In this report, quantitative analysts Victor Ritseson, Balasurya Sivakumar, and Théodore Zitouni used machine learning and traditional trend strategies with sentiment analysis to predict price movements.
The analysis was conducted for three commodities including gold, crude oil and copper. For sentiment analysis news headlines (2010-2019) from leading business newspapers was analysed using VADER library from Python. For commodities that did not have accessible news headlines data the Bloomberg sentiment scores (2014-2021) from top 10 mining companies was used.
The method included creating two different approaches, the first approach consisting of solely using the MACD indicator on sentiment score of commodities, the second approach implemented a machine learning model from random forest.
In conclusion, using sentiment analysis to predict price movements show great potential and especially when using it in combination with the closing price as features for a machine learning model. The machine learning approach showed best promise both from a earnings perspective but also from a risk management perspective, as it successfully predicted downward trends and sold the security accordingly.