Robust Implementation of a Parsimonious Dynamic Factor Model to Nowcast GDP

Abstract

Quarterly GDP figures usually are published with a delay of some weeks. A common way to generate GDP series of higher frequency, i.e. to nowcast GDP, is to use available indicators to calculate a single index by means of a common factor derived from a dynamic factor model (DFM). This paper deals with the implementation stage of this practice. We propose a two-tiered mechanism consisting in the identification of variables highly correlated with GDP as "core" indicators and a check of robustness of these variables in the sense of extreme bounds analysis. Accordingly selected indicators are used in an approximate DFM framework to exemplarily nowcast Spanish GDP growth. We show that our implementation produces more accurate nowcasts than both a benchmark stochastic process and the implementation based on the total set of core indicators.

Keywords: small-scale nowcasting models, Kalman Filter, extreme bounds analysis

JEL Classification: C380, C530