Aventro Quant real-time system dashboard showing live trading signals, portfolio analytics, and risk metrics
Aventro Quant is a fully automated trading system powered by machine learning, built exclusively for XAUUSD (Spot Gold). Every trading day, it collects data from multiple market sources, generates a directional signal (LONG or SHORT) with a calibrated probability score, and executes trades automatically through MetaTrader 5.
The system combines a proprietary ensemble of ML models with an adaptive risk management engine. Position sizes are scaled based on model conviction, delivering consistent returns across five distinct risk profiles.
Performance shown using the Moderate profile ($100K start, Jun 2024 to Apr 2026). All five risk profiles are profitable with the same 72.3% win rate.
The system runs a strict three-stage pipeline once per trading day. Each stage is fully deterministic and reproducible.
Ingests daily data from 5 independent sources: broker feeds, cross-assets, macro indicators, economic calendar, and sentiment.
Computes 50+ quantitative features and passes them through a multi-model ensemble to produce a directional signal with confidence score.
Places the trade on MT5 with dynamically sized positions, volatility-adjusted stop losses, and pre-calculated take profit levels.
The system draws intelligence from five complementary sources to ensure there is no single dependency on any one data provider:
Raw data is transformed into 50+ quantitative features spanning multiple categories including return patterns, momentum dynamics, volatility measures, cross-asset correlations, market regime identification, and calendar-based timing signals.
The model automatically determines which features contribute most to prediction accuracy through built-in importance ranking, eliminating manual feature selection bias.
The prediction engine uses an ensemble of multiple gradient-boosted and linear models that vote on each trade. Their individual probabilities are combined through a weighted consensus mechanism to produce a final directional signal.
No single model excels in all market conditions. By blending multiple architectures, each with different strengths, the system reduces variance, captures both linear and non-linear patterns, and smooths predictions across market regimes.
| Single Model Risk | Ensemble Mitigation |
|---|---|
| Tree models can overfit noisy periods | Linear models regularize the consensus |
| Linear models miss non-linear interactions | Tree models capture complex feature relationships |
| Any single model has high forecast variance | Weighted blending produces stable, reliable predictions |
Each day, the ensemble outputs a probability score (0 to 100%) expressing confidence in a LONG outcome. Scores above 50% produce a BUY signal; below 50% produce a SELL signal. This probability feeds directly into the risk sizing engine, where higher confidence results in proportionally larger position sizes.
The system does not try to predict the market perfectly. Instead, it aims to be right often enough, and to size positions correctly, so that winning trades more than compensate for losing ones.
Aventro Quant features a full Risk Profile Engine with five built-in profiles and the ability to create custom ones. Each profile defines a unique mapping from model confidence to position size. The system allocates more capital to high-conviction trades and less to borderline ones.
| Profile | Risk Range | Use Case |
|---|---|---|
| Prop Firm Safe | Ultra-conservative | Funded account evaluations with strict DD limits |
| Conservative | Fixed low risk | Capital preservation, steady equity curve |
| Moderate ★ | Balanced | Featured profile: growth with controlled drawdowns |
| Kelly Dynamic | Full Kelly Criterion | Mathematically optimal sizing for maximum growth |
| Aggressive | Maximum utilization | Personal accounts with high risk tolerance |
Stop loss and take profit levels are calculated dynamically using real-time market volatility (ATR). This ensures positions adapt to current conditions: wider stops in volatile markets, tighter stops in calm markets. The system maintains a consistent 3:1 risk-to-reward ratio across all conditions.
Period: June 6, 2024 – April 16, 2026 (22 months) | Starting Capital: $100,000 | Total Trades: 240
All five profiles were backtested on the exact same signals and trades. The only variable is position sizing — proving that the ML signal is robust and profitable across all risk levels.
| Profile | Return | Final Capital | Max DD | Profit Factor | Sharpe |
|---|---|---|---|---|---|
| Prop Firm Safe | +165.3% | $265,285 | 2.5% | 3.07 | 4.52 |
| Conservative | +178.5% | $278,501 | 3.1% | 3.23 | 5.56 |
| Moderate ★ | +615.6% | $715,635 | 5.6% | 3.09 | 4.30 |
| Kelly Dynamic | +1,131.8% | $1,231,801 | 7.5% | 3.10 | 3.67 |
| Aggressive | +1,954.3% | $2,054,251 | 11.1% | 3.21 | 3.51 |
Monte Carlo simulation is a statistical stress test. It takes the actual trade results and asks: if these same trades happened in a different random order, would the strategy still be profitable?
By shuffling the trade sequence 1,000 times using out-of-sample data (Jun 2024 to Apr 2026, Moderate risk profile, 227 trades), we generate a probability distribution of outcomes. This reveals whether the edge is structural, meaning the trades themselves are profitable, or merely lucky, meaning profitability relied on a specific favorable sequence of trades.
Average maximum drawdown across all 1,000 simulations: 7.0%. Even the absolute worst-case drawdown was 16.3%, and the system's hard 15% production limit would intervene well before extreme tail scenarios.
A profitable backtest means nothing if the model is overfitted to historical data. Aventro Quant employs a multi-layer validation framework specifically designed to detect and prevent overfitting.
The model is trained on expanding windows of historical data and tested on unseen future data. This mimics real deployment: the model never sees future prices during training.
A strict temporal split ensures the final backtest period was never used during any model training phase. Test performance closely matches training performance — a key indicator of generalization.
The model is retrained with different random seeds. Consistent results across seeds prove performance isn't dependent on initialization luck.
1,000 random trade orderings confirm the edge is structural (see Section 06).
Real-account trading performance confirms that model behavior is consistent between backtest and live environments.
Aventro Quant is a production-grade quantitative trading system that combines machine learning signal generation with institutional-quality risk management. The system has demonstrated:
The system is designed for deployment across multiple risk mandates, from conservative prop firm evaluations to aggressive personal accounts, all powered by the same underlying ML signal.
[email protected] · WhatsApp: +971 55 398 2882
Past performance does not guarantee future results. All figures presented are from backtested data. Trading involves risk of loss. This document is confidential and intended for the recipient only.