Abstract :

In this paper, we assess four models for estimating monthly GDP in Israel using quarterly GDP data and a group of monthly explanatory variables. The models are mixed frequency models that directly characterize the connection between quarterly GDP and the monthly variables and do not require the prior totaling of the monthly variables into quarterly variables. One of the advantages of these models is that they force the estimated monthly GDP data to total the quarterly GDP data. 
As a byproduct of this estimation, we also present nowcasting estimates of monthly GDP.  The nowcasting estimates are used when the GDP of the latest quarter has not yet been obtained, but the data on the monthly explanatory variables has already been obtained. 
Since monthly GDP is an unobserved variable, we cannot directly know which of the models is best for estimating it. We therefore examined the models indirectly through a quarterly aggregation of out-of-sample monthly nowcasting estimates and comparing them to the quarterly GDP data. We found that Model 1 (Mariano and Murasawa, 2010) is preferable to the other models, including the Benchmark Model.
1 We would like to thank Dr. Karnit Flug for the idea and the initiative. We would also like to thank Tsahi Frankovits, Ben Z. Schreiber, Yair Haim, and the participants of the joint seminar of the Research Department and the Information and Statistics Department for their helpful comments.