On Neural Forecasting and News Emotions: The Case of the Spanish Stock Market

Abstract

We provide an overview on the preliminary development of a neural machine translation and deep learning approach to extract the emotional content of economic and financial news from Spanish journals. To this end, we exploit a dataset of over 14 million articles published in Spanish newspapers over the period from 1st of July 1996 until 31st of December 2019. We examine the role of these news-based emotions indicators in forecasting the Spanish IBEX-35 stock market index by using DeepAR [], an advanced neural forecasting method based on auto-regressive Recurrent Neural Networks operating into a probabilistic setting. The DeepAR model is trained by adopting a rolling-window approach to best accounting for non-linearities in the data, and employed to produce point and density forecasts. After providing an overview of the methodology under current development, some preliminary findings are also given, showing an improvement in the IBEX-35 index fitting when the emotional variables are included in the model, and an overall validity of the employed approach.