Mean Break Tests in the Presence of Non-Stationarity

[alexecon]

This paper proposes break tests that are, apparently, robust to possible non-stationarity of a series regressed on a constant. This is the case not included in Kejriwal and Perron's (2010) paper of breaks in models with cointegrating series. It would supplement nicely the suite of RATS code for break tests.

Harvey, D. I., Leybourne, S. J., & Taylor, a. M. R. (2010). Robust methods for detecting multiple level breaks in autocorrelated time series. Journal of Econometrics, 157(10), 342–358. doi:10.1016/j.jeconom.2010.02.003

Thanks. Alex

[TomDoan]

This is now a probably 7 or 8 year old paper. Are there any interesting papers that have applied it to actual data?

[alexecon]

Yes, it's not a new paper. But it does address the question of what to do if one wants to test for possible mean shifts in a potentially non-stationary series using a constant-only model specification. Bai and Perron (2003) use that spec with the U.S. rate but do not address non-stationarity in the empirics. Kejriwal and Perron (2010) provide updated critical values for tests in models featuring cointegrated series which, of course, do not apply to a model with just a constant.

Not sure what qualifies as "interesting" but I have seen a couple of papers applying the test. One, e.g. is this one:

María José Presno, Manuel Landajo, Paula Fernández, Non-renewable resource prices: A robust evaluation from the stationarity perspective, Resource and Energy Economics, Volume 36, Issue 2, May 2014, Pages 394-416, ISSN 0928-7655, http://dx.doi.org/10.1016/j.reseneeco.2014.01.003.
(http://www.sciencedirect.com/science/ar ... 5514000207)
Keywords: Non-renewable resource prices; Structural changes; Sequential procedure; Stationarity testing

[TomDoan]

For a paper putting forward an "improved" test, I would like to see someone show that either (a) it gives different results than what's out there in an important case (and "different" can be shown to be "better") or (b) that it gets the same qualitative results as existing tests but is easier to use. I cringe when someone asks for code to accomplish someone out of JOE, as so many of its papers are theoretically correct and empirically worthless.