Monetary policy uncertainty in Mexico: an unsupervised approach

We study and measure uncertainty in the minutes of the meetings of the board of governors of the Central Bank of Mexico and relate it to monetary policy variables. In particular, we construct two uncertainty indices for the Spanish version of the minutes using unsupervised machine learning techniques. The first uncertainty index is constructed exploiting Latent Dirichlet Allocation (LDA), whereas the second uses the Skip-Gram model and K-Means. We also create uncertainty indices for the three main sections of the minutes. We find that higher uncertainty in the minutes is related to an increase in inflation and money supply. Our results also show that a unit shock in uncertainty leads to changes of the same sign but different magnitude in the inter-bank interest rate and the target interest rate. We also find that a unit shock in uncertainty leads to a depreciation of the Mexican peso with respect to the US dollar in the same period of the shock, which is followed by appreciation in the subsequent period.