Publicação

Comparative Analysis of GDP Forecasting using Ensemble Tree Regression Models: Machine Learning vs. Econometric Models

Ver documento

Detalhes bibliográficos
Resumo:This thesis evaluates whether machine learning can have better results than well-established institutions in forecasting Portugal’s GDP growth using ensemble tree regression models, with the OECD’s economic outlook forecasts for Portugal serving as a benchmark based on data from 1962 to 2022 recovered from OECD and the Federal Reserve’s economic data. The findings reveal that, in general, machine learning did not surpass the forecasts of the OECD. However, machine learning demonstrated the potential for better accuracy at many points, despite a higher propensity for larger errors compared to traditional methods. Among the ensemble tree regression models tested, the gradient boosting regressor consistently provided the best forecasts across all horizons, outperforming the random forest, extreme gradient boosting, and light gradient boosting machine models. The results also suggest that machine learning performs better with a larger volume of data than with higher dimensionality, even if some data points seem irrelevant to forecast future values. This thesis highlights the potential of machine learning in time series forecasting as a complementary tool to traditional methods, rather than a complete replacement.
Autores principais:Coelho, Ricardo Paulo Barbosa de Carvalho Almeida
Assunto:Machine learning Forecasting GDP Portugal Gradient Boosting Regressor SDG 8 - Decent work and economic growth
Ano:2024
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:Universidade Nova de Lisboa
Idioma:inglês
Origem:Repositório Institucional da UNL
Descrição
Resumo:This thesis evaluates whether machine learning can have better results than well-established institutions in forecasting Portugal’s GDP growth using ensemble tree regression models, with the OECD’s economic outlook forecasts for Portugal serving as a benchmark based on data from 1962 to 2022 recovered from OECD and the Federal Reserve’s economic data. The findings reveal that, in general, machine learning did not surpass the forecasts of the OECD. However, machine learning demonstrated the potential for better accuracy at many points, despite a higher propensity for larger errors compared to traditional methods. Among the ensemble tree regression models tested, the gradient boosting regressor consistently provided the best forecasts across all horizons, outperforming the random forest, extreme gradient boosting, and light gradient boosting machine models. The results also suggest that machine learning performs better with a larger volume of data than with higher dimensionality, even if some data points seem irrelevant to forecast future values. This thesis highlights the potential of machine learning in time series forecasting as a complementary tool to traditional methods, rather than a complete replacement.