Browsing by Author "Ahonen, Elena Venla Maria"

The aim of this thesis is to demonstrate the importance of selecting feasible and, preferably data-based prior assumptions for the Bayesian time-varying coefficient vector autoregressive model (TVC VAR model for further reference) of Primiceri (2005) and Del Negro and Primiceri (2015). The TVC VAR model would be suitable for quantifying, for example, the impacts of different monetary policy or fiscal policy regimes. The biggest advantage of the TVC VAR model is that it takes into account both changes in economic policy and in the private sector behaviour. The latter feature makes the model very compelling to use, because the private sector plays an important role in facilitating mote stable change in monetary and fiscal policy regimes. In complex mathematical models, such as the TVC VAR model, the objectiveness of the model may be compromised by deliberate selection of parameters. The TVC VAR model uses the Bayesian approach, which means that the researcher’s choice for the prior assumptions for the model plays an important role in the estimation. Unfortunately, Primiceri’s (2005) approach for selecting hyperparameters for the model is poorly explained and difficult to follow. Given that the model depends only for a small number of hyperparameters, it might be possible that the model can be tuned in a predefined way. To investigate whether the TVC VAR model can be tuned according to a researcher’s preferences, I design a proof of concept approach for optimising the hyperparameters of the model according to a set of predefined results. In other words, my research question is: could one tune the TVC VAR model to produce results according to the researcher’s bias? In my proof of concept approach I tune the TVC VAR model for six different targets for the Finnish government consumption multiplier. Given that Finland is a small open economy, Primiceri’s (2005) original hyperparameter values for the United States are not feasible and other values have to be used. The results from my proof of concept analysis show that the TVC VAR model can be tuned for predefined results, which shows that the practical reliability of the model can be easily compromised. My findings highlight the need for a comprehensible, data-based approach for selecting the hyperparameters for the model.