Loading...
Publication
Inteligência artificial aplicada na previsão de mercados de ações
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.65 MB | Adobe PDF |
Authors
Abstract(s)
A previsão dos mercados financeiros tem sido historicamente um grande desafio da área de Finanças. Esta dissertação investiga a aplicação de técnicas de Inteligência Artificial, nomeadamente os modelos Support Vector Regression (SVR) e Long Short-Term Memory (LSTM), na previsão dos retornos diários do índice S&P 500. Recorrendo à metodologia CRISP-DM, foram conduzidas experiências que avaliam, por um lado, a eficiência fraca do mercado através de modelos que geram previsões baseados no histórico do próprio índice e, por outro, a eficiência semi-forte com a introdução de variáveis externas selecionadas por correlação. Um aspeto central
foi a resolução temporal, onde se exploraram janelas de diferentes granularidades e se construíram ensembles de modelos preditivos com segmentação temporal, permitindo captar padrões sazonais e dependências específicas. Embora os resultados mostrem ganhos apenas marginais na introdução de variáveis externas, confirmando as limitações da hipótese semi-forte, a segmentação temporal revelou-se determinante para melhorar a robustez preditiva. Conclui-se que a análise e modelação da dimensão temporal, aliada a modelos de machine learning, constitui o principal contributo desta dissertação e reforça a importância deste enfoque no estudo da previsão financeira.
Financial market forecasting has historically been a major challenge in the field of Finance. This dissertation investigates the application of Artificial Intelligence techniques, namely Support Vector Regression (SVR) and Long Short-Term Memory (LSTM) models, in predicting the daily returns of the S&P 500 index. Following the CRISP-DM methodology, experiments were conducted to assess, on the one hand, the weak form of market efficiency through models that generate forecasts based on the index’s own historical data and, on the other hand, the semi-strong form with the introduction of external variables selected through correlation analysis. A central focus was temporal resolution, where different granularities were explored and predictive models ensembles with temporal segmentation were developed, allowing the capture of seasonal patterns and specific dependencies. Although the results show only marginal gains from the introduction of external variables, thereby confirming the limitations of the semi-strong hypothesis, temporal segmentation proved decisive in enhancing predictive robustness. It is concluded that the analysis and modeling of the temporal dimension, combined with machine learning models, constitutes the main contribution of this dissertation and reinforces the importance of this perspective in the study of financial forecasting.
Financial market forecasting has historically been a major challenge in the field of Finance. This dissertation investigates the application of Artificial Intelligence techniques, namely Support Vector Regression (SVR) and Long Short-Term Memory (LSTM) models, in predicting the daily returns of the S&P 500 index. Following the CRISP-DM methodology, experiments were conducted to assess, on the one hand, the weak form of market efficiency through models that generate forecasts based on the index’s own historical data and, on the other hand, the semi-strong form with the introduction of external variables selected through correlation analysis. A central focus was temporal resolution, where different granularities were explored and predictive models ensembles with temporal segmentation were developed, allowing the capture of seasonal patterns and specific dependencies. Although the results show only marginal gains from the introduction of external variables, thereby confirming the limitations of the semi-strong hypothesis, temporal segmentation proved decisive in enhancing predictive robustness. It is concluded that the analysis and modeling of the temporal dimension, combined with machine learning models, constitutes the main contribution of this dissertation and reinforces the importance of this perspective in the study of financial forecasting.
Description
Keywords
Previsão do S&P 500 Aprendizagem automática Séries temporais Sazonalidade Regressão Linear SVR LSTM S&P 500 CRISP-DM Artificial intelligence Machine learning Financial forecasting