Five years, eight months, out-of-sample.
This page publishes the validation of Black System over January 2021 – August 2026: methodology, results, Monte Carlo, robustness checks — and the limits. The same document every buyer receives, in the open.
What was tested, and how honestly
Black System is a four-module systematic portfolio on XAUUSD: a swing breakout engine (R), a slow trend-follower with volatility targeting (T), and two mean-reversion grids confined to dedicated capital pockets (Q, Z). The test period contains four distinct regimes: the 2021 post-peak consolidation, the 2022 bear market under Fed tightening, the 2023 range, and the 2024–26 rally.
Year by year — four regimes
| Year | Regime | Return | R | T | Q | Z |
|---|---|---|---|---|---|---|
| 2021 | Post-peak consolidation | +20.3% | +27,420 | -10,418 | +2,087 | +1,209 |
| 2022 | Bear market, Fed hiking | +11.0% | +5,292 | +5,973 | +320 | +1,606 |
| 2023 | Range, then Q4 breakout | +37.7% | +47,044 | -3,478 | +5,604 | +1,177 |
| 2024 | Trend resumes | +29.7% | +31,284 | +9,421 | +11,824 | +2,155 |
| 2025 | Strong rally | +59.5% | +70,641 | +34,711 | +28,352 | +8,278 |
| 2026 | Rally + volatility (8 mo) | +63.3% | +146,059 | +17,217 | +61,115 | +16,310 |
| Total | 100,000 → 621,200 USD | +521% | +327,740 | +53,427 | +109,301 | +30,733 |
Returns on balance with compounding. 2022 is the test that matters: gold fell ~20% peak-to-trough while rates rose — R stays positive, T is short and profitable, the grids survive. T loses in the two range years (2021, 2023): the expected cost of a trend-follower, and the reason it's in the portfolio.
The four modules over 68 months
| Module | Net USD | Trades | Win | PF | Max DD | Years + |
|---|---|---|---|---|---|---|
| R · Swing breakout | +327,740 | 5,138 | 63% | 1.72 | -19,991 | 6/6 |
| T · Time-series momentum | +53,427 | 167 | 56% | 2.22 | -17,838 | 4/6 |
| Q · Grid mean reversion | +109,301 | 3,773 | 76% | 3.30 | -6,124 | 6/6 |
| Z · Prior-day levels | +30,733 | 660 | 81% | 2.70 | -9,198 | 6/6 |
| Portfolio | +521,200 | 9,738 | 69% | 1.92 | -17,168 | 6/6 |
T is close to uncorrelated with everything (R–T monthly correlation 0.12): it is the portfolio's second leg, losing in ranges and paying in the bear year. R and Q are correlated (0.77) and treated as one risk block; budgets are set accordingly.
Drawdown, exactly
The binding figure is the relative equity drawdown of 10.04% — measured on the equity curve including floating P&L, peak to trough. It happened once in 68 months. Closed-trade (balance) drawdown peaked at 7.45%. The portfolio's hard limit is set exactly at 10%: on this history it would have been touched once.
Monte Carlo — 5,000 resampled paths
- Median max drawdown 6.7% · 95th percentile 13.3% · 99th percentile 17.9%
- P(DD > 10%) ≈ 16% over 5.7 years — the 10% limit is expected to trigger roughly once every five to six years
- P(DD > 15%) ≈ 3% · paths ending in loss: 0% (median final balance 570,923 USD)
Method: block bootstrap (blocks of 20, preserving short-term dependence) with 10% of trades randomly skipped to mimic missed executions.
What was discarded
A module is admitted only if its thesis has support in the academic literature and it passes a criterion declared before testing (profit factor ≥ 1.2 in each year with one parameter set). Three strategies failed and were removed during development: an RSI recovery grid, a session breakout with too few trades, and a jump-detection module that proved a coin-flip in five configurations. We publish this because a portfolio is defined as much by what was cut as by what stayed.
Limits of this study — read before buying
- Single instrument. All modules trade XAUUSD; in a single stress day they can lose together. The 10% portfolio limit is the answer to that day, not the correlation matrix.
- Data quality. 72% of the history is real ticks; the earliest years are partly generated from M1 bars, on which grid strategies look marginally better than they are.
- Single broker. Spreads, swaps and commissions are those of one broker; costs on another venue change the grids' economics first.
- Grid tail risk. Q and Z have never exhausted their pockets in six years; the pocket stop is the only guarantee they cannot lose more than budgeted.
- Simulated results. Liquidity effects, requotes and broker-side latency under stress are not reproduced. Past performance, simulated or real, is not indicative of future results.
Research foundation
Moskowitz, Ooi & Pedersen (2012), Time Series Momentum, JFE · Hurst, Ooi & Pedersen (2017), A Century of Evidence on Trend-Following · Moreira & Muir (2017), Volatility-Managed Portfolios, JoF · Corsi (2009), HAR-RV · Grossman & Zhou (1993), drawdown control · Crabel (1990), opening range breakout · Zarattini, Barbon & Aziz (2024), SFI · Leung & Li (2015), optimal mean reversion · Kaminski & Lo (2014), stop-loss rules · Bailey, Borwein, López de Prado & Zhu (2014), Probability of Backtest Overfitting · López de Prado (2018), Advances in Financial Machine Learning · Harvey, Liu & Zhu (2016), RFS.