Random-Forest Machine Learning Strategy

Strategy Overview

The Random-Forest Machine Learning Strategy combines the power of artificial intelligence with robust order management and risk control, using a pre-trained Random Forest machine learning model (generated for multiple instruments — MES, ES, MNQ, NQ...) to predict future price movements.

Unlike traditional rule-based strategies that rely on fixed technical indicators and predefined patterns, the strategy learns from historical price data to identify complex, non-linear relationships in market behavior. The model analyzes multiple features — daily price levels, support and resistance zones, market structure changes, and volatility indicators — to generate highly contextualized trading signals.

The strategy is designed to be fully automatic and can scale into multiple instances on the same or multiple accounts, with robust order management (including connection-loss recovery), precise risk control via dynamic position sizing, and multiple exit strategies including ATM Strategy integration and manual management options.

How It Works

The strategy operates in several stages:

1. Feature Engineering

The strategy constructs a comprehensive feature set from market data:

• Daily Levels: distance from the previous day's high/low and the current day's high/low/open prices
• Higher Timeframe (HTF) Levels: distance from 30-minute and 4-hour timeframe previous highs/lows, capturing medium-term structure
• Volatility Indicators: distance from SuperTrend upper/lower bands, for adaptive volatility measurement
• Volume Dynamics: volume differences between the current and previous bar, capturing momentum shifts
• Price Momentum: close price differences over the last 3 bars, capturing recent directional bias
• Time of Day: encoded time features capturing intraday seasonality patterns

2. Model Prediction

The pre-trained Random Forest model processes the feature set to predict the close price for a specified prediction horizon (default: 20 bars ahead on the 1-minute timeframe), generating a prediction score indicating the expected price movement's magnitude and direction.

3. Signal Generation

When the predicted price movement exceeds the prediction threshold (in points), the strategy generates a trading signal:

• A long signal is generated when the model predicts upward price movement
• A short signal is generated when the model predicts downward price movement
• Signals are optionally filtered using the SuperTrend indicator to reduce false signals in choppy markets

4. Entry Management

The strategy confirms and executes trades based on confirmed signals:

• Only one position per direction can be active at a time
• Market orders are used for instant execution (no pending orders)
• Entry quantity is determined by the selected quantity strategy (Fixed or Risk Adjusted based on the stop-loss level)
• Trades are only allowed within the specified trading time window

As with all our strategies, this strategy implements two methods for setting the quantity of open positions:

• Fixed Quantity: uses the same quantity for every position. If the stop loss is variable between positions, win/loss amounts will vary too.
• Dynamic Risk-Adjusted Quantity: calculates each position's quantity from its stop loss, the maximum risk (in $) allowed per position, and the maximum allowed quantity — producing equal win/loss amounts across all positions regardless of stop loss width.

5. Position Management

Once a position is opened, the strategy manages it using one of several configurable methods:

• ATM Strategy: delegates position management to your chosen ATM strategy template
• SuperTrend: uses the SuperTrend indicator to trail the stop loss and manage exits dynamically

The strategy also allows setting the stop loss based on multiple strategies and the take profit based on a ratio — see the Parameters section for details.

6. Exit Strategies

The strategy supports multiple exit methods:

• Stop Loss: fixed currency amount, points, SuperTrend levels, swing prices, or the opposite of the predicted target
• Take Profit: fixed currency, points, stop-loss ratio, or the predicted price target
• Session Closing: automatically closes all positions at the end of the trading session
• Daily Limits: stops trading when the daily profit target is reached or the daily loss limit is hit
• Trade Count Limits: can stop after reaching maximum daily winning or losing trade counts

Entry Setup Examples

Example #1: Reversal Short Entry

Running on MES on the 1-minute timeframe, the strategy generates a short signal from the model's prediction. Reversal signals are generated in the opposite direction following a Fair Value Gap (FVG). With the SuperTrend entry filter applied to reversal signals, the strategy only takes the reversal signal if SuperTrend is green for bearish signals and red for bullish signals. The same parameters are used for the next examples.

Example #2: Trend Long Entry

Same instrument, timeframe, and parameters as example #1. A first trend-long signal is suppressed because SuperTrend was in a red state at the time; a second trend-bullish signal is taken because SuperTrend was green. The take profit is the model's predicted target, and the stop loss mirrors that same distance from entry, producing a 1:1 risk/reward ratio on this trade.

Example #3: Reversal Entry Mode — Wait for SuperTrend Flip

A reversal signal is dispatched while SuperTrend is bullish (green). With Immediate entry mode, this would trigger an immediate short entry — a loss, since price went higher right after the signal. With entry mode set to Wait for SuperTrend Flip, the strategy instead waits for SuperTrend to flip to bearish (red), which happens a few bars later — at which point the strategy enters short and wins the trade.

Features

The Random-Forest Machine Learning strategy includes a comprehensive set of advanced features:

Artificial Intelligence-Powered Predictions

• Pre-trained Random Forest model for robust and adaptive price predictions
• Machine Learning model updated every week

Advanced Order Management

• No pending orders — only market orders, for guaranteed execution
• Robust in-house order manager, resistant to connection losses
• ATM Strategy integration for user-defined custom position management
• Support for both Rithmic and NinjaTrader brokerage/Tradovate accounts

Entry Filtering and Confirmation

• Optional SuperTrend entry filter to confirm directional bias
• Configurable prediction threshold to filter low-confidence signals
• Trading time window restrictions

Precise Backtesting

• Compatible with NinjaTrader's official backtesting best practices
• Tick-level execution for highly accurate results
• Support for optimization across all configurable parameters

Parameters

The strategy provides comprehensive parameter control while keeping the most critical settings easily accessible.

General Settings
Execute Historical Trades	Enables executing trades when the strategy runs in Historical mode — useful to test performance on past days, but should be unchecked when running Live.
Daily TakeProfit Target	Dollar amount target for daily profit — trading stops once reached after a position closes (0 = disabled).
Daily StopLoss Limit	Dollar amount limit for daily losses — the active trade closes and trading stops once unrealized PnL reaches this limit (0 = disabled).
Stop @ Winning Count	Stops trading after this many winning trades per day (win/loss thresholds are ±$10 PnL). Ignored when set to 0.
Stop @ Losing Count	Stops trading after this many losing trades per day (win/loss thresholds are ±$10 PnL). Ignored when set to 0.
ML Settings
Prediction Threshold Points	Minimum predicted points movement to generate a trading signal, filtering low-confidence predictions — recommended around 10 for MES, higher for MNQ.
Model Directory	Path to the folder containing the pre-trained Random Forest model .zip files. The strategy automatically loads the model suited to the current running day; during backtesting, it dynamically loads the model trained on data prior to the running day, guaranteeing it always runs on out-of-sample unseen data. Leave empty to be prompted with a folder-selection dialog.
Trading Session
Time Zone	The time zone used for the Trade Time parameter below. Futures markets are open/closed based on US Eastern time (EST); this parameter lets you specify the trade time window in EST, CET, or your local machine's time zone, without needing to manually convert. EST, CET, or Local Time
Trade Time (hh:mm-hh:mm,...)	Trading hours in 24-hour format — e.g. "09:30-16:00". Multiple comma-separated windows are supported; each window must start and end on the same day, so a window spanning midnight (e.g. 18:00 to 16:00 the next day) should be split like "18:00-23:59,00:01-16:00".
Entry Setup
Order Label	A label attached to orders placed by the strategy, useful for tracking and filtering.
Quantity Strategy	Fixed Quantity: fixed quantity per trade; skipped if the evaluated risk exceeds Target Risk Per Trade ($) Risk Adjusted: dynamic contract size, calculated from the stop loss price and Target Risk Per Trade ($)
Max Qty	Maximum allowed quantity when Risk Adjusted is selected — useful since some prop firms enforce order-quantity limits that can invalidate an account.
Fixed Quantity	The quantity used when the Quantity Strategy is Fixed Quantity.
SuperTrend Entry Filter	Enables/disables the SuperTrend confirmation filter — only enters trades in the direction confirmed by SuperTrend (reversal trades are confirmed by the inverse of SuperTrend's direction).
Enable Reversal Trades	Enables/disables considering reversal signals for entries. When unchecked, only trend-following trades are opened.
Reversal Entry Mode	Immediate: enters at the first closing bar in the reversal signal's direction, even if SuperTrend is the opposite color of the entry Wait For Supertrend Flip: waits for SuperTrend to flip in the reversal signal's direction before entering
Order Management
Order Management Strategy	ATM Strategy: delegates position management entirely to a NinjaTrader ATM strategy template of your choice, selected via the Chart Trader panel SuperTrend: closes trades using the SuperTrend indicator — the stop loss is set one tick above/below the wick of the bar that broke the supertrend level None: the stop loss defined by Target Risk Per Trade ($) is used instead
ATM Strategy Template Name	The template name used when Order Management Strategy is set to ATM Strategy.
StopLoss Strategy	Fixed Currency Value Points SuperTrend, a dynamic stop based on SuperTrend bands Prediction Target Opposite, a stop set opposite to the predicted target Last Swing Price, a stop at the most recent swing high/low
Swing Points Period	The period used for identifying swing highs/lows when the stop loss strategy is Last Swing Price.
TakeProfit Strategy	Fixed Currency Value: based on Take Profit Currency Points: a fixed number of points from entry Stop Loss Ratio: a manually set win/loss ratio Predicted Target: based on the model's predicted target
Stoploss/TakeProfit Ratio	Used with the Stop Loss Ratio take-profit strategy — the ratio of take profit to stop loss distance (e.g. 2.0 means TP = 2 × SL).
Risk Per Trade ($)	Maximum risk, in dollars, per position — used before opening a trade to verify the calculated stop loss respects this limit; if not, the trade is skipped.
Max Allowed Risk Per Trade ($)	Only used with Risk Adjusted quantity — controls how far the real calculated risk is allowed to deviate from the Target Risk (since an exact match is rarely possible, as futures contract sizes aren't fractional).
StopLoss Points	Stop loss in points, when using the Points-based stop loss.
StopLoss Offset Points	Additional points added/subtracted to the calculated stop loss, as a safety margin.
TakeProfit Currency	Target take-profit amount in dollars, when using Fixed Currency take profit.
TakeProfit Points	Target take profit in points, when using Points-based take profit.
Enable Price Breakeven	Enables/disables the price-based breakeven mechanism.
Price Breakeven Strategy	Points: breakeven triggers based on points Percentage: breakeven triggers based on percentage of take profit reached
Price Breakeven Trigger	Available if price-based breakeven is enabled — the trigger value (in points, or a percentage from 1-100) needed to move the stop loss to breakeven.
On Opposite Signal	Action taken when an opposite-direction signal occurs: Breakeven Position, Close Position, or Do Nothing.
SuperTrend Settings
SuperTrend Mode	ATR: ATR-based calculation Adaptive: based on candle close/high/low
Period / Multiplier / Moving Avg. Type / Smooth	Standard SuperTrend calculation parameters — period, band multiplier (higher = wider bands), moving-average type, and smoothing period.
Command Buttons
The strategy exposes manual command buttons on the Chart Trader panel — make sure it's displayed on your active chart.
Arm/Disarm Buttons	Manually enable/disable entering long or short trades — useful if you have a directional bias.
Close Position Button	Immediately flattens the position on the active instrument without restarting the strategy, which keeps looking for new trades afterward.
Increase Qty Button / Decrease Qty Button	Adjusts the current active position's quantity by +1/-1 contract.
Set Breakeven Button	Moves the open position's stop loss to breakeven, if doing so would improve it.

Performance Reports

Past results are not a guarantee of future results. All backtesting results are for indicative purposes only.

• Commissions are included, using NinjaTrader's Lifetime license commission schema, for realistic results.
• Since the AI model requires the previous day's data, signals only start triggering from the second trading day of the selected interval — a single-day backtest will show no trades. Make sure Execute Historical Trades is enabled.
• These are out-of-sample results — each backtest period loads the model trained on prior data and runs on unseen data, reflecting live/real-time forward-testing conditions.
• We follow NinjaTrader's official guide for precise intrabar-granularity backtesting using additional tick-level data series, and only use market orders — never pending orders — in both real-time and backtest modes.