Forex Trend Prediction After Strong Movements
Machine learning models for predicting currency pair trends following significant market movements
Abstract
The foreign exchange (Forex) market is the largest financial market by trading volume, with billions of dollars exchanged daily. While classical financial theory postulates that markets are efficient—making prediction impossible—the success of quantitative funds and high-frequency traders suggests otherwise.
This project explores the use of machine learning algorithms to predict trends on the Forex market after significant price movements. We developed classification models using Random Forest and XGBoost to analyze 9 major currency pairs (EUR/USD, GBP/USD, USD/JPY, AUD/USD, EUR/JPY, GBP/JPY, NZD/USD, USD/CAD, USD/CHF).
Our approach involved extensive feature engineering with over 15 technical indicators covering volatility (ATR, Bollinger Bands), trend (ADX, Moving Averages), and momentum (MACD, RSI). A rigorous methodology was implemented to select strong movements and avoid data leakage during model training.
Backtesting results demonstrated strong risk-adjusted performance, with Sharpe ratios reaching up to 2.96 on certain currency pairs. The models achieved accuracy rates between 54% and 65% depending on the pair and prediction horizon, with particularly strong results on USD/JPY and NZD/USD pairs.