Prices in efficient markets are influenced by trading based on past pattern
s in the series. This induces parameter instability and near-random-walk be
haviour in any time-series model of such data. Simulation results suggest t
hat this parameter instability makes stationary series more likely To be er
roneously classified as nonstationary, according to standard unit root or s
tationarity tests. It is shown that individual real exchange rare series ap
pear individually non-stationary, especially for tests based on a null of s
tationarity, even though they appeal stationary when treated as a panel.