Forecasting quarterly movements of the lira/pound-sterling exchange rate: Random walks, drift, seasonality and variable parameters

Abstract

Quarterly logarithmic changes of the lira/pound-exchange rate for the period 1973 Q1 to 1989 Q1 are examined and the forecasting performance of some simple time series models is evalutated. The performance of the random walk model with drift is examined as a method of forecasting one quarter ahead lira/pound-sterling exchange rate changes. This model is then extended to incorporate seasonal movements. The assumptions of constant drift and seasonal patters are relaxed and the models are estimated using variable parameter regression based on state-space modelling using the Kalman filter. The within and out-of-sample performance of the models illustrates that an improvement in forecast accuracy is obtained by including seasonal variation. However, the results do not appear to be improved by allowing the parameters to follow a random walk, AR(1) or AR(2) process. The random walk model with drift and seasonal patterns performs better at predicting exchange rate movements than predictions based on the forward rate.