We explore the application of Machine Learning for predicting bilateral Foreign Exchange Rates utilizing the sentiment from news articles and prominent macroeconomic indicators. Using a random forest regressor, we were able to predict foreign exchange rates with an average error of 7.8%. In addition, news articles were an important feature in the majority of these random forest regressions. The novelty of our project relies on utilizing the Latent Dirichlet Allocation to cluster news articles into topics and further understand the semantic meanings behind each topic and applying it to foreign exchange rates.
Task
We explore the application of Machine Learning for predicting bilateral Foreign Exchange Rates utilizing the sentiment from news articles and prominent macroeconomic indicators.
Solution
Use news article sentiment to predict FER.
Collect and Cluster news articles from past 40 years.
Used proportion of articles per year falling in each cluster as an indication of sentiment to predict the next year’s exchange rate.