Relative Efficiency of Univariate and Cointegration based Forecasting
by
Chandra M. Gulati
School of Mathematics and Applied Statistics, University of Wollongong
Coauthors: Michael McCrae,Dept. of Accounting and Finance, University of Wollongong, Yan-Xia Lin,School of Mathematics and Applied Statistics, University of Wollongong, Daniel Pavlik,School of Mathematics and Applied Statistics, University of Wollongong

For cointegrated time series, the inclusion of long run error corrections in cointegration-based error correction models (ECMs) should provide more accurate forecasting ability relative to univariate Box-Jenkins type models, which do not contain these terms. But the available empirical evidence is equivocal. This paper examines the comparative forecasting accuracy of the univariate Box-Jenkins approach BJ) and cointegration based Error Correction Models (ECM) for five Aisan excahnge rates. Out of sample forecasting through a rolling window technique permits multiple sampling of accuracy from one to forty steps ahead. Comparative forecasting accuracy between the models is found to depend upon both the nature of the exchange rate series and the forecast horizon. ARIMA models perform better over shorter term horizons for series with moving-average terms of order greater than one and. ECMs perform better over longer time horizions for series with no moving average terms. The results suggest a need to distinguish between 'sequential' and 'synchronous' forecasting ability in such comparisons.

Date received: November 13, 2000

Copyright © 2000 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cafr-99.