Launch Price Early access pricing — next price rise in 00h·00m·00s Limited time pricing
XAUUSD • H1 • 10 Years Backtested

Karat Killer

Professional Gold Trading Expert Advisor

Advanced algorithmic trading system designed for XAUUSD with robust risk management and consistent performance across market conditions.

€722,988
Net Profit
3.31
Profit Factor
11.2%
Max Drawdown
22.24
Recovery Factor
10.00
Sharpe Ratio

Performance Metrics

€722,988.07
Total Net Profit
3.31
Profit Factor
988
Total Trades
66.0%
Win Rate
€731.77
Expectancy
22.24
Recovery Factor
10.00
Sharpe Ratio
8.17
Sortino Ratio
Equity Curve in EUR (2016-2026) — €10,000 → €732,988

Risk Metrics

11.2%
Equity DD %
8.7%
Balance DD %
217d
Max DD Days
9.31
Calmar Ratio
1.72
Risk/Reward
18
Max Win Streak
5
Max Loss Streak
0.29
Avg Days/Trade

Trading Period

XAUUSD
Symbol
2016-01-01
Start Date
2026-02-02
End Date
$10,000
Initial Deposit

Validation Philosophy

"A model that looks too good on historical data is almost certainly wrong."

In quantitative finance, genuine edges are small. Our methodology is designed to detect and eliminate any artificial inflation of performance metrics, accepting modest but real predictive power over spectacular but illusory results.

What We Optimise For

Consistency across market regimes, realistic out-of-sample metrics, and robustness under transaction costs. Every design decision prioritises real-world viability over backtest aesthetics.

What We Reject

Any result that exceeds established plausibility thresholds is automatically flagged for review. Features without economic rationale are excluded regardless of their in-sample predictive power.

Walk-Forward Validation

Traditional train/test splits or k-fold cross-validation randomly shuffle temporal data, allowing the model to "peek" at future information. Walk-forward validation respects the arrow of time: the model is always trained on past data and tested on strictly future, unseen data — exactly as it would operate in live trading.

Test Period
Multiple Windows
Non-overlapping test sets
Data Span
10+ Years
2016–2026 validation
Design
Expanding Window
Training grows each window
Expanding Window Design
Window 1:  [====== TRAIN ======] [TEST]
Window 2:  [========= TRAIN =========] [TEST]
Window 3:  [============ TRAIN ============] [TEST]
  ...              ...
Window N:  [==================== TRAIN ====================] [TEST]

Each [TEST] = strictly unseen future data. No shuffling. No overlap.

Per-Window Isolation

Within each walk-forward window, the following operations are performed independently, using only that window's training data:

  • Feature selection — Statistical relevance computed exclusively on training samples
  • Model training — All models fitted only on the current window's training set
  • Threshold calibration — Decision threshold optimised on temporal holdout from training data
  • Performance evaluation — Metrics computed on the out-of-sample test set only

Data Leakage Prevention

Data leakage occurs when information from the future inadvertently contaminates the training process. Even subtle forms — a feature computed from an unclosed candle, a cross-timeframe alignment error — can produce dramatically inflated backtest results that collapse in live trading.

1

Temporal Integrity of Entry Points

Trade entries use prices available at the moment of the signal, never prices that would only be known after the fact.

2

Past-Only Feature Computation

Every feature is computed using data with timestamps strictly before the event timestamp. Multi-timeframe features use dedicated offsets ensuring only completed bars are used.

3

Completed Periods Only

All reference levels are derived from fully completed periods. No intra-period data is used for level calculation.

4

Mechanically Predictive Features Removed

Features mechanically correlated with the target variable — not because they capture market dynamics, but because they encode trade outcome information — are permanently removed.

5

Event Deduplication

Strict deduplication rules ensure no market event generates multiple correlated training samples, which would artificially inflate apparent model performance.

Leakage Audit Results

#CheckMethodStatus
1Walk-forward temporal validation (no random splits)Structural✓ PASS
2Entry price uses Open (not Close) of signal barCode audit✓ PASS
3All features use data with timestamp < event time (past only)Code audit✓ PASS
4Higher-timeframe features use completed-bar offsetsCode audit✓ PASS
5Reference levels from completed periods onlyCode audit✓ PASS
6Feature selection computed per-window on training data onlyStructural✓ PASS
7Event deduplication prevents correlated sample inflationStructural✓ PASS
8Mechanically predictive features removed from pipelineManual review✓ PASS
9Low-sample event categories removed (no statistical significance)Statistical✓ PASS

Anti-Overfitting Measures

Model Regularisation

Multiple regularisation techniques constrain model complexity, preventing memorisation of noise and forcing generalisable patterns.

Feature Space Control

From dozens of candidates, only top-ranked features are selected per window using statistical relevance computed exclusively on training data.

Economic Rationale Filter

Every feature must have a plausible economic explanation. Features with statistical significance but no logical mechanism are excluded.

Gradual Learning

Models learn slowly and incrementally, reducing sensitivity to individual training examples and improving generalisation.

Cross-Regime Robustness

The 2016–2026 validation period spans fundamentally different market environments. The model is evaluated across all of these regimes, ensuring it does not rely on a single market condition.

2016–2019
Low-Volatility Pre-Pandemic
Mar 2020
COVID-19 Crash & Recovery
2021–2022
Inflation & Interest-Rate Cycle
2024–2026
Gold Rally & New Highs

Summary of Evidence

Why we believe this is not overfitted:

  • Multiple non-overlapping OOS windows — A comprehensive sweep across a decade of market data including multiple regimes
  • Expanding window design — The model must generalise to unseen future data at every step
  • Per-window feature selection — No global selection that could leak test information into training
  • Modest, realistic metrics — All performance numbers fall within plausibility thresholds for genuine edges
  • Strict temporal boundaries — All leakage checks passed; features use only completed, past data
  • Realistic transaction costs — Spread and commission applied on every trade in every backtest window

About This Analysis

Why this matters: Many of our clients already trade with popular XAUUSD Expert Advisors like The Gold Phantom or Mad Turtle. A common concern is whether adding Karat Killer to their setup would create overlapping trades, correlated drawdowns, or redundant exposure to the same market conditions.

This correlation analysis is provided for informational purposes only. Its sole objective is to demonstrate that Karat Killer can be deployed alongside two of the most popular XAUUSD Expert Advisors on the MQL5 marketplace — The Gold Phantom and Mad Turtle — within the same trading account or multi-strategy setup, without significant overlap in trading signals or equity curve behavior.

The low correlation coefficients (avg. 0.20) indicate that these strategies operate on different market mechanics and timeframes, making them technically compatible for simultaneous execution. This analysis does not constitute investment advice, portfolio allocation recommendations, or any guarantee of future diversification benefits.

Data source: All equity curves were obtained from backtests conducted on IC Trading broker data. Each strategy was configured with comparable risk parameters to ensure a fair comparison across similar exposure levels.

Portfolio Diversification Analysis

Strategies Compared
3
Expert Advisors
Avg Individual DD
16.0%
Average of all EAs
Combined Max DD
9.4%
Equal-weight portfolio
Diversification Benefit
-41%
DD reduction vs avg
Pearson CorrelationAvg: 0.20
Karat KillerMad TurtleGold Phantom
Karat Killer1.000.210.26
Mad Turtle0.211.000.13
Gold Phantom0.260.131.00
Spearman CorrelationAvg: 0.20
Karat KillerMad TurtleGold Phantom
Karat Killer1.000.240.22
Mad Turtle0.241.000.12
Gold Phantom0.220.121.00
Strong Negative (-1 to -0.5)
Weak Negative (-0.5 to -0.2)
Neutral (-0.2 to 0.2)
Weak Positive (0.2 to 0.5)
Strong Positive (0.5 to 1)
Equity Curves in EUR (€10,000 Initial per Strategy)
Combined Portfolio Equity (€30,000 Initial = €10K × 3 Strategies)

Strategy Comparison

Metric Karat_Killer (XAUUSD) Mad Turtle (XAUUSD) The Gold Phantom (XAUUSD)
EA NameKarat_KillerMad TurtleThe Gold Phantom
SymbolXAUUSDXAUUSDXAUUSD
Period2016-01-01 – 2026-01-282016-01-01 – 2026-02-032016-01-01 – 2026-02-03
CurrencyEUREUREUR
Initial Deposit€10,000.00€10,000.00€10,000.00
Total Trades98819828758
Net Profit+€722,988.07+€365,656.43+€619,719.62
Win Rate66.0%58.8%63.9%
Profit Factor3.311.762.41
Expectancy+€731.77+€184.49+€70.76
Equity DD %11.2%22.6%13.4%
Balance DD %8.7%20.4%7.8%
Max DD Days217d953d81d
Max Stagnation217d953d81d
Sharpe Ratio10.002.944.58
Sortino Ratio8.172.885.09
Calmar Ratio9.312.996.53
Recovery Factor22.2413.7922.25
Risk/Reward1.681.241.37
Max Win Streak181444
Max Loss Streak51025
Avg Days/Trade4.161.850.42