(2021-03-18) Karelaki, Vasiliki; Καρελάκη, Βασιλική; Athens University of Economics and Business, Department of Economics; Tzavalis, Elias; Arvanitis, Stylianos; Dendramis, Yiannis
The main focus of this study lies in forecasting time series using data reduction methods. Our objective is to use various shrinkage methods in order to find the optimal model with the best predictive ability. Firstly, we present the Lasso, Ridge regression and the PCR (Principal Components Regression) method. Continuously, we use these methods to forecast two macroeconomic time series. We perform in sample forecasting. Based on our results, the choice of the hyperparameters was the key factor in the forecasting procedure. It is important to mention that the PCR method has a low predictive performance despite the fact that we used only one component. Based on our conclusions the penalization methods perform better than the PCR method.
