Nonlinearity in real exchange rates: an approach with disaggregated data and a new linearity test

Abstract

Employing disaggregated real exchange rates from nine European counties in 16 goods categories, we assess in this study the nonlinearity in the real exchange rates. Surprisingly, we find evidence for nonlinearity in only four (10) out of 143 series with the linearity test proposed by Harvey et al. (2008) at the 5% (10%) significance level. This result differs greatly from those of Juvenal and Taylor (2008), Imbs et al. (2003), Sarno et al. (2004) and Berka (2009), who report ample evidence for nonlinearity for the same or similarly disaggregated real exchange rate datasets.

Acknowledgements

This work was supported by Research Program 2010 of Kookmin University in Korea. The author thanks an anonymous referee for helpful comments.

Notes

¹Many other linearity tests have been developed in the literature. These tests generally assume that the variables under examination are either stationary or nonstationary . Owing to space considerations, we do not present here the number of other available linearity tests; see Yoon (2009a) for some references.

²To implement the linearity test, I used the GAUSS program, which is available at http://www.bepress.com/snde/vol12/iss3/art2/

³The data series are available at http://www.bepress.com/snde/vol12/iss3/art8/

⁴Their sample period is 1975:01–1996:12.

⁵The countries are the UK, France, Germany, Italy and Japan against the US for nine goods categories for the sample period 1974:Q1–1993:Q4. Two series are unavailable from Japan. Sarno et al. (2004) also contain simulation study on the power of the Hansen's (1997) linearity test, which they find very satisfactory for their data-generating processes.

⁶When the linearity test is applied to the first differenced disaggregated real exchange rates, 11 (23) series are found to be nonlinear at the 5% (10%) significance level.

⁷Note that for the tests different test statistics are found at different significance levels, as a significance level-dependent parameter is involved. The tests also do not require that the order of integration be known a priori.

⁸For the real exchange rates in the first difference, only 18 (37) series are nonlinear with the test.