Using Principal Covariate Regression for Macroeconomic Time Series Forecasting

In the presented master thesis a problem of forecasting U.S. real economic variables growth rates by the means of dynamic factor models is considered. Forecasting horizons vary from 1 month to 1 year. The research is focused on different methods of dynamic factors’ estimation. The following modifications of the standard approach are investigated. Firstly, implementation of analysis and selection of predictor variables prior to a factors’ estimation step. Secondly, use of principal covariate regression instead of more standard principal component regression. Thirdly, consecutive use of variables selection and principal covariate regression methods. Forecasting accuracy conclusions are based on comparison of mean squared prediction errors and recession periods dating. Empirical results stand for introduced modifications and their combination.