Geary's count of sign changes as a test of unit roots: some further evidence

Abstract

In a recent issue of this journal, Honohan (1996) examined the perperformance of Geary's (1970) count of sign changes or x test as a test of unit roots. The test is based on counting the number of times the plot of a series crosses the trend line joining the first and last observations. The null hypothesis of a unit root is rejected when the number of crossings is "large". Honohan argues that the sign change test is a useful informal test mainly because of its simplicity — it can be calculated by hand and the 95 per cent critical values may be approximated by twice the square root of t he sample size. In the abstract to his paper, Honohan suggests that, despite its simplicity, the test has surprisingly good power.