as high calculation times and errors. This study compared operation times and prediction mean square error during rolling periods between two different methods. One was the least squares (LS) method, and the other was the Yule-Walker (YW) method. After the linear system equations were obtained by the least squares method, they were solved and updated using invertibility method. The other linear system equations were obtained by the Yule-Walker method and they were solved using Durbin-Levinson Recursion and invertibility method. The study showed that the first method is the preferred technique for long rolling periods because it has less calculation time and it has same prediction mean square error.

Keywords: Autoregressive process, Estimating parameters, Least squares, Yule-Walker.

Corresponding author: E-mail: g4585017@ku.ac.th and fengpsa@ku.ac.th