Machine Learning Forecasting of Foreign Exchange Markets Trend Based on Order Flow and US Economic News

In this study, the effectiveness of various supervised learning classification techniques (Decision Trees, Support Vector Machines, k-nearest neighbors (KNN), and Ensembles) in forecasting the trend of the US dollar against the currencies of seven developed countries within 24 hours after economic news arrival is considered. This is accomplished by utilizing separate and compounded datasets of US economic news and order flow. We compare the accuracy, sensitivity, and specificity of various classifiers and introduce the most accurate methods. The findings suggest that during times of economic news releases, order flow is a more significant predictor than economic news and that integrating variables can improve the predictive power of the exchange rate trend. This study implies that policymakers, foreign exchange investors, and fund managers can use this method to forecast currency pairs trends based on order flow and economic news to select hedging strategies based on their specific objectives.