A Variational Bayesian Approach to Inflation in Emerging Markets

June 2024 · Working paper.

• inflation regimes
• Hidden Markov Models
• variational Bayes
• Mahalanobis distance
• emerging markets
• Latin America

Abstract

This paper develops a novel framework to identify inflation regimes, their dynamics, persistence, and underlying drivers, across nine Latin American economies over 2008–2023. A Multivariate Gaussian Hidden Markov Model (MGHMM), estimated via variational Bayesian inference, recovers the regimes, while a Mahalanobis distance-based measure quantifies the contribution of five driver categories: monetary policy, international factors, demand-pull factors, expectations, and cost-push factors.

Conclusions

• Monetary policy instruments significantly affect inflation, especially during economic disruptions.
• International factors, notably foreign inflation and exchange rates, are prominent — particularly in Chile, the Dominican Republic, Mexico, and Peru.
• Regime persistence and driver composition vary substantially across countries, cautioning against one-size-fits-all policy prescriptions.

Materials

Code

• Python
• numpy
• pandas
• seaborn