Pooling of Forecasts; does it Improve the Nowcast of Swedish GDP?
Year: 2015
Type: dissertation
Abstract: Even slight improvements in forecast accuracy can have a huge economic impact. Especially, when considering the steady increase of interconnections in the global financial markets, and the ever growing pressure of quick decisions. Empirically, combinations of forecasts from several plausible models usually outperform the forecast of the best fitted model in the presence of model uncertainty. Can the accuracy of the nowcast of Swedish GDP be improved by combining the forecasts of multiple economic models? Different specifications of mixed data sampling (MIDAS) regressions are fitted and used to produce nowcasts of current quarter GDP. The performance of the pooling and the single model selection is evaluated by pseudo out-of-sample forecast MSE. Simple pooling schemes of the forecasts significantly outperforms the benchmark, while very few of the best fitted models achieve significantly improved accuracy over the benchmark. Overall, the pooling of forecasts also produces more stable and lower mean square forecast errors than the other model specifications.
Author: Andreas Johansson
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W1012411904