In this paper the dynamics of foreign exchange rates is sought to be s
tudied via new frequency domain techniques. Stationarity properties of
the rates are analysed via a unit root test as well as a test based o
n the evolutionary spectrum. Linearity and Gaussianity are analysed vi
a bispectral tests and compared with the more frequently employed time
domain tests, such as the McLeod-Li and Tsay tests. Finally, an evalu
ation of the out-of-sample forecasting properties for eight methods-Ra
ndom Walk, ARMA, Bilinear, State dependent model, dynamic linear model
, ARCH, GARCH, and Garch-in-mean-is made. The methods used here seem t
o shed a great deal of light on hitherto neglected aspects of exchange
rate dynamics.