The Effect of Regression Design on Optimal Tests for Finding Break Positions
27 Pages Posted: 10 Nov 2016 Last revised: 1 Feb 2017
Date Written: January 31, 2017
In this paper, we derive an optimal test for determining break positions in Gaussian linear regressions. The procedure is an admissible rule in a multiple decision theory setting and the results are exact and valid in small samples. The analysis indicates that regression design can have a very significant effect on the ability of the optimal test to find the position of the break. Some regression designs make it all but impossible to successfully identify a break location in certain subsections of the sample span. Two graphical devices, the cq and ω-plots are available to identify those subsets of the sample span where locating a break position is difficult or impossible.
Keywords: Structural Change, CUSUM Test, Bayes Rules, Multiple Decision Theory, Regression
JEL Classification: C01, C12, C32, C44
Suggested Citation: Suggested Citation