Prædiktion af privatforbruget ved brug af en Bayesiansk strukturel tidsrækkemodel

Abstract

This thesis investigates to what extent a combination of Google Trend data and economic relevant regressors can be used to help nowcast the Danish private consumption. The underlying modelling framework is a Bayesian Structural Time- Series proposed by Scott and Varian (2013) exactly for the purpose of enriching nowcasting of economic key-indicators. The model incorporates automatic variable selection by Stochastic Search Variable Selection (SSVS). The paper gives a thorough introduction to the framework before evaluating the results when applying the proposed model to the Danish private consumption series based on the period 2004-2015. I do not find any strong evidence for the hypothesis that Google Trend data or economic relevant regressors can contribute to a better prediction of the Danish private consumption. In fact I find that the structural time series model without any exogenous variables captures the dynamic of the consumption seemingly well only failing during the beginning of the financial crisis. Only small improvements in nowcasting economic turning points such as the crisis were made when adding the combination of external regressors to the model, but at the cost of adding more noise as well.