FORECASTING INTERNATIONAL GROWTH-RATES WITH LEADING INDICATORS - A SYSTEM-THEORETIC APPROACH

An approach to estimating multivariate, time-invariant state space models for ARMAX-type processes is considered. It is based on estimated Markov parameters (unit-impulse responses) and involves singular value decomposition techniques and leads to internally balanced state space representations. The method is applied to forecast the growth rate of real output of nine countries using leading indicators. The forecasting performance of this state space approach is compared to the performance of autoregressive models with leading indicators.