A critique of recent methods for computing the distribution of the Durbin-Watson and other invariant test statistics

In recent papers Ansley, Kohn and Shively (SAK, 1990; AKS, 1992; KSA, 1993) have proposed methods for evaluating the null distribution of the Durbin-Watson and related statistics more rapidly than by the methods of Davies (1980) or Farebrother (1980, 1990). In this paper we discuss several problems concerning the numerical accuracy of these new procedures and recommend that practitioners intending to use them should do so with caution.