Using machine learning and Bayesian networks to objectively analyze central bank statements and market sentiment:

The minutes of the meetings of the Swedish Central Bank are analyzed from the other banks manually. In finance the reaction of each event as to be as quick as possible, so the goal is to automate the text comprehension process using machine learning algorithms. The problem can be divided into two main tasks, summarize the speech of each board member and find his sentiment. To retrieve the summary has been used an unsupervised approach based on the Text Rank and the Latent Semantic Analysis algorithms, combined with information about the most discussed topics in this kind of meeting. The sentiment behind this kind of economical text cannot be identified as positive or negative, as in most of the literature, but it can be classified as hawkish or dovish. Several supervised classifiers have been used, like SVM and Bayesian Networks, based on the past analysis of the minutes performed by SEB bank. The accuracy of the results achieved is at the level of the state-of-the-art of this field, also thanks to the stability of the opinion of the board members.