TAR models and real exchange rates

The recent past has seen an increased interest in piecewise linear real exchange rate models. By invoking Heckscher's (1916) 'commodity points' it has been argued that a threshold autoregressive (TAR) model should be used to study movements in the real exchange rate. This paper examines the problems of fitting TAR models to real exchange rates. We find that the power of the tests for TAR behavior can be very low for realistic parameter settings. Moreover the confidence intervalls for the threshold parameter are too wide to be used for economic analysis.