We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
MULTI-STEP ESTIMATION FOR FORECASTING.
- Authors
Clements, Michael P.; Hendry, David F.
- Abstract
The article presents a discussion on multi-step estimation for economic forecasting. Minimizing multi-step (in-sample) criteria for estimating unknown parameters has a long pedigree, although the method does not seem to have been subject to many formal analyses. Economic analysts applied this idea to the exponentially-weighted moving average or integrated moving-average model and considered multi-step estimation criteria for autoregressive models. The intuition is that when a model is not well specified, minimization of 1-step errors need not deliver reliable forecasts at longer lead times, so estimation by minimizing the in-sample counterpart of the desired step-ahead horizon may yield better forecasts. When models are mis-specified for the data generation process (DGP), mean-square forecast error (MSFE) rankings can alter as the forecast horizon increases. Indeed, a necessary condition for such a result in large samples is that the models under consideration are mis-specified. This implication depends only on the DGP providing the correct conditional expectation, which is the minimum MSFE predictor.
- Subjects
ECONOMIC forecasting; BOX-Jenkins forecasting; ECONOMIC indicators; MATHEMATICAL models; TIME series analysis; ECONOMICS
- Publication
Oxford Bulletin of Economics & Statistics, 1996, Vol 58, Issue 4, p657
- ISSN
0305-9049
- Publication type
Article
- DOI
10.1111/j.1468-0084.1996.mp58004005.x