Are PPP tests erratically behaved? Some panel evidence

Abstract

This paper examines whether, in addition to standard unit root and cointegration tests, panel approaches also produce test statistics behaving erratically when applied to tests for Purchasing Power Parity (PPP). We show that if appropriate tests (which are robust to cross‐sectional dependence) are used, any evidence of erratic behaviour disappears, and empirical support is found for PPP.

Keywords:

• Purchasing Power Parity
• PPP
• real exchange rates
• erratic behaviour
• panel tests

JEL Classifications:

• C12
• C23
• F31

Notes

1. See Caporale and Cerrato (2006) for a critical survey of the empirical literature testing PPP by means of panel methods. Another new development in the literature on real exchange rates is the modelling of nonlinearities (resulting, for instance, from transaction costs – see Taylor, Peel, and Sarno (2001)) in mean reversion. Some studies also allow for structural breaks (see, for example, Papell (2002)).

2. See, for example, Breitung and Pesaran (2008) or Gengenbach, Palm, and Urbain (2008) for excellent recent surveys of panel unit tests. It should be noted that most of the tests used here allow for cross‐section dependence using a factor structure and thus are effectively testing a joint null hypothesis of no unit root and no cross‐unit cointegration with an I(1) factor. Thus, if the null is rejected, this also constitutes evidence against cross‐unit cointegration relationships.

3. In practice, one can evaluate the numerical distribution functions obtained by MacKinnon (1994, 1996) via response surface regressions.

4. Although it relies on large‐N asymptotics, simulation evidence reported in Choi (2006) reveals an accurate behaviour of the tests for small N.

5. The findings were very similar when the GDP deflator was used instead of the CPI series (they are not reported here for the sake of brevity).

6. The only exceptions are the test statistics by Moon and Perron (2004), which are not significant over the entire sample period. However, even there, we observe a tendency of the test statistics to approach the rejection region.