Forecasting elections in a multiparty system : the case of Portugal and Brazil

Mestrado Bolonha em Econometria Aplicada e PrevisãoThis work tries to forecast election results in Brazil and Portugal using two bayesian models and one frequentist in order to find out which one has better results. We will use older election‘s results and polls in order to check if there are sistematical biases towards certain parties. We also use macroeconomical data to check how influential this data is to forecast election. The analysis pointed out that there are no sistematical biases for any party in any polling company. We also found out that there is no significant relationship between macroeconomic data and the election results in these countries. Furthermore, the fact that both examples had few elections and have a lot of parties which are constantly being created and dismissed, there is not a "perfect" model, however, they all have very acceptable results.