Academic journal article Journal of Money, Credit & Banking

How Sure Are We about Purchasing Power Parity? Panel Evidence with the Null of Stationary Real Exchange Rates

Academic journal article Journal of Money, Credit & Banking

How Sure Are We about Purchasing Power Parity? Panel Evidence with the Null of Stationary Real Exchange Rates

Article excerpt

This article presents evidence on mean reversion in industrial countries' real exchange rates in a setup that accounts naturally for cross-sectional dependence, is invariant to the benchmark currency, and actually tests for the null of interest, that is, purchasing power parity. Our results are based on the Kwiatkowski et al. (1992) test for the stationarity null generalized in a multivariate random walk plus noise model by Nyblom and Harvey (2000).

FOR PURCHASING POWER PARITY (PPP) to hold in the long run, real exchange rates must be stationary. Permanent shocks to them would imply a permanent tendency for the purchasing power of the currencies to deviate from one another. Whether real exchange rates are stationary or nonstationary matters, since the two alternatives are associated with two quite different long-run economic implications. In the former case, PPP serves as a good first approximation to long-run behavior. This is the view of practicing economists when they base their long-run exchange rate forecasts on some measures of equilibrium real exchange rates, or make decisions on fixing parities between currencies. In the latter case, PPP serves no purpose, even over the long run. The finding of a unit root in real exchange rates would be problematic for many of the established theories. Furthermore, it would make long-run forecasting a useless exercise.

Having said this, does this mean we believe in the possibility of an "eternal" equilibrium real exchange rate? Few economists would go that far. There is evidence that price levels in rich countries tend to be higher than in poor countries when converted to a common currency, due to, for example, the Balassa-Samuelson effect. The evidence for the industrial countries is more debatable, however (see Rogoff 1996). Thus, what we are really testing empirically is whether permanent deviations from PPP are of relatively minor importance. In that sense, the long-run PPP, and its usefulness, is an empirical matter that can be tackled by testing for stationarity of real exchange rates.

Since PPP holds at best only as a long-run relationship, statistical inference on it depends critically on the number of available observations. Therefore, we consider it very important to use as much data as possible. Many of the PPP studies done in the past were based on a single time series due to the lack of suitable panel methods for testing. Tests for unit roots in panels have been developed only in the 1990s (see Im, Pesaran, and Shin 1997; Levin, Lin, and Chu 1997, hereafter LLC(1)). For the PPP tests, this has meant testing for the "wrong" null, that is, the null of the theory not holding. Empirical testing on relatively short univariate time series, for example, on post-1973 data, typically failed to support PPP. The emerging "consensus" of the failure of PPP started to shift back toward its acceptance in the 1990s. This occurred when studies using longer time series or panel methods were able to reject the unit root. Somewhat remarkably, both approaches arrived at similar speeds of adjustment to PPP, the half-life of deviations from PPP being in the range of three to five years [for discussion on the empirical results other than the very latest ones, see Froot and Rogoff (1995) for an excellent survey; for panel studies, see, for example, Coakley and Fuertes (1997), MacDonald (1996), Oh (1996), Wei and Parsley (1995), and Wu (1996)]. Support for PPP was considered to be due to increased power of the unit root tests with more observations and more variation in data.

However, one should be very careful in drawing conclusions based on these tests. The inference can be misleading since little is known about their capabilities of distinguishing the relevant alternatives in particular real-life time series. In testing for PPP, caution is needed in statistical decision making because the two competing processes are very similar in any finite samples, one being highly persistent but stationary and the other nonstationary. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.