Purchasing Power Parity for Traded and Non-traded Goods: A Structural Error Correction Model Approach

When univariate methods are applied to real exchange rates, point estimates of autoregressive (AR) coefficients typically imply very slow rates of mean reversion. However, a recent study by Murray and Papell (2002) calculates confidence intervals for estimates of half-lives for long-horizon and post-1973 data, and concludes that univariate methods provide virtually no information regarding the size of the half lives. This paper estimates half-lives with a system method based on a structural error correction model for the nominal exchange rate, a domestic price index, a foreign price index, and a monetary variable. The method is applied to estimate half lives of real exchange rates based on producer price indices, consumer price indices, and GDP implicit deflators. The idea is that the traded component of the producer price index (PPI) is proportionately larger than that of the consumer price index (CPI). If the convergence rate is faster for traded goods prices than that for non-traded goods prices, half-lives for the real exchange rate based on the PPI should be shorter than those for the real exchange rate based on the CPI and that on the GDP implicit deflator. Our empirical results are consistent with this view.