Predicting economic growth using machine learning techniques and sentiment analysis

The purpose of this paper is to construct sentiment index from financial and economic news and examine its potential relation with main economic and political events that affect economic activity in Turkey. Since there is no effective economic and financial lexicon (as in English) for sentiment analysis in Turkish language, we developed a sentiment index by using machine-learning algorithms for the period 2011-2019. Data set used in this study includes 131.601 news in Turkish, which were selected according to a carefully specified set of words, published in printed media. We classified the semantic orientation of news by a group of experts to construct annotated data set. It is observed that sentiment index covers the important events for Turkish economy. Considering time lag in official statistics, the sentiment index can be used as a leading economic indicator. It is planned to investigate whether the sentiment index increases the explanatory power of the econometrics models explaining economic activity in Turkey as a future work.