According to a recent research paper published in the Journal of the Faculty of Engineering and Architecture of Gazi University, a time series model that uses sentiment analysis and recurrent neural networks (RNNs) can accurately predict the trend of gold prices — including both the direction of the
The study, entitled “The Hybrid Gold Index (XAU/USD) Direction Forecast Model Using Deep Learning,” has been prepared by Onur Kantar (Kocaeli University, Turkey) and Zeynep Hilal Kilimci (Kocaeli University,
Authors have argued that sentiment analysis can capture market sentiment and serve as a leading indicator for stock price movements. They also point out that time-series data provides the ability to glean important insights into historical price patterns.
They set out to test this hypothesis using publicly available data on CNN, Twitter, and other popular news sources frequently used to discuss finance. Next, they apply several DNNs to label the sentiment in gold-related Twitter messages. The next steps were to combine the sentiment analysis results with the time series data to create a prediction model for the Gold Index.
The results demonstrate that the proposed model outperforms state-of-the-art models concerning accuracy. The accuracy in predicting the gold Index path was achieved with a MAPE of 2.86% compared to 3.74% for the best-in-class benchmark model.
The researchers note that this approach should be useful for trading purposes (i.e., prediction of gold index behavior) for investors and analysts alike.
Saqib Iqbal, an analyst at Trading.Biz, commented on the paper with the following arguments:
According to Iqbal, the model has a strong theoretical base. The authors claim that sentiment analysis, as is time series data, is important to make predictions for the direction of the gold index.
The model is highly accurate. The ML algorithm could foresee the course of the Gold Index with MAPE as low as 2.86%, whereas the MAPE for the best previous model was 3.74%.
The model is computationally efficient. The model can be trained to be run locally on a laptop.
The model was built upon a relatively small dataset. The authors made use of a dataset ranging from the year 2015 to the year 2022. Maybe there is some degradation going on with model training on bigger datasets.
The model is not open-source. The model code is unfortunately not publicly available. This makes it hard for other academics to reproduce the experiments or improve our work.
In summary, this research paper is a major milestone in forecasting. The proposed model can be seen as an important new tool for investors and analysts in predicting the movement of the Gold Index.
