Random Forest Classification Trade

Entry Logic

• Long entry is triggered when a trained random forest model classifies the next period's return as 'up'.
• Short entry is triggered when the model classifies the next period's return as 'down'.
• Confirmation requires the model's prediction probability to be above 75%.
• Timeframe is determined by the model's training data (e.g., 15-minute, 1-hour).
• Location context is implicitly learned by the model from the input features.
• Market condition is a feature in the model.

Exit Logic

• Profit target is a fixed risk-multiple (e.g., 1.5R) or when the model generates an exit signal.
• No scaling out.
• A trailing stop of 1 ATR is used after the price moves 1R in favor.
• Exit on signal failure if the model's prediction is incorrect and the stop loss is hit.
• Exit on an opposite signal from the model.
• Exit on time expiration after 8 periods.
• Exit on momentum loss if the price fails to make a new high/low for 3 consecutive periods.

Stop Loss Structure

• Hard stop is placed at 1.5 ATR from the entry price.
• Soft stop is not used.
• Maximum dollar loss is set at $125 per trade.
• Maximum percent loss is 1.25% of account equity.
• Structural stop is not used.

Risk Management Framework

• Risk per trade is 0.6% of account equity.
• Maximum daily loss limit is 2.4% of account equity.
• Maximum weekly loss limit is 6% of account equity.
• Maximum drawdown is 18%.
• Risk-reward ratio requirement is a minimum of 1:1.5.

Position Sizing Model

• Sizing is fixed fractional.
• Volatility adjustment using ATR is applied to normalize position size.
• Conviction sizing is not used.
• No scaling in.
• No scaling out.

Trade Filtering

• The model filters trades based on the learned patterns in the data.
• Avoid trading in market conditions the model was not trained on.
• Instrument selection is based on where the model has shown the best backtested performance.
• Time-of-day restrictions can be included as a feature in the model.
• News avoidance is handled by pausing the strategy during major economic releases.

Context Framework

• The model learns context from input features like moving averages, volatility, and momentum.
• The model determines the importance of each contextual factor.
• Higher timeframe context can be included as features.

Trade Management Rules

• Move stop to breakeven when the trade is 1R in profit.
• No scaling out.
• Do not add to positions.
• The model can be trained to handle different market dynamics.

Time Rules

• Optimal trading window is identified from the model's backtest performance.
• Times to avoid are those where the model has historically underperformed.
• Session-specific behavior can be learned by the model.

Setup Classification

• A+ setup: High probability prediction (>80%) with confirming price action.
• A setup: High probability prediction (>75%).
• B setup: Moderate probability prediction (65-75%).
• C setup: Low probability prediction (<65%).

Market Selection Criteria

• Instruments are those on which the model was trained and validated.
• High liquidity and low transaction costs are essential.
• The model's performance can be market-specific.

Statistical Edge Metrics

• All metrics (win rate, average win/loss, profit factor, expectancy) are derived from out-of-sample backtesting of the random forest model.

Failure Conditions

• The model can fail due to concept drift, where the market dynamics change.
• Overfitting is a major risk that must be addressed during model development.
• Poor data quality can lead to an unreliable model.

Psychological Rules

• Trust the model's signals and avoid emotional overrides.
• Continuously monitor the model's live performance.
• Be prepared to retrain or disable the model if its performance degrades.

Advanced Components

• Feature engineering is crucial for creating a robust model.
• Hyperparameter tuning is necessary to optimize the random forest.
• The model should be tested for robustness using various validation techniques.

Location

• The strategy's effectiveness is tied to the quality of the model and the data.
• It can be applied to any market with sufficient historical data.
• Performance may vary depending on the market regime.