N. Kourentzes and F. Petropoulos, 2014.
Predicting the present state of the economy can be challenging. Recently, econometric models that are capable of using multiple sources of hard and soft information, published at various frequencies and dates, to produce nowcasts of policy relevant variables have been proposed. Following the progress in the modelling aspects of nowcasting, the literature has started exploring new useful inputs. The ‘Big data` and various Internet sources can be used in order to further enhance nowcasting models, by enriching them with new types of information, as well as being more up to date. This research proposes a methodology to create useful inputs by mining news articles, capturing the current sentiment in the media about the state of the economy. News articles are nowadays easy to access and process, as most news outlets provide online versions of them. Furthermore, the high frequency of publication and the representation of the current economic discourse are invaluable for nowcasts. The case study of the Greek economy is used and rolling nowcasts are produced for the period of 2008 to 2013. Empirical evidence suggests that the inclusion of such information improves the accuracy of nowcasts.