| Resumo: | This thesis explores the applicability of CNNs as a price movement forecasting tool for ETFs, using a technical analysis approach and three different image encoding techniques. After developing a general methodology, the thesis focuses on the application to the U.S. financial services sector. Subsequently, the research draws comparisons to results obtained for other U.S. sector ETFs using the same model approach. Overall results show that the CNN models, while proving some potential and exceeding a random model in accuracy, show significant weaknesses for all industries in predicting Buy and Sell signals. Addressing these weaknesses, limitations of the approach are explored to suggest methods for model performance improvements.
|
|---|