Revisiting Non-Parametric Exchange Rate Prediction

Main Article Content

Nikola Gradojevic
Marko Caric


Exchange Rates, Market Microstructure, Artificial Neural Networks, Nearest-neighbors Regression, Forecasting


Given the large body of research addressing exchange rate predictability, the inability of the non-linear model by Diebold and Nason (Journal of International Economics 1990; 28: 315-332) to forecast better than a random walk is puzzling. This paper examines the forecasting performance of Diebold and Nason’s non-parametric model for six major spot Canadian dollar exchange rates for the period 1987-2004. The findings suggest that a more flexible non-parametric estimation technique (artificial neural networks) is required and draw into question the choice of lagged dependent variables as explanatory factors. This paper also proposes a pure microstructure exchange rate model as an alternative to non-linear autoregressive models. Such a model sheds new light on the current evidence on linear/non-linear exchange rate predictability based on market microstructure variables.


Download data is not yet available.
Abstract 84 | PDF Downloads 93