Are disaggregate data useful for factor analysis in forecasting French GDP?

Abstract

This paper compares the GDP forecasting performance of alternative factor models based on monthly time series for the French economy. These models are based on static and dynamic principal components obtained using time and frequency domain methods. We question whether it is more appropriate to use aggregate or disaggregate data to extract the factors used in forecasting equations. The forecasting accuracy is evaluated for various forecast horizons considering both rolling and recursive schemes. We empirically show that static factors, estimated from a small database, lead to competitive results, especially for nowcasting. Copyright © 2009 John Wiley & Sons, Ltd.

A dynamic factor model for nowcasting Canadian GDP growth

“…Following Giannone et al (2008), there is now a large literature on nowcasting with DFM. Recent contributions include: Brazil (Bragoli et al 2015), BRIC countries and Mexico (Dahlhaus et al 2015), China (Giannone et al 2013), France (Barhoumi et al 2010), Indonesia (Luciani et al 2015), Ireland (D'Agostino et al 2013), New Zealand (Matheson 2010), Norway (Aastveit et al 2011 andRicci 2014), and the United States (Giannone et al 2008). A recent paper by Bragoli and Modugno (2016) constructs a DFM for Canada that bears many similarities with the model developed in this paper.…”

Why Didn't the Global Financial Crisis Hit Latin America?

“…However, empirical evidence is mixed. Barhoumi et al (2010) for example conclude that dynamic factor models with a large number of variables do not necessarily produce better forecasting results of French GDP than static models with a small number of variables. Schumacher (2007) also mentions a number of studies with mixed empirical success for the dynamic factor model.…”

“…correlated, which covers cross-correlation and heteroskedasticity between the idiosyncratic errors and correlation between the common components and the idiosyncratic components (see e.g. Barhoumi, Darné and Ferrara 2010).…”

Forecasting US Unemployment with Radial Basis Neural Networks, Kalman Filters and Support Vector Regressions

“…The lower the negative value, the more accurate are the SVR forecasts. The use of MDM test is common practice, because it assesses the significance of observed differences between the performances of two forecasts (Barhoumi et al, 2010). The statistic is measured in the out-of-sample period for the MSE and MAE loss functions.…”