An Efficient Algorithm For Selecting Optimal Configurations Of Ar-Coefficients

INTRODUCTION With respect to Auto-Regressive (AR) modeling we distinguish the correct model, which generated the signal, and the optimal model. It can be a better strategy to estimate the optimal model, which has a simpler structure and can more reliably be estimated. Due to the finite sample size not all aspects of the correct model can be estimated with a sufficient accuracy, so we balance between a minimum estimation error (variance) and a minimum modeling error (bias). Assume an unbiased statistic s, BIAS fsg = 0, so the mean square error is MSE fsg = VAR fsg. Can we construct a better statistic? In mean square sense it can be done by s 0 = s. We search for the minimum mean square error as a function of and find: BIAS fs 0 g = (<F77.