Abstract
This paper presents a dynamic nowcasting model for Quarterly GDP in Israel. Currently, monetary policy in Israel is evaluated and updated on a monthly basis. The recent GDP figure is, however, unavailable for monetary policy makers, at the Bank of Israel, at the month following the end of the quarter, due to a six-week lag of the GDP data publication.
The aim of this nowcasting project is to derive "flash" estimates of GDP at a four-week lag, in order to gain four weeks in terms of data availability when updating the interest rate. This is done by utilizing the information contained within a large group of monthly indicators that are available at the relevant date.
Indicator selection, from a pool of these high frequency series, is applied through a variety of dimension reduction techniques. The ability to apply these techniques while conditioning them on the predicted indicator will be examined and discussed in this article.
The Elastic Net is found to be the most comprehensive model selection technique, generating the lowest mean absolute forecast error of only 1.62%. In addition, the Elastic Net successfully captures the timing and magnitude of the 2008-2009 economic cycle. Notable variables that have model inclusion persistence are: The Price of Oil, Employers Survey, Purchasing Managers Index, Industrial Production Index, and Employed Persons Index in Manufacturing of Electronic Motors, Components, and Transport Equipment.