Forecasting linear dynamical systems using subspace methods

摘要

This article presents a novel approach to predict with subspace methods. It consists in combining multiple forecasts obtained from setting a range of values for a specific parameter that is typically fixed by the user in this literature. Two procedures are proposed. The first one combines all the forecast in a particular range. The second one predicts with a restricted number of combinations previously optimized. Both methods are evaluated using Monte Carlo experiments and by forecasting the German gross domestic product