Testing for unit roots in the context of misspecified logarithmic random walks

We show both analytically and by Monte Carlo that the null distribution of the Dickey–Fuller-test is extremely non-robust to improper transformations of the data.