Series: Occasional Papers. 2514.
Author: José Manuel Carbó, Claudia Toledo and Ángel Iván Moreno
Full document
Abstract
Sentiment dictionaries have been a fundamental tool in the automated analysis of texts. While there are multiple options in English for analysing economic and financial texts, Spanish alternatives are more limited. Two of the most notable are the dictionaries developed by the Banco de España and the Central Bank of Chile, designed to measure sentiment in Financial Stability Reports. This article details the process of creating a first pan-Hispanic consensus dictionary, combining both approaches so that it can be used in different Spanish-speaking countries. The results show that the dictionary resulting from merging the two (the “Unión” dictionary) is an effective and balanced tool, capable of capturing the sentiment of texts in both countries. To validate its robustness, we demonstrate that the Unión dictionary is resistant to the removal of terms and that its results are similar to those obtained with Large Language Models (LLMs). Overall, we show that dictionaries, and particularly the Unión dictionary, can achieve results comparable to those of LLMs for sentiment in financial stability texts, due to the structured and specific nature of these documents.