Nonstationary statistical tests in time-scale space
Ahmed Hambaba · 2002
Detection and estimation of abrupt changes in a nonstationary environment is an important and challenging problem. A statistical test for selecting the order of a nonstationary AR model is presented based on wavelet vanishing moment and predictive least squares principle. The order of the nonstationary AR is estimated at a different resolution level, which makes the order reliable. Its confidence interval is also determined. The test is derived based on the order probability distribution of the wavelet approximation sequence.