DFROG: Dutch Forecasting Model for Real Time Output Growth

The model uses 80 monthly series (surveys, production and sales figures, financial market indicators, prices) to make a forecast for the previous, current and next quarter. The forecasts are updated at least six times a year after the release of new quarterly GDP figures and at the start of the bi-annual DNB forecasts. The choice for the current model is based on previous research on the optimal model type and specification. It turns out that the dynamic factor model is a very competitive short-term forecasting method. See Jansen, Jin and de Winter (2016) for a comparison of the dynamic factor model with other popular nowcasting methods, and Jansen and de Winter (2018) for a comparison of the forecasting accuracy of the short-term forecasts of dynamic factor models versus professional analysts. See Koopman, Hindrayanto and de Winter (2016) for a recent contribution on the optimal model specification of a dynamic factor model. Alongside DFROG several research projects are currently ongoing, investigating: (a) the merits of out-of-the box machine-learning models (random forest, long-term short term models), (b) topic models based on newspaper texts, and (c) new innovative data-sources, i.e. Google Trends data. 

• Jansen, W.J.J., Jin, X. and J.M. de Winter, 2016. Forecasting and nowcasting real GDP: Comparing statistical models and subjective forecasts, International Journal of Forecasting, 32(2), 411-436 (link) 

• Jansen, W.J.J. and J.M. de Winter, 2018. Combining Model‐Based Near‐Term GDP Forecasts and Judgmental Forecasts: A Real‐Time Exercise for the G7 Countries, Oxford Bulletin of Economics and Statistics, 80(6), 1213-1242 (link) 

• Hindrayanto, I., Koopman, S.J. and J.M. de Winter, 2016, Forecasting and nowcasting economic growth in the euro area using factor models International Journal of Forecasting, 32(4), 1284-1305 (link)