Description

Through the application of textual border analysis tools, this work presents the construction of a high-frequency indicator, generated in real time, based on the computerized reading of the news from the main print media from January 2015 to December 2020 in Chile. This indicator captures the emotional tone of economic and opinion news by making use of an extensive —and novel— purpose-built dictionary in Spanish. This lexicon of words was subjected to challenging statistical tests of robustness, complemented by tests of predictive precision. The latter were carried out by comparing the degree of similarity between a classification by automated means and another by manual means in a random sample. The economic application shows that the constructed indicator has a high correlation with confidence indicators based on surveys, and a high predictive capacity in the face of shock phenomena hitting the economy.