Methodology
Walk-forward validation, no cheating
Backtesting means replaying a strategy over history as if you were playing it back then. To be honest it must obey two hard rules: no lookahead bias (no future data used to decide the past) and no repainting (signals are not rewritten after the fact).
We use walk-forward validation: the model learns on a past window and is tested on the immediately following one, then the window advances. On each fold we measure hit rate, simulated ROI and drawdown — and we also show the strategies that perform poorly, with no cherry-picking.
Walk-forward validation
Simulated equity (example)
Illustrative example: train and test advance over time; the equity includes drawdowns, not only the positive stretches.
Step by step
How we validate
- 1Temporal train/test split: we split history by date, never at random, so the test stays in the future of the train.
- 2Walk-forward with no lookahead: the model sees only data available at that date; the window advances fold by fold.
- 3Per-fold metrics: on each window we compute hit rate, simulated ROI and drawdown, drawdowns included.
- 4Baseline comparison: results are measured against chance and null strategies, with no cherry-picking.
Transparency
What you can verify
- Walk-forward with no lookahead bias
- Strict temporal train/test split, no repainting
- Transparent hit rate, simulated ROI and drawdown
- Weak strategies stay visible too
What a backtest does NOT guarantee
Past results don't repeat in the future: a backtest measures a strategy's historical consistency, not its future profitability. It stays an analysis tool, not a promise — and the game's expected value stays negative.