Browsing by Author "Österman, Esa-Petter"

This thesis studies the extent to which a state of the art Bayesian model, Bayesian predictive synthesis, can improve the nowcasting of Finnish GDP. GDP is the most looked after and thus the most nowcasted economic variable. The true values for GDP are published with a significant delay and revised later on, which highlights the necessity of well-performing forecasting models. In this thesis I replicate an existing study on Bayesian predictive synthesis in Finnish setting and then extend the framework to study forecast accuracy for GDP change. In the first part of the thesis, the theoretical background of the BPS model framework is studied in detail after which the application follows. In the empirical study six projection models are formed as dynamic linear models which are then synthesized with the novel Bayesian predictive synthesis. These results are benchmarked against the existing Bayesian VAR model that is used by Bank of Finland. In the empirical application I find that the Bayesian predictive synthesis is unable to improve the projection models' forecast accuracy. I also find that the synthesis for GDP levels performs better than the synthesis for GDP change. In the literature the synthesis model is mainly used to project nonstationary series. Results from this study support the assumption that BPS framework applies better for nonstationary projections. Also, the model is found to be very sensitive to the set of projections and parameter selection, which highlights the need for expert opinion in choosing the right models to synthesize for each application. Based on this research, this stand-alone BPS framework is not suitable to replace existing models for GDP nowcasting. For future research, it is recommended that BPS model is used to synthesize the already existing nowcasting models for improving these models.