Testing for Trend in the Presence of Autoregressive Error: A Comment

Roy, Falk and Fuller (2004) presented a procedure aimed at providing a test for the value of the slope of a trend function that has (nearly) controlled size in autoregressive models whether the noise component is stationary or has a unit root. In this note, we document errors in both their theoretical results and the simulations they reported. Once these are corrected for, their procedure delivers a test that has very liberal size in the case with a unit root so that the stated goal is not achieved. Interestingly, the mistakes in the code used to generate the simulated results (which is the basis for the evidence about the reliability of the method) are such that what they report is essentially equivalent to the size and power of the test proposed by Perron and Yabu (2009), which was shown to have the standard Normal distribution whether the noise is stationary or has a unit root.