We derive formulae for the asymptotic density and distribution functions of
the t-statistic for autoregressive unit roots based on M-estimators. These
formulae depend upon a nuisance parameter. Consequently, new critical valu
es for this test would have to be generated for each different M-estimator
that is used. We therefore provide a methodology for tractable and accurate
analytic approximations of quantiles based on asymptotic expansions of fun
ctions, We use it to derive simple yet accurate approximations for the asym
ptotic distribution of these unit root M-tests. Using these asymptotic appr
oximations, critical values of the tests can easily be obtained without eve
n resorting to a computer. The approximation requires no new tabulation, an
d the resulting cumulative distribution function (cdf) has a maximum absolu
te error of 0.002 for typical quantiles (i.e. 1-10% quantiles). (C) 2000 El
sevier Science S.A. All rights reserved. JEL classification: C22.