A note on forecasting in co-integrated systems

Abstract

This note compares the forecasting performance of four estimation methods of a cointegrated system by the use of Monte Carlo simulation. The methods are a VAR in levels, a VAR in first differences, and the Engle-Granger two-step estimator and the nonlinear least squares estimator of the error correction representation of the model. It is found that the NLS estimator provides the most accurate forecasts of the system, which supports the theoretical result of it being an efficent full information systems estimation method. Response surfaces are also estimated in order to summarise the results of the Monte Carlo simulations.