## Bent Nielsen

### 14 May 1997

Abstract

Asymptotic analyses of unit root tests in autoregressive time series are usually based on the assumptions that the number of unit roots is known and that the remaining characteristic roots are stable. The last assumption seems not to be necessary. This is stated more precisely for two examples. One is a unit test in a univariate second order model, the other a test for at most one cointegrating relation in a bivariate first order model.