The rank-based unit root tests proposed by Hasan and Koenker (1997,Econometrica 65,133-161)have power equal ato size for normal innovations.Unit root tests based on M-estimators exhibit the same behavior.The problem occurs because the test statistics are transformed to obtain computationally convenient critical values.I describe a convenient way to compute critical values without transforming the test statistics.The resulting tests are almost as powerful as least squaresbased tests for normal errors and much more powerful for thicker tailed distributions.
展开▼