Ensaios sobre modelos de machine learning aplicados à econômia
| Ano de defesa: | 2026 |
|---|---|
| Autor(a) principal: | |
| Orientador(a): | |
| Banca de defesa: | |
| Tipo de documento: | Tese |
| Tipo de acesso: | Acesso aberto |
| Idioma: | por |
| Instituição de defesa: |
Não Informado pela instituição
|
| Programa de Pós-Graduação: |
Não Informado pela instituição
|
| Departamento: |
Não Informado pela instituição
|
| País: |
Não Informado pela instituição
|
| Área do conhecimento CNPq: | |
| Link de acesso: | http://repositorio.ufc.br/handle/riufc/87420 |
Resumo: | This study brings together three empirical essays that focus on the application of machine learning models to economic forecasting at both the national and regional levels, seeking to evaluate the extent to which these approaches can complement traditional econometric methods. Despite the recent advances in these techniques, important limitations still remain regarding the economic interpretability of the models, the incorporation of regional heterogeneities, and the comparison with structural approaches. The choice of the sample period is justified by its coverage of major consolidations in the Brazilian macroeconomic environment, particularly the adoption of the inflation-targeting regime, while also encompassing distinct economic regimes, episodes of instability, and changes in the conduct of economic policy. In addition, the temporal delimitation was defined according to the availability, compatibility, and standardization of the databases employed throughout the essays. In this context, the first essay presents a national inflation forecasting exercise based on an extensive set of macroeconomic variables. Additionally, the study aims to identify the key predictors of inflation across multiple forecasting horizons. To this end, machine learning models are employed as the main approach and compared with benchmark models. The empirical findings suggest that machine learning models outperform benchmark models across different forecasting horizons. Furthermore, the results provide evidence of the relevance of production and fiscal sectors, labor market conditions, energy costs, and expectations in explaining inflation forecasts over different periods. The second essay proposes a regional inflation forecasting exercise based on an implicit hierarchical structure, incorporating regional elements into the forecasting process. Furthermore, a dynamic decomposition of inflation predictors is performed across regions and forecasting horizons. The results indicate the presence of significant regional heterogeneity across the states of Northeastern Brazil. The dynamic decomposition also reveals that the determinants of regional inflation gradually change over time, exhibiting patterns consistent with economic and political events at both the national and regional levels. The third essay proposes the use of multivariate neural network models to generate dynamic counterfactuals for Brazilian macroeconomic variables based on shocks applied to selected variables. The simulated trajectories are then compared with the dynamic responses obtained from a structural model, allowing an assessment of the extent to which machine learning-based methods can reproduce propagation and persistence patterns commonly found in traditional econometric models. The findings indicate that the multivariate model responds to shocks in a more moderate manner and in closer alignment with the macroeconomic literature. In particular, the model is able to reproduce propagation patterns similar to those generated by the structural model, although such responses arise from statistical relationships learned from the data rather than from explicitly modeled causal mechanisms. In this sense, the estimated counterfactuals exhibited smaller and more gradual responses at the onset of the shock for all variables when compared with the structural model. Taken together, the results reinforce the potential of machine learning methods not only as forecasting tools, but also as complementary instruments to traditional econometric approaches for analyzing dynamic and complex phenomena, especially in environments characterized by high macroeconomic instability. |
| id |
UFC-7_e33c3fbed8c172e10377d0da25d40a52 |
|---|---|
| oai_identifier_str |
oai:repositorio.ufc.br:riufc/87420 |
| network_acronym_str |
UFC-7 |
| network_name_str |
Repositório Institucional da Universidade Federal do Ceará (UFC) |
| repository_id_str |
|
| spelling |
Moura, Yure Revélles da Silvahttp://lattes.cnpq.br/5185802743634482http://lattes.cnpq.br/8710490356999657Castelar, Luiz Ivan de Melo2026-08-06T19:12:08Z2026-08-06T19:12:08Z2026MOURA, Yure Revelles da Silva. Ensaios sobre modelos de machine learning aplicados à econômia. 2026. 173p. Tese (Doutorado) - FEAAC - Faculdade de Economia, Administração, Atuária e Contabilidade - CAEN - Programa de Pós-Graduação em Economia, Universidade Federal do Ceará, Fortaleza , 2026.http://repositorio.ufc.br/handle/riufc/87420This study brings together three empirical essays that focus on the application of machine learning models to economic forecasting at both the national and regional levels, seeking to evaluate the extent to which these approaches can complement traditional econometric methods. Despite the recent advances in these techniques, important limitations still remain regarding the economic interpretability of the models, the incorporation of regional heterogeneities, and the comparison with structural approaches. The choice of the sample period is justified by its coverage of major consolidations in the Brazilian macroeconomic environment, particularly the adoption of the inflation-targeting regime, while also encompassing distinct economic regimes, episodes of instability, and changes in the conduct of economic policy. In addition, the temporal delimitation was defined according to the availability, compatibility, and standardization of the databases employed throughout the essays. In this context, the first essay presents a national inflation forecasting exercise based on an extensive set of macroeconomic variables. Additionally, the study aims to identify the key predictors of inflation across multiple forecasting horizons. To this end, machine learning models are employed as the main approach and compared with benchmark models. The empirical findings suggest that machine learning models outperform benchmark models across different forecasting horizons. Furthermore, the results provide evidence of the relevance of production and fiscal sectors, labor market conditions, energy costs, and expectations in explaining inflation forecasts over different periods. The second essay proposes a regional inflation forecasting exercise based on an implicit hierarchical structure, incorporating regional elements into the forecasting process. Furthermore, a dynamic decomposition of inflation predictors is performed across regions and forecasting horizons. The results indicate the presence of significant regional heterogeneity across the states of Northeastern Brazil. The dynamic decomposition also reveals that the determinants of regional inflation gradually change over time, exhibiting patterns consistent with economic and political events at both the national and regional levels. The third essay proposes the use of multivariate neural network models to generate dynamic counterfactuals for Brazilian macroeconomic variables based on shocks applied to selected variables. The simulated trajectories are then compared with the dynamic responses obtained from a structural model, allowing an assessment of the extent to which machine learning-based methods can reproduce propagation and persistence patterns commonly found in traditional econometric models. The findings indicate that the multivariate model responds to shocks in a more moderate manner and in closer alignment with the macroeconomic literature. In particular, the model is able to reproduce propagation patterns similar to those generated by the structural model, although such responses arise from statistical relationships learned from the data rather than from explicitly modeled causal mechanisms. In this sense, the estimated counterfactuals exhibited smaller and more gradual responses at the onset of the shock for all variables when compared with the structural model. Taken together, the results reinforce the potential of machine learning methods not only as forecasting tools, but also as complementary instruments to traditional econometric approaches for analyzing dynamic and complex phenomena, especially in environments characterized by high macroeconomic instability.O presente estudo reúne três ensaios empíricos voltados à aplicação de modelos de machine learning na previsão econômica em âmbito nacional e regional, buscando avaliar em que medida essas abordagens podem complementar os métodos econométricos tradicionais. Apesar do avanço recente dessas técnicas, ainda permanecem limitações relacionadas à interpretação econômica dos modelos, à incorporação de heterogeneidades regionais e à comparação com abordagens estruturais. A escolha do período de análise justifica-se por abranger importantes consolidações no cenário macroeconômico brasileiro, especialmente a adoção do regime de metas de inflação, além de contemplar diferentes regimes econômicos, episódios de instabilidade e mudanças na condução da política econômica. Adicionalmente, a delimitação temporal também considerou a disponibilidade, compatibilidade e padronização das bases de dados utilizadas ao longo dos ensaios. Nesse contexto, o primeiro ensaio realiza um exercício de previsão da inflação nacional considerando um conjunto extenso de variáveis macroeconômicas. Adicionalmente, buscou-se identificar os principais preditores da inflação em múltiplos horizontes temporais. Para isso, utilizaram-se modelos de machine learning como abordagem principal, realizando-se uma comparação com modelos benchmark. Os achados empíricos sugerem que os modelos de machine learning superam os modelos benchmark em diferentes horizontes temporais. Além disso, encontraram-se evidências da relevância dos setores produtivo e fiscal, do mercado de trabalho, dos custos energéticos e das expectativas para a previsão da inflação em diferentes períodos. O segundo ensaio propõe um exercício de previsão da inflação regional por meio de uma estrutura hierárquica implícita, incorporando elementos regionais ao processo de previsão. Adicionalmente, realizou-se uma decomposição dinâmica dos preditores da inflação por região e horizonte temporal. Os resultados apontam para uma heterogeneidade regional relevante na previsão entre os estados da região Nordeste. A decomposição dinâmica permitiu identificar que os determinantes da inflação regional se modificam gradualmente, apresentando dinâmicas compatíveis com acontecimentos econômicos e políticos em âmbito nacional e regional. O terceiro ensaio propõe a utilização de modelos de redes neurais multivariadas para gerar contrafactuais dinâmicos de variáveis macroeconômicas brasileiras a partir de choques em variáveis selecionadas. As trajetórias simuladas são então comparadas às respostas dinâmicas obtidas por um modelo estrutural, permitindo avaliar em que medida métodos baseados em machine learning conseguem reproduzir padrões de propagação e persistência presentes em modelos econométricos tradicionais. Os resultados evidenciaram que o modelo multivariado responde a choques de forma mais moderada e alinhada com a literatura macroeconômica. Em especial, o modelo conseguiu reproduzir trajetórias de propagação semelhantes às observadas no modelo estrutural, ainda que tais respostas decorram de relações estatísticas aprendidas a partir dos dados, e não de mecanismos causais explicitamente modelados. Nesse sentido, os contrafactuais estimados apresentaram respostas menores e mais graduais no início do choque para todas as variáveis, em comparação ao modelo estrutural. Em conjunto, os resultados reforçam o potencial dos métodos de machine learning não apenas como ferramentas de previsão, mas também como instrumentos complementares às abordagens econométricas tradicionais na análise de fenômenos dinâmicos e complexos, especialmente em ambientes marcados por elevada instabilidade macroeconômica.Ensaios sobre modelos de machine learning aplicados à econômiainfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/doctoralThesisMachine LearningMacroeconomiaPrevisãoInflaçãoMachine LearningMacroeconomicsInflationForecastCNPQ::CIENCIAS SOCIAIS APLICADAS::ECONOMIAinfo:eu-repo/semantics/openAccessporreponame:Repositório Institucional da Universidade Federal do Ceará (UFC)instname:Universidade Federal do Ceará (UFC)instacron:UFCORIGINAL2026_tese_yrsm.pdf2026_tese_yrsm.pdfapplication/pdf3423998http://repositorio.ufc.br/bitstream/riufc/87420/1/2026_tese_yrsm.pdf9edd8329578a81c11345b7135163a96aMD51LICENSElicense.txtlicense.txttext/plain; charset=utf-81748http://repositorio.ufc.br/bitstream/riufc/87420/2/license.txt8a4605be74aa9ea9d79846c1fba20a33MD52riufc/874202026-08-06 16:12:09.675oai:repositorio.ufc.br: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Repositório InstitucionalPUBhttp://www.repositorio.ufc.br/ri-oai/requestbu@ufc.br || repositorio@ufc.bropendoar:2026-08-06T19:12:09Repositório Institucional da Universidade Federal do Ceará (UFC) - Universidade Federal do Ceará (UFC)false |
| dc.title.pt_BR.fl_str_mv |
Ensaios sobre modelos de machine learning aplicados à econômia |
| title |
Ensaios sobre modelos de machine learning aplicados à econômia |
| spellingShingle |
Ensaios sobre modelos de machine learning aplicados à econômia Moura, Yure Revélles da Silva CNPQ::CIENCIAS SOCIAIS APLICADAS::ECONOMIA Machine Learning Macroeconomia Previsão Inflação Machine Learning Macroeconomics Inflation Forecast |
| title_short |
Ensaios sobre modelos de machine learning aplicados à econômia |
| title_full |
Ensaios sobre modelos de machine learning aplicados à econômia |
| title_fullStr |
Ensaios sobre modelos de machine learning aplicados à econômia |
| title_full_unstemmed |
Ensaios sobre modelos de machine learning aplicados à econômia |
| title_sort |
Ensaios sobre modelos de machine learning aplicados à econômia |
| author |
Moura, Yure Revélles da Silva |
| author_facet |
Moura, Yure Revélles da Silva |
| author_role |
author |
| dc.contributor.authorLattes.pt_BR.fl_str_mv |
http://lattes.cnpq.br/5185802743634482 |
| dc.contributor.advisorLattes.pt_BR.fl_str_mv |
http://lattes.cnpq.br/8710490356999657 |
| dc.contributor.author.fl_str_mv |
Moura, Yure Revélles da Silva |
| dc.contributor.advisor1.fl_str_mv |
Castelar, Luiz Ivan de Melo |
| contributor_str_mv |
Castelar, Luiz Ivan de Melo |
| dc.subject.cnpq.fl_str_mv |
CNPQ::CIENCIAS SOCIAIS APLICADAS::ECONOMIA |
| topic |
CNPQ::CIENCIAS SOCIAIS APLICADAS::ECONOMIA Machine Learning Macroeconomia Previsão Inflação Machine Learning Macroeconomics Inflation Forecast |
| dc.subject.ptbr.pt_BR.fl_str_mv |
Machine Learning Macroeconomia Previsão Inflação |
| dc.subject.en.pt_BR.fl_str_mv |
Machine Learning Macroeconomics Inflation Forecast |
| description |
This study brings together three empirical essays that focus on the application of machine learning models to economic forecasting at both the national and regional levels, seeking to evaluate the extent to which these approaches can complement traditional econometric methods. Despite the recent advances in these techniques, important limitations still remain regarding the economic interpretability of the models, the incorporation of regional heterogeneities, and the comparison with structural approaches. The choice of the sample period is justified by its coverage of major consolidations in the Brazilian macroeconomic environment, particularly the adoption of the inflation-targeting regime, while also encompassing distinct economic regimes, episodes of instability, and changes in the conduct of economic policy. In addition, the temporal delimitation was defined according to the availability, compatibility, and standardization of the databases employed throughout the essays. In this context, the first essay presents a national inflation forecasting exercise based on an extensive set of macroeconomic variables. Additionally, the study aims to identify the key predictors of inflation across multiple forecasting horizons. To this end, machine learning models are employed as the main approach and compared with benchmark models. The empirical findings suggest that machine learning models outperform benchmark models across different forecasting horizons. Furthermore, the results provide evidence of the relevance of production and fiscal sectors, labor market conditions, energy costs, and expectations in explaining inflation forecasts over different periods. The second essay proposes a regional inflation forecasting exercise based on an implicit hierarchical structure, incorporating regional elements into the forecasting process. Furthermore, a dynamic decomposition of inflation predictors is performed across regions and forecasting horizons. The results indicate the presence of significant regional heterogeneity across the states of Northeastern Brazil. The dynamic decomposition also reveals that the determinants of regional inflation gradually change over time, exhibiting patterns consistent with economic and political events at both the national and regional levels. The third essay proposes the use of multivariate neural network models to generate dynamic counterfactuals for Brazilian macroeconomic variables based on shocks applied to selected variables. The simulated trajectories are then compared with the dynamic responses obtained from a structural model, allowing an assessment of the extent to which machine learning-based methods can reproduce propagation and persistence patterns commonly found in traditional econometric models. The findings indicate that the multivariate model responds to shocks in a more moderate manner and in closer alignment with the macroeconomic literature. In particular, the model is able to reproduce propagation patterns similar to those generated by the structural model, although such responses arise from statistical relationships learned from the data rather than from explicitly modeled causal mechanisms. In this sense, the estimated counterfactuals exhibited smaller and more gradual responses at the onset of the shock for all variables when compared with the structural model. Taken together, the results reinforce the potential of machine learning methods not only as forecasting tools, but also as complementary instruments to traditional econometric approaches for analyzing dynamic and complex phenomena, especially in environments characterized by high macroeconomic instability. |
| publishDate |
2026 |
| dc.date.accessioned.fl_str_mv |
2026-08-06T19:12:08Z |
| dc.date.available.fl_str_mv |
2026-08-06T19:12:08Z |
| dc.date.issued.fl_str_mv |
2026 |
| dc.type.status.fl_str_mv |
info:eu-repo/semantics/publishedVersion |
| dc.type.driver.fl_str_mv |
info:eu-repo/semantics/doctoralThesis |
| format |
doctoralThesis |
| status_str |
publishedVersion |
| dc.identifier.citation.fl_str_mv |
MOURA, Yure Revelles da Silva. Ensaios sobre modelos de machine learning aplicados à econômia. 2026. 173p. Tese (Doutorado) - FEAAC - Faculdade de Economia, Administração, Atuária e Contabilidade - CAEN - Programa de Pós-Graduação em Economia, Universidade Federal do Ceará, Fortaleza , 2026. |
| dc.identifier.uri.fl_str_mv |
http://repositorio.ufc.br/handle/riufc/87420 |
| identifier_str_mv |
MOURA, Yure Revelles da Silva. Ensaios sobre modelos de machine learning aplicados à econômia. 2026. 173p. Tese (Doutorado) - FEAAC - Faculdade de Economia, Administração, Atuária e Contabilidade - CAEN - Programa de Pós-Graduação em Economia, Universidade Federal do Ceará, Fortaleza , 2026. |
| url |
http://repositorio.ufc.br/handle/riufc/87420 |
| dc.language.iso.fl_str_mv |
por |
| language |
por |
| dc.rights.driver.fl_str_mv |
info:eu-repo/semantics/openAccess |
| eu_rights_str_mv |
openAccess |
| dc.source.none.fl_str_mv |
reponame:Repositório Institucional da Universidade Federal do Ceará (UFC) instname:Universidade Federal do Ceará (UFC) instacron:UFC |
| instname_str |
Universidade Federal do Ceará (UFC) |
| instacron_str |
UFC |
| institution |
UFC |
| reponame_str |
Repositório Institucional da Universidade Federal do Ceará (UFC) |
| collection |
Repositório Institucional da Universidade Federal do Ceará (UFC) |
| bitstream.url.fl_str_mv |
http://repositorio.ufc.br/bitstream/riufc/87420/1/2026_tese_yrsm.pdf http://repositorio.ufc.br/bitstream/riufc/87420/2/license.txt |
| bitstream.checksum.fl_str_mv |
9edd8329578a81c11345b7135163a96a 8a4605be74aa9ea9d79846c1fba20a33 |
| bitstream.checksumAlgorithm.fl_str_mv |
MD5 MD5 |
| repository.name.fl_str_mv |
Repositório Institucional da Universidade Federal do Ceará (UFC) - Universidade Federal do Ceará (UFC) |
| repository.mail.fl_str_mv |
bu@ufc.br || repositorio@ufc.br |
| _version_ |
1881776040142962688 |