A non-linear model of the real US/UK exchange rate

Abstract

This paper provides a framework for building and estimating non‐linear real exchange rate models. The approach derives the stationary distribution from a continuous time error correction model and estimates this by MLE methods. The derived distribution exhibits a wide variety of distributional shapes including multimodality. The main result is that swings in the US/UK rate over the period 1973:3 to 1990:5 can be attributed to the distribution becoming bimodal with the rate switching between equilibria. By capturing these changes in the distribution, the non‐linear model yields improvements over the random walk, the speculative efficiency model, and Hamilton's stochastic segmented trends mode..