In this thesis we develop a multivariate timeseries classification model using machine learning techniques and the FinBERT model for extracting sentiment on daily oil news headlines gathered from oilprice.com. The model improves classical methods for predicting brent oil futures by using natural language processing techniques to capture daily market sentiment on oil and by using machine learning techniques and models to capture non linearity in the data. Using model assessment techniques, we choose an ensemble model to conduct simple trading strategies to show how the developed models can be used in the financial markets. In addition, we explore how news sentiment about oil affects the return of brent oil futures under extreme events by conduction an event study. This thesis finds that by using the developed model in trading brent oil futures with shortsaleconstraints, one can outperform a simple buyandhold trading strategy in a bull market. As the model shows a clear bias towards predicting positive future trading days, limitations have to be set on how the model would perform in a bear market.
Date of Award | 2 Feb 2023 |
---|
Original language | English |
---|
Awarding Institution | - Universidade Católica Portuguesa
|
---|
Supervisor | Dan Tran (Supervisor) |
---|
- Machine learning
- Quantitative finance
- NLP
- FinBERT
- Oil
- News
- Sentiment
Predicting brent oil futures returns with machine learning and text data
Meland, A. N. (Student). 2 Feb 2023
Student thesis: Master's Thesis