Test for mean reversion in real exchange rates under the current floating rate system
This dissertation explores the stationarity properties of real exchange rates under the current floating rate system using panel data of G8 countries. Two data-dependent model selection methods, the general-to-specific (GS) and the Schwarz information criterion (SIC), are used alternatively to select the lag length in the Augmented Dickey-Fuller test. Two panel data tests, the GLS and the Seemingly Unrelated Regression (SUR), are conducted. Homogeneity of the convergence speed is imposed on the GLS model. Within the SUR models, homogeneity is imposed on the restricted model and heterogeneity is allowed for the unrestricted model. For the CPI-constructed real exchange rates, the homogeneity models can reject the unit root null hypothesis at the 5% significance level. Monte Carlo simulations show that these tests provide considerable power against the unit root null. The heterogeneity model gives mixed results against the unit root and it has limited power. These findings are consistent with the results reported in the literature, for examples, Papell (1997), Taylor and Sarno (1998), and Higgins and Zakrajsek (2000). The power of test is compared between two lag selection methods as well as among the three models. Monte Carlo simulations indicate that the SIC is slightly more powerful than the GS and that the more restricted models perform better than less restricted models in term of rejecting the unit root null. For the WPI-constructed real exchange rates, the evidence against the unit root null hypothesis is not as strong as that from the CPI models. Only the SIC selected GLS model can reject the null at the 5% significant level while others have less significant results. Monte Carlo simulations show that the WPI-based models have lower test power against the unit root null than their CPI counterparts, perhaps because the WPI constructed real exchange rates are shorter time series. As in the CPI models, the SIC selected models have slightly stronger power against the unit root null than the GS selected ones.
