A carregar...
Publicação
A Transformer-based Approach to Time-Series Forecasting: The Galp Example
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.09 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Financial time-series forecasting, more concretely stock price forecasting, has been a
highly studied problem since the beginning of trading. Throughout the decades, the
evolution of time-series forecasting models has led to more precise and consistent
solutions to this problem. However, the traditionally used models have only been able
to sustain this precision on a shorter forecasting horizon. This study aims to assess
the performance of Transformer-based models on the stock price forecasting context,
compared to other most commonly used models such as RNN-based models or CNNbased models. For this specific experiment, five models were selected: Informer,
Autoformer, PatchTST, TimesNet and LSTM. Using Galp Energia S.A.’s seventeenyear historic stock price data, these models will produce comparable results, evaluated
using Mean Average Error (MAE), Mean Squared Error (MSE) and Root Mean
Squared Error (RMSE). The final results are expected to provide valuable insights on
whether the Transformer architecture is the next step on the time-series models’
evolution.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science
Palavras-chave
Transformer CNN RNN Long Time-Series Forecasting Galp Energia SDG 8 - Decent work and economic growth