Method

A strategy reaches the site only after it has held up on data it has never seen, net of costs. Once deployed, its parameters are no longer touched.

DATA

We work on hourly (H1) bars of futures, gold, indices and FX, with several years of history. The same strategies are also tested on data from different brokers, because spreads, trading hours and price feeds change results more than one might expect.

TWO MODEL FAMILIES

• Machine learning. Gradient boosting models estimate the direction of the next bar; every entry has a take profit and a stop loss.
• Systematic rules. Trend, mean reversion and breakout logic, written explicitly and verifiable.

VALIDATION

1. Walk-forward training: the model only learns from data prior to the period it is evaluated on.
2. A final out-of-sample period is kept separate during development and used to check the model on new data.
3. One configuration is chosen per market; after deployment its parameters stay frozen.
4. Every backtest includes estimated costs and slippage.

LIMITATIONS

A backtest is still a simulation: markets change, and a strategy that performed well in the past may underperform or lose. Many ideas fail the tests; we record them as negative results instead of quietly discarding them. How every figure is computed is explained in the reporting standard.