Forecasting of the variation of the Banxico reference rate using a text classification model of the monetary policy committee minutes

Abstract

This study proposes a classification model that combines Natural Language Processing (NLP) and Machine Learning techniques to analyze 96 documents published by Banxico concerning Monetary Policy decisions related to adjusting the reference rate. The document is organized as follows. The introduction highlights the critical role of the reference rate set by Banxico in ensuring the stability and functioning of the Mexican economy. It also reviews the application of NLP in financial documents over last past decade, with a focus on research related to Monetary Policy decisions. The methodology section outlines the CRISP-DM framework used to develop the classification model. It details the steps taken, including the use of Python code to expand the training dataset and the model construction process. The result section evaluates the model’s performance, which is deemed satisfactory based on the predefined evaluation metrics. Finally, the conclusion discusses the model's limitations, suggests potential future research, and emphasizes the importance of this research in enhancing financial decisions-making through the use of the Mexican reference rate.