Random Forest Forex Forecasting with Technical and Twitter Sentiment Features

Summary

This project builds a random forest regression model to estimate the next day’s EUR/USD closing price from daily price data, technical indicators, and Twitter sentiment. Predictors include OHLCV values, short and long EMAs, RSI, OBV, and daily mean sentiment derived from historical tweets. The workflow computes indicators, splits the data into training and untouched test portions, cross-validates the training set, and evaluates predictions and a trading strategy against a buy-and-hold benchmark.

The project reports favorable model scores and a strategy Sharpe ratio near two in its tests, with returns often exceeding the benchmark. These results are exploratory: the sample covers about two years, and the final usable tweet set is limited. The author notes that data splitting further reduces statistical power and that obtaining sentiment data consistently may be difficult for live trading. Further work on longer histories, feature selection, indicator settings, and real-time data would be needed before drawing robust conclusions.

Key ideas

• The model combines daily forex prices, EMA, RSI, OBV, and Twitter sentiment to predict the next session’s close.
• The workflow uses cross-validation on training data and retains an untouched set for evaluation.
• Reported strategy performance is promising in this sample, but it is not evidence of a robust live trading edge.
• Limited tweet coverage, short historical data, and data availability create substantial constraints on inference and deployment.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.