Research
Backtest crypto trading strategies without fooling yourself
Use built-in and custom strategies, realistic fills, professional metrics, held-out data and walk-forward tests in Basis.
Start with causal rules and explicit execution
A valid backtest computes each signal from information available at that moment and applies a declared fill model. Basis strategies expose entries, exits, stops and metrics on the chart so the path can be inspected rather than trusted as one final number.
- No future-bar inputs or centered calculations.
- Signal timing separated from fill timing.
- Fees, slippage and position sizing included in the scenario.
- Trades and chart markers available for manual inspection.
Read a metric set, not one headline number
| Metric | Why it matters |
|---|---|
| Net return | Outcome, but sensitive to size and period |
| Sharpe / risk-adjusted return | Return relative to variability |
| Maximum drawdown | Observed capital and psychological pressure |
| Profit factor | Gross win relative to gross loss |
| Expectancy | Average payoff per trade |
| Trade count | Whether the sample can support the conclusion |
| Exposure | How much time and capital the strategy actually used |
Separate discovery from evaluation
- 01
Define the hypothesis
Write why the rule could work before optimizing parameters.
- 02
Choose a training period
Use it to develop and reject variants.
- 03
Lock the rule
Stop changing it before opening held-out data.
- 04
Evaluate out of sample
Judge degradation, trade count and regime sensitivity.
- 05
Walk forward
Repeat train/test windows to see whether ranking persists through time.
Reject beautiful tests with weak evidence
- Very few trades or one exceptional winner.
- Large performance drop after realistic fees.
- One exact parameter works while nearby values fail.
- All profit comes from one symbol or regime.
- Hundreds of trials were searched but not reported.
- The held-out period changes the sign of expectancy.
Important
Backtesting estimates historical behavior under assumptions. It does not prove future profitability.