Publicação

Dynamic factor models and deep learning architectures in GDP nowcasting: comparative evidence and macro momentum applications

Detalhes bibliográficos
Resumo:	This thesis compares Dynamic Factor Models and deep learning architectures (LSTM, MLP) for GDP nowcasting and tests their influence in a macro momentum strategy using U.S. macroeconomic data from 2016-2025. While deep learning models achieved comparable accuracy in stable periods, DFMs proved more robust during volatile phases where deep learning models struggled to adapt. Strategies on equities showed weak Sharpe Ratios with severe drawdowns, while strategies on treasuries were more stable yet inconclusive. In a macro momentum framework, GDP-based signals alone do not deliver abnormal returns and higher forecast accuracy does not necessarily lead to better trading performance.
Autores principais:	Vordermark, Timm
Assunto:	GDP nowcasting Dynamic factor model Long short-term memory network Multi layer perceptron Deep learning Macro momentum Trading strategies
Ano:	2026
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 compares Dynamic Factor Models and deep learning architectures (LSTM, MLP) for GDP nowcasting and tests their influence in a macro momentum strategy using U.S. macroeconomic data from 2016-2025. While deep learning models achieved comparable accuracy in stable periods, DFMs proved more robust during volatile phases where deep learning models struggled to adapt. Strategies on equities showed weak Sharpe Ratios with severe drawdowns, while strategies on treasuries were more stable yet inconclusive. In a macro momentum framework, GDP-based signals alone do not deliver abnormal returns and higher forecast accuracy does not necessarily lead to better trading performance.

Não foram encontrados projetos financiados.