Does the use of a big data variable improve monetary policy estimates? Evidence from Mexico

We analyse the performance improvement on a monetary policy model of introducing non-conventional market attention (NCMA) indices generated using big data. To address this aim, we extracted top keywords by text mining Banco de Mexico’s minutes. Then, we used Google search information according to the top keywords and related queries to generate NCMA indices. Finally, we introduce as covariates the NCMA indices into a bivariate probit model of monetary policy and contrast several specifications to examine the improvement in the model estimates. Our results show evidence of the statistical significance of the NCMA indices where the expanded model performed better than models only including conventional economic and financial variables.