Meta-Labeling RSI Trades for Signal Filtering and Position Sizing

Summary

The article applies a secondary machine-learning classifier to RSI crossover signals. RSI remains responsible for trade direction, while a Random Forest estimates whether each signal is likely to succeed using 27 contextual features measured at the signal bar. Triple-barrier labels identify outcomes, and predicted probabilities are used both to skip lower-confidence trades and scale approved position sizes. Features include RSI depth, trend and volatility measures, distance from recent highs or lows, cyclical time encodings, and session volatility.

The study uses seven years of EURUSD hourly data, with a test period from 2022 through 2024. It reports that plain RSI struggled during the 2022 downtrend and that the meta-labeling approach reduced drawdown through lower exposure. The article cautions that this reduction came from taking fewer and smaller trades, not demonstrated predictive skill. Results are specific to the stated sample and setup; the backtest does not establish that the model will generalize to other periods or markets.

Key ideas

• The RSI rule supplies trade direction while a secondary classifier filters and sizes its signals.
• Triple-barrier labels distinguish successful trades from stop-loss or time-expiry outcomes.
• The classifier uses price, volatility, trend, and time-context features measured at each signal.
• Probability-based sizing increases exposure as estimated success probability rises.
• The reported drawdown reduction is attributed to reduced exposure rather than proven predictive skill.

Tags

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