Additional Info

• File: Link
• Published in: ADVANCES IN DECISION SCIENCES, vol. 24, pp. 1-38 (ISSN 2090-3367)
• Year: 2020
• Abstract: Factor models (FM) are now widely used for forecasting with largeset of time series. Another class of models, which can be easily estimated and usedin a large dimensional setting, is multivariate autoregressive models (MAR), whereindependent autoregressive processes are assumed for the series in the panel. Whenapplied to big data, the estimation, model selection and combination of both modelscan be time consuming. We assume both FM and MAR models are misspecified andprovide a scoring rule which can be evaluated on an initial training sample to eitherselect or combine the models in forecasting exercises on the whole sample. Somenumerical illustrations are provided both on simulated data and on well known largeeconomic datasets. The empirical results show that the frequency of the true positivesignals is larger when FM and MAR forecasting performances differ substantiallyand it decreases as the horizon increases.