It is common to interpret rejections of the unit-root null hypothesis in favor of a trend stationary process with possible trend breaks as evidence that the data are better characterized as stationary about a broken trend. This interpretation is valid only if the model postulated under the alternative hypothesis is the only plausible alternative to the model postulated under the null. We argue that there are economically plausible models that are not well captured under either the null hypothesis or the alternative hypothesis of these tests. We show that applied researchers who ignore this possibility are likely to reject the unit-root null with high probability in favor of a trend stationary process with possible breaks. Our evidence shows that this potential pitfall is both economically relevant and quantitatively important. We explore the extent to which applied users may mitigate inferential errors by using finite-sample and bootstrap critical values.
