to investigate the validity or otherwise o f the PPP hypothesis within the SACU area. We are able to find evidence in support of the PPP hypothesis using Larsson et al. (1998Larsson et al. ( , 2001. The Larsson et al.'s (1998Larsson et al.'s ( , 2001 procedure is an extension of Johansen's (1988Johansen's ( , 1995
methodology that allows for estimation of the number of co-integrating relations. This offers an interesting alternative to the residual based co-integration tests.
IntroductionThe Purchasing Pow er Parity (PPP) hypothesis is the hypothesis that exchange rates betw een currencies are determ ined in the long ru n by the a m o u n t of goods and services that each can buy. A basic n o tio n is that in the absence of trade im pedim ents, if the price of tradables were low er in one country than another, trad ers could gain by buying goods in the cheaper country an d selling in the dearer. Consequently, relative price levels determ ine the equilibrium exchange rate. The first person to treat PPP as a practical em pirical theory is Cassel (1921Cassel ( , 1922. PPP can be used for a w id e range of applications. For instance, in: choosing the right initial exchange rate for a new ly independent country; forecasting the real exchange rate an d adjusting for price differential in international com parisons of incomes.This p ap er em ploys panel data unit root and co integration tests to investigate the validity of PPP hypothesis in SACU area. The rationale for using panel data unit root and co-integration techniques is that they offer a significant improvement in addressin g the low -pow er problem of the conventional tests, especially, in developing countries where there are serious data limitations; and facilitate pooling of long ru n inform ation contained in the panel, while permitting the short ru n dynam ics and fixed effects to be heterogeneous am o n g different m em bers of the panel. Moreover, long-run relations betw een tw o integrated panel vectors can exist even though there is no individual time series co-integration (Phillip an d Moon, 1999). Most of the conventional tests suffer from size distortions (probability of falsely rejecting the null of "non-stationarity" is high, especially, w h en the true data generating process (d.g.p) is a nearly stationary process). There is a trade off between size and pow er in that un it root tests m ust have either high probability of falsely rejecting the null of non-stationarity when the true data generating process is a nearly stationary process or low p o w er against any stationary alternative (Blough, 1992). They are very sensitive to the choice of lag length, and the inclusion or otherw ise of constant and trend terms. M oreover, there is supporting evidence that it is the span of data not the frequency of data that m atters for the power of these tests (Pedroni, 1997). Therefore, unit root and co-integration tests based on quarterly data do not necessarily have better pow er than those based on corresponding annual data. In the light of these data li...