Modeling and forecasting financial time series through hybrid methodologies

Abstract

This project aims to develop and apply hybrid methodologies for modeling and forecasting financial time series. The focus of the project is to use both features extracted through classical models such as (V)ARMA, M(GARCH), M(SV), etc., as well as machine learning models such as neural networks, ensemble models, machine vector support, among others. By studying both approaches, an improvement in the quality of forecasts is expected (both from a statistical and economic point of view), which will be evaluated directly and indirectly through Monte Carlo experiments and empirical data referring to Brazilian markets. and international. The results obtained will be published in scientific journals with a selective editorial policy and the implemented codes will be made available free of charge in a public repository on GitHub.