Forecasting Dutch GDP

To facilitate stable economic development and prevent bankruptcy both business and governmental institutions need reliable estimates of current GDP. In the Netherlands, however, official GDP estimates have a significant publication lag of 45 days after the end of a quarter. The EICIE model developed by de Groot and Franses (2006) made the first step by using staffing data from a temp agency. This data is available a week after the end of a quarter. To improve the predictive accuracy of EICIE, a novel approach to utilize news articles by applying sentiment analysis techniques is developed in this thesis. The approach reduces the publication lag while improving the accuracy of the estimates. Moreover, some dimensionality reduction and variable selection methods have been shown to considerably boost predictive performance, when faced with the sparse data sets generated by the approach adopted in this thesis. The prediction accuracy is significantly improved upon with a 65.4% improvement compared to the EICIE model if both models are re-estimated after each forecast and an 81.6% improvement in prediction accuracy if the models are only estimated once with the training data.