We show that, in practice, the standard unit root tests, cointegration tests, and similar tests are unreliable. This conclusion is more generally applicable to other regression-based tests. In particular, these tests attempt to solve a problem by creating another problem