Abstract
Empirical studies have flourished on long-run PPP using panel unit root tests, which supposedly have high power. In this paper, we adopt, in a panel data context, a nonlinear multiple-regime model, namely Threshold Autoregression (TAR), and perform a panel unit root test for each regime in the TAR. This new procedure takes advantage of two existing approaches: the inference for a TAR model with a unit root, and the panel unit root tests with the augmented panel Dickey-Fuller regression. The real exchange rate dynamics in a panel of 17 OECD countries over the recent floating exchange rate period are investigated. Three distinct regimes are identified. In particular, the support for long-run PPP is much stronger in the post-Maastricht-Treaty period than the period before; in the preMaastricht-Treaty period, there is some evidence for long-run PPP when the sterling-dollar appreciated, while the evidence is weak when the sterling-dollar depreciated. Our results are robust to expanding the data set to 2012, using the euro foreign exchange reference rates.