Forecasting one day stock returns in Latin American markets: a horserace

• Autores: Sampron Noel, Alfredo Ignacio
• Año de publicación: 2025
• Idioma: inglés
• Tipo de recurso: tesis de maestría
• Colaborador/a o director/a de tesis: García Cicco, Javier
• Institución: Universidad de San Andrés

Fil: Sampron Noel, Alfredo Ignacio. Universidad de San Andrés. Departamento de Economía; Argentina.
We investigate the predictive power of Generalized Autoregressive Conditional Heteroskedasticity (GARCH), Vector Autoregression (VAR), and Hidden Markov Models (HMM) for forecasting stock returns in Argentina, Brazil, and Mexico. Our research extends prior work by considering the impact of volatility and foreign exchange (FX) variations, including the implicit exchange rate between American Depository Receipts (ADRs) and local stock prices, particularly relevant for Argentina's capital controls. We address three key questions: which model offers superior predictive accuracy, whether incorporating exchange rates enhances predictive power, and which return denomination (local currency or USD) is easier to predict. Findings reveal that model rankings remain consistent across local currency and USD-denominated assets. Broad market indices are best captured by VAR models. Our results align with the finding that more sophisticated models tend to outperform benchmarks, yet performance varies significantly.