Quantile unit root test and PPP: evidence from 23 OECD countries

ABSTRACT

Application of six different univariate unit root tests to real effective exchange rates of 23 OECD countries supports their stationarity or Purchasing Power Parity theory (PPP) only in five countries, a result consistent with previous research that is known as PPP puzzle. However, descriptive statistics of each effective rate reveals a clear sign of non-normal distribution. To account for this, we use quantile unit root test which allows impact of different shocks to be realized at different quantiles. When we applied this new test to the same rates, number of countries in which PPP is supported increased to 16. Apparently, incorporating effects of shocks improves testing efficiency and provides more support for the PPP and reduces the severity of the puzzle.

KEYWORDS:

• PPP
• quantile unit root test
• OECD

JEL CLASSIFICATION:

• F31

Acknowledgements

Valuable comments of two anonymous referees are greatly appreciated. Remaining errors, however, are our own.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ For a few review articles, see Sarno (2005), and Bahmani-Oskooee and Hegerty (2009).

² Note that similar quantile tests have been applied by Nikolaou (2008) to the real bilateral exchange rates between euro and the US dollar, British pound and the US dollar and Japanese yen and the US dollar and by Ferreira (2011) to pairwise real exchange rates between the Italian lira, French Franc, Deutsch mark and the British pound. Our study is more comprehensive than these not only in terms of country coverage but also in terms of using real effective exchange rates rather than bilateral rates. For another application of quantile test to persistence of the US inflation, see Wolters and Tillman (2015).

³ The data come from Bank for international Settlement, http://www.bis.org/statistics/eer/index.htm.

⁴ To save the space, we do not report the results for break dates. They are available from the authors upon request.

⁵ See Gospodinov (2004) for more details.

⁶ To save the space, we did not report the results for optimum lag(s). They are available from authors upon request.

⁷ Grey dashed lines show 95% bootstrap confidence intervals of ${\hat{\alpha }}_0(\tau )$ which are obtained from bootstrapping procedure.

⁸ Note that the real effective exchange rate data are in index form. Thus with re-basing the series the intercept may get different sizes. Hence, we cannot reliably equate the size of intercept to the size of shock when comparing across countries. We thank the referee for this comment. Furthermore, it should be pointed out that the findings here could be sensitive to the well-known quantile-crossing problem.