Previsibilidade da taxa de câmbio com sentimento das notícias

Abstract

This paper aims to study the out-of-sample predictability of nominal exchange rate variations with news sentiment, we use financial news from The New York Times and The Wall Street Journal to construct these sentiments for the period from january 1980 to may 2017. The results suggest a strong predictive power of the news sentiment generated with the time variant dictionary, which can be used as the only predictor or as an additional predictor to the model based on the conventional Taylor rule. Besides demonstrating that the time-varying dictionary method is more suitable for large textual samples compared to fixed dictionaries when the objective of the work passes through predictive power. When used as the only predictor, superiority of performance was observed in the prediction on random walk for all analyzed countries and on the conventional Taylor rule for 7 of 9 countries. For extended Taylor rule models with news sentiment, superior performance was observed for all countries in relation to the conventional Taylor rule models. All of these results at 1% significance.