Univariate exchange rate forecasts : approximating nonlinearity of unknown form

Abstract

This thesis aims to improve exchange rate forecasts by using nonlinear meth ods. Most studies of exchange rates are based on linear models or linear approximations to nonlinear theories. This is because standard practice as sumes that a linear functional form can provide a satisfactory approximation to the data generating process. However, there is no a priori reason to pre fer linear to nonlinear methods. The popularity of the linear approach is due in part to its simplicity. The basic linear stochastic model immediately evokes the idea of a stable equilibrium continually disturbed by exogenous shocks. In addition, the linear model has several advantages compared to the nonlinear alternative: it reduces the computational burden of estima tion, simplifies interpretation and builds on an established body of statistical results. However, despite these advantages, there are both empirical and theoretical reasons why the linearity assumption does not serve economists studying exchange rates as well as it has in other fields.