Deep-factors: a deep learning portfolio theory extension for Brazilian stock market
Carregando...
Arquivos
Data
2023-07
Autores
Orientador(res)
Matsumoto, Élia Yathie
Kitani, Edson Caoru
Métricas
Título da Revista
ISSN da Revista
Título de Volume
Resumo
The usage of pricing models based on factors (i.e., CAPM, 3-FAMA-FRENCH) to systematically model an asset return, has played an important role for investment strategies in recent decades. In spite of that, due to their linear nature, such models are subject to unrealistic assumptions in order to remain consistent. Thanks to the recent advancements in computational methods, specially in deep learning, pricing models have been leveraged to overcome such a constraints by enabling them to include non-linear dynamics that are more accurate to model the real market. Experiments in this topic have exposed remarkable results compared with statistical-linear approaches by leading to factors with higher returns. This a-priori evidence is the main motivation for this research which is based on a 5-step algorithm that is used to build deep-factors that are expected to beat latent-linear factors within the Brazilian stock market. Results of this research showed that despite the exposed evidence from the reference experiment, deep-factors were not able to outperform latent-factors in the studied market. Nevertheless, we contribute to the literature by shedding light about the insights and pitfalls of this method applied to a real finance setting.