Rolling window selection for out ofsample forecasting in panel data modelswith time varying parameters
This thesis proposes a novel way to handle the presence of time varying parameters in economic time series by applying the method of Inoue et al. (2016), a method that finds an optimal window of observations to include when performing out of sample forecasts, on panel data models. In order to make the newly created combination more widely applicable a dynamic cluster method is introduced, which clusters units while accounting for the time varying property of the individual units in panel data. Performance is measured using the Root Mean Squared Forecast Error (RMSFE) on a data set of fifteen OECD countries for forecasting GDP growth. RMSFE is improved from 0.738 to 0.427 when introducing dynamic clusters and the optimal moving window method of Inoue et al. (2016). As results show slight improvements compared to other methods, this paper indicates a potential benefit of combining optimal moving window methods with panel data models.
|Thesis Advisor||Wang, W.|
Jongsma, M.L. (2019, November 12). Rolling window selection for out ofsample forecasting in panel data modelswith time varying parameters. Econometrie. Retrieved from http://hdl.handle.net/2105/50487