Out-of-sample exchange rate forecasting structural and non-structural nonlinear approaches

Forecasting foreign exchange rates is a perennial dilemma for exporters, importers, foreign exchange rate traders, and the business community as a whole. Foreign exchange rate models using popular linear and non-linear specifications do not produce particularly accurate forecasts. In point of fact, these models have not improved much upon the random walk model, especially in out-of-sample forecasting. Given these results, this dissertation constructs and evaluates new forecasting models to generate as accurate as possible out-of-sample forecasts of foreign exchange rates. The information content of futures contracts on foreign exchange rates is investigated and used to forecast future exchange rates using alternative techniques, both structural (econometric) and non-structural (fuzzy) models. The results of two specifications of a structural model are compared against the well-known random walk model. The first specification assumes future exchange rates are determined by futures prices and a lagged structure of spot rates. The second specification assumes that future spot rates are a function of only a lagged structure of the futures prices. The forecasting accuracy of the models is tested for both in-sample and out-of-sample periods; out-of-sample tests range from the short term to the long term (30- to 180-day forecasts). The results indicate that the random walk model remains a competitive alternative. In out-of-sample predictions, however, we can improve upon it in certain cases. The results also show that the predictive accuracy of the models is better in the short term (30 to 60 days) than in the longer term (180 days)