Exportar Publicação

A publicação pode ser exportada nos seguintes formatos: referência da APA (American Psychological Association), referência do IEEE (Institute of Electrical and Electronics Engineers), BibTeX e RIS.

Exportar Referência (APA)
Ramos, F.R. (2021). Data Science na modelação e previsão de séries económico-financeiras: das metodologias clássicas ao Deep Learning.
Exportar Referência (IEEE)
F. R. Ramos,  "Data Science na modelação e previsão de séries económico-financeiras: das metodologias clássicas ao Deep Learning",, 2021
Exportar BibTeX
@null{ramos2021_1733592339045,
	year = "2021",
	url = "http://hdl.handle.net/10071/22964"
}
Exportar RIS
TY  - GEN
TI  - Data Science na modelação e previsão de séries económico-financeiras: das metodologias clássicas ao Deep Learning
AU  - Ramos, F.R.
PY  - 2021
DO  - 10.13140/RG.2.2.14510.02887
UR  - http://hdl.handle.net/10071/22964
AB  - The articulation of statistical, mathematical and computational techniques/tools, in the process of analysis, modelling and forecasting time series, manifests clear support for decision making. The constant challenge in the quest for the most accurate results possible has led researchers not only to improve the existing techniques, but also to invest in the search for alternative methodologies. Specifically, for economic and financial series, the application of methodologies based on Artificial Intelligence, in particular Deep Learning, has been pointed out as a promising option. This study makes a critical comparison of the results obtained by applying classical forecasting methodologies (namely autoregressive models and exponential smoothing) and Deep Learning (through the implementation of some neural network architectures). The empirical study focused on four economic-financial series with different characteristics: Consumer Price Index for All Urban Consumers: All Items in U.S. City Average (CPIAUCSL); Vehicle-Miles Travelled (VMT); Portuguese Stock Index 20 (PSI 20) and Standard & Poor's 500 Exchange-Traded Fund (SPY). The comparative analysis is made based on both predictive quality and computational cost associated with each of the forecasting models. Recognized the advantages in the application of Deep Learning methodologies, we discuss some changes to introduce in the existing models to improve their predictive quality while reducing computational execution time. The changes introduced in neural network models proved to be promising in reducing the associated computational time but and the values of the error metric used. This success is especially evident in series with ‘irregular’ dynamics, as is the case with financial series.
ER  -