Testing for unit roots based on sample autocovariances

Summary We propose a new unit root test for stationary null hypothesis H0 against alternative H1. Our approach is nonparametric as only assumes that the process concerned I(0) without specifying any parametric forms. The based on fact sample autocovariance function converges to finite population an while it diverges infinity with roots. Therefore rejects large values of function. To address technical challenge how large, we split and establish appropriate normal approximation null-distribution statistic. substantial discriminative power statistic rooted from takes value under This allows us truncate critical so has asymptotic one. It also alleviates loss due sample-splitting. implemented in user-friendly R