In this article, we carry out unit root tests on real exchange rates recursively as in Caporale et al. (2003), but, following Arghyrou and Gregoriou (2007), we adjust the residuals for non-normality and heteroscedasticity using a wild bootstrap method. The results are striking: this correction sharply increases the rejection percentages of the unit root null and attenuates the erratic behaviour of the t-statistic, providing evidence in favour of purchasing power parity (PPP) and suggesting that such a correction might at least go some way towards solving the 'PPP puzzle'.